Mechanical Engineering

International Review of

Mechanical Engineering (IREME) Contents High Temperature Air Combustion: Sustainable Technology to Low NOx Formation by Seyed Ehsan Hosseini, Mazlan Abdul Wahid, Abuelnuor Abdeen Ali Abuelnuor

947

Resolution of an Inverse Problem in Thermal Diffusion Process for the Identification of a Heat Flux by Abdelkarim Maamar, Touhami Y., Bounegta B., Khadir M.

954

Energy Efficiency in the Ceramic Industry: Recovery Heat of Combustion of Smoke Oven for a Spray-Drayer by V. Bristot, V. Bristot, L. Schaeffer, V. Gruber, J. Alves, J. Mangili, R. Tassi

959

Flow of a Second Order/Grade Fluid Induced by a Pull of Disks in an Orthogonal Rheometer Under the Effect of a Magnetic Field by H. Volkan Ersoy

966

Impact of Liquid Pressure Losses and Solid-Phase, in the Performance of a Three-Phase Flow Air-Lift Pump by Dimitrios N. Androulakis, Apostolos N. Vlachos, Dionissios P. Margaris

972

Design of Compliant Mechanisms by Topology Optimization for Strain Actuators and Engineering Support by G. Arunkumar, J. Santhakumar

979

PID Controller Tuning for Magnetic Suspension System Using Evolutionary Algorithm by V. Rajinikanth, K. Latha

988

Optimization of Deep Drawing Process Parameters Using Design of Experiments by J. Santhakumar, G. Arunkumar

996

Finite Element Simulation and Experimental Evaluation on Superplastic Forming Process of Aluminium Alloy Sheet by G. Kumaresan, K. Kalaichelvan

1001

Wall-Models for LES of Channel Flows by Taieb Nehari, Lotfi Tefiani, Driss Nehari

1005

Modelling of Centrifugal Compressor Impellers Using Adaptive Neuro- Fuzzy Inference Systems (ANFIS) by Layth H. Jawad, S. Abdullah, R. Zulkifli, W. M. F. W. Mahmood

1011

Numerical Study of Heat and Mass Transfer in Membrane Distillation for Desalination of Seawater by A. Rachdi, R. Sebai, F. Bouslama, R. Chouikh

1018

Supervision and Control Architecture Proposal for Automation and Robotics Training on Platform by Ricardo A. Castillo, João M. Rosário, Oscar F. Aviles

1025

Injector Nozzle Spray on Compressed Natural Gas Engines: A Technical Review by Semin

1035

Design and Development of Camless Valve Train for I.C. Engines by R. Shriram, M. Mari Ram Kumar, T. Vignesh Aadhithya, B. Vijaya Ramnath

1044

(continued)

Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

An Analytical Investigation of Overall Thermal Transfer Value on Commercial Building in Malaysia by M. F. Sukri, M. A. Salim, M. A. Mohd Rosli, S. B. Azraai, R. Mat Dan

1050

Analysis of a Diesel Engine with Developed Multi-Zone Combustion Model by Considering Heat Transfer between Zones by Reza Akbarpour Ghiasi, Yahya Ajabshirchi

1057

Prediction of Machining Parameters of Surface Roughness of GFRP Composite By Applying ANN and RSM by S. Ranganathan, T. Senthilvelan

1068

Characterization of Two Sprays Interacting by A. Amoresano, C. De Nicola, F. De Domenico

1074

A Sensitive Methodology for the EGR Optimization: a Perspective Study by A. Amoresano, V. Niola, A. Quaremba

1082

Numerical Analysis of Anti-Icing and De-Icing Thermal Systems by F. De Domenico, A. Amoresano, C. De Nicola

1089

Errata coririge

1096

International Review of Mechanical Engineering (I.RE.M.E.), Vol. 6, N. 5 ISSN 1970 - 8734 July 2012

High Temperature Air Combustion: Sustainable Technology to Low NOx Formation Seyed Ehsan Hosseini, Mazlan Abdul Wahid, Abuelnuor Abdeen Ali Abuelnuor Abstract – In recent decade, more stringent laws have been ordained to cope with environmental issues and global warming. Industrial sectors have been urged to substitute new combustion methods to decline their emissions, but the cost of pollutant reduction in traditional combustion is efficiency abatement. In the other word, emission and fuel consumption cannot be declined simultaneously by conventional combustion. High temperature air combustion (Hitac) is an innovative substitution for conventional combustion which has been developed to increase combustion efficiency and to decline pollutant formation contemporaneously. Recently, some valuable experimental and numerical analysis have been done to study the variety aspects of Hitac and to study the reasons of the compatibility of high efficiency and low NOx production in Hitac area. The outstanding characteristic of Hitac is its sustainability under low oxygen concentration when the combustion air is preheated more than the fuel auto-ignition temperature. Therefore, it can be observed that thermal NOx is suppressed due to lack of oxygen concentration. This paper is concerned with NOx formation reduction in Hitac systems via physical and chemical analysis. Chemical kinetic, heat transfer concepts, simulation studies and experimental investigations have been employed to analyze NOx formation mitigation in Hitac method. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved. Keywords: Hitac, NOx Formation, Preheated Air, Dilution, Flame

Respiratory irritation, asthma attacks, aggravation of heart diseases and lung damage are the main consequences of ground level ozone. . Indeed, acid rain formation, visibility reduction due to increase smog and particulate matter are other disadvantages of raising NOx production [3]. Furthermore, N2O is capable to convert to NOx in suitable conditions and it has been reported that the effect of N2O on global warming is 310 times more than dioxide carbon; consequently NOx has great potential to increase global warming indirectly [4]. By invoking to aforementioned disadvantages of NOx, it can be claimed that NOx is a kind of hazardous gases formed in combustion and it can endanger animal, vegetable and humankind life. Therefore, comprehensive investigation should be done to reduce NOx formation in combustion methods. Approximately, 79% by volume of air is nitrogen, thus in combustion systems which air is applied as an oxidizer, the presence of nitrogen in combustion process is unavoidable. However, under the specific circumstances in the furnace NOx is formed due to oxygen and nitrogen reaction [5]. Thermal NOx, N2O intermediate NO formation mechanism, prompt NOx, and fuelbound nitrogen are named as the main mechanisms for NOx formation in different combustion methods [6].

Nomenclature Ai, Bi, Ci K1,K2,K3 K-1,K-2,K-3 Kv Me Mf Ma

Reaction constant Reaction constant for forward reactions Reverse rate constants Recircultion ratio Flow rate of exhaust gases Flow rate of the fuel Flow rate of oxidizer

I.

Introduction

Generally, NOx which is an abbreviation for the combination of NO2 and NO is usually constituted in presence of nitrogen and oxygen within a locally high temperature conditions. Atmosphere can be jeopardized by increasing NOx formation in industrial sectors. Particularly, acid rain, ozone depletion and smog are mentioned as the main consequences of more NOx production [1]. Ground level ozone or so-call bad ozone is usually formed in presence of NOx according to following reactions [2]: NO2 O +O2

NO+O ∆H=57kJ, ∆G= -51.3 kJ

(1)

O3 ∆H=142 kJ, ∆G = 163.2 kJ

(2)

Manuscript received and revised June 2012, accepted July 2012


947

Seyed Ehsan Hosseini, Mazlan Abdul Wahid, Abuelnuor Abdeen Ali Abuelnuor

II.

II.3.

NOx Formation Mechanisms II.1.

Malte and Pratt [13] introduced N2O intermediate NO formation mechanism which is occurred in lean fuel, moderate temperatures and low pressure combustion conditions. In these circumstances N2O is converted to NO via following formulas [14]:

Thermal NOx

Oxygen and nitrogen in extremely high temperature can react inside the combustion furnace according to following reactions which are called Zeldivich formulation [7]: O + N2

NO + N

(3)

N + O2

NO + O

(4)

N + OH

NO + H

(5)

N2O Intermediate NO Formation

N2 + O + M

N2O + M

(8)

N2O + O

N2 + O2

(9)

N2O + O

NO + NO

(10)

Following processes show the presence of H2O impurities conspicuously affect the N2O decomposition [15]:

where K1, K2, K3 are forward rate constants and K-1, K-2, K-3 are the reverse rate constants according to Table I [2]. Thermal NOx formation is accelerated exponentially according to formula (6) at temperatures more than 1500oC [9], [10]:

H2O + O

OH + OH

(11)

N2O + OH

N2 + H2O

(12)

(6) In Eq.(8), M is general third body. The N2O which was constituted in Eq.(8) decomposes by Eq.(9),(11). Eq.(13) shows the chemical kinetics law for the rate of NOx formation via N2O intermediate mechanism:

The rate of NO formation is achieved by formula (7) when the nitrogen radical (N) assumed in steady state conditions: 1

(13)

1

The reaction rate constants are calculated by Eq.(14):

2

(7)

2

exp 0 then Eq (14) can be written as:

If where T is temperature(kelvin) and the reaction constant , , where taken in Table I. Form formulas (6) and (7) it can be seen when the rate of reaction decreases, NOx formation declines because combustion takes place in limited time in the furnace. Moreover, NOx constitution mitigates in low temperatures.

The concentration of radicals like O, OH and H affect the concentration rates and NO constitution [1]. The N2O intermediate NO formation is accelerates in low oxygen concentration conditions [16].

TABLE I THERMAL NOX REACTION RATE CONSTANTS 1.8

K1

10

[m3/gmol s]

K-1

3.8

10

[m3/gmol s]

K2

1.8

10

[m3/gmol s]

3.8

K-2

7.1

K3

1.7

K-3

II.2.

10 10 10

(14)

II.4.

[m3/gmol s]

Fuel Bond NOx Formation Mechanism

Fuel-bound NOx formation mechanism is occurred when the molecular structures of the fuel are constituted of nitrogen species [17]. In the combustion of these fuels the nitrogen atoms are decomposed to intermediate products form which can react with NOx [18].

[m3/gmol s] [m3/gmol s]

Prompt NOx

Prompt NOx formation mechanism was introduced by Fenimore in 1971. Prompt or Fenimore NOx formation occurs in fuel rich conditions (equivalence ratio greater than 1.2) [11]. The rate of prompt NOx constitution augments near equivalence ratio of 1.4 [12].

III. Hitac Hitac technology, emerged from 1990, and it has been successfully applied, specially, in metallurgy and steel industries of some developed countries. Flameless


International Review of Mechanical Engineering, Vol. 6, N. 5

948


Oxidation (FLOX) in Germany, also known as High Temperature Air Combustion (Hitac) in Japan, Moderate and Intensive Low oxygen Dilution (MILD) combustion in Italy [19], or Colorless Distributed Combustion (CDC) [21], Low NOx Emission Injection in the US is a new combustion system which accomplishes low NOx emissions and high efficiency among several techniques. Postponed mixing of fuel and air and flue gas utilization in the flame zone are the fundamentals of Hitac [22]. The application of high temperature air combustion has been investigated experimentally [23][28] and numerically [29]-[35]. Fundamentally, Hitac is identified by aspects of turbulence and chemistry strongly [36]. The characteristics of flameless oxidation for various gas type fuels like methane or ethane [37], also for mixtures of gaseous hydrocarbons and hydrogen [38], and biogas [39], [40] has been analyzed. In addition, this method has been applied successfully for liquid fuels [41]-[43] and solid fuels [44], [45].

IV.

Fig. 1. Methane Hitac formation conditions

It can be seen that the high temperature reactants and low temperature products are the most important items in Hitac formation.

V.

In order to maintain the temperature of exhaust gases, the regenerative or recuperative heat exchangers are used in Hitac systems and the energy of flue gases is absorbed by this equipment to increase the temperature of the combustion air [50]. Higher efficiency is achieved by applying this secondary air [53]. According to experimental results NOx constitution decreases from 1500 ppm in conventional combustion to 100 ppm in Hitac by applying recuperative heat exchangers [54]. In Hitac systems the fuel nozzle surrounds by regenerators which are made by honeycomb [55]. These honeycombs recover around 72% exhaust gases energy [56].

Flameless Formation

In Hitac the chemical reaction is occurred between the fuel and highly diluted air inside the combustion furnace in temperature above the self-ignition of the fuel [46], [47]. In these circumstances the structure of the flame is altered and become pale [48], [49]. The energy of exhaust gas which is wasted from chamber stack is applied in Hitac as exhaust gas recirculation (EGR) system. Therefore, the combustion efficiency increases drastically in Hitac systems [40]. The performance of Hitac systems is evaluated by recircultion ratio (Kv) in Eq (15): Kv =Me/ (Ma +Mf)

Hitac Heat Exchangers

VI.

The Role of Diluted Air in Hitac

The color, size, luminosity, visibility and lift of distance of the flame are changed by the amount of oxygen in combustion air. Luminosity and visibility of the flame decrease when the oxygen concentration declines in the oxidizer. Lift-off flame and a large ignition delay occurred in conventional flame due to diluted oxidizer with nitrogen [57], [63]. An ultra-lean mixture is defined as a fuel/air lean premixed gas mixture with the fuel concentration near to the lower flammability limit [65]. Fuel ultra-lean mixture combustion cannot take place in the low concentration of oxygen; therefore air preheating method is the best option to accelerate the reactions [20]. NOx formation is significantly increases when air concentration increases due to flame temperature enhancement. Fig. 2 shows the linearity relation between NOx formation and oxygen availability [18]. The rate of NO production diagram in CFD simulation is more similar to reality when N2O-intermediate mechanism is taken into account. Experimental investigations confirm that Hitac systems are reachable with diluted oxidizer. Low oxygen concentration is the main factor to achieve the flameless combustion regime.

(15)

In this formula Me is the flow rate of exhaust gases before reaction, Mf is the flow rate of the fuel and Ma is the flow rate of oxidizer [50]. In order to have flameless mode it is necessary to heat up the chamber over the selfignition of the fuel. Therefore, the system should be run by traditional combustion at the first step. In order to transient from conventional flame to flameless mode the preheated and diluted oxidizer should be charged to the chamber by very high velocity. In these conditions the visible and audible flame becomes disappear and the reaction region spreads in whole of the combustion chamber [51]. As a result the uniform temperature is observed inside the chamber and hot spots are eliminated and thermal NOx is suppressed [7]. The recirculation ratio should be more than 2.5 and the chamber temperature more than 1100 in Hitac conditions [52]. Moreover, the rate of concentration of oxygen decreases by EGR application [27]. The suitable conditions for Hitac constitution has been shown in Fig. 1.



949

Seyedd Ehsan Hosseeini, Mazlan Abdul A Wahid, Abuelnuor A Abbdeen Ali Abuelnuor

To obtainn diluted oxidizer in Hitac systems, fluee gas recirculation by high mom mentum air innjection is onne of the best optioons [50]. Alsoo, highly prehheated air which is constituted a large amountt of inert gasees such as N2, H2O and CO2 hass been appliedd to obtain Hitac H conditionns as shown in Figg. 3.

VIII.

The Effects E of Heat Transfe fer on NOx Formation in Hitacc

The T role of heat transferr in differentt combustionn furn naces have beeen investigateed by Khalil [65] by CFD D metthod. In flameeless mode, heeat transfer viia radiation iss don ne by suspendded particles iinside the furrnace and thee cham mber walls which w were hheated during conventionall com mbustion by luuminous flam me. Figure 5 illustrates i thee heatt transfer mechanisms in a Hiitac furnacee scheematically [588].

F 2. Excess air effects on NO formation Fig.

Fig. 5. Heaat transfer mechaanisms in Hitac fu urnace

Nabil N Rafidi et e al [59] stipulated that th he amounts off partticles in Hitaac furnace aare more thaan traditionall com mbustion. Theerefore, the rate of heat transfer viaa radiiation mechannism from pparticles incrreases. As a resu ult, hot spots are a omitted annd the uniform m temperaturee can be observedd inside the furnace due to dispersedd reacction zone. In n these circuumstances the pale flamee temperaturee redu uces and therrmal NOx is ssurpassed beccause thermall NOx is constituted in very highh temperature.. Also oxygenn conccentration proomotes thermaal NOx. Indeeed, the role off resid dent time in thermal NOx formation is unavoidable.. Uniform temperaature inside H Hitac furnace is one of thee pow wer points off this combusstion method. Moreover,, dilu uted high tempperature air is injected insside the Hitacc cham mber in high velocity. v Therefore, T it iss concluded thhat thermal NO N x formationn is suppressed. s T The role of particles insside Hitac iss conspicuously nootable to provvide moderatee temperaturee enviironment. Moore than 90% oof NOx production in Hitacc is reelated to N2O-intermediate mechanism [60]-[62]. It iss noteeworthy that the t reactions ((14), (15), (16 6) involve thee oxy ygen radical, thhus this mechhanism is favorable in Hitacc with h low oxygeen concentrattion atmosph here. Figure.66 show ws the summ mery of Hitaac specificatio ons and NOx form mation mechanisms and confirms that the N2O-intermediate mecchanism is doominance in NO N formationn in Hitac. It is i noteworthhy that in conventionall com mbustion prehheated air inncreases the efficiency off com mbustion but NO N x formationn increases siimultaneouslyy due to increase thhermal NOx foormation.

Fig. 3. Schem matic of air preheeating and dilutioon system of MIL LD combusttion technology

Experimenntal investiigations connfirm that air combustion dilution d with H2O shows more m compatibbility with low NOx constitutioon and likew wise dilution with CO2 is more in agreementt with NOx foormation reducction rather than N2 diluter, because the speciific heat of H2O is more than N2 and CO2. Figure 4 depicts the NOx formation trrend in Hitacc system in different diluution conditions [558].

fects of dilution inn various conditioons on NOx formaation Fig. 4. The effe

Copyright © 20012 Praise Worthyy Prize S.r.l. - Alll rights reserved

Internationnal Review of Mecchanical Engineeering, Vol. 6, N. 5

950


[4]

However, in Hitac high temperature reactant not only increases the efficiency of combustion but also reduces NOx formation dramatically [58].

[5]

[6]

[7]

[8]

[9]

[10] Fig. 6. Hitac specifications and NOx formation mechanisms [11]

VIII. Conclusion Thermal NOx which is named as the most probable mechanism in combustion phenomena is suppressed in Hitac due to low resident time, low oxygen concentration and moderate temperature inside the chamber. In traditional combustion the efficiency of furnace increases significantly by using preheated air in combustion process. However, NOx formation decreases drastically. Hitac has many features that are superior to traditional combustion. Not only fuel consumption but also the nitrogen oxide (NOx ) formation is kept at very low level due to EGR application in Hitac. These exhausted products are led into the fresh reactants inside the chamber; therefore high peak temperature is eliminated. As a result, NOx formation via thermal NOx mechanism is omitted, and other inconspicuous NOx formation methods in conventional combustion are remained. N2O-intermediate NOx formation mechanism is dominance method for NOx formation in Hitac due to its moderate temperature and lean fuel condition. The dual role of small particles inside the Hitac chamber is important. In one hand, heat transfer via radiation is occurred due to presence of these particles; therefore the temperature of pale flame declines dramatically and thermal NOx suppressed. On the other aspect, the presence of these radicals in mediate temperature and lean fuel condition lead the system to N2O-intermediate NO formation mechanism. The fuel consumption and NOx formation in Hitac systems reduce around 30% and 70% respectively compared to conventional mechanisms.

[12]

[13]

[14]

[15]

[16]

[17]

[18] [19]

[20]

[21]

[22]

[23]

References [1]

[2] [3]

Abbas Khoshhal, Masoud Rahimi, Ammar Abdulaziz Alsairafi, CFD study on influence of fuel temperature on NOx emission in a HiTAC furnace, International Communications in Heat and Mass Transfer 38 (2011) 1421–1427. Sandun Fernando, Chris Hall, and Saroj Jha, NOx Reduction from Biodiesel Fuels, Energy & Fuels 2006, 20, 376-382. Environmental Protection Agency. AIR Nows Ozone and Your Health. 2004 (cited March, 2012); www.airnow.gov.

[24]

[25]

[26]


International Carbon Bank and Exchange. Calculating Greenhouse Gases Emissions, 2000 www.icbe.com/emissions/calculate.asp. Reed, R.J., North American Combustion Handbook, third edition, Vol. II, The North American Manufacturing Company, Ltd., 1997, 153–176. J.A. Miller, C.T. Bowman, Mechanism and modeling of nitrogen chemistry in combustion, Progress in Energy and Combustion Science 15 (1989) 287–338. Bowman C. Kinetics of pollutant formation and destruction in combustion. Progress in Energy and Combustion Science 1975; 1:33-45. Yeshayahou Levy , Valery Sherbaum, Patric Arfi , Basic thermodynamics of FLOXCOM, the low-NOx gas turbines adiabatic combustor, Applied Thermal Engineering 24 (2004) 1593–1605. Dean A, Bozzelli J. Combustion chemistry of nitrogen in gasphase combustion chemistry. Gardiner Jr WC, editor. New York: Springer-Verlag; 2000.p. 125-341. Lavoie G, Heywood J, Keck J. Experimental and theoretical study of nitric oxide formation in internal combustion engines. Combustion Science and Technology 1970; 1:313-26. Fenimore, C. P., (1971) "Formation of Nitric Oxide in Premixed Hydrocarbon Flames." Proceedings of the 13l Symposium (International) on Combustion Pittsburg, PA, The Combustion Institute. Pillier, A. E., Mercier, X. R., Pauwels, J. F., and Desgroux, P., (2005), "Influence of C-2 and C-3 Compounds of Natural Gas on NO Formation: An Experimental Study Based on LIF/CRDS Coupling", 30th Symposium (Intrnational) Combustion, Vol. 30, pp. 1183-1191. P. C. Melte and D. T. Pratt, Measurement of Atomic Oxygen and Nitrogen Oxides in Jet Stirred Combustion, Proc. 15th Int. Symposium on Combustion, The Combustion Institute, pp. 1061– 1070, Tokyo, Japan, 1974. G. Lfffler, V.J. Wargadalam, F. Winter, H. Hofbauer, Decomposition of Nitrous Oxide at Medium Temperatures, Combustion and Flame 120 (2000) 427. Johnsson, J. E., Glarborg, P., and Dam-Johansen, K., TwentyFourth Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, 1992, p.917. Weihong Yang, Wlodzimierz Blasiak, Mathematical modeling of NO emissions from high temperature air combustion with nitrous oxide mechanism, Fuel Processing Technology 86 (2005) 943 – 957. Flanagan, P., Gretsinger, K., Abbasi, H. A., and Cygan, D., (1992), Factors Influencing Low Emission Combustion, PD-Vol. 39, Fossil Fuels Combustion, ASME. Glassman, Irvin, (1977). "Combustion" Third Edition, Academic Press, New York. Özdemir ÍB, Peters N. Characteristics of the reaction zone in a combustor operating at mild combustion. Exp Fluids 2001; 30:683–95. 330. Fethi Bouras, Azeddine Soudani, Mohamed Si-Ameur. Beta-Pdf Approach for Large-Eddy Simulation of Non-Premixed Turbulent Combustion, International Review of Mechanical Engineering (IREME), March 2010, Vol. 4. n. 3, pp. 358-363. Arghode VK, Gupta AK. Development of high intensity CDC combustor for gas turbine engines, Applied Energy 2011; 88:963– 73. Cho E-S, Chung SH. Improvement of flame stability and Nox reduction in hydrogen-added ultra-lean premixed combustion. Journal of Mechanical Science Technology 2009; 23:650–8. M. Mortberg, W. Blasiak, A.K. Gupta, Experimental investigation of flow phenomena of a single fuel jet in cross-flow during highly preheated air combustion conditions, J. Eng. Gas Turbines Power 129 (2007) 556-564. T. Hasegawa, S. Mochida, A.K. Gupta, Development of advanced industrial furnace using highly preheated air combustion, J. Propul. Power 18 (2) (2002) 233-239. P. Sabia, M. de Joannon, S. Fierro, A. Tregrossi, A. Cavaliere, Hydrogen-enriched methane mild combustion in a well stirred reactor, Exp. Therm. Fluid Sci. 31 (2007) 469-475. T. Plessing, N. Peters, J.G. Wünning, Laser optical investigation of highly preheated combustion with strong exhaust gas


951


[27]

[28]

[29] [30]

[31]

[32]

[33]

[34]

[35]

[36]

[37]

[38]

[39]

[40]

[41]

[42] [43]

[44]

[45]

[46]

[47]

recirculation, Twenty-Seventh Symp. (International) Combust.2 (1998) 3197-3204. E.-S. Cho, B. Danon, W. de Jong, D.J.E.M. Roekaerts, Behavior of a 300 kWth regenerative multi-burner flameless oxidation furnace, Applied Energy 88 (2011) 4952–4959. B.B. Dally, A.N. Karpetis, R.S. Barlow, Structure of turbulent non-premixed jet flames in a diluted hot coflow, Proc. Combust. Inst. 29 (2002) 1147-1154. P.J. Coelho, N. Peters, Numerical simulation of a mild combustion burner, Combut. Flame 124 (2001)503-518. M. Mancini, P. Schwoppe, R. Weber, S. Orsino, on mathematical stabilization of flameless combustion, Combust. Flame 150 (2007) 54-59. M. Mancini, R. Weber, U. Bollettini, Predicting Nox emissions of a burner operated in flameless oxidation mode, Proc. Combust. Ins 29 (2002)1155-1163. E. Oldenhof, M.J. Tummers, E.H. van Veen, D.J.E.M. Roekaerts, Role of entrainment in the stabilization of jet-in-hot-coflow flames, Combust. Flame158 (8) (2011) 1553-1563. B. Danon, W. de Jong, D.J.E.M. Roekaerts, Experimental and numerical investigation of a FLOX 369 combustor firing low calorific value gases, Combustion Science and technology 182(9) pp 1261-1278. A. De, E. Oldenhof, P. Sathiah, D.J.E.M. Roekaerts, Numerical simulation of Delft-jet-in-hot coflow (DJHC) flames using the Eddy dissipation concept model for turbulence-chemistry interaction, Flow Turbulence Combust. Doi: 10.1007/s10494-0119337-0. A. Hamidane, H. Benmoussa, S. Belhas, Numerical Study of a Turbulent Flow in Flameless Combustion Chamber, International Review of Mechanical Engineering (IREME), February 2012, Vol. 6 N. 2, pp. 205-208. J. Aminian , C. Galletti, S. Shahhosseini , L. Tognotti, Key modeling issues in prediction of minor species in dilutedpreheated combustion conditions, Applied Thermal Engineering 31 (2011) 3287-3300. Effuggi A, Gelosa D, Derudi M, et al. Mild combustion of methane-derived fuel mixtures: Natural gas and biogas. Combust SciTechnol, 2008, 180(3): 481–493. Derudi M, Villani A, Rota R. Sustainability of mild combustion of hydrogen-containing hybrid fuels.Proc Combust Inst, 2007, 31(2): 3393–3400. Sheng Chen ,Chuguang Zhenga, Counter flow diffusion flame of hydrogen-enriched biogas under MILD oxy-fuel condition. International journal of hydrogen energy 36 (2011)15403-15413. Weihong Yang , Anna Ponzio, Carlos Lucas, Wlodzimierz Blasiak, Performance analysis of a fixed-bed biomass gasifier using high-temperature air, Fuel Processing Technology 87 (2006) 235 – 245. Marco Derudi, Renato Rota Politecnico di Milano, Experimental study of the mild combustion of liquid hydrocarbons, Proceedings of the Combustion Institute 33 (2011) 3325–3332. S.R. Wu, W.C. Chang, J. Chiao, Low NOx heavy fuel oil combustion with high temperature air, Fuel 86 (2007) 820–828. Weihong Yang, Wlodzimierz Blasiak, Numerical simulation of properties of a LPG flame with high-temperature air, International Journal of Thermal Sciences 44 (2005) 973–985. N. Schaffel, M. Mancini , A. Szlek , R. Weber b, Mathematical modeling of MILD combustion of pulverized coal, Combustion and Flame 156 (2009) 1771–1784. Hannes Stadler , Dobrin Toporov, Malte Forster, Reinhold Kneer, On the influence of the char gasification reactions on NO formation in flameless coal combustion, Combustion and Flame 156 (2009) 1755–1763. Frassoldati, A., Cuoci, A., Faravelli, T.,Ranzi, E., Astesiano, D., Valenza, M.,Dei Lambruschini, C., Sharma, P. Experimental and modeling study of low- NOx industrial burners. MPT Metallurgical Plant and Technology International 31 (6), pp. 4446. C. Pfeiler, H. Raupenstrauch. The Influence of Different Turbulence Models on the Predictions for Turbulent NonPremixed Flames. International Review of Mechanical Engineering (IREME), February 2010, Vol. 4. n. 2, pp. 201-207.

[48] Kiomars Abbasi Khazaei, Ali Asghar Hamidi, and Masoud Rahimi . Numerical Investigation of Fuel Dilution Effects on the Performance of the Conventional and the Highly Preheated and Diluted Air Combustion Furnaces, Chinese Journal of Chemical Engineering, Volume 17, Issue 5, October 2009, Pages 711-726. [49] Dally, B.B., Riesmeier, E., Peters, N, Effect of fuel mixture on moderate and intense low oxygen dilution combustion, Combust. Flame, 137 (4), 418-431 (2004). [50] Wünning JA, Wünning JG. Flameless oxidation to reduce thermal NO formation. Progress Energy Combustion Science 1997; 23:81–94. [51] Szegö G G, Dally B B, Nathan G J. Operational characteristics of a parallel jet MILD combustion burner system. Combust Flame, 2009, 156(2): 429–438. [52] MI Jianchun, LI Pengfei and ZHENG Chuguang , Numerical Simulation of Flameless Premixed Combustion with an Annular Nozzle in a Recuperative Furnace, Chinese journal of chemical engineering 18(1) 338 pp255-269. [53] Hai Zhang , Guangxi Yuea, Junfu Lu , Zhen Jiaa, Jiangxiong Mao , Toshiro Fujimori , Toshiyuki Sukob, Takashi Kiga, Development of high temperature air combustion technology in pulverized fossil fuel fired boilers, 2007 Proceedings of the Combustion Institute 31 II , pp. 2779-2785. [54] C. Galletti, A. Parente, L. Tognotti, Numerical and experimental investigation of a mild combustion burner, Combust. Flame 151 (2007) 649-664. [55] Tsuji H, Gupta AK, Hasegawa T, Katsuki M, Kishimoto K, Morita M. High temperature air combustion. CRC Press; 2003. [56] P.M. Park, H.C. Cho, H.D. Shin, Unsteady thermal flow analysis in a heat regenerator with spherical particles, Int. J. Energy Res. 27 (2003) 161–172. [57] Kishimoto K, Watanabe Y, Kasahara M, Hasegawa T, Tanaka R. Observational study of Chemiluminescence from flames with preheated and low oxygen air. The First Asia-Pacific Conference on Combustion, May 12–15, Osaka, Japan. 1997, p. 468–71. [58] H.Oryani, Sh.Khalilarya , S.Jafarmadar, H.Khatamnezhad and S.Majidyfar, Numerical Investigation of Influence of Dilution in Air and Fuel Sides on MILD Combustion Burner, Australian Journal of Basic and Applied Sciences, 5(10): 272-279, 2011 ISSN 1991-8178. [59] Nabil Rafidi , Wlodzimierz Blasiak , Heat transfer characteristics of HiTAC heating furnace using regenerative burners, Applied Thermal Engineering 26 (2006) 2027–2034. [60] P. Glarborg, J. E. Johnsson, and K. Dam-Johansen, Kinetics of Homogeneous Nitrous Oxide Decomposition, Combust. Flame, vol. 99, pp. 523–532, 1994. [61] C. Galletti, A. Parente, M. Derudi, R. Rota, and L. Tognotti, Numerical and Experimental Analysis of NO Emissions from a Lab-Scale Burner Fed with Hydrogen-Enriched Fuels and Operating in MILD Combustion, Int. J. Hydrogen Energ., vol. 34, pp. 8339–8351, 2009. [62] R. C. Steele, P. C. Malte, D. G. Nichol, and J. C. Kramlich, NOx and N2O in Lean-Premixed Jet-Stirred Flames, Combust. Flame, vol. 100, pp. 440–449, 1995. [63] V. Rambabu, V. J. J. Prasad, T. Subramanyam, B. Satyanarayana, Evaluation of Performance, Combustion Characteristics and Emissions of DI- Diesel Engine Fueled with Preheated Cotton Seed Methyl Ester, International Review of Mechanical Engineering (IREME), July 2010, Vol. 4. n. 5, pp. 502-506. [64] Essam E. Khalil, Combustion and Heat Transfer in Furnaces and Combustion Chambers, International Review of Mechanical Engineering (IREME), February 2012, Vol. 4 N. 2, pp. 142-149. [65] Allouis, C., Beretta, F., Amoresano, A., Experimental study of lean premixed prevaporized combustion fluctuations in a gas turbine burner, (2008) Combustion Science and Technology, 180 (5), pp. 900-909.



952


Authors’ information Seyed Ehsan Hosseini was born in Esfahan IRAN at 23 August 1978. He recieved his Bachelor of Mechanical Engineering, heat transfer and fluid mechanics from Semnan University in Iran at 2001and Master of Mechanical Engineering from Universiti Teknologi Malaysia at 2012. He has around ten years work experiences in steel industries in IRAN. His research interests are conventional and MILD combustion, alternative fuel, renewable and sustainable energies and environmental issues. Mr. Hosseini is one of the members of High Speed Reacting Flow Laboratory (HiREF) in Universiti Teknologi Malaysia. Mazlan Abdul Wahid (Corresponding Author) was born at first of October 1966 in Malaysia. He received his PhD in 2003 from State University of New York at Buffalo and Master Degree in 1994 from University of Leeds, UK. He has co-authored two books and over 80 papers. He currently served as the Head of High Speed Reacting Flow Laboratory at Faculty of Mechanical Engineering, UTM and spearheading fundamental and engineering applications research of high speed flows in the presence of different physical phenomena such as combustion and heat transfer. He is also involved in the research related to turbulence combustion modeling, pulse combustion and supersonic reacting shock wave phenomena. Dr. Mazlan Abdul Wahid was the Chairman of the 10th Asian International Conference on Fluid Machinery, Kuala Lumpur, Malaysia 2009 and 4th International Meeting on Advances of Thermofluids, Melaka, Malaysia 2011. Abuelnuor Abdeen Ali Abuelnuor was born in Argo Sudan at 03 November 1970. He received his High Diploma in Mechanical Engineering University of the Nile Valley Atbara Sudan 1994, Bachelor in mechanical engineering Sudan University of Science and Technology, 1999, M.Sc. in Laser application in mechanical engineering, Sudan University of Science and Technology (SUST) 2007 and Currently he is working toward the Ph.D. degree at the Faculty of Mechanical Engineering (FKM), in Universiti Teknologi Malaysia (UTM) under Dr Mazlan Abdul Wahid supervision. He has around fourteen years work experiences in teaching in Sudan University of Science and Technology. His research interests are conventional and flameless combustion, Heat and mass transfer and Laser application in mechanical engineering.Mr. Abuelnuor is one of the members of High Speed Reacting Flow Laboratory (HiREF) in Universiti Teknologi Malaysia.



953


Resolution of an Inverse Problem in Thermal Diffusion Process for the Identification of a Heat Flux Abdelkarim Maamar, Touhami Y., Bounegta B., Khadir M.

Abstract – In this work we propose a temporal determination of a heat flow, within a diffusive 2d linear medium with resolution of an opposite problem. Here 2 different methods of resolution are suggested and a comparison between the two is made. Regarding the first method of inversion that is based on a convolution integral for that it is necessary first of all to solve the direct problem which requires the construction of the step responses and the temperature measurements using the finite differences method. In the reversal phase, the linear system is considered under the form of single-input (the flow to be identified), multi-output (temperature measurements). The 2nd method is the regularizing method of the combined gradient, where the direct problem with all the assumed known parameters has an analytical solution to be used as observation to estimate the unknown parameters with resolution of an inverse problem. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Linear Thermal Diffusion Process, Opposite Problem, Heat Flux, Convolution Integral, Combined Gradient, Finished Difference

Nomenclature a cp eT eSp h H(t) M(t) P u(t), y(t) d n +1 FLi

(

J u, p n

)

δT R

I.

Thermal diffusivity, m2s-1 Specific heat, J kg-1K-1 Average standard diversion of the temperature,°C Average standard difference of heat flow, W m-3 Convection coefficient, W m-2K-1 Impulse responses Indexical Responses Heat source, W m-3 Vectors Descente direction The heat flow, W Functional gradient

In several industrial applications, a direct measurement of a surface heat flow or a temperature (inner wall of a combustion chamber of an engine or a piping of a steam generator, interface between plate and disc of brakes, external surface of a spacecraft penetrating the atmosphere, etc...) is delicate and sometimes impossible to realize. The determination of these surface sizes starting from transitional measures of temperatures carried out inside and with the faces of the solid constitutes an inverse problem of heat conduction. In spite of their practical interest, the methods able to solve inverse problems of heat transfer can be numerous in literature. Most techniques which made it possible to solve the inverse problems successfully is based on procedures for minimizing a standard deviation criterion [1][2][3].

Sensitivity function No future time

Greek Letters

λ ρ

∆ γ n +1

II.

Thermal conductivity m-1K-1 Density, kg m-3 Laplacian Operator Depth descent

index/Exhibitors

Introduction

The System Studies and its Modelling

The geometry studied is L-shaped plate, shown in Fig. 1 it presents the results related to a building material of the concrete type. On all the plate sides, the boundary conditions of the flow field are imposed. One gives the law of variation of flow on the time interval [ 0, 6000s ], the evolutions of these three flows are represented on the figures (Fig. 2, Fig. 3 and Fig. 4). Here:

i, j, k, n, m



954

Abdelkarim Maamar, Touhami Y., Bounegta B., Khadir M.

FL1= 0.02 t 0< t < 6000s 100 0 < t < 4000s ⎧ FL2 = ⎨ ⎩50 4000 < t < 6000s FL3 = 200 0 < t < 6000s

225

FLUX 3 (W/m)

200

175

To model this system by finite deference, we chose a plate grid with: ∆X = 0,01 m and ∆Y = 0,005 m. To quantify this surface heat flow, a direct simulation allowing the storage of measurements is carried out on the sides and inside this plate. The evolutions of some selected temperatures are represented on Fig. 5 and the propagation sight overview of heat in the plate is shown in Fig. 6.

FL 3 Exacte

150

125

100

1000

2000

3000

4000

5000

6000

Time (s)

FL3 Ti2

Tf3

Ti3

Tf1

Ti1

FL2

45

Tf4 FL3

FL2

35

Ti4 Tf2

Ti1 Exacte Ti2 Exacte Ti3 Exacte

40

Temperature ( C)

FL1

Fig. 4. Evolution of three heat flow according to time

30 25

FL1

20

Fig. 1. Boundary conditions

15

140 130

10

120 FL 1 Exacte

110

5

100

0

FLUX 1 (W/m)

90

0

1000

2000

3000

4000

5000

6000

Time (s)

80 70

Fig. 5. Evolution of internal temperatures

60 50 40 30 20 10 0

0

1000

2000

3000

4000

5000

6000

Time (s)

Fig. 2. Evolution of three heat flow 1 according to time 120 110 100 90 FLUX 2 (W/m)

80 70

Fl 2 Exacte

60 50

Fig. 6. Fields u (x, t) solution of the direct problem at the final instant

40 30

III. Methods of Inversions

20 10 0

III.1. Convolution Integral 0

1000

2000

3000

4000

5000

6000

Time (s)

Having Known the H(t) matrix, it is then traditional to write the system response at any vector u(t) in the form

Fig. 3. Evolution of three heat flow 2 according to time



955


(

where:

t

y ( t ) = H ( t − τ ) u ( τ ) dτ

∫

(1) β = n

t0

where u(t) is the entries vector (dim·p), y(t) the outputs vector (dim q), H(t) the impulse responses matrix (dim.qxp), T is time, t 0 the initial time for which y(t0) = 0. By using this equation the matrix of the step responses M(t), we get: t

y (t ) = −

∫

dM ( t − τ )

t0

dτ

)

d n +1 = ∇J T,P n + β n d n

of the convolution integral, [4]:

u (τ ) dτ

(

∇J T,P n

(

)

∇J T,P n +1

2

)

(6) et β = 0 0

2

The depth of descent is defined at each iteration by:

{(

λ n +1 = Arg min J P n − γ d

(2)

n+1

)}

(7)

The iterative process of the conjugate gradient consists of [6]: initially, at the iteration n=0, select an initial vector, P 0 .

Thus the final result is:

(

Calculate the gradient of the functional J T ,P n

y = M u(t)

(

)

)

and

knowing that d 0 = ∇J T ,P 0 calculate:

(3)

with:

(

)

d n +1 = ∇J T ,P n + β n d n , n ≥ 0

⎡ y ( F ) − y* ( F ) ⎤ ⎢ ⎥ ⎢ y ( F + 1) − y* ( F + 1) ⎥ Y =⎢ ⎥ ⎢ ................................ ⎥ ⎢ ⎥ * ⎣ y ( F + R ) − y ( F + R )⎦

∫

with β n

calculated as previously. Find the depth of descent γ n +1 checking:

(

)

{(

J p n − γ n +1d n +1 = min J p n − γ d n +1

⎡ M (1) − M ( 0 ) ⎤ ⎢ ⎥ ⎢M ( 2) − M ( 0) ⎥ M =⎢ ..................... ⎥ ⎢ ⎥ ⎢⎣ M ( R ) − M ( 0 ) ⎥⎦

γ ≥0

)}

Calculate the new value of the vector p to the iteration n+1: P n +1 = P n − γ n +1d n +1

This relation constitutes a system of q*(R+1), the equations to P is unknown. In general this system is overdetermined, i.e. that q*(R+1)>p. The exact resolution is then not possible, by choosing a quadratic norm, then this is the solution to the least squares sense, thus:

u(F) = (MT M)-1 MT Y

Calculate the value of the quadratic functional: if

(

continue and returns to 2.

IV.

(4)

Result

According to Figs. 7 to 10, the flow is identified with the conjugate gradient method compared to the method of convolution integral with or without noise. According to Figs. 11 and 12 one the change of the temperatures of the internal nodes is represented according to simulated and identified measurements.

III.2. Method of the Combined Gradient (MGC) The iterative method consists in approaching the new iterate p n +1 from the reiterated preceding one P n with n the iteration count according to the following formula [5]: P n +1 = P n − γ n +1d n +1 (5)

(

)

J T ,P n +1 ≤ J stop the iterative process stops if not,

V.

Discussions

To justify our choice, all the obtained values of esp and eT in Tables I and II respectively are gathered, for the both inverse methods used, starting from the temperatures without and with noise, the values of the convolution integral method without noise and with noise measurement and a a pitch of future time R=4. Note that the conjugate gradient method shows the best results for flow identification.

)

where in the conjugate gradient method, γ n +1 > 0 the n +1

descent direction, d is the depth of descent. The direction must be built in such a way that the directions of successive descent are combined between them:



956


140

140

130

130 120

120 110 100

FL 1 Exacte FL 1 Ident IC

110

Ep = 1.97 w/m

100

esp = 1,97 W/m

Es = 1.76 W/m

90

FLUX 1 (W/m)

FLUX 1 (W/m)

90

80

80

70

70

60

60

50

50

40

40

30

30

20

20

10

10 0

FL 1 Exacte FL 1 Ident GC Bruit= 0.5, R=4

0

0

1000

2000 3000 Time (s)

4000

5000

6000

0

1000

2000

3000

Time (s)

4000

5000

6000

Fig. 10. Comparison of a stream of origin and identified noisy (noise = 0.5 ° C), (C. G)

Fig. 7. Comparison of a stream of origin and identified (Intégrale de convolution) 40 140 130

Ti1 Ti1 Ti2 Ti2 Ti3 Ti3

35

120

Flux 1 Exacte Flux 1 Ident GC

100

30 Temperature ( C)

110

Es= 0.06 W/m

25

FLUX 1 (W/m)

90 80

Exacte Ident GC Exacte Ident GC Exacte Ident GC

Et = 0.19

C

20

70 60

15

50 40

10

30

5

20 10 0

0 0

1000

2000

3000

4000

5000

6000

0

1000

2000

3000

4000

5000

6000

Time (s)

Time (s)

Fig. 11. Evolution of Temperature trends in internal nodes Comparison of Simulated and identified (GC

Fig. 8. Comparison between the exact solution and identified (conjugate gradient)

45 FL 1 Exacte 0 ,5 °C, R=0 FL 1 Ident Bruit = 0.5

140 130

Ti1 Ti1 Ti2 Ti2 Ti3 Ti3

40

Ep= 6.16 w/m

esp = 6,16 W/m

120

35 Temperature ( C)

110

30

100 FLUX 1 (W/m)

90

Et = 0.96

70

C

eT = 0,96 °C

25

80

Exacte Ident Exacte Ident Exacte Ident

20

60

15

50 40

10

30 20

5

10 0

0 0

1000

2000

3000

4000

5000

6000

0

1000

2000

3000

4000

5000

6000

Time (s)

Time (s)

Fig. 12. Evolution of Temperature trends in internal nodes. Comparison of Simulated and identified. (IC)

Fig. 9. Comparison of a stream of origin and identified noisy (noise = 0.5 ° C), (I. C)



957


[4]

If one looks at the squared deviations, on the temperatures which are the best criterion to justify our choice of the inverse method, one also notes that the conjugate gradient method shows the best results, for example for the internal nodes, without noise eT = 0,19 °C for the conjugate gradient method against eT = 0,96°C for the convolution integral method.

[5]

[6]

TABLE I VALUES OF ESP (W/M) WITHOUT NOISE AND WITH MEASUREMENT NOISE FOR THE TWO INVERSE METHODS WITH MEASUREMENT NOISE WITHOUT NOISE 0,5°C GC 0,06 1,76 FLOW1 IC (R=4) 1,97 2,56

Authors’ information 1,2,3,4

Laboratory of Energetics in Arid Regions, University of Béchar, Algeria.

TABLE II VARIOUS SQUARED DIFFERENCES: ET (IN °C) CALCULATED ON THE TEMPERATURES WITH NOISE WITHOUT NOISE 0,5 GC 0,19 0,26 0,38 IC 0,96 1,16 1,87

VI.

First Name: Abdelkarim Seconde Name : Maamar Born: 02/11/1963 in Ras El Ma Sidibelabbes (ALGERIA) Status: Married Children: 05 Grade: Teacher Assistant Course Professional Address: University of Bechar Algeria Staff Address: BP.417 Bechar (08000 E-mail: [email protected] Tel : 06 62 85 07 90, Tel / Fax Professional: 049 81 52 44 UNIVERSITY STUDIES 1984: BAC (Math) 1989: Engineering degree status in Mechanical Engineering (Energy) 2004: Diploma of Magister energy physics 2010: PhD thesis State University of Bechar LANGUAGES Arabic, French, English PROFESSIONAL EXPERIENCE 1990 - 2000: Teacher of Math Education 1997 - 2000: Associate at CUB (module Heating, Ventilation and Air Conditioning. 2000-2003: Professor at the Engineering Center University of Bechar. 2003-2006: Lecturer at University Center of Bechar. 2006 to date: Lecturer at University Center of Bechar. 2003 2008: Teacher of Math UFC. 2004-2009: Math Teacher Training Institute. AREAS OF RESEARCH 1. Thermal Transfer, Heating, Air Conditioning, The inverse method, renewable energy. MEANS OF SIMULATION USE 1. Finite volume method. 2. Method of finite differences 3. Method of element fnite 4. Fortran 5. Fluent. TOPICS OF SUPERVISION OF ENGINEERS • Ten subjects project graduation for students in Thermal Engineering (long cycles). • Ten subjects project graduation for students of environmental engineering (long cycle). • Topics Project graduation for LMD • Co-framer about magister. CONTRIBUTIONS TO THE SEMINARS: 1. Five participations (SIPE). 2. Two participations (ICHMT). 3. Two entries with the Tunisian Society of Physics. 4. A study day at Constantine "Code of fluid mechanics Transat.

Conclusion

In this work, we presented two methods of dealing with inverse problems in thermal conduction, the first method is based on convolution integral like the type Beck, the second is the conjugate gradient. From the perspective of the exploitation of digital simulation, one can underline the following aspects: All curves present good agreement when the measurements are noise-free simulated with two inverse methods. When measurements are noisy and on the convolution method gives a poor fit, and of course there is the presentation jump phenomenon and skew for R = 0 which will improve with the increase of the future time step. This phenomenon does not arise in the identification of the heat flux with the conjugate gradient method, the curves are "found" even with different measurement noise. Finally and to justify our choice between the two methods, one has to calculate each time the squared difference between the identified curves and the origin and according to the results found one notes well that of regularization of the iterative conjugate gradient the method presents the best variations compared to the convolution integral method for measurements with and without noise.

References [1]

[2]

[3]

Y. Touhami, Identification spatio-temporelle d’une source de chaleur dans un milieu diffusif par résolution d’un problème inverse, Thèse de doctorat de l’université de Provence, France, 1996. A. J. Silva Neto, M. N. Ozisik, Two-dimensional inverse heat conduction problem of estimating the time-varying strength of line heat source, J. Appl. Phys, vol. 71, n° 11, pp 123 - 134, juin 1992. S. Rouquette, Identification des transferts thermiques par méthode inverse dans un procédé de PACVD, approche méthodologique de la modélisation d'un processus complexe régi par un système d'équations aux dérivées partielles non linéaires, Thèse de doctorat, Université de Perpignan France, 19 décembre 2003.

G. Blank, M. Raynaud, A guide for the use of the function specification method for 2D inverse heat conduction problems, Rev. Gén. de thermique, vol 37, n° 4, pp 3 - 17, 1998. A. M. Osman, K. J. Dowding, J. V. Beck, Numerical solution of the general two-dimensional inverse heat conduction problem (IHCP) Trans. of the ASME, vol. 119, n°9, pp 38 - 44, 2004. N. Daouas, Analyse d'un problème inverse de conduction de chaleur bidimensionnelle non linéaire à l'aide du filtre de Kalman discrect, Congrès français de thermique, congrée SFT, 3- 6 juin, France, 2002.



958


Energy Efficiency in the Ceramic Industry: Recovery Heat of Combustion of Smoke Oven for a Spray-Drayer V. Bristot1, V. Bristot2, L. Schaeffer3, V. Gruber4, J. Alves5, J. Mangili6, R. Tassi7

Abstract – The expansion of the ceramic coating market in decades boosted the development and improvement of manufacturing techniques for many companies, resulting in a highly competitive market driven by low manufacturing costs. The Strategies to stay alive and even leading reference, are companies perform constant new product launches, by analysis of plans to reduce costs. It is known that the energy matrix of the ceramic tile industry Wet process corresponds approximately to the range of 22 to 30% of the composition manufacturing costs, where this percentage, about 37% of the consumption of thermal energy is directed to only one device, the spray-dryer. On the basis of these values in the last 10 years companies have maintained their focus directed to the strategic management energy efficiency, which depart from the principles cogeneration projects and recovery heat, to replace equipment with more efficient ones. The main objective this article is to demonstrate the planning, execution and outcome of a recovery project of heat from the flue gas scrubbing combustion furnace of a roll as an air dilution a spray-dryer in the coatings industry. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved. Keywords: Energy Efficiency, Furnace Rolls, Heat Recovery, Coatings Ceramic, Spray-Dryer

I.

installation high-speed burner for increasing the efficiency of heat exchange, and the control and regulating the temperature and pressure in the burning zone [6], [11]. Project recovery of waste heat from the production process are, were and are always optimal strategies to reduce the consumption of thermal energy, even when used on machines with high efficiency, although they are always recoveries heat which would otherwise be sent to the atmosphere [7].

Introduction

The thermal energy costs represent one of the largest portions of the costs ceramic coating production in Brazil and worldwide, according to Boschi et al. (2010) [1]. Enrique (2006) [2], said the increase in fuel consumption for power generation thermal and consequently production of ceramic tiles, as has long been an indirect indicator of increased production plants, assuming that knowing how much the competitor consumes fuel, represents the ability to know production thereof. Currently, large companies have invested in dedicated professionals exclusively to the management area of energy efficiency, aiming to find solutions energy and make acquisitions of reliable data on fuel consumption and productivity of each machine. These data are important for future strategic plans, aimed at increasing productivity, reducing costs, investments in new equipment and even the replacement [1]. The strategies and measures to reduce consumption have evolved over the years. First step was to contain the waste of energy during the manufacturing process, performing the recovery of energy lost in heat engines, especially in the furnace firing, taking the hot air to the spray-dryers, dryers and even for heating combustion air. Therefore, attention was directed to science and technology materials, with the adoption, since the design of thermal machines, devices to increase efficiency and reduce energy consumption, for example, the specification of materials low-density refractory to reduce the thermal mass of the structure of the oven, the

II.

Ceramic Wet Process

Nowadays, there are a wide variety of ceramic tiles of various shapes, sizes and aesthetic characteristics. Basically a number ceramic coating can be determined as a base overlaid with a glass cover, this support consisting of raw material source clay [9]. Regardless of the type of coating and/or shapes and dimensions, the manufacturing process of the wet coating may be explicit quite short and generic as shown in Fig. 1. Briefly the steps are defined as: 1- Box of raw material: covers the supply of raw materials used in Fabrication of the support composition. 2- Milling: a stage at which water and chemical solutions are added together to compositions of the raw materials so that in a ball mill the mixture is homogenized, and that the disintegration occurs method of reducing the size of particles.



959

V. Bristot, V. Bristot, L. Schaeffer, V. Gruber, J. Alves, J. Mangili, R. Tassi

support, making the ask for a harder material, resistant to water and chemicals, reduced porosity, formation of the vitreous layer and excellent technical characteristics. The heat necessary for promote the reaction is coming from the combustion of many burners along the side the oven. In Figure 3 there is a schematic of a furnace roller. 8- Rating: sector where there is a selection of ceramic pieces have characterized, distinguishing parts Class A and Class Commercial for visual and dimensional defects.

Fig. 1. Flowchart of the manufacturing process of ceramic tile wet

It appears at this stage that a ceramic slurry, aqueous solution + clay, called slip. 3- Spray: The spray-dryer, also known as the Brazilian industry atomizer, it is an equipment consisting of a huge drying chamber hot gas with temperatures around ± 600 ° C where the suspension to dry, slip, is injected at high pressure (25-30 kgf / cm ²), passing through nozzles that provide droplet formation, which in contact with the body heat in the chamber, dry, forming spherical particles of diameter ± 1.0 to 4.0 tenths of a millimeter and moisture content between 5-7% called atomized powder. Drying occurs by convection flows from the hot gas coming from combustion of natural gas in a heat generator diluted with ambient air at room temperature. In Figure 2 it can be see in detail the main components of an atomizer.

Fig. 3. Detailed diagram of the zones of an oven to roll

III. Fluid Mechanics All internal flows are accompanied by dissipation of energy and These, depending on the geometric shape of the components, the type and form flow fluid. The inevitable losses aggregated to energy transfer between the units make up the system is undesirable from the standpoint of efficiency systematically, in which forces affecting the flow are due to body forces such as gravity, the inertia, friction, surface tension, magnetic and electric fields. Knowledge of characteristics of fluid flow is of fundamental importance for information overall behavior of the system and above all to establish and control functions defined and low energy loss [3].

IV.

Determining the Flow

The flow can be characterized in an orderly fashion, moving in parallel plates, called laminar flow, or disorderly manner, resulting in random fluctuations and macroscopic speed calling turbulent flow [4]. The Reynolds number, ℜ e represents the relative influence of forces inertial and viscous forces [3], and dimensionless parameter characterized the flow, defined as: v ⋅ di ℜe = (1)

Fig. 2. Detailed diagram of an atomizer

4- Pressing: aims to give shape and dimension to the material, it is that first formats the ceramic piece emerge. 5- Drying: this is where the increase in mechanical strength and moisture removal parts so that they can be decorated and burned without the occurrence of cracks due to the rapid extraction of moisture during the process of burning or surface defects. 6- Enameling: a process which parts receive inputs for surface decoration. 7- Burns: It is undoubtedly one of the most important stages of the process of obtaining ceramic tile, because it is of fundamental changes that occur physico / chemical components and inputs decoration

ν

where: v [m/s]: the mean flow velocity di [m]: inside diameter of pipe υ [m²/s]: kinematic viscosity of the fluid Thus, for flows where ℜe is less than 2.300, called runoff laminar flow with to and greater than 4.000 ℜe, the flow is turbulent. Reynolds ranges of between 2.300 and 4000 there characterization system, since the flow in this range is characterized by being highly unstable [4].



960


V.

Loss of Load Pipeline

VII.

The pressure drop is the pressure drop associated with loss of energy due to friction between the fluid and the duct walls, which can be calculated with the aid of Darcy-Weisbach [4]: ∆p = ∫

lv 2 ⋅ ρ ar 2 ⋅ di

[ Pa ]

Many Brazilian companies confidentially store the consumption data and thermal energy production, measured in daily enjoy the valuable equipment without information that may be provided with the preparation of derivatives of these calculations, making them mostly just another routine action plant [1]. Thermal energy is one of the basic inputs used in the manufacture of coatings ceramic and also a major contributor to the high cost of production. as Figure 5, which shows the energy matrix of a company in the southern region of Santa Catarina consumer gas and make it will work, it can be stated that about 40% of the fuel refer to use in atomisers, whereas 15% and 45% are directed to the use of dryers and ovens respectively. These slices may undergo minor changes dependent on the type of coating, it is intended to Wall and / or floor.

(2)

where: ∫ [---]: coefficient of friction l [m]: length of straight duct v [m / s]: the mean flow velocity di [m]: inside diameter of pipe ρar [kg / m³]: density of air The friction factor ∫, which is determined experimentally, is a function of ℜe and ε roughness, expressed as the relative roughness ε/d, which is determined by Moody diagram [3].

VI.

Energy Industry Analysis Ceramic Wet

Flow Measurement - Pitot-Prandtl Tube

Pitot-Prandtl is a tool for measuring velocity of a fluid at a given gas flow. The operating principle is based on the measurement of two opposed pressures, total pressure and static pressure. In Fig. 4, can be observed that the inner tube of smaller diameter measures the total pressure of the flow, since that the section of this feeling must always be in parallel to the flow, although, the tube larger diameter of the external measures the static pressure, located perpendicularly to the flow [7]. The pressure difference between the extremities as measured with aid of a differential pressure gauge, resulting in pressure flow dynamics, where the average flow velocity is given by: v=

2 ⋅ Pd

ρ ar

[ m/s]

Fig. 5. Distribution of heat energy consumption for process step

In the analysis of the manufacturing process of ceramic tiles, the oven is thermal equipment which is the highest consumption. In traditional models, up to 23% of energy introduced into the kiln is lost through the stack gas emissions and other 55% transported with air cooling [5]. Noting the information of energy distribution of a roller oven perceives that 14-23% of the total energy used in this step is dispersed to the atmosphere, ie from 6.3 to 10.4% of the total energy consumed in the manufacturing process, from the atomiser to the burning. In return, the atomizer and utilizing atmospheric air to promote from combustion of natural gas, employs air at room temperature for diluting gases generated by the hot combustion, resulting in a large volume of air at 600 ° C, this temperature dependent on the process conditions. Dilution enables forming a mass of hot air for drying the slip to occur by convection, as quickly as possible, avoiding defects that may occur in the dust atomized.

(3)

where: Pd [Pa]: pressure fluid dynamics ρar [kg / m³]: density of air

VIII. E Methodology and Design The energy generated by natural gas combustor is directed to the atomizer formation of a hot air mass homogenized formed by the combustion gases elevated temperatures and atmospheric air, as outlined above.

Fig. 4. Prandtl-Pitot tube



961


The energy required to promote the temperature rise of cold air used as the dilution air temperature around 600 ° C is undoubtedly a major factor in the high consumption of Fuel this device, ie the greater the mass of the dilution air which enters the spray dryer, the greater the energy required from the combustion of natural gas. Therefore, knowing that all dry and hot combustion gases of the furnace is directed to the atmosphere without prior use, the objective is to reuse all this heat and direct you to the dilution air into the atomizer, gaining some energy and thus reducing the fuel consumption of the equipment. A simple sketch is schematically shown in Fig. 6, where a pipe properly sized, insulated and made of carbon steel, is connected to the chimney combustion gases of two furnaces, directing the flow of air to the atomizer through a pressurizing centrifugal fan of the burner installed below the atomiser. The existence of a stack between the air inlet and outlet of the burner fan prevents that during the stop of the spray to cool the pipe and may cause hazards training acids due to condensation of the vapors of combustion by a sudden drop of temperature.

higher thickness, the investment would be high and the heat loss would not be so expressive. The fan specified to ensure the pressure loss in the stretch pipe between the furnace and atomizer to carry the larger mass of hot air available on the furnace stack has the following characteristics: • Flow [m³ / h]: 55,000 • Total Pressure [mmCA]: 202 • Power [kWh]: 37 Saving data, shown in Table I, are designed based reducing fuel consumption and the increase of power consumption caused the new engine to be installed to drive the centrifugal fan, providing 16.1% reduction of manufacturing costs. The logical explanation why the electric power consumed is less than the installed due to the fact that the fan is designed to work with air 25 ° C as the temperature of arrival of the air is high, the density decreases and the same therefore the power consumed, which is proportional also decreases. TABLE I ECONOMIC FEASIBILITY

The Fig. 7 is an illustration of a constructive way to the pipe lead the hot air furnace to the atomizer.

Fig. 6. Schematic of the layout of the recovery of the hot air furnace atomizer

IX.

Withdrawals of Technical Feasibility and Economic Data

Data collection for the beginning of the project part of the knowledge of equipment under study, as well as features construction techniques, use of fuel productivity, streams of hot air, the science of work routines and settings. In turn, the design involving pipes, specification of the thermal insulation selection of centrifugal fan and auxiliary systems are carried out taking into account reads the data and raised in their own equipment. All settings accessories and materials were based on the principle of more cost-effective. In summary, the mixture of hot air ovens of the two results in flow characteristics mass of 21,286 kg / h and a temperature of 241.3 ° C, which when directed through atomizer Pipe previously isolated with ceramic fiber blanket 50 mm thickness causes a drop in temperature of 19.2 ° C. The dimensions of the insulation is based on the cost front better heat loss x investment, which thickness the adopted lower heat loss and to be raised

Fig. 7. Representation of the pipe carrying hot air pre-insulated

X.

Logic Operation

During the initiation of the production cycle of the sprayer, ie, until the temperature of the hot air chamber reaches the temperature of work flow and establish a continuous production, the valve "A" will be 100% open directing the flow of hot air to the atmosphere and the valve "B" remains 100% closed preventing the entry of



962


hot air the atomiser. After reaching the working temperature, a gauntlet is thrown to the reversing valves, the valve fully opening "B" and closing the valve "A" in order to direct all flow to the atomizer as dilution air. If there is need for the atomizer off, the valves by means of the handle are displaced to the starting position start of cycle, such action maintains the transport tubing hot air from the furnace to atomizer, minimizing the possibility of degrading the probability of pipe condensation of water along with some elements present in the composition of gas combustion acids which can form under the temperature drop.

XI.

burner responsible for conducting the combustion of natural gas being working at full capacity, ie, be providing all the energy to perform which was drying in the project to generate. However, by introducing hot air ovens, the amount of energy generated for the same production is lower, a fact proven by specific consumption decrease, so the combustor ceased to operate at full capacity for a moment, allowing the increase of production and thus returning to full load condition. The increase in production enabled the company to ease the care of type of pasta for specific atomizers, namely to promote the continuous streams of Working atomizers, thus avoiding waste with setups. Meanwhile the return Financial recovery is provided by the house of U$ 116.999 monthly value significant that throughout the year is about U$ 1.403.988 less in the composition of basket cost and more as capital for future investments. The project photos can be seen in Figs. 9.

Analysis of Results

The Fig. 8 shows the evolution of energy indicator rate to the atomizer Over the period of start-up project to the full functioning of the system heat recovery, which can be seen that from the day 10, the indicators tends to improve, confirming the objectivity of the project.

XIII. Conclusion Power management is indeed an area of great concern to management and control by the high representative of the production costs of an industrial unit. the controls fuel consumption and productivity of the equipment should fail to routines are only manufacturing and become a source of valuable information for control energy efficiency, productivity and statements to make decisions. Alternatives for reducing energy consumption are very concerned and sought by companies in the ceramic sector. The use and recycling of hot air ovens available in the equipment itself or in dried and atomizers are gaining dedicated space in the development of new projects and industrial lay-outs. The heat of combustion and chemical reactions during firing of pieces which until then was exhausted to atmosphere, is now directed to the atomizer, representing savings significant fuel. The savings from the recovery arrives home 19% of all natural gas consumption of the equipment and/or about 3% of all thermal energy consumption of the industrial park, as seen by the outcome of the project. The return on investment of these projects is right and in most cases time is less than 06 months, which like many companies to adopt as a priority the development of these works. Unfortunately, many of these projects still end up being left behind by the high cost of deployment, and even having a guaranteed payback and quick stop be encouraged by the competent bodies and by their own plants. Therefore, it is wide importance, the search for new methods and materials that provide a reduction in costs with the implementation of similar projects because in addition to these economic energy, the company is gaining respect for the environment by reducing greenhouse gas pollutants and fuel distribution utilities with the greater availability of fuel possible growth of other industries.

Fig. 8. Indicator energy [kcal/lH2O evaporated]

Viewing the data in a more macro and performing new measurements after operating at full system, in Table II can compare planned results x actual results of the specific average consumption atomizer throughout the month of October, and as productivity and economy. All data were the same method records data and passed by corrections standards for calculating the volume of fuel consumed. TABLE II COMPARISON OF PREDICTED RESULTS X REAL

In comparative results may be noted that in addition to the improvement in energy performance in 18.8% over the period to which heat recovery is not was being used, there was a simultaneous increase in productivity equipment around 16% capacity which was being worked on. This increase is justified in the fact that the Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved


963


(a) (c)

(d)

(b)

(e)

Figs. 9. (a) Out of the Oven; (b) Atomizer and Centrifugal Fan; (c) Out of the Oven and Interconnection; (d) Air transport tubing Isolated; (e) Air conveying pipeline [9]

Nassetti, G., et al. Piastrelle Ceramiche & Energia. Centro Cerâmico. Bologna, Itália, 1998. [10] Bazzo, E. Geração de Vapor. 2 ed. - Florianópolis: Ed. da UFSC. 216 p. 1995. [11] Allouis, C., Beretta, F., Amoresano, A., Experimental study of lean premixed prevaporized combustion fluctuations in a gas turbine burner, (2008) Combustion Science and Technology, 180 (5), pp. 900-909.

References [1]

[2]

[3] [4]

[5] [6] [7] [8]

Cassani, F. Recuperando Energia dos Fornos, Secadores e Atomizadores. Revista Cerâmica Industrial – Volume 14 – n. 3 – Maio/Junho 2009. Boschi, A. O., Alves, H. J., Meichiades, F. G., Brito, H. B. Análise do Consumo de Energia Térmica no Setor Brasileiro de Revestimentos Cerâmicos. Revista Cerâmica Industrial – Volume 15 – n. 4 – Julho/Agosto 2010. Clezar, C. A., Nogueira, A. C. R.. Ventilação Industrial. 2 ed. rev. - Florianópolis: Ed. da UFSC. 240 p. 2009. Enrique, J. E., et al. Evolución de los consumos de energía térmica y eléctrica en el sector de baldosas cerámicas. Técnica cerámica, n. 246, España, 2006. Kreith, F. Princípios da Transmissão de Calor. 3 ed. Editora Edgard Blucher. 550p. 1977. Linsingen, I. V. Fundamentos de Sistemas Hidráulicos. 3 ed. rev. - Florianópolis: Ed. da UFSC. 399 p. 2008. Modesto, C. O., Júnior, J. C. B. Material Cerâmico. Colégio Maximiliano Gaidzinski. 227p.2001. Nassetti, G. Como Melhorar a Eficiência Energética na Indústria de RevestimentosCerâmicos. Revista Cerâmica Industrial – Volume 15 – n. 1 – Janeiro/Fevereiro 2010.

Authors’ information 1

UFRGS, Federal University of Porto Alegre, Brazil. E-mail: [email protected] 2 IMG, Institute Maximiliano Gaidzinski Cocal do Sul, Brazil. E-mail: [email protected] 3 UFRGS, Federal University of Porto Alegre, Brazil. E-mail: [email protected] 4 SATC Faculty Criciúma, Brazil. E-mail: [email protected]



964


5 UFSC, Federal University of Santa Catarina, Brazil. E-mail: [email protected]

João Bosco da Mota Alves - Graduated in Electrical Engineering from Universidade Federal do Pará (1971), Masters in Electrical Engineering from Universidade Federal de Santa Catarina (1973) and Ph.D. in Electrical Engineering from Universidade Federal do Rio de Janeiro (1981). He was professor at the Federal University of Pará from 1973 to 1989, and Professor from 1986 to 1989. He was Professor at the Federal University of Uberlandia from 1989 to 1985. Professor, Federal University of Santa Catarina from 1996 until August 2008 when he retired. Currently serves on the Graduate Program in Engineering and Knowledge Management, Federal University of Santa Catarina. He has experience in Computer Science with an emphasis on Intelligent Robots, acting on the following topics: Remote Experimentation, Remote Systems, Distance Education, Accessibility, Computers in Education, General Systems Theory, Interdisciplinarity and Systemic Vision in Organizations.

6 SATC Faculty Criciúma, Brazil. E-mail: [email protected] 7 IMG, Institute Maximiliano Gaidzinski Cocal do Sul, Brazil. E-mail: [email protected]

Vilson Menegon Bristot - PhD in the Engineering of Mines Metallurgy and Materials, Federal University of Rio Grande do Sul, master's at Mechanical Engineering from Federal University of Rio Grande do Sul (2008) and Bachelor's at Engenharia Agrimensura from University do Extremo Sul Catarinense (2003),. He is currently professor / researcher at the Faculty SATC, a professor/ researcher at the University Barriga Verde (UNIBAVE), a professor/ researcher at the University do Extremo Sul Catarinense (UNESC) and professor at the Institute Maximiliano Gaidzinski.

João Luís Brunel Mangili – Graduated in Mechanical Engineering from Faculty SATC (2012). He has experience in energy efficiency and energy saving system at ceramics industry.

Vilmar Menegon Bristot - Bachelor's at Engineering Agrimensura from University do Extremo Sul Catarinense (1993), master's at Electric Engineering from Federal University of Santa Catarina (2002) and PhD student in the Engineering of Mines Metallurgy and Materials, Federal University of Rio Grande do Sul. He is currently professor /researcher at the Faculty SATC, a professor/ researcher at the University do Extremo Sul Catarinense and director / professor at the Institute Maximiliano Gaidzinski.

Reginaldo Tassi – Bachelor´s at computer science from University do Extremo Sul Catarinense(2002), master´s at Engineering of Mines Metallurgy and Materials, Federal University of Rio Grande do Sul,(2010). He is currently professor / researcher at the University Barriga Verde (UNIBAVE) and Professor of electromechanical technician at the Institute

Lirio Schaeffer - Ph.D. in Mechanical Forming. Rheinisch-Westfalischen Technischen Hochschule/Aachen, R.W.T.H.A., Germany. Professional performance: Coordination of Improvement of Higher Education Personnel, CAPES, Brazil. 2003 - Present – Relationship: Employee Department of Metallurgy, UFRGS, Brazil.1974 - Present - Public Servants, Functional Placement: Professor, Exclusive Dedication.

Maximilian Gaidzinski

Vilson Gruber - Graduated in Data Processing from the University Santana of São Paulo (1996), specialization in Business Management in Telecommunications from the School of Business Paulista - São Paulo (2000), specialization in Educational Psychology from the University Castelo Branco, Rio de Janeiro (2006) specialization in Project Management at Faculty SATC, Criciúma (2010), Ph.D in Engineering of Mines Metallurgy and Materials, Federal University of Rio Grande do Sul - Porto Alegre (2007). He is currently a professor / researcher at the Faculty SATC, develops research projects and coordinates undergraduate courses in Telecommunications Systems and Postgraduate Courses. He has experience in Telecommunication Systems, Computer Networks acting on the following themes: Networks and Mobile Phones, Projects, Embedded Systems, Digital Systems, Remote Experimentation, Accessibility and Technology, Computer Systems and Digital Inclusion.



965


Flow of a Second Order/Grade Fluid Induced by a Pull of Disks in an Orthogonal Rheometer Under the Effect of a Magnetic Field H. Volkan Ersoy

Abstract – The flow of a second order/grade fluid produced by a pull with constant velocities of disks in an orthogonal rheometer with the application of a magnetic field is studied. It is investigated the effect of the magnetic field exerted perpendicularly to the disks on the flow for both positive and negative values of the material modulus α1 . It is shown that the flow in the presence of magnetic field has thinner boundary layers than that in the absence of magnetic field. It is found that the y- component of the dimensionless force per unit area exerted by the fluid is greater than the x- component for all values of the Hartmann number M. Increasing the magnetic field causes an increase in the x- component but a decrease in the y- component for the elasticity of fluid with positive and negative sign. It is obtained that the x- component corresponding to α1 > 0 is larger than that corresponding to α1 < 0 . An opposite effect is seen for the ycomponent. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved. Keywords: Magnetic Field, Orthogonal Rheometer, Pull, Second Order/Grade Fluid

ρ σ

Nomenclature A1 , A2 B B0

D / Dt E h I J M p R T Txy , Txz

First and second Rivlin-Ericksen tensors Magnetic induction Applied magnetic induction Material time derivative operator Electric field Half the vertical separation of the disks Unit tensor Current density Half the eccentricity distance Hartmann number Pressure Reynolds number Cauchy stress tensor Stress tensor components

Ω

ζ

I.

v Vx , V y

x, y, z z ′ , z ′′ α1 , α 2 β

µ µm

Introduction

Magnetohydrodynamics (MHD) is the study of electrically conducting fluids in the presence of magnetic and electric fields and examines the phenomena associated with electro-fluid-mechanical energy conversion. Its subject consists of five classes of the engineering sciences, namely, fluid mechanics, thermodynamics, mechanics, materials, and the electrical sciences. The MHD applications deal with astronomy, geophysics and engineering. Its applications in engineering are MHD generators, pumps, bearings flow meters, plasma jets, fusion machines for power, electric arc heaters and space power generators [1]-[4]. The MHD flows of non-Newtonian fluids have attracted the attention of engineers, physicists and mathematicians. It has been a subject of great interest due to its applications in cooling systems with liquid metals, purification of crude oil and polymer technology. In recent years, it has generally been recognized that in industry non-Newtonian fluids are more appropriate than Newtonian fluids. That non-Newtonian fluids are finding increasing application has given impetus to many researchers. The governing equations for non-Newtonian fluids such as polymer solutions, greases, melts, muds,

Txy , Txz Dimensionless stress tensor components

u, v, w

Density of the fluid Electrical conductivity of the fluid Common angular velocity of the disks Non-dimensional vertical distance

Velocity components in Cartesian coordinate system Velocity vector Dimensionless velocities in the x- and ydirections Cartesian coordinates Rotation axes of the upper and bottom disks Material moduli of the fluid Elastic parameter of the fluid Dynamic viscosity Magnetic permeability



966

H. V. Ersoy

emulsions, paints, blood, jams, soaps, shampoo and certain oils are much more complicated than the NavierStokes equations. Non-Newtonian fluids have a nonlinear relationship between the stress and the rate of strain at a point and thus the resulting differential systems are highly nonlinear. It is difficult to suggest a single model which exhibits all properties of non-Newtonian fluids. In order to predict various flow characteristics of these fluids, many fluid models have been proposed as evidenced in the literature. The second grade model has received special attention amongst the many models which have been used to describe the non-Newtonian behavior. Although the second grade model for steady flows is used to predict the normal stress differences, this model is not capable of describing the shear thinning and thickening phenomena. Second grade fluids are dilute polymeric solutions (e.g., poly-iso-butylene, methyl-methacrylate in n-butyl acetate, polyethylene oxide in water) [5]. While there are a large number of solutions that form the basis of several viscometers and rheometers, only a few of them are actually dynamically possible. One such motion is the flow that occurs in an orthogonal rheometer consisting of two parallel disks rotating with the same angular velocity about non-coicident axes. This instrument was originally developed by Maxwell and Chartoff [6]. In this domain, Abbott and Walters [7] obtained an exact solution for the flow of a Newtonian fluid. In the case of a viscoelastic fluid, they employed a perturbation analysis and assumed that the offset between the axes is small. Later, Berker [8] showed that there is the existence of an infinite number of non-trivial solutions to the Navier-Stokes equations in the orthogonal rheometer. This motion falls under the category of “pseudo-plane motions”. A single solution requires a symmetric flow for both coaxial and noncoaxial rotations. For the coaxial case, the solution becomes the rigid body motion. Rajagopal and Gupta [9] both established an exact solution for the flow of a second grade fluid between eccentric rotating disks and studied the stability of this flow. Rajagopal [10] also studied the problem for a second order fluid whose normal stress moduli do not obey the relations α 1 > 0 and α1 + α 2 = 0 . Rajagopal [11] showed that the motion represented by Berker [12], who considered the solutions that are not axially symmetric when the disks rotate about a common axis, is one with constant stretch history. This result was also established, independently, by Goddard [13]. The flow induced by applying a magnetic field in an orthogonal rheometer has also attracted the interest of many investigators. Mohanty [14] found an exact solution for the MHD flow of a Newtonian fluid. Rao and Rao [15] examined the MHD flow for a second grade fluid. Kasiviswanathan and Gandhi [16] studied the MHD flow of a micropolar fluid. Ersoy [17] investigated the flow of an Oldroyd-B fluid in the presence of a magnetic field. We refer the reader to the

paper by Ersoy [18] for a detailed list of references related to the flows in an orthogonal rheometer. Ersoy [19] was the first to consider the flow induced by a pull of eccentric rotating disks. He investigated the unsteady flow of a Newtonian fluid and obtained an exact solution. Asghar et al. [20] studied the effects of Hall current and heat transfer for an Oldroyd-B fluid in the steady flow. This flow was extended to a Burgers’ fluid by Siddiqui et al. [21] and to a generalized Burgers’ fluid by Hayat et al. [22]. In this paper, our main purpose is to extend the paper [23] to the case of magnetohydrodynamic flow. The variations of the dimensionless components of the translational velocity and the x-, y- components of the dimensionless force per unit area exerted by the fluid for both a second order fluid and a second grade fluid are analyzed when a uniform magnetic field is applied to the disks that is pulled with constant velocities in an orthogonal rheometer. It is shown that the solution for [23] can be recovered by taking M = 0 .

II.

Basic Equations and Solution

The flow field of the problem is bounded by two insulated disks located at z = h and z = − h . The top disk rotates about the z ′ -axis and the bottom disk rotates about the z ′′ -axis with the same angular velocity Ω . The distance between the axes of rotation is shown in the y- direction by 2 . The region between the disks is occupied by an incompressible second order/grade fluid. The top and bottom disks are pulled with the constant velocities U and −U , respectively. The velocity U has two components, U x in the x-direction and U y in the ydirection. A uniform magnetic field of strength B0 is applied in the z- direction. We assume that the magnetic Reynolds number for the flow is very small so that the induced magnetic field is negligible in comparison with the applied magnetic field. The geometry of the problem is shown in Fig. 1. z z=h

z′ Ω

z ′′ Ω

y z = −h

Fig. 1. Flow geometry

The Cauchy stress T in an incompressible and homogeneous second order/grade fluid is given by Rivlin and Ericksen [24]: T = − pI + µ A1 + α1 A2 + α 2 A12


(1)


967

H. V. Ersoy

where p is the pressure, µ the dynamic viscosity of the fluid, α1 and α 2 the material module which are usually referred to as the normal stress coefficients. In the above representation, I is the identity tensor, and the kinematical tensors A1 and A2 are defined through: A1 = ( grad v ) + ( grad v )

T

A2 =

where ρ is the density of the fluid, J the current density, B the magnetic induction, µm the magnetic permeability, E the electric field, and σ is the electrical conductivity of the fluid. The boundary conditions can be written as: u = −Ω ( y −

(2)

) + U x , v = Ωx + U y ,

(12)

w = 0 , at z = h

DA1 T + A1 ( grad v ) + ( grad v ) A1 Dt

(3)

u = −Ωy , v = Ωx, w = 0 , at z = 0 u = −Ω ( y +

) − U x , v = Ωx − U y , w = 0,at

(13)

z = − h (14)

where v is the velocity vector and D / Dt the material time derivative. We notice that if α1 = α 2 = 0 , the model (1) reduces to the classical linearly viscous fluid model [28]. The thermodynamical principles impose some restrictions on α1 and α 2 [25]. In particular, the Clasius-Duhem inequality implies that:

where u , v , w denote the x -, y -, z - components of the velocity, respectively. We seek a solution of the form:

µ ≥ 0 α1 + α 2 = 0

which satisfy Eq. (7). The appropriate boundary conditions for f ( z ) and

u = −Ωy + f ( z ) , v = Ωx + g ( z ) , w = 0

(4)

g ( z ) from Eqs. (12)-(15) are:

and the requirement that the specific Helmholtz free energy be a minimum in equilibrium implies that:

α1 ≥ 0

f (h) = Ω + U x , g (h) = U y

(5)

f (0) = 0 , g (0) = 0

Dv = ∇ ⋅T + J × B Dt

From Eqs. (1)-(3), (6) and (15), we have:

(7)

∇⋅B = 0

(8)

∂p = ρ Ω ( Ωx + g ) + µ f ′′ + α1Ωg ′′ + J y B0 ∂x

(17)

∂p = − ρ Ω ( −Ωy + f ) + µ g ′′ − α1Ωf ′′ − J x B0 ∂y

(18)

∂p = 2 ( 2α1 + α 2 )( f ′f ′′ + g ′g ′′ ) ∂z

(19)

where a prime denotes differentiation with respect to z . Using Eq. (11), we obtain:

(6)

∇⋅v = 0

(16)

f ( − h ) = −Ω − U x , g ( − h ) = −U y

The fluids characterized by above restrictions are called the second grade fluids in the literature. On the other hand, Eq. (1) is called a second order fluid model ( α1 < 0 and α1 + α 2 ≠ 0 ), which is in good agreement with experimental results, if it is not required to be compatible with thermodynamics [26]. Therefore, it is clear that the results established for the case α1 > 0 have more value than the solution α1 < 0 . In this study, we consider both positive and negative values of α1 . For example, an examination of a problem depending on the sign of α1 can be found in the papers by Ersoy [23], [27]. The governing equations are:

ρ

(15)

J x = σ ( Ex + vB0 ) ,

(

)

J y = σ E y − uB0 , J z = σ Ez

(20)

∇ × B = µm J

(9)

Bearing in mind that the disks are insulated and using Eqs. (9)-(10), (17)-(20), we have:

∇× E = 0

(10)

ρΩg + µ f ′′ + α1Ωg ′′ − σ B02 f = constant

(21)

J = σ ( E + v × B)

(11)

− ρΩf + µ g ′′ − α1Ωf ′′ − σ B02 g = constant

(22)



968

H. V. Ersoy

Defining F ( z ) = f + ig ( i = −1 ), Eqs. (21) and

1

R = 10 β = 0.2

(22) can clearly be combined as: ⎛ σ B 2 + i ρΩ ⎞ F ′′ − ⎜ 0 F = constant ⎜ µ − iα Ω ⎟⎟ 1 ⎝ ⎠

0.5

M = 0, 1, 2

Vx = 1

(23)

Vy =1

ζ 0

with the conditions: -0.5

F ( h ) = ( Ω + U x ) + iU y F ( 0) = 0

(24)

-1

F ( − h ) = − ( Ω + U x ) − iU y

-1

The solution in non-dimensional form of Eq. (23) under the conditions (24) is:

f g +i = ⎡⎣(1 + Vx ) + iVy ⎤⎦ Ω Ω

sinh sinh

M 2 + iR ζ 1 − iβ M 2 + iR 1 − iβ

1

R = 10

0.5

ζ 0 -0.5

-1 -2

(26)

-1

0

f /Ω

1

2

versus ζ for M=0, 1, 2

Fig. 3(a). Variations of f / Ω

( R = 10 , β = −0.2 , Vx = 1 , Vy = 1 )

1

are

R = 10

β = − 0 .2 Vx = 1

M = 0, 1, 2

Vy =1

M = 0,1, 2

ζ 0

Vx = 1

0.5

M = 0,1, 2

Vx = 1

0.5 R = 10 β = 0.2


Vy =1

The effects of parameters on f / Ω and g / Ω displayed graphically in Figs. 2(a)-3(b). 1

1

β = −0.2

(25)

2

Uy U z Vx = x , V y = ,ζ = h Ω Ω

0.5

( R = 10 , β = 0.2 , Vx = 1 , Vy = 1 )

dimensionless velocities in the x- and y- directions, respectively, and ζ the non-dimensional disk gap, all of which are defined below:

αΩ σ ρΩh , β= 1 B0 h , R = µ µ µ

0

g /Ω

Fig. 2(b). Variations of g / Ω

where M is the Hartmann number, R the Reynolds number, β the elastic parameter, Vx and V y the

M =

-0.5

Vy =1

-0.5

ζ 0

-1

-0.5

-1

-1 -2

-1

0

f /Ω

Fig. 2(a). Variations of f / Ω

1

-0.5

Fig. 3(b). Variations of

2

g/Ω

0

g /Ω

0.5

1

versus ζ for M=0, 1, 2 ( R = 10 ,

β = −0.2 , Vx = 1 , V y = 1 )


From Eqs. (1)-(3), we find:

( R = 10 , β = 0.2 , Vx = 1 , Vy = 1 )

Txz = µ f ′ + α1Ωg ′, Tyz = µ g ′ − α1Ωf ′


(27)


969

H. V. Ersoy

Since the stress tensor components Txz and Tyz do

III. Conclusion

not depend on x and y , they correspond to the x - and y - components of the force per unit area exerted by the fluid on the disks, respectively. The dimensionless stress tensor components on the disks are:

(Txz + iTyz )ζ =∓1 =

)

and y = f ( z ) / Ω as the z =constant plane shifts from

M2 +βR +

(

+i R − β M

2

M 2 + iR 1 − iβ

tanh

where Txz =

(

⎡⎣(1 + Vx ) + iVy ⎤⎦

The flow in an orthogonal rheometer is a motion wherein streamlines in any z =constant plane are concentric circles. The locus of the centers of these circles is a curve in space described by x = − g (z ) / Ω z = − h and z = h . In other words, the fluid layer in each z =constant plane rotates as if a rigid body about the point in which the plane and the curve intersect. The velocities f / Ω and g / Ω correspond to the dimensionless translational velocities in the x- and ydirections, respectively. When the disks are pulled, we assume that these circles are the curves defined in the same manner but these are affected by pulling the disks. In this study, the flow induced by a pull of disks in an orthogonal rheometer in the presence of a uniform applied magnetic field is examined. Magnetohydrodynamic phenomena result from the mutual effect of a magnetic field and a second order/grade fluid. An electromagnetic force is produced in the fluid when the magnetic field is applied. This force that is also called the Lorentz force has the tendency to slow down the flow. Our main aim is to examine the influence of the magnetic field for both positive and negative values of the material modulus α1 of the fluid. Figs. 2(a)-3(b) reveal that increasing the Hartmann number M has the effect of decreasing both f / Ω and g / Ω for positive and negative values of β . The dimensionless

) (28)

Tyz Txz , Tyz = . Figs. 4-5 show the µΩ / h µΩ / h

variations of (Txz )

ζ =∓1

( )ζ =∓1 .

and Tyz

7 6.5 (T yz ) ζ = ∓1

6

R = 10 β = 0.2 Vx = 1

5.5 5 4.5

Vy =1

4 3.5

(Txz ) ζ = ∓1

3 2.5 0

0.5

1

1.5

components of the translational velocity, f / Ω and g / Ω , have smaller values for β < 0 when compared with those for β > 0 . It is observed that the boundary layer thickness decreases with the increase of the Hartmann number. Figs. 4-5 illustrate the effect of the Hartmann number on the x- and y- components of the dimensionless force per unit area exerted by both a second order fluid ( β < 0 ) and a second grade fluid ( β > 0 ). This figures disclose that the y- component is greater than the xcomponent for all value of the Hartmann number. The increase of the Hartmann number produces an increase in the x- component but a decrease in the y- component for both a second grade fluid and a second order fluid. The x- component for a second grade fluid is larger than that for a second order fluid. On the other hand, an opposite behavior is observed for the y- component.

2

M

( )ζ

Fig. 4. Variations of ( Txz )ζ = ∓1 and Tyz

=∓1

versus M

( R = 10 , β = 0.2 , Vx = 1 , Vy = 1 )

7 (T yz ) ζ = ∓1

6

R = 10

5

β = −0.2

4

Vx = 1 Vy =1

3

(T xz ) ζ = ∓1

2 1 0

0.5

1

1.5

References

2

M

[1]

( )ζ

Fig. 5. Variations of ( Txz )ζ = ∓1 and Tyz

= ∓1

versus

[2]

M

( R = 10 , β = −0.2 , Vx = 1 , Vy = 1 )

[3]


G.W. Sutton, A. Sherman, Engineering Magnetohydrodynamics, (Dover Publications, 2006). A. B. Çambel, Plasma Physics and Magnetofluid-Mechanics, (McGraw Hill, 1963). K. R. Cramer, S. Pai, Magnetofluid Dynamics for Engineers and Applied Physicits, (Scripta Publishing Company, 1973).


970

H. V. Ersoy

[4] [5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22]

[23]

[24]

[25]

W. F. Hughes, F. J. Young, The Electromagnetodynamics of Fuids, (John Wiley &Sons, 1966). S. Asghar, S. Nadeem, K. Hanif, T. Hayat, Analytic Solution of Stokes Second Problem for Second-Grade Fluid, Mathematical Problems in Engineering, Vol. 2006, pp. 1-8, 2006 B. Maxwell, R. P. Chartoff, Studies of a Polymer Melt in an Orthogonal Rheometer, Transactions of the Society of Rheology, Vol. 9, pp. 41-52, 1965. T. N. G. Abbott, K. Walters, Rheometrical Flow Systems: Part2. Theory for the Orthogonal Rheometer, Including an Exact Solution of the Navier-Stokes Equations, Journal of Fluid Mechanics,Vol. 40, pp. 205-213, 1970. R. Berker, An Exact Solution of the Navier-Stokes Equation: The Vortex with Curvilinear Axis, International Journal of Engineering Science, Vol. 20, pp. 217-230, 1982. K. R. Rajagopal, A. S. Gupta, Flow and Stability of a Second Grade Fluid between Two Parallel Plates Rotating about Noncoincident Axes, International Journal of Engineering Science, Vol. 19, pp. 1401-1409, 1981. K. R. Rajagopal, The Flow of a Second Order Fluid between Rotating Parallel Plates. Journal of Non-Newtonian Fluid Mechanics, Vol. 9, pp. 185-190, 1981. K. R. Rajagopal, On the Flow of a Simple Fluid in an Orthogonal Rheometer, Archive for Rational Mechanics and Analysis, Vol. 79, pp. 39-47, 1982. R. Berker, A New Solution of the Navier-Stokes Equation for the Motion of a Fluid Contained between Two Parallel Plates Rotating about the Same Axis, Archiwum Mechaniki Stosowanej, Vol. 31, pp. 265-280, 1979. J. D. Goddard, The Dynamics of Simple Fluids in Steady Circular Shear, Quarterly of Applied Mathematics, Vol. 31, pp. 107-118, 1983. H. K. Mohanty, Hydromagnetic Flow between Two Rotating Disks with Noncoincident Parallel Axes of Rotation, Physics of Fluids., Vol. 15, pp. 1456-1458, 1972. R. Rao and P. R. Rao, MHD Flow of a Second Grade Fluid in an Orthogonal Rheometer, International Journal of Engineering Science, Vol. 23, pp. 1387-1395, 1985. S. R. Kasiviswanathan, M. V. Gandhi, A Class of Exact Solutions for the Magnetohydrodynamic Flow of a Micropolar Fluid, International Journal of Engineering Science, Vol. 30, pp. 409417, 1992. H. V. Ersoy, MHD Flow of an Oldroyd-B Fluid between Eccentric Rotating Disks, International Journal of Engineering Science, Vol. 37, pp. 1973-1984, 1999. H. V. Ersoy, On the Locus of Stagnation Points for a Maxwell Fluid in an Orthogonal Rheometer, International Review of Mechanical Engineering (IREME), Vol. 3, pp. 660-664, 2009. H. V. Ersoy, Unsteady flow due to a sudden pull of eccentric rotating disks, International Journal of Engineering Science, Vol. 39, pp. 343-354, 2001. S. Asghar, M. R. Mohyuddin, T. Hayat, Effects of hall current and heat transfer on flow due to a pull of eccentric rotating disks, International Journal of Heat and Mass Transfer, Vol. 48, pp. 599-607, 2005. A. M. Siddiqui, M. A. Rana, N. Ahmed, Effects of hall current and heat transfer on MHD flow of a Burgers’ fluid due to a pull of eccentric rotating disks, Communications in Nonlinear Science and Numerical Simulation, Vol. 13, pp. 1554-1570, 2008. T. Hayat, K. Maqbool, M. Khan, Hall and heat transfer effects on the steady flow of a generalized Burgers’ fluid induced by a sudden pull of eccentric rotating disks, Nonlinear Dynamics, Vol. 51, pp. 267-276, 2008. H. V. Ersoy, Flow of a Second Order/Grade Fluid Induced by a Pull of Disks in an Orthogonal Rheometer, International Review of Chemical Engineering (IRECHE), Vol. 3, pp. 585-591, 2011. R. S. Rivlin, J. L. Ericksen, Stress-Deformation Relations for Isotropic Materials, Journal of Rational Mechanics and Analysis, Vol. 4, pp. 323-425, 1955. J. E. Dunn, R. L. Fosdick, Thermodynamics, Stability and Boundedness of Fluids of Complexity Two and Fluids of Second Grade, Archive for Rational Mechanics and Analysis, Vol. 56, pp. 191-252, 1974.

[26] R. L. Fosdick, K. R. Rajagopal, Anomalous Features in the Model of Second Order Fluids, Archive for Rational Mechanics and Analysis, Vol. 70, pp.145-152, 1979. [27] H. V. Ersoy, MHD Flow of a Second Order/Grade Fluid due to Non-coaxial Rotation of a Porous Disk and the Fluid at Infinity, Mathematical and Computational Applications, Vol. 15, pp. 354363, 2010. [28] Amoresano, A., Niola, V., Langella, F., Experimental analysis of the behavior of the droplets of high viscous fluids impacting on a flat heated surface, (2011) 8th WSEAS Intl. Conf. on Fluid Mechanics, FM'11, 8th WSEAS Intl. Conf. on Heat and Mass Transfer, HMT'11, 8th WSEAS Intl. Conf. on Mathematical Biology and Ecology, MABE'11, pp. 105-110.

Authors’ information Yildiz Technical University, Department of Mechanical Engineering, Yildiz 34349 İstanbul, Turkey. E-mail: [email protected] H. Volkan Ersoy (Associate Professor) was born in Tekirdağ, Turkey. He studied Mechanical Engineering and received BS, MS and PhD degrees from Istanbul Technical University. His areas of interest are mechanics of fluids, mechanics of rigid bodies and mechanics of deformable bodies.



971


Impact of Liquid Pressure Losses and Solid-Phase, in the Performance of a Three-Phase Flow Air-Lift Pump Dimitrios N. Androulakis, Apostolos N. Vlachos, Dionissios P. Margaris

Abstract – The aim of the paper is to present the results of an experimental investigation of a three-phase flow air-lift pump lab-scale installation. The existing lab-scale installation was modified, in order to create new potentials of further investigation concerning two new downcomers. In order to certify the installation, several experiments took place and the results were compared with former ones of the same conditions. Experiments were held in 28 mm and 40 mm diameter pipes with defined points of submergence ratio, gas flow rate and solid-phase mass. The impact of liquid pressure losses via a downcomer and the impact of the solid-phase are to be investigated. The measured points of three-phase flow are then presented on operational curves, concerning superficial velocities. Finally, experimental results are presented in a suitable flow regime map. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Multi-Phase Flows, Air-Lift Pumps, Superficial Velocity, Operational Curves, Regime Map

In the literature, there is a wide variety of definitions for the flow regimes; this derives partly from the subjective nature of defining and choosing common measures and properties and partly from the incapability of the flow models and the numerical formulas to precisely incorporate the local properties and specificities. Knowing and predicting the exact flow pattern is important as far as it inflicts the proper use of the corrective factors. Hewitt and Roberts [4] designated five (5) basic patterns for vertical upward flows, named as followed: 1. Bubble flow 2. Slug or plug flow 3. Churn flow 4. Annular flow, and 5. Wispy annular flow. Initially, the air-lift pump was thought to be appropriate only for simple uses such as pumping water in inhabited areas or buildings and furthermore in industries. Through time, air-lift pumps proved to be applicable in more demanding conditions including food industries (flours or nuts transportation), oil/water mining, deep sea mining (for manganese and other ores), dangerous fluids transportation (corrosive, radioactive, explosive, poisonous) and as a mean of conveying every kind of slurries in mining. More recently, it has been reported by Kamata and Ito [5] that in the steel making process, although the principle of the air-lift pump is applied only to an RH vacuum degasser to circulate molten steel and to remove hydrogen gas, carbon and unmetallic inclusions in molten steel, the simplicity of the equipment may make it applicable for the

Nomenclature D JG JL JS L P P1 QG QL QS Q G* S S/L

Diameter of air-lift pump upriser, m Gas superficial velocity, m/s Liquid superficial velocity, m/s Solid superficial velocity, m/s Height of the air-lift pump upriser from the airinjection point, m Measured gas pressure, bar Environmental pressure, bar Normal gas volume flow rate, m3/h Liquid volume flow rate, m3/h Solid volume flow rate, m3/h Measured gas volume flow rate, m3/h Height of multi-phase column in the air-lift pump upriser from the air-injection point, m Submergence ratio, the height of the multi-phase column divided by the total height of the air-lift pump upriser

I.

Introduction

Superficial velocity and void fraction are the prime requisites from the physical properties of multi-phase flows. Flow patterns and regimes are characterized using these terms. In the literature, theoretical and experimental data concerning the air-lift pump performance are given as functions of superficial velocities and void fraction. Samaras and Margaris [1], [2], [3] used these parameters for presenting their proposed flow regime maps of the air-lift pump, concerning the regime transitions. Manuscript received and revised June 2012, accepted July 2012


972

D. N. Androulakis, A. N. Vlachos, D. P. Margaris

transportation of molten iron/steel between different refining processes. Summarizing, the benefits of the air-lift pumps compared with other types of pumps have to do with the flexibility of adapting in different demands (m3/day), with better durability concerning corrosion and with the simplicity of the construction and the installation of an air-lift pump. On the other hand, air-lift pumps face difficulties when it comes to small drillings, high pressures and pumping emulsions or heavy crude oil. In addition to the last problems, also comes the not exactly precise theoretical model to predict and describe the flow conditions and characteristics in full details. This has to do both with the transition of the flow patterns of gas-phase and with the not always accurate mathematic formulas. Several experimental studies have been made with airlift pump installations, with a variety of different constructional attributes. Yoshinaga and Sato [6] used an air-lift pump with 6.74 m length upriser and tube diameter of 40 mm and 26 mm . Chi Ho Yoon et al. [7] used a 3.64 m length upriser and tube diameter of 30 mm and 20 mm. Khalil et al. [8] studied the performance of an air-lift pump with a 2 m length upriser and tube diameter of 25.4 mm. All the above installations worked with submergence ratio 0.8. In the present investigation, the experimental process is held in a 4.08 m length upriser and in tube diameters of 40 mm and 28 mm. The lab-scale installation comprises two downcomers, in order to investigate the impact of the liquid pressure losses, alongside with the solid-phase effects, in the air-lift pump performance. Experiments are held with defined points of submergence ratio, gas flow rate and solid-phase mass. The data collected are then presented and evaluated on operational curves concerning superficial velocities. Finally, some experiments are presented and compared with others from the literature in a regime map proposed by Samaras-Margaris [1], [2].

II.

frequency inverter monitored pump in order to stabilize the submergence ratio during experiments. The air is injected in three different levels, thus achieving several combinations of submergence ratios. For the expansion of the structure’s capabilities, there was an expanded modification in the horizontal pipes that are used for the connection of the main water tank and the suction box, especially the area of the devices 11-15 (Fig. 1, Table I). New electronic-signal rotameters were installed and the control panel was replaced. The new control panel comprises more flexible, useful and leak-safe water and air connections. Along, was installed the electronic rotameter device, analog air rotameters and electric devices for safe cut-off the pump’s electric power in an emergency situation. The new form of the structure is exposed in Table I, Fig. 2, Table II and Fig. 3. Three ways of water connection were created. Direct horizontal connection (through valves 2 and 3), horizontal connection with one downcomer (through valves 4, 6 and 7) and horizontal connection with two downcomers (valves 4, 5 and 7 are open). Regulating valves are presented with numbers in brackets, (1) to (7), in Table II and Fig. 3. The purpose of this expansion was to explore the reactions of the air-lift pump under the situation of pressure losses between the “water tank” (simulating sea, lake or other water quantity) and the upriser pipe of an air-lift device. Also the solid-lifting capability is an important attribute that was examined under the new pressure conditions.

Lab-scale Air-Lift Pump Installation

The experimental installation that is described in this chapter is established in the Fluid Mechanics Laboratory, Mechanical Engineering and Aeronautics Department of Patras University, Greece. The purpose of the experiments for this paper is to explore the possibilities of expanding the faculties of the existing installation, by inserting new attributes. The purpose is to achieve experiments of two and three-phase flow under a variety of new conditions, which have never been tested on this device. The installation –on its former condition Fig. 1is constituted by two upriser tubes, 40 mm and 28 mm diameters, the possibility of two and three-phase flow experiments (with several particle diameters and densities), the extra possibility of forced flow via auxiliary pump (in order to realize bubble flow regime, impossible for simple air-lift pump performance) and a

Fig. 1. Schematic presentation of the air-lift pump lab-scale installation. Numbers are explained in Table I

For the needs of the experiments, there were made assumptions of ideal gas, adiabatic flow, homogenous pressure and temperature in the gas pipe. Steady gas flow was forced (controlled via specific manometer) in the



973


upriser pipes of 40 mm and 28 mm. In the two-phase flow each experiment, for every measured point, was kept for 20 seconds, while in the three-phase (using solid particles of 1.06 g/cm3 density) was kept for 60 seconds if 0.5 kg of solid particles were laid in the water tank, 90 seconds for 1 kg of solid particles and 120 seconds for 1.5 kg of solid particles laying in the water tank. The duration of each experiment was differently defined, in order to achieve sufficient flow stability through the up risers. The flow through the pipes and the oscillations of the liquid was recorded for every experiment, with a 30 fps digital camera.

A/A 1 2 3 4 5 6 7 8

TABLE I AIR-LIFT PUMP INSTALLATION PARTS LIST Description A/A Description Main water tank (400 mm) 9 Auxiliary water tank Suction box 10 Pump Upriser (40 mm - 28 mm) 11 Air rotameters Solid particles box 12 Water rotameters Air injection points 13 Air compressor Air-water-solids separator 14 Water valves Water-solids separator 15 Control panel Solid particles scale 16 Air-water separator

Fig. 3. Water connections ground plan. Numbers are explained in Table II

The equation used to export further numerical results, from the collected experimental data, is listed below: QG =

( P + P1 ) Q* P1

(1)

G

where QG* stands for the gas volume flow rate, as measured by the rotameters in m3/h, for the existing pressure in the installation pipes, P, which is measured in bar. P1 stands for the environmental pressure, measured also in bar, and QG stands for the corrected gas volume flow rate in the installation pipes measured in m3/h. The gas volume flow rate or the liquid volume flow rate Qi, measured in m3/h, is given by equation (2):

(

3

Qi m / h Qi = J i A ⇒ J i =

Qi A

⇒u=

3600 π d j2

) (2)

4

where d stands for the diameter of the pipe-upriser in mm, and j takes the values j = 1, 2 for d = 40 and 28 mm respectively. A stands for the surface of the section of the pipe, measured in m2. Ji = u stands for the superficial velocity of the liquid or gas phase (m/s), i stands for L = liquid or G = gas phase. Superficial velocity, measured in m/s, is the velocity that one phase would have, if the phase were flowing alone in a pipe and expanded in all the diameter of the pipe.

Fig. 2. Photo of the air-lift pump lab-scale installation. Numbers are explained in Table I TABLE II PART LIST OF THE ADDITIONAL PARTS A/A Description [1] Main water tank (400 mm) [2] Suction box 17 Water rotameter 1 18 Water rotameter 2 19 1st Downcomer 20 2nd Downcomer (1) to (7) Regulating valves

III. Presentation of Experimental Results After the modifications of the installation took place, it was necessary to certify the installation and to secure



974


that everything was going to work precisely. This was achieved by performing some indicative experiments of two-phase liquid-gas flow. The results of these experiments were compared with former ones, Mitkas et al. [9]. For sure, it was not expected to have the same numerical results, because the installation changed (pipes, downcomers, one-way valves junctions, rotameters, panel instruments, new gas and liquid flow indicators) and experiments were held in different environmental conditions, but concerning the form the operational curves extracted, the results were consistent with former experimental and theoretical investigations of Samaras-Margaris [2] to predict the air-lift pump performance. In addition, to secure the precision of the data collected, all experiments were defined and categorized concerning finite consecutive conditions and measured points and all experiments were held twice. Referring to the whole experimental procedures, due to the installation (machinery and measuring devices), the definition of the specific measured points was based in the gas-phase conditions. This definition was achieved consecutive for each specific point, by means of using two manometers and a flow meter for the air before entering the injection point in the upriser. For this reason, the former investigation and comparisons concerning the operational curves were referred considering same gas superficial velocity values. Furthermore, experiments were categorized concerning the submergence ratio S/L, which is the height of the multi-phase column divided by the total height of the airlift pump upriser, both counting from the air-injection point. For example, Fig. 4 presents three-phase flow experiments. It shows the performance of the 40 mm upriser. The upper curves represent the S/L = 0.8 experiments. The lower curves present the S/L = 0.7 experiments. All four experiments deal with Js = 0.0083 m/s solid phase superficial velocity. As it was expected the liquid superficial velocity is higher with greater submergence ratio, due to the fact that the multi-phase column faces lesser difficulties to reach the upmost point of the installation. In addition, comparing experiments with the same submergence ratio, the ones with the onedowncomer connection achieve lower liquid superficial velocity due to the pressure loses concerning the connection type. All four experiments have similar conditions of 1012 mbar environmental pressure, 25 o C temperature and 74% relative humidity. In Fig. 5 the affection of the solid phase superficial velocity is presented. Someone may discern that for the same connection type the experiment with the lowest solid phase superficial velocity has higher liquid superficial velocity. The two upper curves present the no-downcomer experiments and the two lower curves present the experiments with one-downcomer.

Experiments were held in the D = 40mm air-lift pump upriser. The conditions for all four experiments were 25oC temperature, 1012 mbar environmental pressure and 74% relative humidity. D=40mm, Js=0.0083m/s

1: S/L=0.8, no-downcomer 2: S/L=0.8, 1-downcomer 3: S/L=0.7, no-downcomer 4: S/L=0.7, 1-downcomer

Liquid superficial velocity, JL (m/s)

0,70

0,65

1

0,60

2

0,55

0,50

3 4

0,45

0,40 0

2

4

6

8

10

12

Gas superficial velocity, JG (m/s)

Fig. 4. Experimental results presentation on a JL = f(JG) operational curve (different submergence ratios) D=40mm, S/L=0.8


0,70

0,65

1 0,60

3 2

0,55

4 1: Js=0.0063m/s, no-downcomer 2: Js=0.0063m/s, 1-downcomer 3: Js=0.0083m/s, no-downcomer 4: Js=0.0083m/s, 1-downcomer

0,50

0,45

0,40 0

2

4

6

8

10

12


Fig. 5. Operational curve JL = f(JG) for different solid-phase superficial velocity

In Fig. 6, the operational curves for four experiments are presented. The D = 28 mm air-lift pump upriser is used for these experiments. The two upper curves present the experiments with submergence ratio S/L = 0.8, while the lower curves present the experiments with submergence ratio S/L = 0.7. The solid-phase superficial velocity is 0.0013 m/s for all four experiments. Someone can discern from the graph that the experiments with the greatest submergence ratio achieve higher values of liquid superficial velocity, and for the same submergence ratio the experiment with the no-downcomer connection has greatest liquid superficial velocity for the same value of gas superficial velocity. The conditions for all experiments were 1012 mbar environmental pressure, 25 oC temperature and 74% relative humidity. Fig. 7 depicts the operational curves of two experiments in the D = 40 mm air-lift pump upriser and two experiments in the D = 28 mm air-lift pump upriser. The total solid mass lifted was 0.5 kg for both uprisers.



975


The curve showing the operation of the D = 28 mm air-lift pump upriser is higher because of the formula, used to calculate the mass superficial velocity, that gives greater superficial velocity value to the solid phase lifted by the smaller diameter upriser. Someone could notice that the experiments with the no-downcomer connection are higher than the ones with the one-downcomer connection and other conditions the same.

experiment with 3.64 m length upriser and 20 mm diameter upriser. The submergence ratio for curves 1 and 2 is S/L = 0.8. Someone may discern that the operational curves of all six experiments expand from the slug flow area, to the churn flow area until they reach the annular flow area. This change comes as a result of the increasing gas superficial velocity –or general flow velocity-that is responsible for the coalescences and the oscillatory motion of the flow. It is obvious from the graph that the experiments with the greatest submergence ratio achieve higher values of liquid superficial velocity. Furthermore, for the same submergence ratio the experiment with the nodowncomer connection has greatest liquid superficial velocity for the same value of gas superficial velocity. The conditions for our experiments were 1012 mbar environmental pressure, 25 oC temperature and 74% relative humidity.

D=28mm, Js=0.0013m/s


0,77

1: S/L=0.8, no-downcomer 2: S/L=0.8, 1-downcomer 3: S/L=0.7, no-downcomer 4: S/L=0.7, 1-downcomer

0,70

0,63

1 0,56

2 3

0,49

D=40mm

4 3,0

0,42

bubble 2

4

6

8

10

12

14

16

18

20

22


0


Fig. 6. Operational curve JL = f(JG) for different submergence ratio S/L=0.7


0,70

1: D=40mm, Js=0.0063, no-downcomer 2: D=40mm, Js=0.0063, 1-downcomer 3: D=28mm, Js=0.0013, no-downcomer 4: D=28mm, Js=0.0013, 1-downcomer

0,65

0,60

wispy annular

2,5

1: Yoshinaga and Sato [6], S/L=0.8 2: Chi Ho Yoon et al. [7], S/L=0.8 3: S/L=0.8, Js=0.0083,no-downcomer 4: S/L=0.8, Js=0.0083,1-downcomer 5: S/L=0.7, Js=0.0083,no-downcomer 6: S/L=0.7, Js=0.0083,1-downcomer

2,0

slug 1,5

1,0

annular 1

2 3 4 5 6

0,5

churn 0,0 0

1

2

3

4

5

6

7

8

9

10

11

12

13

14


0,55

0,50

1 0,45

Fig. 8. Presentation, in a regime map proposed by Samaras-Margaris [3], of the operational curve JL = f(JG) for different submergence ratio experiments and the same solid-phase superficial velocity 0.0083 m/s

3 4

2

IV.

0,40 0

2

4

6

8

10

12

14

16

18

20

22

Results and Discussion

The valuation of this experimental investigation has to do with the general needs of simulating and optimizing an issue. Firstly, it was a challenge to construct a labscale installation with better specifications than the former one [9], [11]. The enhancements of the new installation have to do with the import of new and more precise parts, such as the one-way valve junctions, the adaptation of new liquid and gas flow meters and last but not least the construction of two downcomers. The downcomers give the capability of achieving various conditions simply by the flick of a switch. By these means we can investigate the reactions of the air-lift pump under the situation of pressure losses between the “water tank” (sea, lake or other water quantity in mining) and the upriser pipe of an air-lift device. In real-scale airlift pumps the pressure losses may occur due to various reasons, such as wrong estimation of conditions, use of improper machinery, even loss of water pressure due to


Fig. 7. Operational curve JL = f(JG) for different solid-phase superficial velocity, in different diameter pipe uprisers

In Fig. 8, the operational curves for four experiments are presented in a regime map introduced by MargarisSamaras [1], [2], alongside with one experiment of Yoshinaga and Sato [6] (curve 1) and one experiment of Chi Ho Yoon et al. [7] (curve 2). Our experiments took place in the D = 40mm air-lift pump upriser. The curves 3 and 4 present the experiments with submergence ratio S/L = 0.8, while the lower curves 5 and 6 present the experiments with submergence ratio S/L = 0.7. The solid-phase superficial velocity is 0.0083 m/s for all four experiments. Curve 1 presents an experiment with a 6.74m length upriser and 26 mm diameter upriser. Curve 2 presents an



976


mining itself and reducing the water deposits. Especially for air-lift pumps that are used even when trying to lift solid parts along with slurries or liquids it is important not only to study the effects of different liquid pressures when entering the suction box via the downcomers but to study the effect of the solid-phase too. Secondly, for the needs of optimization many comparisons took place between the submergence ratios, between the downcomers and between the diameters of the upriser pipes. As presented in Fig. 4 and Fig. 6 the greater the submergence ratio the higher the value achieved for the liquid superficial velocity for defined value of gas superficial velocity. This is obvious in all graphs and has to do with the extra difficulties the flow has to overcome, with lower submergence ratio S/L, in order to reach the uppermost point of the installation where submergence ratio is S/L = 1. In Fig. 5 someone would observe the importance of the solid-phase flow. The greater the solid superficial velocity the lower the liquid superficial velocity for certain values of gas superficial velocity. This happens because part of the momentum of the liquid phase is consumed by the solid particles to be lifted. This is why the curves representing the experiments with greater solid superficial velocity are lower in the graph. Of course, for the same solid superficial velocity the nodowncomer experiment curves are higher. These results come to agreement with former experiments and theoretical predictions from Samaras-Margaris [10]. Fig. 7 combines two experiments of D = 40 mm diameter upriser with two of D = 28 mm diameter uprisers. In this graph the curves representing the small upriser are the highest. This has to do with the formula used to calculate the superficial velocity (both for liquid and gas phase). The actual liquid flow rate that is divided by the intersection of the upriser pipe, gives greater ratio divided by smaller intersection (D = 28 mm). Finally, Fig. 8 presents some characteristic operational curves of submergence ratio S/L = 0.8 and S/L = 0.7 on a regional map proposed by Samaras-Margaris [1], [2], [3]. Someone could observe that the experiments cover the regions of slug, churn and annular flow consecutively. The flow regime has to do with the general velocity characteristics of the flow in the air-lift pump upriser. These experiments expand in the same field of values and regimes, with the ones representing the greater submergence ratio S/L standing higher.

V.

follow in order to develop a computational method for the precise prediction of a three-phase flow air-lift pump performance.

References [1]

V.C. Samaras and D.P. Margaris, Predicting Three-Phase Air-Lift Pump Performance, International Review of Mechanical Engineering, vol. 3, n. 3, pp. 339-344, 2009 [2] V.C. Samaras and D.P. Margaris, Investigating Experimentally Flow Regimes in Three-Phase Air-Lift Pumps, International Review of Mechanical Engineering, vol. 4, n. 6, pp. 726-732, 2010. [3] V.C. Samaras and D.P. Margaris, Predicting the slug-churn transition in Air-Lift pump two phase flow, Proceedings 6th World Conf. On Experimental Heat Transfer,Fluid Mechanics, Matsushima, Miyagi, Japan, 2005. [4] G. F. Hewitt, and D. N. Roberts, Studies of two-phase flow patterns by simultaneous X-ray and flash photography, UKAEA Report AERE-M2159, 1969. [5] C. Kamata and K. Ito, Cold model experiments on the application of gas lift pump to the transportation of molten meta,. ISIJ International, vol. 35, pp. 859-865, 1995. [6] T. Yoshinaga, and Y. Sato, Performance of an Air-Lift Pump for Conveying Coarse Particles. Int. Journal of Multiphase Flow, vol. 22, pp. 223-238, 1996. [7] Chi Ho Yoon, Kwang Soo Kwon, Ou Kwang Kwon, Seok Ki Kwon, In Kee Kim and Dong Kil Lee, An experimental Study on Lab Scale Air-Lift Pump Flowing Solid-Liquid-Air Three-Phase Mixture, Proc. International Offshore and Polar Conference, Seattle, USA, 2000, pp. 515-521. [8] M.F. Khalil, K.A. Elshorbagy, S.Z. Kassab, and R.I. Fahmy, Effect of air injection method on the performance of air lift pump, Int. Journal of Heat and Fluid Flow, vol. 20, pp. 598-604, 1999. [9] K.D. Mitkas, D.P. Margaris, V.C. Samaras, and N.A. Avgerinos, Predicting the slug-churn transition in two-phase and three-phase air-lift pump installations, 3rd International Conference on Experiments/Process/System modeling/Simulation/Optimization (3rd IC-EpsMsO), Athens Greece, 8-11 July, 2009. [10] V.C. Samaras and D.P. Margaris, The Influence of a High Density Phase in a Multiphase Air-Lift Pump Performance, International Review of Chemical Engineering, vol. 2, n. 2, pp. 240-245, 2010. [11] D.N. Androulakis, A.N. Vlachos, K.D. Mitkas, and D.P. Margaris, Experimental Investigation of an Air-lift Pump Lab-scale Installation, 4th International Conference on Experiments/Process/System modeling/Simulation/Optimization (4th IC-EpsMsO), Athens Greece, 6-9 July, 2011.

Authors’ information Mechanical Engineering and Aeronautics Department, University of Patras, Post Code 26500, Rio, Patras, Greece Dimitrios N. Androulakis was born in 07 July 1986 in Heraklion city in Greece. He is a Mechanical Engineer studied in Mechanical Engineering and Aeronautics Department at University of Patras, Greece. His research activity up to now is experimental investigation in multiphase flows concerning lab-scale installations. In his diploma thesis he was dealing with the experimental investigation of an air-lift pump lab-scale installation. Dipl. Eng. Androulakis is scientific member of TCG (Technical Chamber of Greece).

Conclusion

The investigation, presented in this paper, comes in agreement with former ones [11] and deals with greater variety of possible features affecting the performance of an air-lift pump. The potentials of the existing installation are capable of giving more information about multi-phase flows in air-lift pumps. The experimental investigation is going on and a theoretical one will



977


Apostolos N. Vlachos was born in 06 March 1987 in Athens city, Greece. He is a Mechanical Engineer studied in Mechanical Engineering and Aeronautics Department at University of Patras, Greece. His research activity is experimental investigation in multiphase flows concerning lab-scale installations. In his diploma thesis he was dealing with the experimental investigation of an air-lift pump lab-scale installation. Dipl. Eng. Vlachos is now attending a Masters Degree Program in Marine Architecture & Science at National Technical University of Athens. Dionissios P. Margaris was born in Zakynthos island, Greece on September 28th, 1953. He is Associate Professor in Mechanical Engineering and Aeronautics Department at the University of Patras, Patras, Greece.His research activities/fields are multiphase flows of gasliquid-solid particles, gas-liquid two-phase flow, air-lift pump performance, centrifugal and Tjunction separation modeling in gas-liquid two-phase flow, experimental and theoretical investigation of hot air dehydration of agricultural products, experimental and theoretical investigation of capillary pumped loops, steady and transient flows in pipes and network and numerical simulation of centrifugal pump performance. Also he is dealing with fluid dynamics analysis of wind turbines and aerodynamic installations, aero-acoustic analysis and environmental impacts of wind turbines. He is participating in over 90 international conferences on the above scientific areas and has over 50 publications on high-interested impact factor journals. Prof. Dionissios P. Margaris is participating in several research projects supported by HAI, GSRT, CEC-THERMIE. Also he is member of AIAA, AHS, ASME and EUROMECH unions as well as of TCG (Technical Chamber of Greece).



978


Design of Compliant Mechanisms by Topology Optimization for Strain Actuators and Engineering Support G. Arunkumar1, J. Santhakumar2

Abstract – Compliant mechanisms are the focus of active research because of the flexibility, stability and unitized construction. It is a single elastic continuum used to transfer the force and motion mechanically by elastic deformation without any links and joints. Authors proposed a topology optimization method for designing a compliant amplifier and the steps for conducting the numerical experiments for any shape of design domains with different types of constraint. The paper narrates the design of the compliant amplifier mechanism from the basic geometrical shapes using topology optimization which can be used for amplifying the output displacement of the strain actuators. Numerical experiments are carried out for different basic configurations to design the compliant amplifier. The compliant amplifier is used for amplifying the displacement and stroke performance of the strain actuators when integrated with piezo actuators. The objective is to maximize the geometric advantage of the compliant amplifier. The maximization of the geometric advantage is accomplished by efficient design of compliant mechanisms using the topology optimization approach. The analysis results help to select the best compliant amplifier design from the basic configurations. The outcome of this work will be useful for all types of strain actuators. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved. Keywords: Compliant Mechanisms, Geometric Advantage, Topology Optimization, Numerical Experiments

I.1.

Nomenclature V U V Fd K ∆in fa a ∆out b A

Compliant mechanisms may be considered for use in a particular application for a variety of reasons. The advantages of compliant mechanisms are considered in cost reduction and simplified manufacturing process. The cost reduction by part-count reduction reduced assembly time and simplified manufacturing processes. The increased performance by means of increased precision, increased reliability, reduced wear, reduced weight, and reduced maintenance [1]-[3]. The advantage of compliant mechanisms is the potential for a dramatic reduction in the total number of parts required to accomplish a specified task. Some mechanisms may be Manufactured from an injection-moldable material and be constructed of one piece [1].

Applied voltage Deflection vector at ‘a’ Deflection vector at ‘b’ Dummy unit force (N) Global stiffness matrix Input deflection (m) Input force (N) Input point Output deflection (m) Output point Vector of design variable

I.

Advantage of Compliant Mechanisms

Introduction

The mechanism is a mechanical device used to transfer the force, motion or energy from input to output. Compliant mechanism is a flexible structure that elastically deforms without joints to produce a desired force or displacement at the output [1]. Unlike rigid-body mechanisms, compliant mechanisms gain some of their mobility from the deflection of flexible members rather than from movable joints only [1].

II.

New Age Industry Applications

Compliant mechanisms are used for manufacturing the car wiper, steering part, gears, brakes, over running clutches, centrifugal clutches, orthoplaner flat spring, and tyres in the automotive sector, the adaptive compliant wing, compliant adaptive rotor blades, proportional valve in the aerospace industry, the compliant gripper, compliant heart valve, medical invasive surgery, surgical



979

G. Arunkumar, J. Santhakumar

tools, precision monitoring equipment etc., in the medical field, the antilock breaking system, air bag deployment in the flight safety, the seat belt and helmets in the highway safety. In other sectors the high precision equipment parts, compliant die grippers, robot end effectors, fish hook pliers are made by compliant mechanisms for new age industries.

IV.1. Geometrical Advantage The force provided by the stack actuator to the compliant amplifier is modeled as a point load at ‘a’ is shown in Fig. 1. This approach is not restricted to symmetric problems however it can be applied to any shape and size design domain with any set of loading and geometrical constraints [6],[8]. Also the magnitude of the force provided by the stack is proportional to the voltage applied to it and the direction is along the longitudinal strain [6], [11]. Max (GA) is the ratio of the output displacement of actuator at ‘b’ to the displacement at the point ‘a’:

III. Design Engineering Support The engineering support for compliant design by topology optimization is to a great extent assisted by computer simulations of compliant mechanism response Subjected to real life loading conditions. Computer simulations are an excellent means of understanding the complex mechanical behavior and different topology of the domain using the visualization and graphic capabilities available with FE tools such as Ansys, Hyper mesh and MSC Nastran etc.

max ( GA ) =

∆ out V T KU = ∆ in U T KU

III.1. Topology Optimization It is a material distribution method used to find the best use of material for a body or domain, also to find the optimum shape and size of a linearly elastic structure. The topology optimization method can be used to design compliant mechanical amplifiers with any direction of force and motion transmission [1],[12]. The compliant mechanical amplifiers are the compliant mechanisms which are used to amplify the force and displacement at the output [5]-[6].

Fig. 1. Compliant mechanism with unconstraint output

Constraint equations are given as subjected to: Fin = KU, Fd = KV Volume ≤ Vo where K is the global stiffness matrix, U is the deflection vector at ‘a’, and V is the deflection vector at ‘b’. The output displacement is equal to mutual stain energy.

IV. The Need for Displacement Amplification of a Strain Actuator One type of smart material actuator typically used in adaptive structures is an induced-strain piezoceramic stack actuator. Piezoceramic (PZT) actuators are solidstate devices, which offer the advantages of high energy density and high output force when compared with conventional methods of actuation [4], [6]. One limitation of PZT actuators, however, is that they are capable of producing only 0.1 percent strain, resulting in a restricted range of motion. Thus limiting the stroke Performance. To increase the effective stroke of the actuator, compliant amplifiers have been designed to provide mechanical amplification. However it can be applicable for the strain actuators like piezoceramic actuators, and piezoelectric actuators [6], [8]. Designers of these mechanical amplifiers typically employ a singlepiece compliant amplifier as a coupling structure with the PZT actuator to avoid problems associated with clearances and backlash in hinge joints [9]-[10]. Numerical experiments on compliant mechanisms made by polypropylene integrated with strain actuator have been considered for displacement amplification of the actuators using topology optimization method.

V.

Design of Compliant Mechanism Using Rectangular Geometry V.1.

Selection of the Design Domain

The rectangular domain of size (600 X 500 mm) is selected as a basic geometrical domain for integrating with strain actuators like piezo electric and piezoceramic actuators is shown in Fig. 2. The domain is made by polypropylene material. The cross-sectional area of the element is considered for FE discretization is 25 mm2. All corners of the domains fixed (Point constraints) has been considered as boundary conditions. The input force applied at the input point is 300 N. It has been determined based on mechanical design calculation. Input and output displacement are determined by these equations. ∆out = VbT KUa; ∆in = FinT ∆in = UaT KUa. The ANSYS topology optimization tool has been used for finding the topology of the compliant amplifier. The topology optimization steps for rectangular domain with point constraints are shown in the Fig. 5(a)-(e) and 5(j)(m) for domain with surface constraints.



980


V.2.

Grid Dependency Check of the Design Domain

Grid refinements may affect the output displacement. In order to obtain a grid size that is almost independent of grid refinement, the grid dependency check is carried out. Initially the domain is divided into (m) × (n) a division is shown in Fig. 5(a). The output displacement is obtained. Then the domain is refined as (2m × 2n), (3m × 3n), (4m × 4n) and so on. In each refinement, the output displacements are obtained and are compared with previous one. Such grid refinements are carried out till the percentage variation in the output displacement is less than or equal to 5% between successive refinements [8], [13].

Fig 2. Basic Design Domains for Compliant Amplifier

Fig. 3(a). Rectangular domain with point constraints

Fig. 4(a). Rectangular domain with surface constraints

Fig. 3(b). Taper domain with point constraints

Fig. 4(b). Taper domains with surface constraints

Fig. 3(c). Hexogonal domain with point constraints

Fig. 4(c). Hexogonal domain with surface constraints

Figs. 3 (a), (b), (c) Geometry taken for analysis with point constraints

Figs. 4(a), (b), (c) Geometry taken for analysis with surface constraints



981


Fig. 5(a). Meshed domain

Fig 5 (b). Design domains with boundary Condition and applied force

Fig 5 (c). Converged topology plot

Fig 5(d). Material density plot

Fig. 5(e). Topology optimized domain

Fig. 5(f). Displacement plot 5

6

4.7977 4.4291

4

GEOMETRIC ADVANTAGE (GA

OUTPUT DISPLACEMENT (mm)

5

4.4291

3.9671

3.9671

3.3804

3.3723

3 2

1.7557 1.9044

1.9044

1 0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

4

3.938 3.607

3.16

2.599

2.593

2 1.339

1.428

1

0 0

0.9

3.607

3.16

3

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

INPUT LOCATION (Xi)

INPUT LOCATION (Xi)

Fig. 5(h). Effect of (Xi) on geometrical advantages

Fig. 5(g). Effect of input locations



982


MECHANICAL ADVANTAGE (MA)

1.2 1.04

1 0.9

0.9

0.82

0.8

0.82

0.7 0.6

0.7

0.6

0.6

0.52 0.4 0.2 0 0

0.2

0.4

0.6

0.8

1

INPUT LOCATION (Xi)

Fig. 5(i). Effect of (Xi) on mechanical advantage

Fig. 5(j). Design domains with BC

Fig. 5(k). Converged topology plots

Fig. 5(l). Material density plot

Fig. 5(m). Topology optimized domain

Fig. 5(n). Displacement plot 7

5 4.5548

4.7977

O UTP UT DIS P LACE M E NT (m m )

O U T PU T D ISPL A C EM EN T (m m )

6

4.5548

4 3.5507

3.5507

3.133

3

3.133

2.5915

2.5915

2 1.0988

1 0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

6

5.8623

5

4.7977 3.9981

4 3.1984

3 2.3988 2 1.5992 1 0 0

0.9

50

100

150

200

250

300

350

400

INPUT FORCES (N)

OUTPUT LOCATION (X0 )

Fig 5(o). Effect of output location

Fig 5(p). Effect of applied force



983


EFFECT OF THICKNESS (T)

18 16.107



16 14 12 10 8.6914

8 6.8863

6.1565

6 4.7977

4

3.58

2 0 0

0.5

1

1.5

2

2.5

3

3.5

9.5953

8 6 4.7977 4 3.1984 2.3988

2

0

1.9191

0.25

0.5

0.75

1

1.25

1.5

THICKNESS OF THE DOMAIN (mm)

Figure 5(q). Effect of aspect ratio

Fig 5(r). Effect of thickness

0.833 and the results are shown in the Figure 6 (c). The Fig. 6(c) shows the maximum output displacement is obtained at (Xo = xo/xm) is 7.48 mm.

Analysis of Compliant Mechanisms with Different Basic Configurations

Three types of basic geometric configurations (Rectangular, Taper and Hexogonal domain) with point constraints (all corners of the respective domain is fixed) has been taken for the analysis is shown in Fig. 3. Three types of configurations (Rectangular, Taper and Hexogonal domain) with surface constraints left and right end of the surface is fixed has been taken for analysis is shown Fig. 4.

VI.

10

0

ASPECT RATIO OF THE DOMAIN

V.3.

12

VI.3. Effect of Input Force (Fi) To study the effect of input force, the input location (Xi = xi/xm), the output locations (X0 = x0/xm) are kept constant. The magnitude of input force (Fi) =100 N, 150 N, 200N, 250 N, 300 N and 350 N applied at (Xi = xi /xm = 0.5). The output displacement values obtained are shown in Fig. 6(d). The Fig. 6(d) shows, while increasing the magnitude of forces applied at the input location the output displacement also increased linearly.

Analysis of Geometry with Point Constraints

VI.4. Effect of Aspect Ratio (A)

VI.1. Effect of Input Location (Xi)

To study the effect of aspect ratio, the aspect ratio of the domain has been varied from 1.2, 1.5, 2, 2.4 & 3.0. Fig. 6(e) shows, when the aspect ratio increases, the output displacement are found to increase because the size of the domain decreases the output displacement will get increased. The Fig. 6(e) shows, maximum output displacement are obtained, when the aspect ratio value is considered as 3.

To study the effect of input location, input force (Fi) = 300N, output location (X0) = x0/xm = 0.5 are kept constant. Input force is applied at various input locations (Xi = xi/xm) and the results are shown in Figure 6 (a). From Fig. 6(a) the maximum output displacement is obtained at (X0= x0/xm = 0.5) is 7.48 mm. From the Fig. 6 (b) the geometric advantage will be maximum at the input location Xi = (xi)/ (xm) = 0.55 and the corresponding value is 5.32. When the input location is varied from the left end to the middle, the output displacement continuously increases and reaches a maximum around the midspan and then decreases towards the right end for all the three geometries. The configuration at input locations is similar to a fixed beam with a point load in between fixed supports. Hence the deflection will be maximum near the middle. For any input location, taper domain with point constraints give highest output displacement and the hexogonal domain with point constraints give the lowest output displacement. But the rectangular domain with point constraints gives the output displacement values in between.

VI.5. Effect of Thickness (T) To study the effect of thickness, the input force (Fi), the input location (Xi = xi/xm) and the output locations (X0 = x0/xm) are kept constant. The thickness of the domain is varied as 0.25 mm, 0.75 mm, 1 mm, 1.25 mm and the results of output displacement are shown in the figure 6 (f). Increase in domain thickness increases the cross sectional area of the element over which the input forces is applied. The reduced stress results in lower input deflection in thickness of compliant mechanism reduce the output deflection for the given applied force.

VII. Analysis of Geometry with Surface Constraints

VI.2. Effect of Output Locations (X0)

VII.1. Effect of Input Location (Xi)

To study the effect of output locations, the input force (Fi), the input location Xi = (xi)/ (xm) are kept constant. The output locations (X0 = x0/xm) is varied as 0.083, 0.167, 0.25, 0.33, 0.417, 0.5, 0.583, 0.667, 0.75 and

To study the effect of input location, the input force (Fi) =300N, Output location (X0 = X0/Xm = 0.5) are kept constant. Input force is applied at various input locations



984



(Xi = Xi/Xm) and the results are shown in Fig. 7(a). The Fig. 7(a) shows the maximum output displacement is obtained at (X0 = x0/xm = 0.5 mm) is 5.45 mm. From the Fig. 7(b) the Geometric Advantage will be maximum at the input location Xi = (xi)/ (xm) = 0.55 and the corresponding value is 4.80 mm.

25 20 15

Rectangular Domain Taper Domain

10 5 0 0

0.5

1

1.5


8

2

2.5

3

3.5

ASPECT RATIO

7 6

Fig. 6(e). Effect of aspect ratio (A)

5


4

Hexogonal Domain

3

OUTPUT DISPLACEMENT(mm)

2 1 0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

INPUT LOCATION

Fig. 6(a). Effect of (Xi) on output displacement

16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0

Rectangular Domain Taper Domain Hexogonal Domain

0

0.25

0.5

0.75

1

1.25

1.5

THICKNESS OF THE DOMAIN(mm) GEOMETRIC ADVANTAGE (GA)

6 5 4

Fig. 6(f). Effect of thickness


3

Hexogonal Domain

When the input location is varied from the left end to the middle, the output displacement continuously increases and reaches a maximum around midspan and then decreases towards the right end for all the three geometries. Hence the deflection will be maximum near the middle. For any input location, taper domain with surface constraints give highest output displacement and the hexagonal domain with surface constraints give the lowest output displacement. But the rectangular domain with surface constraints gives the values in between.

2 1 0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

INPUT LOCATION

Fig. 6(b). Effect of (Xi) for Geometrical advantage


8 7 6 5

Rectangular Domain

4

Taper Domain

VII.2. Effect of Output Locations

Hexogonal Domain

To study the effect of output locations, the input force (Fi), the input location (Xi = xi/xm) are kept constant. The output locations (X0 = x0/xm) is varied as 0.083, 0.167, 0.25, 0.33, 0.417, 0.5, 0.583, 0.667, 0.75 and 0.833 and the results are shown in the Fig. 7(c). The Fig. 7(c) shows the output displacement maximum at the output location (Xo= xo/xm = 0.5) and the corresponding value is 5.45 mm.

3 2 1 0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

OUTPUT LOCATION


Fig. 6(c). Effect of output locations (Xo)

10

VII.3. Effect of Input Force (Fi)

9 8

To study the effect of input force, the input location (Xi = xi/xm), the output locations (X0 = x0/xm) are kept constant. The magnitude of input force (Fi) 100 N, 150 N, 200 N, 250 N, 300 N and 350 N applied at (Xi =xi /xm = 0.5). The output displacement values obtained are shown in Fig. 7(d). The Fig. 7(d) shows, while increasing the magnitude of forces applied at the input location the output displacement also increased linearly.

7 Rectangular Domain

6

Taper Domain

5

Hexogonal Domain

4 3 2 1 0 0

50

100

150

200

250

300

350

400

INPUT FORCE (N)

Fig. 6(d). Effect of Input force (Fi)



985


VII.4. Effect of Aspect Ratio (A)

with two types of constraints are compared and found that the taper domain with point constraints gives the maximum.

To study the effect of aspect ratio, the aspect ratio of the domain has been varied from 1.2, 1.5, 2, 2.4 & 3.0. Fig. 7(e) shows, when the aspect ratio increases, the output displacement are found to increase because the size of the domain decreases the output displacement will naturally get increased. The Fig. 7(e) shows, the maximum output displacement is obtained, when the aspect ratio value is considered as 3.


6

VII.5. Effect of Thickness (T)

Taper Domain Hexogonal Domain

2

1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

OUTPUT LOCATION

Fig. 7(c). Effect of output locations (Xo)

7

5

6 5 Rectangular Domain

4

Taper Domain Hexogonal Domain

3 2 1 0

Rectangular Domain

0

Taper Domain

3

Rectangular Domain 3

0

6

4

4

0



To study the effect of thickness, the input force (Fi), the input location (Xi = xi/xm) and the output locations (X0 = x0/xm) are kept constant. The thickness of the domain is varied as 0.25 mm, 0.75 mm, 1 mm, 1.25 mm and the results of output displacement are shown in the Fig. 7(f). Increase in domain thickness increases the cross sectional area of the element over which the inputs force is applied. The reduced stress results in lower input deflection in thickness of compliant mechanism reduce the output deflection for the given applied force.

5

50

100

150

200

250

300

350

400

INPUT FORCE (N)

Hexogonal Domain

2

Fig. 7(d). Effect of Input force (Fi)

1 0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

INPUT LOCATION

OUTPUT DEFLECTION (mm)

30

Fig 7(a). Effect of (Xi) on output displacement

25 20 Taper Domain Rectangular Domain

15

GEOMETRIC ADVANTAGE (GA)

6

10

5

5 0

4 Rectangular Domain

0

0.5

1

1.5

Taper Domain

3

2

2.5

3

3.5

ASPECT RATIO

Hexogonal Domain

2

Fig. 7(e). Effect of aspect ratio

1 0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

INPUT LOCATION OUTPUT DISPLACEMENT(mm)

12

Fig 7 (b). Effect of (Xi) for geometrical advantage

VIII. Results and Discussion The analysis conducted for Rectangular, taper and hexagonal design domain with two different boundary conditions point and surface constraints (Both ends and all corners of the design domain fixed). The output displacement obtained at the output point of the Rectangular domain, taper and Hexogonal domain

11 10 9 8 Rectangular Domain

7

Taper Domain

6

Hexogonal Domain

5 4 3 2 1 0 0

0.25

0.5

0.75

1

1.25

1.5

THCIKNESS OF THE DOMAIN (mm)

Fig. 7(f). Effect of thicknesses



986


Effect of Input Locations (Xi) The taper design domain with point constraints (all corners fixed) gives maximum output displacement 7.48 mm at the output point.

[3]

[4]

Effect of Geometric Advantage (GA) The geometrical advantage (GA) or amplification factor is the ratio of output displacement to input displacement of the domain. The taper domain with point constraints (All corners of the domain are fixed) gives maximum Geometric Advantage 5.32. The amplification factor is 3.

[5]

[6]

[7]

Effect of Output Locations (X0) The taper domain with point constraints gives the maximumoutputdisplacement 7.48 mm.

[8]

Effect of Input Forces (Fi) The taper domain with point constraints gives maximum output displacement. The output displacement of the domain when the force applied at 300 N is 7.48 mm. Also it shows while increasing the magnitude of forces applied at the input location the output displacement also increased linearly.

[9]

[10]

[11]

Effect of Aspect Ratio (A) The taper design domain with point constraint gives maximum output displacement with respect to corresponding aspect ratio is given in Fig. 7(e). If we increase the aspect ratio the output displacement also increased.

[12]

[13]

Effect of Thickness (T) The taper domain with point constraints gives maximum output displacement at various thickness of the design domain is represented in Fig. 7(f). The output displacement 7.48 mm when the thickness of the domain 0.5 mm.

[14]

Journal of Mechanical Design, Transactions of ASME, Vol. 116, pp. 270-279 (1994). Anandhasuresh G.K. and Saxena A, On an Optimal Property of Compliant Topologies, Journal of Structural Multidisc Optimization, Vol. 19, pp. 36-49 (2000). Ashok Midha and Howell L.L, The Effects of a Compliant Workforce on the Input/output Characteristics of Rigid Link Toggle Mechanisms, Journal of Mechanism and Machine Theory, Vol. 30, No. 6, pp. 801-810 (1995). Krishnan, G. and Ananthasuresh, G. K, Evaluation and Design of Compliant Displacement Amplifying Mechanisms for Sensor Applications, Journal of Mechanical Design, Volume 130, Issue 10, pp.1-9 (2008). Canfield S. and Fracker M, Topology Optimization of Compliant Mechanical Amplifiers, Journal of Structural Multidisc Optimization, Vol. 20, pp. 269-279 (2000). Murphy, M.D., Midha, A., and Howell, L.L, The Topological Synthesis of Compliant Mechanisms, Mechanism and Machine Theory, Vol. 31, No. 2, pp. 185-199 (1996). Arunkumar,G and Srinivasan, Design of Displacement Amplifying Complaint Mechanism with integrated Strain Actuator using Topology Optimisation, International Journal of Mechanical Engineering Science, Vol.220, Issue 10, pp.12191228 (2006). Howell, LL and Midha, A, A method for the Design of Compliant Mechanisms with Small-Length Flexural Pivots, ASME Journal of Mechanical Design, Vol.116, pp.280-290 (2006). Kota, S and Hetric J, Synthesizing high-performance compliant stroke amplification systems for MEMS, Proceedings of the IEEE Micro Electro Mechanical systems, pp.164-169 (2000). Boyle, C., Howell, L.L., Magleby, S.P., and Evans, M.S, Dynamic Modeling of Compliant Constant-Force Compression Mechanisms, Mechanism and Machine Theory, Vol.38, No. 12, pp. 1469-1487 (2003). Ananthasuresh G.K.,Anupam Saxena, A Computational Approach to the Number of Synthesis of Linkages, Journal of Mechanical Design, Transactions of ASME,Vol.125,pp. 110-118 (03). Tenek, L.T. A Beam Finite Element Based on the Explicit Finite Element Method, International Review of Mechanical Engineering, Vol. 2 n. 1, pp. 122 – 131(2008). Niola, V., Quaremba, G., Amoresano, A., A study on infrared thermography processed trough the wavelet transform, (2009) Proceedings of the 8th WSEAS International Conference on System Science and Simulation in Engineering, ICOSSSE '09, pp. 57-62.

Authors’ information IX.

Dr. G. Arunkumar1 is working as Professor and Head of Mechanical Engineering at Saveetha School of Engineering, Saveetha University, Chennai – 602 105, Tamilnadu, India. Author has been published many papers in refreed International, National journals and conferences. E-mail: [email protected]

Conclusion

In this work three regular shapes of the design domains has been considered, however this method is applicable for any shape and size of the design domain. In the example problem the linear analysis has been considered, due to the requirement of topology optimization and minimum displacement amplification at the output point. The amplified displacement can be used for applications like opening and closing of proportional valves and vibrating a small chute etc. Authors approached new age industries in Chennai and the work was initiated for the betterment of future industry.

Mr. J. Santha Kumar2 is working as Assistant Professor of Mechanical Engineering at SRM University, SRM Nagar, Chennai, Tamilnadu, India. E-mail: [email protected]

References [1] [2]

Howell L.L, Complaint Mechanisms, Willey Publications, (New York, 2001, pp 5-25). Midha, A., Norton, T.W., and Howell, L.L, On the Nomenclature, Classification, and Abstractions of Compliant Mechanisms,



987


PID Controller Tuning for Magnetic Suspension System Using Evolutionary Algorithm V. Rajinikanth1, K. Latha2

Abstract – In this paper, a model based intelligent PID controller tuning is attempted for a Magnetic Suspension System (MSS) using Bacterial Foraging Optimization (BFO) algorithm. MSS is an Electro- Mechanical system with highly non linear dynamics and exhibits unstable steady state behavior around the nominal operating range. In this work, Integral Time Absolute Error (ITAE) based performance criterion is employed to guide the BFO to search the optimized controller parameters such as Kp, Ki, and Kd. The work is substantiated through a comparative study with Particle Swarm Optimization (PSO) algorithm by using PID and modified PID controller structure. The effectiveness of the proposed scheme has been validated with a simulation study and the results show that the proposed method provides better result in the control of the unstable MSS system with effective setpoint tracking and load disturbance rejection. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Magnetic Suspension System, Unstable System, PID controller, Bacterial Foraging Optimization

stability, smooth setpoint tracking, load disturbance rejection and measurement noise attenuation. It also provides optimal and robust performance for a wide range of operating conditions for stable, unstable and non linear processes. Further, it supports tuning and retuning based on the performance requirement of the system to be controlled. Most of the real time mechanical loop has electrohydraulic [1], mechatronic [2], and electro-mechanical systems. The above systems can help to effectively control the mechanical loops using a closed loop control system. Designing a robust controller for such system is also necessary to increase the efficiency [3]. In this paper, a modified structure PID controller tuning for the Magnetic Suspension System (MSS) is discussed. The MSS has high nonlinear system dynamics which can be modeled around the operating point as an open-loop unstable system with or without a measurement delay. MSS have practical significance in engineering areas such as high-speed ground transportation systems, frictionless bearings, levitation of wind tunnel model, vibration isolation in machinery, levitation of molten metal in induction furnaces, and levitation of metal slabs during manufacturing. Recent research reports the requirement to develop control methodologies for magnetic suspension system. Marjan and Boris [4] developed a linearized mathematical model for a laboratory scale electromagnet-ball suspension system and proposed a decomposed fuzzy PID controller. Ishtiaq and Akram [5] discussed a linear and nonlinear state space model based controller design. Muthairi and Zribi [6] examined a

Nomenclature A,B,C R L L0 i(t) m C g x0 Xm Xg Kp Ki Kd r(t) β,γ y(t) t e(t) Ess Mp ts Jcc j, k, l

Matrix parameters Internal resistance of coil Self inductance of coil Incremental inductance of coil Coil current Mass of the ball Magnetic force constant Gravitational constant Equilibrium position Distance between magnet and ball Distance between ball and ground Proportional gain Integral gain Derivative gain Reference input Setpoint weighting parameters Process output Time Error Steady state error Peak overshoot Settling time Cost function BFO algorithm variables

I.

Introduction

In industries and laboratories, Proportional + Integral + Derivative (PID) and modified structure PID controllers are widely used to provide closed loop



988

V. Rajinikanth, K. Latha

static and dynamic Sliding Mode Controller (SMC) and presented the simulated result to validate the performance of the proposed method. A phase-lead compensation based controller design for a linearized system model was developed by Shiao [7]. Hassanzadeh et al. [8] proposed binary and real-coded Genetic Algorithm (GA) based controller and with experimental results, they confirmed that the binary coded GA works well than the continues GA. Michail et al. [9] proposed H∞ controller design for Maglev suspension system using a sensor optimization concept. Soft computing based optimization is a powerful tool for finding the solutions for complex engineering problems. Evolutionary methods such as Genetic Algorithm [10],[11], Particle Swarm Optimization [12],[13], and Bacterial Foraging Optimization [15],[16] are sufficiently discussed in the literature. The evolutionary approach based PID controller auto tuning has attracted the control engineers and the researchers due to its simplicity, high computational competence, easy execution and stable convergence. In this work, the PID controller tuning is proposed for the Magnetic Suspension System (MSS) using the Bacteria Foraging Optimization (BFO) algorithm introduced by Passino [15]. It is a biologically inspired computation technique based on mimicking the foraging activities of Escherichia coli (E.coli) bacteria and it is successfully used in various engineering applications. The literature gives the application details of BFO in PID controller tuning for a class of stable systems [16] and unstable systems [17], [18]. In this work, the performance of the proposed tuning technique is evaluated with a simulation work. A comparative study also carried out with PID and modified PID controller structures tuned with the BFO and Particle Swarm Optimization (PSO) algorithms. The remaining part of the paper is organized as follows: Description of Magnetic Suspension System is provided in section 2, section 3 presents the overview of Bacterial Foraging Optimization algorithm. PID controller tuning is discussed in section 4. Section 5 gives the simulated results followed by the conclusion of the present research work in section 6.

II.

to gravity, L0 is the incremental inductance of the magnetic coil due to equilibrium position x0. From Eq (1), it is observed that, the coil inductance is the function of ball position ‘x’ and the approximate inductance at ‘x0’ is: L0 x0 x

L ( x) = L +

(1)

the electric current through the coil until the electromagnetic force holds the ball in an equilibrium state. Ysp(t)

Electromagnet

i Ym(t)

Current Controller

Ball Position (Feedback)

f

Xm

Ball Xg Photo Detector

Optical Source

m *g

Ground Fig. 1. Schamatic of Magnetic Suspension System

II.1.

Mathematical Model

The mathematical model for the above system can be developed by analyzing the electrical and mechanical sections separately. • Voltage applied to the coil is: V (t ) = R i (t ) + L

Magnetic Suspension System

d i (t ) dt

(2)

• Force by the electromagnet is:

In this work, Magnetic Suspension System consisting of a ferromagnetic ball suspended in a controlled magnetic field as in Figure.1 is considered [5]. It is an Electro-Mechanical system and the coil acts as an actuator to position the ball based on the reference signal. An optical position sensor is used in the feedback loop to sense the actual position of the suspended ball. An optimally tuned controller is then used to regulate where, x = Xm is the distance between object and magnet, i = current through the coil, L = self inductance of the coil, R= internal resistance of the coil, m= mass of the object, C= magnetic force constant, g= acceleration due

2

L X ⎛i⎞ f ( x,i ) = C ⎜ ⎟ and C = 0 0 2 ⎝ x⎠

(3)

• Mechanical force on the ball is: m

d2x

⎛i⎞ = (m ⋅ g ) − C ⎜ ⎟ dt 2 ⎝ x⎠

2

(4)

Eq. (3) has the nonlinear dynamics, and the linearized form around the operating position is:



989


m

⎛i ⎞ dx 2 = −C ⎜ 0 ⎟ dt ⎝ x0 ⎠

2

⎡ i ( t ) x ( t ) ⎤ ⎪⎫ ⎪⎧ − ⎨1 + 2 ⎢ ⎥ ⎬ + mg x0 ⎦ ⎪⎭ ⎪⎩ ⎣ i0

The transfer function model of the system is widely considered to design the PID controller. The transfer function model of the system obtained from Eq. (9) is presented in Eq (10):

(5)

The suspended ball reaches the equilibrium state at G (s) =

2

⎛i ⎞ C⎜ 0 ⎟ = m⋅ g ⎝ x0 ⎠

(6)

⎡ 2.842 x10−14 s 2 + ⎤ ⎢ ⎥ −13 ⎢⎣ −0.905 x10 s − 2333⎥⎦ G (s) = 3 x e −td S s + 100 s 2 + 1088s − 163300

2

2i C 2i C dx 2 = − 02 i ( t ) + 0 3 x ( t ) dt x0 x0

(7)

where x(t) and i(t) are the incremental displacement and incremental magnet current respectively. The MSS parameters considered in this work is presented in Table 5 [5].

(11)

III. Bacterial Foraging Optimization Bacteria Foraging Optimization (BFO) algorithm is a nature inspired stochastic search technique based on imitating the foraging (methods for locating, handling and ingesting food) behaviour of Escherichia coli (E.coli) bacteria. This algorithm was initially proposed by Passino [15], to design a model based adaptive controller. A detailed description of the BFO algorithm is already discussed in the literature [19], [20]. During foraging, a bacterium can exhibit two different actions: tumbling or swimming. The tumble action modifies the orientation of the bacterium. During swimming (chemotactic step) the bacterium will move in its current direction. Chemotactic movement is continued until a bacterium goes in the direction of positive nutrient gradient. After a certain number of complete swims, the best half of the population undergoes reproduction, eliminating the rest of the population. In order to escape local optima, an elimination-dispersion event is carried out where, some bacteria are liquidated at random with a very small probability and the new replacements are initialized at random locations of the search space.

The linearized state-space model using Eqn.1, 3 and 5 is given below:

(8)

In the state space model, the state equation is ‘Ax + Bu’ and the output equation is ‘Cx’. The actual model of the system derived from Table I values are presented in Eq. (9): 1 0 ⎤ ⎡ 0 ⎢ A = ⎢1633.33 −23.33⎥⎥ 0 ⎢⎣ 0 116.66 −100 ⎥⎦ ⎡ 0 ⎤ B = ⎢⎢ 0 ⎥⎥ ; C = [1 0 0] ⎢⎣100 ⎥⎦

(10)

where ‘td’ is the measurement delay by the position sensor. In this paper, the delay is considered as ‘0.05s’. The above transfer function model has two stable complex poles (-65.6797 + i29.8920, -65.6797 i29.8920), an unstable pole (31.3595) and positive zeros. From the above observation, the MSS is open loop unstable by design and the system should be operated in closed loop along with optimally tuned controller.

TABLE I MSS SYSTEM PARAMETERS Parameter Value m 0.05 Kg g 9.81 m/s2 L 0.01H R 1Ω C 0.0001 x01 0.012 M 0 M/s x02 0.84A x03

1 0 ⎤ ⎡ 0 ⎢ 2 ⎥ Cx Cx03 ⎥ A = ⎢ 03 − ; 0 2 2 ⎥ ⎢ Mx3 Mx01 01 ⎢ ⎥ Cx03 ⎢ R ⎥ − 2 2 ⎢ 0 L ⎥⎦ Lx01 ⎣ ⎡ 0 ⎤ B = ⎢⎢ 0 ⎥⎥ ; C = [1 0 0] ⎢1 ⎥ ⎣ L⎦

s 3 + 100 s 2 + 1088s − 163300

For stability reasons, a small dead time is considered with the G(s) and the final model of the MSS will be:

with respect to the above condition, Eqn.5 can be formed as: m

2.842 x10−14 s 2 − 0.905 x10−13 s − 2333

III.1. Chemo-Taxis This is the initial stage of BFO search. During this process, the bacteria can move towards the food location with the action of swimming and tumbling via flagella. Through swimming, it can move in a specified direction and during tumbling action, the bacteria can modify the direction of search.

(9)



990


These two modes of operation is continuously executed to move in random paths to find adequate amount of positive nutrient gradient. These operations are performed in its whole lifetime.

START Initialization: Assign values for the BFO parameters.

III.2. Swarming Allocate the multi objective function for the optimization search.

In this process, after the success towards the best food location, the bacterium which has the knowledge about the optimum path will attempt to communicate to other bacteria by using an attraction signal. The signal communication between cells in E.coli bacteria is represented by the following equation:

Assign the maximum and minimum values for the search parameters. Generate initial positions for Bacteria in the ‘ND’ dimensional search space.

n

Jcc (θ, P (j, k, l)) = ∑ J cc (θ, θi (j, k, l)) = A + B

(12)

i =1

Evaluation

where A and B are given below:

Initiate Chemotaxis Tumble / Run

P ⎡ ⎛ d exp W − − ⎢ ⎜ ∑ ⎢ attractant ⎜ attractant ∑ θ m − θ i m i =1 ⎣ m =1 ⎝

(

n

)

2

⎞⎤ ⎟⎟ ⎥ ⎠ ⎥⎦

Is End of Chemo taxis?

No

Yes

and:

Start Reproduction P ⎡ ⎛ ∑ ⎢⎢hrepellant exp ⎜⎜ −Wrepellant ∑ θm − θ i m i =1 ⎣ m =1 ⎝ n

(

)

2

⎞⎤ ⎟⎟ ⎥ ⎠ ⎦⎥

Is End of Reproduction?

III.3. Reproduction

No

Yes

In swarming process, the bacteria accumulated as groups in the positive nutrient gradient and which may increase the bacterial density. Later, the bacteria are sorted in descending order based on its health values. The bacteria which have the least health will expire and the bacteria with the most health value will split into two and breed to maintain a constant population.

Start Elimination / Dispersal

Is End of Elimination / Dispersal?

No

Yes No Is optimised values of Kp, Ki , Kd are available?

III.4. Elimination-Dispersal

Yes

Based on the environmental conditions such as change in temperature, noxious surroundings, and availability of food, the population of a bacteria may change either gradually or suddenly. During this stage, a group of the bacteria in a restricted region (local optima) will be eliminated or a group may be scattered (dispersed) into a new food location in the search space. The dispersal possibly flattens the chemo-taxis advancement. After dispersal, sometimes the bacteria may be placed near the good nutrient source and it may support the chemo-taxis, to identify the availability of other food sources. The above procedure is repeated until the optimized solutions are achieved. Fig. 2 shows the basic flow chart of the BFO algorithm.

STOP Fig. 2. Flow chart for Bacterial Foraging algorithm

IV.

PID Controller Tuning

In closed loop systems, the main objective of the controller is to make the peak overshoot (Mp), settling time (ts) and final steady state error (Ess), as small as possible. In BFO based tuning approach, the Cost Function (CF) is used to appraise the performance of the closed loop system during the optimization search. Integral



991


Time Absolute Error (ITAE) criterion shown in Eq. (14) is preferred as the ‘CF’ in this study: ITAE =

T

∫0 t ⋅ e ( t )

dt =

T

∫0 t ⋅ ⎣⎡ r ( t ) − y ( t )⎦⎤ dt

During the search, the range of search space in BFO algorithm will influence the performance of the optimization algorithm, such as convergence speed and accuracy in optimized controller parameters. If the search space is small (ie. lesser than 100), then the evolutionary algorithm will identify the accurate parameters within lesser search time. In the proposed method, the three dimensional search space is defined as follows:

(13)

where e(t) = error, r(t) = reference input, and y(t) = process output. The controller tuning problem is to find the optimal values of Kp, Ki and Kd form search space that minimizes the objective function (Eq.(13)). Fig. 3 illustrates the basic block diagram of a setpoint weighted PID controller proposed by Chen et al. [16].

E(s)

R (s)

Uc(s)

+

Gp(s)

Kp + Ki + Kd _

Kp = -100 to 100; Ki = -10 to 10 and Kd = -10 to 10 If the search does not provide the optimized PID parameter, then the range of Kp and Ki is to be adjusted. The retuning is performed for Kp and Ki values without changing the Kd. In this case, the search dimension for the BFO algorithm is two (Kp, Ki) and the retuned values are presented in Table II.

Y(s)

_ Kp1 + Kd1

BFO Algorithm ITAE

Fig. 3. PID-PD controller structure

Ki

R(s)

Kd Uc (s)

P+I+D

Y(s) MSS

‐

‐

The controller output for the above block is represented in Eq. (14): Uc (t) = Uc1 – Uc2

K p

P + D

(14) Fig. 4. PID-PD controller tuning using BFO

where: ⎪⎧ U C1 = ⎨ K p β e ( t ) + K i ⎩⎪

and:

t

∫0 e ( t ) dt + K d γ

d e ( t ) ⎪⎫ ⎬ dt ⎭⎪

V.

Results and Discussions

The higher order unstable transfer function model given in Eq. (11) is considered in this study. BFO based auto tuning is initially proposed with a basic PID controller. The overshoot produced by the system with PID controller is large and the final convergence of the BFO algorithm does not provide the optimized PID parameter. Hence a BFO based I-PD controller tuning is proposed for the MSS system. In the beginning of the optimization search, the BFO algorithm assigns a lower positive arbitrary value for the controller parameters based on the search space and calculates the performance index (ITAE) which guides the search. The initial controller values are randomly adjusted until the search converges with a minimised ITAE. The final convergence of the Kp, Ki and Kd are shown in Figs. 5 (5(a) Convergence of Proportional gain, (b) Integral gain and (c) Derivative gain). In the BFO tuned I-PD, the controller parameter search value has converged at 259th iteration and the values are given in Table II.

d y ( t ) ⎪⎫ ⎪⎧ U C 2 = ⎨ K p1 (1 − β ) y ( t ) + K d 1 (1 − γ ) ⎬ dt ⎪⎭ ⎪⎩

where : ‘β’ and ‘γ’ are setpoint weighting parameters. When β = γ = 1, the structure will provide a basic PID controller. When β = 0 and γ = 0, the above structure will provide a modified structure PID (I-PD) controller. The modified structure PID (I-PD) controller tuning is similar to a basic PID controller tuning. Before the optimization search, it is necessary to assign the following algorithm parameters: Dimension of search space is three; number of bacteria is chosen as sixteen; number of chemotactic steps is set to eight; number of reproduction steps and length of a swim is considered as four; number of elimination-dispersal events is two; number of bacteria reproduction is assigned as eight; probability for elimination – dispersal has a value of 0.3. Other additional parameters are assigned as, dattractant =0.3, Wattractant = 0.5, hrepellant =0.6 and Wrepellent = 0.6. Fig. 4 shows the block diagram of the BFO algorithm based PID-PD controller tuning for the unstable MSS system.

Method BFO PSO BFO (Retuned)


TABLE II PID CONTROLLER PARAMETERS Kp Ki -72.0480 -3.0225 -90.8401 -7.9225 -87.0501 -11.0307

Kd -2.0193 -3.7720 -2.0193


992


20

4

0

2 0 Ki

Kp

-20 -40

-2

-60 -4

-80 -100

-6

0

50

100

150 Iteration

200

250

300

0

50

100

150 Iteration

(a)

200

250

300

(b) 2 1 0 Kd

-1 -2 -3 -4 -5 0

50

100

150 Iteration

200

250

300

(c) Fig. 5. BFO based auto tuning of I-PD controller 1.2

35 BFO PID

30

1

PSO PID Servo Response

Response

25 20 15 10

BFO I-PD PSO I-PD

0.6 0.4 0.2

5 0

0.8

0

0

5

10

15

0

5

10 Time,s

Time,s

1.5

1.2

1 Regulatory Response

Step Response

1 0.8 BFO I-PD PSO I-PD

0.6 0.4 0.2

0.5 0

BFO I-PD PSO I-PD

-0.5 -1 -1.5

0

10

20

30

40

-2 0

50

10

20

40

50

(d)

(c) 1.2

1.4

1

1.2 Regulatory Response

Regulatory Response

30 Time,s

Time, s

0.8 BFO I-PDr PSO I-PD

0.6 0.4 0.2 0

20

(b)

(a)

0

15

1 0.8

BFO I-PD PSO I-PD

0.6 0.4 0.2

0

10

20

30

40

0

50

Time, s

0

10

20

30

40

50

Time, s

(e)

(f)

Fig. 6. Responses of the MSS with PSO and BFO tuned PID and modified structure PID controller



993


1.2

Response

1 0.8

Reference Signal BFO I-PD PSO I-PD

0.6 0.4 0.2 0

0

10

20

30

40

50

60

70

80

90

100

Time,s Fig. 7. Servo and Regulatory responses for the MSS with BFO and PSO tuned controllers

Fig. 6(a) and 6(b) shows the servo response of the Magnetic Suspension System with a PID and I-PD controller structure respectively. From the illustrations and its corresponding results presented in Table 3, it is observed that, due to large overshoot, the response is unsatisfactory for PID and to enhance the setpoint tracking performance, it is necessary to apply a modified PID structure (I-PD). Multiple setpoint tracking response with I-PD structure is depicted in Figure 6(c). A change in setpoint of 0.1 (10% of the input) is introduced at 25s. The result of this proposed method presents an improved performance than PSO tuned I-PD.

t=50s respectively. A load change of 0.05 (5%) is introduced at 75s. The results show that, the BFO tuned I-PD provides an improved performance than the PSO tuned I-PD.

VI.

Conclusion

In this paper, we have attempted a basic BFO algorithm to optimize the controller parameters for an unstable Magnetic Suspension System. In this work, ITAE minimization is highly prioritized as a performance measure. It monitors the optimization algorithm until the parameters converge to a minimized value. The simulation result shows that the I-PD structure provides the optimal value than the basic PID controller. The BFO based controller tuning approach is a simple auto tuning approach and improves the performance of the process in terms of setpoint tracking, multiple setpoint tracking and load disturbance rejection.

TABLE III PERFORMANCE COMPARISONS IAES ISER IAER Method ISES BFO PID 458.1 23.19 PSO PID 16.59 8.846 BFO I-PD 0.469 0.674 5.149 2.997 PSO I-PD 1.326 2.628 1.492 3.519 BFO I-PD (Re) 0.788 1.538 0.938 2.187 (the subscript ‘S’ is for servo tracking and ‘R’ is for regulatory response)

References [1]

A load disturbance of 0.1 (10%) is introduced at 25s and its regulatory response with I-PD structure is presented Fig. 6(d). The observations shows that, the proposed method give large undershoot compared to PSO based I-PD. These results show that, obtained BFO based I-PD controller parameters need further retuning for better load disturbance rejection. Fig. 6(e) and 6(f) shows the regulatory response of the MSS with a I-PD controller for a load change introduced at 25s with a value of 0.1 (10%) and 0.02 (2%) respectively. The BFO I-PD with the retuned Kp and Ki helps to get a reduced undershoot for the load disturbance rejection and provides improved result when the load disturbance is 2%. Setpoint tracking, multiple setpoint tracking and load disturbance rejection with retuned controller parameter is depicted in Fig. 7. A change in setpoint of 0.1 (10% of the input) and -0.1 is introduced at t=25s,

[2]

[3]

[4]

[5]

[6]


S. F. Rezeka, A.Khalil, A.Abdellatif, “Parametric Study of Electro-Hydraulic Servo Valve Using a Piezo-Electric Actuator”, International Review of Mechanical Engineering (I.RE.M.E.), Vol. 5 N. 5,2011, pp. 961-967. Wajdi S. Aboud, Sallehuddin Mohamed Haris, “A Study on Load Dependent Controller Performance for Mechatronic Suspensions”, International Review of Mechanical Engineering (I.RE.M.E.), Vol. 6 N. 3, (Part B), 2012, pp. 611-616. Ali Zolfagharian, Mohd Zarhamdy Md. Zain, Abd Rahim Abu Bakar, Azizan As’arry, “Suppressing Chatter Noise in Windscreen Wiper Operation Using a Robust Hybrid Controller”, International Review of Mechanical Engineering (I.RE.M.E.), Vol. 6 N. 3, (Part B), 2012 pp. 630-635. Marjan Golob, Boris Tovornik, “Modeling and control of the magnetic suspension system”, ISA Transactions, 42, 2003, pp. 89–100. Ishtiaq Ahmad, Muhammad Akram Javaid, “Nonlinear Model & Controller Design for Magnetic Levitation System”, in the proceedings of Proceedings of the 9th WSEAS International Conference on Signal Processing, Robotics and Automation (ISPRA '10). University of Cambridge, UK, pp. 324-328 N.F. Al-Muthairi and M. Zribi, “Sliding Mode Control of A Magnetic Levitation System, Mathematical Problems in


994


[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22]

[23]

Engineering, 2, 2004, pp.93–107. Ying-Shing Shiao, “Design and Implementation of a Controller for a Magnetic Levitation System”, Proc. Natl. Sci. Counc. ROC (D), Vol. 11, No. 2, 2001. pp. 88-94. I.Hassanzadeh, S.Mobayan and G. Sedaghat, “Design and Implementation of a Controller for Magnetic Levitation System Using Genetic Algorithms”, Journal of Applied Sciences, 8(24), 2008, pp. 4644 - 4649. Konstantinos Michail, Argyrios Zolotas, Roger Goodall, and John Pearson, “Sensor Optimisation via H∞ Applied to a MAGLEV Suspension System”, World Academy of Science, Engineering and Technology, 41,2008, pp.171-177. D. Mačiūnas, R. Belevičius, “Multi-objective Optimization of Grillages Using Adaptive Genetic Algorithm”, International Review of Mechanical Engineering (I.RE.M.E.), Vol. 6 N. 3,(Part A), 2012, pp. 432-439. J.L. Marcelin, “ Optimization of the Boundary Conditions by Genetic Algorithms”, International Review of Mechanical Engineering (I.RE.M.E.), Vol. 6 N. 1, 2012,pp. 50-54. Arash Mohammadzadeh, A.Ghoddoosian, M. Noori-Damghani, “Balancing of the Flexible Rotors with Particle Swarm Optimization Method”, International Review of Mechanical Engineering (I.RE.M.E.), Vol. 5 N. 3, 2011,pp. 490-496. Arash Mohammadzadeh, N. Etemadee, “Optimized Positioning of Structure Supports with PSO for Minimizing the Bending Moment”, International Review of Mechanical Engineering (I.RE.M.E.), Vol. 5 N. 3, 2011, pp. 422-425. Rosario Russo, Alberto Clarich, Marco Carriglio, “A Multiobjective Optimization of Engine Crankshaft Design using modeFRONTIER”, International Review of Mechanical Engineering (I.RE.M.E.),Vol. 6 N. 3, (Part B), 2012, pp. 574-577. Kevin M. Passino., “Biomimicry of bacterial foraging for distributed optimization and control”, IEEE Control Systems Magazine, 2002, pp.52-67. Wael M. Korani, Hassen Taher Dorrah, Hassan M. Emara. “Bacterial foraging oriented by particle swarm optimization strategy for PID tuning”, Proceedings of the 8th IEEE international conference on Computational intelligence in robotics and automation, 2009, pp. 445-450. Rajinikanth.V., Latha. K., “Bacterial Foraging Optimization Algorithm based PID controller tuning for Time Delayed Unstable System”, The Mediterranean Journal of Measurement and Control, Vol 7, Issue 1 , pp. 197-203, 2011 V. Rajinikanth and K. Latha, “I-PD Controller Tuning for Unstable System Using Bacterial Foraging Algorithm: A Study Based on Various Error Criterion”, Applied Computational Intelligence and Soft Computing, Hindawi Publishing Corporation, doi:10.1155/2012/329389 Hanning Chen, Yunlong Zhu, and Kunyuan Hu, “Cooperative Bacterial Foraging Optimization”, Discrete Dynamics in Nature and Society, Hindawi Publishing Corporation, doi:10.1155/2009/815247. Hanning Chen, Yunlong Zhu, and Kunyuan Hu, “Adaptive Bacterial Foraging Optimization”, Abstract and Applied Analysis, Hindawi Publishing Corporation, doi:10.1155/2011/108269. Rajinikanth,V., and Latha, K.,“Identification and Control of Unstable Biochemical Reactor”, International Journal of Chemical Engineering and Applications, Vol 1, Issue 1, 2010, pp.106 -111. Chan-Cheng Chen, Hsia-Ping Huang and Horng-Jang Liaw. “SetPoint Weighted PID Controller Tuning for Time-Delayed Unstable Processes”, Ind.Eng.Chem.Res, Vol. 47, No. 18, 2008, pp.6983-6990. Niola, V., Quaremba, G., Amoresano, A., A study on infrared thermography processed trough the wavelet transform, (2009) Proceedings of the 8th WSEAS International Conference on System Science and Simulation in Engineering, ICOSSSE '09, pp. 57-62.


Department of Electronics and Instrumentation Engineering, St. Joseph’s College of Engineering, Chennai - 600119, Tamilnadu India. E-mail: [email protected] 2 Department of Instrumentation Engineering, Madras Institute of Technology Campus, Anna University, Chennai- 600 044, Tamilnadu India. E-mail: [email protected]

V.Rajinikanth received B.E. in Instrumentation and Control Engineering from the University of Madras (1999) and M.E. in Process Control and Instrumentation from Annamalai University (2002). Currently he is working as an Assistant Professor in the Department of Electronics and Instrumentation Engineering in St.Joseph’s College of Engineering, Chennai. His research interests are Unstable System Identification and control, and Soft computing. Dr. K. Latha received Ph.D in Electrical Engineering from the Anna University, Chennai in 2006. Currently she is working as an Associate Professor in the Department of Instrumentation Engineering, Madras Institute of Technology Campus, Anna University, Chennai, India. Her research interests are Process modeling, Dynamic system Modeling and Control, and Soft computing.



995


Optimization of Deep Drawing Process Parameters Using Design of Experiments J. Santhakumar1, G. Arunkumar2 Abstract – Sheet metal forming is a technique by which most automotive body parts are produced in automobile industries. In sheet metal forming, a thin blank sheet is subjected to plastic deformation using forming tools to confirm to a designed shape. During this process, if the process parameters are not selected properly, the blank sheet is likely to develop defects. Optimization of process parameters in sheet metal forming is an important task to reduce manufacturing cost. The purpose of this investigation is to determine the optimum values of the process parameters. It is essential to find their influence on the deformation behavior of the sheet metal. Four process parameters are taken to optimize the deep drawing process, namely Plastic Anisotropy (R), Strain Hardening Coefficient (n), Tensile Strength (TS) and Friction Coefficient (µ). Numerical simulation is nowadays a modern engineering practice for sheet metal product and tool design developments using the finite element method. Present work examines the accuracy of numerical simulation results of deep drawing testing of steel sheets using the finite element software (Ls-Dyna) to reproduce the design of experiment (DOE). Three level and four factors are used in the DOE. Historically, sheet metal formability has been assessed by tensile testing and biaxial stretching such as the Erichsen simple test. Lately, the concept of experimental forming limit curve for strains, Forming Limiting Curve (FLC) and the numerical simulations were developed to evaluate sheet metal formability and it’s forming operations by predicting the onset of local necking and fracture. In the present work, we use ANOVA method to compute the influence and contribution of the above parameters. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Deep Drawing, LS-Dyna, DOE, ANOVA, Numerical Simulation

I.

All these key elements must be considered simultaneously in a study of formability because they all interact in the process of forming parts.

Introduction

Deep drawing is a metal forming process in which sheet metal is stretched into the desired part shape. A tool pushes downward on the sheet metal, forcing it into a die cavity in the shape of the desired part. The tensile forces applied to the sheet cause it to plastically deform into a cup-shaped part. Deep drawn parts are characterized by a depth equal to more than half of the diameter of the part. These parts can have a variety of cross sections with straight, tapered or even curved walls but cylindrical or rectangular parts are most common [9]. Deep drawing is most effective with ductile metals, such as aluminum, brass, copper and mild steel. Examples of parts formed with deep drawing include automotive bodies and fuel tanks, cans, cups, kitchen sinks, pots and pans. I.1.

Sheet Metal Formability

Fig. 1. Schematic diagram of Forming Limit Diagram

Formability can be defined as the ability of a sheet metal to be deformed by a specific sheet metal forming process from its original shape to the defined final shape without failure. The three key elements of this definition are Material, Process and shape.

Fig. 1 shows Forming Limit Diagram (FLD) indicates the limiting strains that sheet metals can sustain over a range of major to minor strain ratios which can potentially lead to the formation of wrinkles [4]. The FLD shows the strain combinations that produce



996

J. Santhakumar, G. Arunkumar

TABLE II L9 ORTHOGONAL ARRAY OF TAGUCHI METHOD [4] Column numbers of factors assignment Experiment No A(r) B(n) C(TS) D(µ) 1 1 1 1 1 2 1 2 2 2 3 1 3 3 3 4 2 1 3 2 5 2 2 1 3 6 2 3 2 1 7 3 1 2 3 8 3 2 3 1 9 3 3 1 2

instability or fracture and those which are permissible in forming operations. The location of Formability Limiting Curve (FLC) is determined by (FLC0), which is defined as the lowest point on the FLC and typically occurs in plain strain deformation (when minor strain is zero). FLC0 is a function of initial shell thickness and “n” value and it is given by FLD = n / [23.3 + 14.1 t] The “n” value has greater impact on value of FLC0 than the thickness (t). Measured strains that plot higher than the FLC are said to be in the failure zone. These locations indicate are likely will exhibit localized thinning, necking or splitting.

II.

In the present study, Taguchi method of experimental design used to plan the numerical simulations. In Taguchi design, there are many factors both process and material parameters which influence deep drawing process [9]. In this study, deep drawing of LPG bottle was considered as a case and hence the choice of a high corrosion resistant and heat resistant material such as AISI 304 stainless steel. Among all process parameters, Plastic Anisotropy (r), Strain Hardening Coefficient (n), Tensile Strength (TS) and Friction Coefficient have played an important role in the quality of the formed component and hence the above parameters are considered in this study. To evaluate, three levels are chosen for each parameter. The levels are based on the process window and confirm to low, medium and high feasible values. Other process parameters such as punch radius, draw depth etc., were fixed to the recommended values in the simulations. Table I shows the chosen process parameters and their levels used in the FE simulations. The high order interactions among the above four factors are assumed negligible and the information on the main effects can be obtained by running 34 = 81 experiments. However, the appropriate Taguchi orthogonal array for the above four parameters with three levels is L9 to conduct nine simulations, Table II. The first column represents the number of simulation and the subsequent columns represent the process parameters and the rows represent simulations with the levels of each parameter.

Taguchi Technique

Finite element method combined with Taguchi technique form a refined predictive tool to determine the influence of forming process parameters [9].The Taguchi method employed to identify the relative influence of each process parameter considered in a study. Taguchi method has been applied in forming studies to design the experiments and determine the influence of process parameters on characteristics of the formed part. However, questions concerning the influence of factors on the variation of thickness in terms of discrete proportion can only be obtained by performing analysis of variance (ANOVA). Although average calculations are relatively simple, it does not capture the variability of results with a trial condition. The following process parameters are selected and their levels have represented in Table I. [4] TABLE I Process parameter A. Plastic Anisotropy (r) B.Strain Hardening Coefficient (n) C.Tensile Strength (TS) (MPa) D. Friction Co-efficient (µ)

II.1.

1 1.35

Level 2 1.90

3 2.35

0.19

0.23

0.27

274

333

392

0.10

0.17

0.25

Ex. no 1 2 3 4 5 6 7 8 9

Approach of Orthogonal Arrays

An experiment during the product design stages involves the materials used in manufacturing the experimental product which affects the final quality outcome. Factors such as variations in the chemical ratio, the level of ingredients used and how the product is formed together will contribute to the variation in the targeted value of the final product. Orthogonal Arrays (OA) are a special set of Latin squares constructed by Taguchi to layout the product design experiments [4],[9]. By using this table, an orthogonal array of standard procedure can be used for a number of experimental situations. Consider common 3-level and 4 factors OA as shown in Table II.

TABLE III L9 FACTORS FOR CUP DRAWING Plastic Strain Tensile anisotropy hardening strength 1 0.25 274 1 0.30 333 1 0.35 392 1.4 0.25 333 1.4 0.30 392 1.4 0.35 274 1.8 0.25 392 1.8 0.30 274 1.8 0.35 333

Friction coefficient 0.10 0.17 0.25 0.25 0.10 0.17 0.17 0.25 0.10

The experiments are designed with various combinations of process parameter levels, FE simulations were carried out to predict the deformation behavior of the blank sheet. The results obtained from the FE simulations were treated using statistical approach



997


namely, ANOVA method. The purpose of using ANOVA is to elucidate the parameters that govern the deep-drawing process that markedly influence the thickness distribution [9]. This will yield information about the impact of each parameter on the results predicted by the finite element method. Consequently, the degree of importance of each process parameter in the deformation behavior of the blank sheet can be determined. Fig. 3. Solid modeling of deep drawing

III. Deep-Drawing Simulations of Stainless Steel Cup FEA simulation provide insights on the necessary process parameters loading paths of part geometry and part formability by analyzing the thinning, thickening and strain distribution in the deformed sheets. This information facilitates to optimize the process parameters as well as part geometry modification. The Geometric model is generated using SOLID WORKS and the same is represented in Fig. 3.The geometric model is meshed in HYPERMESH and the FE model is represented in Fig. 4. FE simulation of deep drawing process has been performed using a dynamic explicit commercial FE code LSDYNA3D and preprocessing done using Hyper form. Fig. 2 shows geometry of forming tools and Fig. 4 shows a FE model of deep drawing setup using Hyper form. The geometrical FE model, where the Z-axis denote the axial direction of the sheet. The sheet was located above the die and axial load was given by punch. The sheet was meshed and represented by a total of 8947 nodes and 8730 elements (342 in triad type and 8388 in quad type). Both the punch and die were modeled as rigid elements. The other two parameters required for the simulation are the Elastic modulus (E) and Poisson ratio (n) which are 200Gpa and 0.3 respectively. After completing meshing, boundary conditions were applied. All the degrees of freedom (DOF) have applied for die. The punch has allowed moving only in z-axis and all other axis is constrained. Load conditions were applied using deep drawing setup option in the pre-processor, which includes velocity of the punch, total punch travel initial and final force of the punch have applied. The model imported and solved in LS-DYNA. The Fig. 2 shows the geometry of forming tools [9].

Fig 4. FE model of deep drawing setup using Hyper form

An austenitic grade AISI 304 stainless steel is used for this study. The work hardening behavior has been considered isotropic with the plastic anisotropy described by the Hill48’s quadratic yield criterion. The elastic properties are Young’s modulus (E) and Poisson ratio. The values are 200 Mpa and 0.33 respectively. The hardening parameters are strength coefficient (K) and the strain-hardening exponent n. The values are 1330 MPa and 0.35 respectively. The initial blank sheet has a radius of 320 mm and a thickness of 4 mm. The blank is meshed with eight node solid finite elements. An inplane average FE mesh size of 7 mm was used with two layers through thickness. III.1. Finite Element Analysis Results This simulation is performed using AISI 304 Stainless steel, young’s modulus is 200 GPA. This simulation is solved in dynamic explicit solver Ls-Dyna (970). Optimal values of process parameters and their shell thickness results in high quality parts are represented in Fig. 5. III.2. FLD Plot for Deep Drawing Process The Fig. 6 shows the first part simulation results for the axisymmetric cup. The first part simulation result ensures the formability of axisymmetric cup is in safe limit with reference to Fig. 1

IV.


IV.1. Design of Experiments Results Design of experiments of deep drawing process conducted with L9 Orthogonal Array. The total no of levels used is three and total no of factors used is four.

Fig. 2. Geometry of forming tools



998


Fig. 5. Shell Thickness plot

Optimal values of process parameters and their combination results in high quality parts show the thickness variation in the deep drawing part in all axis (X, Y and Z) direction. Wall thinning occurs at the bottom of the cup and thickening occurs near the top of the flange. At the top of the cup section and at the flange, blank thickening occurs due to the friction at die blank sheet interface and the circumferential forces. The thickness variation is limited in experiments where a favorable combination of process parameters and their levels were used. The maximum thinning was observed to be less than 16% of the initial thickness. The Table IV shows the minimum shell thickness value based on the simulation results. The Table V shows the mean value of the deep drawing process for the three levels and four factors.

Fig. 6. FLD Plot

The DOE orthogonal arrays for deep drawing process are shown in the Table IV. The quality characteristics have been chosen for the experiment should reflect as accurately as possible the design parameters under study. Thickness is one of the major quality characteristic in sheet metal formed part, especially for the studied geometry which is analogous to cup forming. The thickness is unevenly distributed in the part after deep drawing [9]. The thickness is uniform at the bottom face of the punch, minimum at the punch radius and vertical surface and thicker at the flange area. Existence of thickness variation from the production stage may cause stress concentration in the part leading to acceleration of damage mechanism.

Ex. no 1 2 3 4 5 6 7 8 9


TABLE IV TAGUCHI ORTHOGONAL TABLE (L9) Plastic Strain Tensile Friction anisotropy hardening strength coefficient 1 0.25 274 0.10 1 0.30 333 0.17 1 0.35 392 0.25 1.4 0.25 333 0.25 1.4 0.30 392 0.10 1.4 0.35 274 0.17 1.8 0.25 392 0.17 1.8 0.30 274 0.25 1.8 0.35 333 0.10

Thickness 1.8400 0.4996 0.1753 0.2070 2.8270 1.0990 3.1210 0.4391 3.1270


999


Level 1 2 3 Delta Rank

TABLE V MEAN TABLE OF DEEP DRAWING PROCESS Plastic Strain Tensile anisotropy hardening strength 0.8383 1.7227 1.126 1.3777 1.2552 1.2779 2.229 1.4671 2.0411 1.3907 0.4674 0.9151 2 4 3

Friction coefficient 2.598 1.5732 0.2738 2.3242 1

IV.2. ANOVA Calculation of results and determination of average for factor level effects involve simple arithmetic operations, which produce answers to major questions that were unconfirmed in the preliminary stages of the investigation. However, questions concerning the influence of factors on the variation of thickness in terms of discrete proportion can only be obtained by performing analysis of variance (ANOVA) [9]. The traditional method of calculating average factor effects and thereby determining the desirable factor levels is to look at the simple averages of the results. A better way to compare the sheet metal behavior to deep drawing is to use the mean squared deviation, which combines effects of both average and standard deviation of the results. The percentage contribution of individual process parameter on the deep-drawing process can be calculated by one way ANOVA. The same have been determined using Minitab software. The Table VI shows the percentage contribution the process parameters on the deep drawing process.

Source Degree of freedom Sum of squares Mean square F ratio % contribution

Plastic anisotropy

TABLE VI Strain hardening

Tensile strength

Friction coefficient

2

2

2

2

2.95

0.33

1.44

8.141

1.47 1.47

0.16 0.08

0.72 0.38

4.070 5.17

22.93%

2.56%

11.22%

63.29%

Friction Co-efficient (63.29%), Plastic anisotropy (22.93%), Tensile strength (11.22%), Strain hardening co-efficient (2.56%). From deep drawing simulation, it is established that the friction coefficient and the plastic anisotropy parameter at the tool sheet interface are the major factors affecting the deep draw ability. The tensile strength and the strain-hardening coefficient are not as important. The analysis of variance by the use of Taguchi method shows that the most significant parameter is the friction coefficient.

References [1]

Taguchi, G and Konishi, S Taguchi method, orthogonal arrays and linear graphs, Tools for Quality Engineering, American Supplier Institute, 35–38 (1987). [2] Duan, X and Sheppard, T, Influence of forming parameters on the final subgrain size during hot rolling of aluminium alloys, J. Mat. Process Technol. 245–249 (2002). [3] Lee, S.W, Study on the forming parameters of the metal bellows, J.Mater. Process. Technol. 47–53 (2002). [4] Park, K and Kim, Y, The effect of material and process variables on the stamping formability of sheet materials, Journal of Mater. Process. Technol. 64–78 (1995). [5] Colgan, M and Monaghan, J, Deep drawing process: analysis and experiment, J. Mater. Process. Technol. 132 (2003) 35–41. [6] Obermeyer, E.J and Majlessi, S.A, A review of recent advances in the application of blank-holder force towards improving the forming limits of sheet metal parts, J. Mater. Process. Technol. 222–234 (1998). [7] Yoshihara, S and Manabe H Nishimura, Effect of blank holder force control in deep drawing process of magnesium alloy sheet, J. Mater. Process. Technol. 579–585 (1998). [8] Leu, D.K, The limiting drawing ratio for plastic instability of the cup drawing process, J. Mater. Process. Technol. 168–176 (1999). [9] Padmanabhan, R, Influence of process parameters on the deep drawing of stainless steel, Finite Elements in Analysis and Design 1062 – 1067 (2007). [10] Sheng, Z.Q and Jirathearanat T. Altan, Adaptive FEM simulation for prediction of variable blank holder force in conical cup drawing, Int. J.Mach. Tools Manuf. 487–494 (2004). [11] Sriram Srinivasan and R.Sathyanarayan Optimization of Tool Life Using Linear Regression Analysis, International Review of Mechanical Engineering, Vol. 6 N. 3, pp. 405-410 (2012).

Authors’ Information V.

Conclusion

The use of FEM with Taguchi technique to determine the proportion of contribution of four important process parameters in the deep drawing process namely Plastic Anisotropy (R), Strain Hardening Coefficient (n), Tensile Strength (TS), Friction Coefficient (µ). Finite element simulations using Ls-Dyna 3D for deep drawing are carried out to study the factor effects on thickness and formability of deep drawing process. Taguchi technique forms an effective tools combination to predict the influence of process parameters. The analysis of variance (ANOVA) carried out to examine the influence of process parameters on the quality characteristics (thickness variation) of the circular cup and their percentage contribution are calculated.


Dr.G.Arunkumar is working as Professor and Head of Mechanical Engineering at Saveetha School of Engineering, Saveetha University, Chennai – 602 105, Tamilnadu, India. Author has been published many papers in refreed International, National journals and conferences. E-mail: [email protected] Mr.J.Santha Kumar is working as Assistant Professor of Mechanical Engineering at SRM University, SRM Nagar, Chennai, Tamilnadu, India. E-mail: [email protected]


1000


Finite Element Simulation and Experimental Evaluation on Superplastic Forming Process of Aluminium Alloy Sheet G. Kumaresan1, K. Kalaichelvan2

Abstract – Superplasticity is the ability of certain materials to undergo large elongation at the proper temperature and strain rate. This work focused on enhance superplastic behavior of Aluminium alloys 7075 by uniform thinning at reduced flow stress and optimum strain rate at constant temperature. The experiments were conducted to obtain a rectangular shaped superplastically formed component by varying process parameters such as strain rate, temperature and pressure. This work is made to explore the superplastic deformation behavior of the aluminium alloys 7075 into a rectangular shaped product by numerical simulation using standard finite element code ABAQUS. It is shown that the proposed optimization approach compares the characterics of deformation and failure during superplastic forming and to find suitable process parameters such as temperature, pressure and time for enhanced uniform thinning. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Finite Element Modeling, Superplastic Forming, Rectangular Box Forming, Thinning, Optimization

I.

Introduction

Superplastic alloys can be formed by such bulk deformation processes as compression molding, closeddie forging, coining, hubbing, and extrusion. Sheet forming of these materials can also be carried out using such operations as thermoforming, vacuum forming, and blow forming. Superplastic sheet metal forming allows the production of complex parts that are not formable under normal conditions.Superplastic materials are polycrystalline solids which have the ability to undergo large uniform deformation prior to failure, and elongation in excess of 200% is usually indicated as superplasticity. Superplasticity is a neck free elongation process at very low flow stress by maintaining temperature to be above half of the melting point of the materials. Superplasticity is observed in fine grained alloys (grain size less than 10µm) at controlled strain rate (ranging from 10-4 to 10-2S-1) [1]-[4]. Aluminium and titanium alloys are commercially used in air craft and automobile industries because of their property of light weight and high strength. Hence the main application is the development of complex shaped parts with simple tooling. The problems related to the non-uniform thickness distribution and cavitations often occur during the superplastic forming of Al alloy sheets, leading to a degradation of the mechanical properties of the superplastically formed parts. The most common constitutive model to describe the behaviour is the material power law given by: σ=k

m


where σ is effective flow stress (N/m2), is strain rate, (s-1), K is material constant depends upon temperature and grain size and m is strain rate sensitivity index. The value of ´m´ has a controlling influence on the stability of superplastic flow. The value of ´m´ lies between 0.3 and 0.9 for most of the superplastic materials [5]-[10]. The high value of ´m´ imports to the material resistance to localized deformation such as necking and thinning so that the material can undergo large deformation without failure. A high value of ´m´ causes the flow stress to be highly sensitive to the strain rate. An ideal value of m=1 would correspond to Newtonian viscous flow which leads to complete neckfree tensile deformation [14]. An attempt has been made by the present author to develop close form solutions of the rectangle shape. In order to prove the correctness of the Finite Element model, experiment works have been performed with 7075 Al alloy sheets. In this study an attempt was made to study the various process parameters in the prediction of the thinning characteristics in a rectangular die.

II.

Finite Element Simulation

Superplastic forming is a complicated process involving large strain and material non-linearity. The deformation is dependent on boundary conditions. The superplastic behaviours of materials is characterized by the rigid viscoplastic. The simulation of superplastic forming is performed ABAQUS. The behaviour of superplastic is generally characterized by a relationship


1001

G. Kumaresan, K. Kalaichelvan

between the von-mises equivalent stress and the equivalent strain rate. II.1.

Configuration of the Sheet and Die

The biaxisymmetric models were performed to reduce the complexity of the analysis and to reduce computational time to determine the influences of various factors on the thickness distribution of the model. The Die is an analytical rigid body and the sheet is deformable body with shell element. Die and blank assembled view as shown in the Fig. 1. Meshing and visualization view as shown the Fig. 2. Fig. 3. First stage forming

Fig. 1. Die and Blank assembled view

Fig. 4. Second stage forming

III. Experimental Work III.1. Selection and Processing of Specimen Material

Fig. 2. Meshed view

II.2.

Contact Algorithm

Superplastic blow forming process, contact between the sheet and the die results in friction. The contect algorithm between the die and the sheet is simulated by means of a rigid surface and a deformable surface. The contact between sheet and die result in changes in contact pressure during forming and the analysis consider the modeling with the cotact algorithm. The contact pressure clearance relationship between the sheet and die was established to monitor the contact interaction. Friction between the die and work piece is modeled using coulomb friction model. During forming the first stage and second stage as shown in Figure 3 and Fig. 4 respectively.


The 7075 Al alloy has been used in this study and the analysed chemical composition (in wt %) 5.6-Zn, 2.5Mg, 1.6-Cu, 0.23-Cr, balance Al. The as - received plates of 5mm thickness were homogenized at 500˚C for 1 hour. The fine grain microstructure in aluminium alloys required for superplasticity could be obtained through static or dynamic recrystallization. Static recrystallization forms a fine grain structure prior to superplastic deformation, where as dynamic recrystallization forms a fine grain structure during the early stages of superplastic deformation [11], [12]. A static thermomechanical treatment process was carried out to produce a very fine grained microstructure, by using the modified Taharsahraoui method [13]. The thermomechanical treatment process parameters used in this study are shown in Table I. TABLE I THERMOMECHANICAL TREATMENT PROCESS Conditions Stage Temperature Time Solution treatment Overaging Warm rolling Recrystallization Aging

500°C 380°C 180°C 500°C 180°C

1h Furnace cooling to 380°C 2h Furnace cooling to 190°C 65-85% Reduction of thickness 0.5 h Water quench 0.5h Water quench


1002


III.2. Experimental Setup The experimental setup consists of a split furnace, an air compressor with tank, and a thermocouple to measure the die temperature. The forming chamber consists of top and bottom dies, and the recess is provided in the bottom die to hold the thermomechanically processed 2mm thick sheet blank. The top die is a rectangular shape. The superplastic forming process experimental setup is shown in Fig. 5.

than by the elongation of the grain them. The sample 1 that was formed under constant pressure of 0.2MPa and the forming time of 120 minutes, and the sample formed under this pressure and forming time having comparatively lower thinning factor of 0.96 compared to the sample 2. The sample 2 that was formed under constant pressure of 0.2MPa and the forming time of 130 minutes, and the sample formed under this pressure and forming time having comparatively higher thinning factor of 0.98 compared with the samples 1 and 3. The sample 3 that was formed under constant pressure of 0.2MPa and the forming time of 140 minutes, and the sample formed under this pressure and forming time having comparatively lower thinning factor of 0.94 compared with the samples 1 and 2. From this experiments the sample 2 gives high thinning factor because the 130 minutes forming time gives the better formability, after increased the forming time the grains are not stable due to the high forming time. Figure 6 and 7 shows the formed component and the different places of the thickness measurement in the components respectively. Table II exhibits the thinning factor of sample 1, 2 and 3. Fig. 8 shows the Thickness distribution of the formed component.

Fig. 5. Experimental setup

III.3. Experimental Procedures The specimens used for the bulge forming were 80mm in diameter. The bulge forming was conducted by placing the specimens over a stainless steel die, the constant pressure bulge forming were performed at 530˚C under selected gas pressure of 0.2MPa. Sample 1 was formed under a forming time of 120 minutes, sample 2 was formed under forming time of 130 minutes and the sample 3 was formed under a forming time of 140 minutes. The deformed samples were taken out from the die setup, and the thickness distribution was measured, by using a Digital micrometer.

IV.

Fig. 6. Formed component


IV.1. FEM (ABAQUS) Result FE Simulation (ABAQUS) carried out using visco plastic material property, sheet thickness using 2mm, many numbers of experiment were generated and visualization, to vary the process parameters such as temperature, pressure and forming time. Finally to optimize the process parameters for temperature is 530°c, Pressure is 2 bar (0.2Mpa) and forming time is 120 minutes.

Fig. 7. Thickness measurement in the component

Sample no

IV.2. Experimental Result Strain is accumulated during superplastic deformation primarily as a result of grain boundary sliding, rather


1 2 3

TABLE II THINNING FACTOR OF SAMPLE 1, 2 AND 3 Thickness measurement at Average various point in mm thickness in Point1 Point 2 Point 3 mm 1.892 1.765 1.828 1.828 1.74 1.695 1.721 1.719 1.89 1.73 1.862 1.827

Thinning factor 0.96 0.98 0.94


1003


[8]

[9]

[10]

[11]

Fig. 8. Thickness distribution of the samples

V.

[12]

Conclusion

[13]

A detailed FE analysis of superplastic forming was carried out to find an optimum process parameter of such as, temperature and pressure. FE analysis identified the optimum process parameter such as, 530 °C and 0.2MPa. Based on this results the first experiment were carried out with in 120min (forming time) to obtained the forming in rectangular shape at high thinning factor of 0.96. The second experiment was carried out at 530°C and 0.2 MPa after 130min (forming time), to obtained the good forming, when compared to the first sample and very high thinning factor of 0.98. The Third experiment was carried out 530°C and 0.2 MPa after 140min (forming time), to obtained the good forming, when compared to the first sample and less thinning factor of 0.94 In the study of FEA simulations results and experimental results, the variations occurs from FEA results to experimental results on the forming time is 7.69%, this small variation may be due to the selection of material model and conditions.

[14]

N. Mezghani, H. Salhi, M. Ayadi, A. Cherouat, Experimental and numerical simulation of hydroforming process, International review of mechanical engineering 2, 839-844, 2008. J. Rezaeepazhand, N. Irani, L.Pahlavan, Comparison of explicit and implicit finite element estimation of springback in metal sheets stamping, International review of mechanical engineering 3, 670-673, 2009. R. Alipour, finite element analysis of elongation in free explosive forming of aluminum alloy blanks using CEL method, International review of mechanical engineering 5, 1039-1042, 2011. Smolej, M.Gnamus, E.Slacek, The influence of the thermo mechanical processing and forming parameters on superplastic behaviour of the 7475Aluminum alloy, Journal of Materials processing Technology 118, 397-402, 2001. Prabhakar Reddy, Abhijit dutta, Amit kumar, M.Komaraiah, Thermo mechanical treatment of 7010 Al alloy for achieving fine grain size, 20th AIMTDR, Birla institute of technology, 743-747, 2003. Taharsahraoui, Mohamedhadji, Nacerbacha, Riadbadji, Superplastic deformation behavior of 7075 aluminum alloy, The International Journal of Materials Engineering and performance 12 , 398-401, 2003. Amoresano, A., Niola, V., Langella, F., Experimental analysis of the behavior of the droplets of high viscous fluids impacting on a flat heated surface, (2011) 8th WSEAS Intl. Conf. on Fluid Mechanics, FM'11, 8th WSEAS Intl. Conf. on Heat and Mass Transfer, HMT'11, 8th WSEAS Intl. Conf. on Mathematical Biology and Ecology, MABE'11, pp. 105-110.

Authors’ information G. Kumaresan, M.E, (Ph.D) is a Teaching Research Associate in the department of production technology, MIT Campus, Anna University, Chennai. With over 8 years of teaching experience. His areas of special interest include materials and metal forming. K. Kalaichelvan, Ph.D is a Associate Professor in the department of production technology, MIT Campus, Anna University, Chennai. With over 20 years of teaching experience. His areas of special interest include materials, metal forming, finite element analysis, MEMS and NEMS.

References [1] [2]

[3] [4]

[5]

[6]

[7]

K.A. Padmanaban, G.J. Davies, Superplasticity Springer-Verlag Berlin Heidelberg, New York. 1980. Xing Huilin, Zhang Kaifeng, Qiao Yu, Z.R.Wang, an advanced superplastic sheet – forming machine controlled by microcomputer, Journal of Materials processing Technology 55, 43-47, 1995. Kuniaki Osada, Commercial applications of superplastic forming, Journal of Materials processing Technology 68, 241-245, 1997. A.J.Barnes, Superplastic forming 40 Years and still growing, Journal of Materials Engineering and performance 16, 440-454, 2007. A.Mohammad nazzal, K.Marwam khraisheh, M.Basil darras, Finite element modeling and optimization of Superplastic Forming Using variable strain rate approach, Journal of Materials Engineering and performance 13, 691-699, 2004. O.F.Yenihayat,A.Mimaroglu , H.Unal, Modelling and tracing the super plastic deformation process of 7075aluminium alloy sheet use of finite element technique, The international journal of Materials and Design 26, 73-78, 2005. P.S.Bate, N.Ridley, B.Zhang, S .Dover, Optimisation of the superplastic forming of aluminium alloys, Journal of Materials processing Technology 177, 91-94, 2006



1004


Wall-Models for LES of Channel Flows Taieb Nehari, Lotfi Tefiani, Driss Nehari Abstract – The present paper describes the analysis and the implementation of simple wall models for Large Eddy Simulations of high Reynolds numbers flows. The comparison with direct numerical simulation (DNS) in channel flow at Reτ=950 was carried out. The results obtained from wall models compare well with numerical data. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Turbulent Flow, Large Eddy Simulation, Wall model, Channel

I.

Introduction

The numerical prediction done by using the large eddy simulation (LES), have proven to be powerful prediction tool for a variety of different flows. However, at high Reynolds number the near-wall region contains small vortical structures (streaks) that are dynamically important to the flow, but which have dimensions that scale are comparable with the viscous scale. Unfortunately, for such case the LES should be sufficiently refined to resolve the near wall structures, however, this making it too expensive at high Reynolds numbers. From the last three decades many research have done to overcome this problem by approximating the overall dynamic effects of the streaks on the larger outer scales through an appropriate boundary conditions without resolving the inner viscous region. Where the first grid point is usually taken form the logarithmic layer above the viscous sublayer. Thus, the sublayer is not resolved and the boundary condition must account for all of the effects of the small-scale turbulence generated at the near-wall region. However, the typical no-slip boundary condition is often replaced with a condition on the wall shear stress. The simplest wall stress models are analogous to the wall functions commonly used in Reynolds-averaged Navier-stokes (RANS) approaches except that they are applied in an instantaneous sense. The wall function provides an algebraic relationship between the local wall stress and the tangential velocities at the first grid points. This approach was first employed by Deardorff (1970) [1] for a channel flow simulation. The goal of the present research is to study the applicability of a simple near-wall model, based on the local equilibrium hypothesis and to analyze its effect on the flow dynamic by simulating no-separated unsteady turbulent boundary layer. The LES code used in the present study was the same one developed by Armenio and Piomilli (2000) [2], but modified to use wall stresses from model routines instead of computing viscous stress directly from the wall gradients.


The code uses the standard dynamic SGS model (Germano et al. 1991[3], Lilly 1992 [4]) with test filtering and averaging on horizontal planes. As the turbulent channel flow (or turbulent Poiseuille flow) is simple and well identified by the literature. The wall-model used in the present study is tested for such flow.

II.

Numerical Implementation II.1.

The Dynamic Model

The dynamic model of Germano et al. (1991) [3] is one of the most successful ones which are used for evaluating subgrid-scale (SGS) model coefficients directly from information contained in the resolved turbulent velocity of large eddy simulation. The model is based on the relation between SGS stresses at different scales (the grid scale ∆ and a test filter scale⎯∆ ) and it is expressed by the below identity: Li, j = Ti, j − τ i, j = ui u j − ui u j

(1)

where, τi,j is the SGS stress tensor at scale ∆, Ti,j is the SGS stress tensor at test-filter scale⎯ ∆ (∆=α∆, where α is the ratio filter, hereafter we use α∆ instead of⎯ ∆) and Li,j is the SGS stress tensor defined from scales intermediate between ∆ and α∆. Hereafter, a tilde (∼) denotes the filtering operation at the grid-scale ∆ and a bar (−) denotes the test-filtering at the test-filter scale α∆. Using the Smagorinsky model for the deviatoric part of SGS stresses we obtain:

τ i,Dj = −2Cs,2 ∆ ∆ 2 S Si, j

(2a)

2 Ti D, j = −2Cs2,α∆ (α∆ ) S Si , j

(2b)


1005

T. Nehari, L. Tefiani, D. Nehari

Replacing the above equations in eq. (1) yields an error in that identity, the least-squares procedure of Lilly (1992) [4] can then be used to yield the Smagorinsky coefficient: < Li , j M i , j > Cs2,∆ = (3) < Mi, j Mi, j >

account the contribution of the two horizontal components of the velocity to the wall stresses for complex geometry flows in the horizontal plan. Following the formulation of the last authors and taking into account that the shear stress at the wall is defined as τw=ρuτ2 (where uτ is the friction velocity and ρ is the density of the fluid) hence:

where Mi,j is given by: M i, j = 2∆ 2 ⎡ S Si, j − α 2 β S Si, j ⎤ ⎢⎣ ⎥⎦

(4)

β is a parameter that account for possible scale dependence of Cs,∆ and is defined as:

β=

Cs2,α∆ Cs,2 ∆

(6)

The results of the semi-dynamic scale-dependent model used by the authors have shown better prediction of the neutral atmospheric flow over heterogeneous land surface. II.2.

The Wall-Model Implementation

The technique of wall modeling was developed in order to use the LES in practical application. By assuming that the turbulence is isotropic in the near-wall layer, Deardorff (1970) [1] has implemented constraints on wall-parallel velocities in terms of the wall-normal second derivatives in order to force the LES to satisfy the log-low in mean. Later, Schumann (1975) [6] has proposed a simple algebraic model by assuming that the wall stresses were in phase with the velocity at the first off wall grid point and that the deviation from their mean was proportional to the deviation of the velocity from its mean. Piomelli and al. (1989) [7] have slightly modified this model taking into account the structure and orientation of the elongated structures present in the near-wall region. Recently Bou-Zeid et al (2005) [8] based on the modified wall-model of the last authors have proposed a modified-new-model that take into Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

2

(7)

here < > represents the horizontal plan average and Vm is the mean value of the velocity in the horizontal plan at the first grid point location Y: Vm2 = um2 + wm2

(5)

Porté-Agel et al. (2000) have implemented a fully dynamic formulation where the scale dependence parameter β is computed through an additional filtering operation (at second test filtering α2∆ by using α=2). This approach uses a planar averaging dynamic model for atmospheric boundary-layer flows. The results obtained by the authors more are accurate than the traditional dynamic model for wall boundary layer. Recently Bou_Zeid et al. (2004) [5] have fitted the profiles of β obtained by last authors along the wallnormal direction by a simple exponential equation:

β = F ( y ) = 1 − 0.65e

⎛V ⎞ < τ w / ρ >= ⎜ m+ ⎟ ⎝U ⎠

(8)

um and wm are respectively the plane average at the first grid point of the streamwise and spanwise velocity components: um =

wm =

1 Lx Lz

1 Lx Lz

∫ ∫ u ( x,Y ,z ) dxdz

(9a)

Lx Lz

∫ ∫ w ( x,Y ,z ) dxdz

(9b)

Lx Lz

On the other hand, U+ is the mean value (in a Reynolds averaging sense) of the non dimensional velocity at the first grid point and is calculated assuming the presence of the Log-profile as follows:

( )

U + = 2.5 ln Y + + 5.2

(10)

where the wall-unit at the first grid point is defined as Y + = Yν / uτ , and ν is the molecular viscosity of the fluid. Subsequently, the stress is partitioned into its streamwise and spanwise components as usually:

τ12,w ( x, y ) = τ 32,w ( x, y ) =

u ( x,Y ,z ) Vm w ( x,Y ,z ) Vm

(11)

(12)

where τ13,w ( x,z ) and τ 23,w ( x,z ) are the instantaneous wall stress, that are a function of the streamwise and spanwise coordinates. In the other hand, u ( x,Y ,z ) and w ( x,Y ,z ) are respectively the resolved streamwise and

spanwise components of velocity at the first grid point off the wall. Note that when the statistical steady state is


1006


reached < u ( x,Y ,z ) >= uτ U + and the Reynolds average value of the wall stress is equal to the imposed one. The vertical velocity (vw(x,0,z)=0) is forced to be zero at the wall. The modified model employed in the present research uses Eqs. (11) and (12), in conjunction with a dynamic model for modeling the SGS stress in the core region. As well known, the Shumann’s model requires the knowledge of the external, imposed pressure gradient that must be in balance with the wall shear stress. Grotzbach (1987) extended the Shumann’s approach to flows where the pressure gradient is not known a priori, computing the mean velocity Vm of Eq. (1) by averaging the resolved velocity at the first grid point over the plane of homogeneity. The wall stress was thus calculated using the log-law and solving the below equation (Eq.(13)) recursively for uτ : ⎛ Yν Vm = 2.5 ln ⎜ uτ ⎝ uτ

⎞ ⎟ + 5.2 ⎠

(

(

))

2

(14)

with A=19, and concerning the coefficient κ, it is extracted using Eq. (14) by performing an averaging along homogeneous direction at the first grid point. As for the present case (channel flow) we have horizontal plan homogenous, this averaging is used too for such coefficient:

κ=

(

< ν SGS >

(

< Y + 1 − exp −Y + / A

III. Turbulence Simulation and Results The computational domain considered for the channel flow is (2δπ, 2δ, δπ) in (x, y, z), where x is the streamwise direction, y is the wall-normal direction and z the spanwise direction. The computational grid used in the present research is uniform along the three-space directions (see Fig. 1).

(13)

The model was implemented directly as imposed fluxes in the discretized equation. It required to ad modifications in the code. In particular, the CrankNicolson part of the code has been modified. At the first grid point the eddy viscosity need to be known as boundary condition, the best way to extrapolate such quantity is to use the approach of Wang and Moin (2002). Where, the eddy viscosity νt is obtained from a simple mixing length eddy viscosity model with near wall damping function:

νt = κ Y + 1 − exp −Y + / A ν

fractional-step herein employed, needs a correction to the diffusive flux at the wall that is indeed zero. Second, a subroutine is called to calculate the coefficient κ in order to inject it in the turbulent boundary condition part of the code. The last modification, concern the use of the semi-dependency model (introduction of β coefficient), this one is only included the part of the code where is computed the coefficients Mij.

))

2

(15)

Fig. 1. Typical computational domain for turbulence simulation

The model has been tested against the database of the numerical results found by Juan et al. (2004) [9] that have used DNS method for the prediction of turbulent Poiseuille flow. The performance of the models has been evaluated at the Reynolds number Reτ=950 (where Reτ=uτδ/ν). Four cases were computed, at the same Reynolds number and the same grid but with different values of α and β, in order, to find the best combination to get accurate results both for the mean flow and for its statistics. Table I, summarize the cases investigated without filtering in the present study with their numerical parameters.

>

The implementation of the above arguments in order to improve the performance of the code for the prediction the flows at high Reynolds number need to introduce some modification on it. First, a subroutine is called that calculates the instantaneous values of the wall stresses, as a function of x and z, these values, multiplied by ∆x∆z, are used as diffusive fluxes at the wall in the explicit part of Crank-Nicolson algorithm. The implicit part does not need any modification, since the ∆u formulation of the


Case A1 A2 B1 B2

TABLE I SUMMARY OF THE COMPUTATIONS MADE FOR REτ=950 WITHOUT FILTERING Grid ∆x+ ∆y+ α β 24*32*24 1 248.72 59.38 √6 24*32*24 F(y) 248.72 59.38 √6 24*32*24 1 248.72 59.38 √4 24*32*24 F(y) 248.72 59.38 √4

∆z+ 124 124 124 124

Fig. 2 shows the mean streamwise velocity profiles of four calculations at Reτ=950 obtained by the present method for different values of α and β and compared against the results obtained by DNS method of Juan et al. International Review of Mechanical Engineering, Vol. 6, N. 5

1007


higher near the wall compared with the DNS results for different values of α and β. This is observed mostly in the first 4 points in the vicinity of the wall. Then from the wall-unit Y+∼200 the solutions of the present code start to be agree with the DNS solution until the location Y+∼800, at the core the code overpredicts slightly the accurate numerical solution (DNS). From the plot it appears that the more accurate solution is for α=√4 and β=F(y). Concerning the vrms , the actual numerical solutions are mostly underpredicted from the near wall region to the core by the code. In the other hand, the numerical solution from the present wallmodeling gives in general good prediction for wrms . 3

(a) DNS LESα=6β=1 LESα=6β=F(y) LESα=4β=1 LESα=4β=F(y)

2.5 2

urms/uτ

(2004) [9] and also against the analytic solution predicted by the Log-law profile. The plot demonstrates that the profiles of the velocity obtained agree well with the literature results, in the other hand, they do not exhibit the same shape. They start with higher values at the first grid points near the wall compared the DNS data, because the imposed velocity at the first grid point is of the standard Log-law (Eq. (13)). However, we observe that all the velocity profiles start from the Log solution at the location Y+∼30, then, they start to overderpredict slightly both the analytical and numerical (DNS) solution until Y+∼100. From the location Y+∼100, the numerical solution of the present code starts to over or under predict the literature solutions depending on the coefficients α and β. The plot shows that α=√6 gives always an underprediction of the solution but the solution for the case β=F(y) seems more accurate than of the case β=1. On the other hand, for α=√4 the solution is always overpredicted for the caseβ=F(y), furthermore, for the case both for β=1 we observe that the solution is too close to the analytical solution then from the Y+∼100 it underpredicts the literature solutions especially at the core region. Thus, from the above plot it appears that the dependency case (β=F(y)) with α=√4 gives more accurate solution for the profile velocity.

1.5 1 0.5 0

0

200

400

+

600

800

y

(b)

2

vrms/uτ

1.5

20

1

0.5

15

u

+

DNS LES α=6, β=1 LES α=6, β=F(y) LES α=4, β=1 LES α=4, β=F(y) Log law

10

0

0

200

400

+

600

800

y 2

5

(c)

y

+

250

wrms/uτ

1.5

0

500 750

Fig. 2. The mean velocity profile for different α and β at Reτ=950. Without filtering and with 24*32*24 grid points

1

0.5

In Fig. 3 the variation of the Reynolds stresses normalized by the friction velocity along the normal wall direction obtained by the present method for different values of α and β are shown and compared against the results obtained by DNS results. From the figure we observe in general a good agreement with the accurate numerical solution especially for the streamwise and spanwinse components fluctuation. The plot 2.a shows that the streamwise velocity fluctuation of the present wall modeling are much too


0

0

200

400

y+

600

800

Fig. 3. The Reynolds stresses for different α and β at Reτ=950. Without filtering and with 24*32*24 grid points

The total shear stress is defined as the sum of the viscous stress and the Reynolds stress d − ρ < uv > ). (τ = ρ dy


1008


As well known from the literature of turbulent channel flow, the viscous stress dominates at near the wall region [15]. In contrast, the Reynolds stress dominates far from the wall and the sum of the two contributions normalized by the shear wall follow a linear relationship (τ/τw=-y/h +1). 1 α=6, β=1 α=6, β=F(y) α=4, β=1 α=4, β=F(y)

0.8

τ/τw

0.6

mean velocity and gives more eddy viscosity compared for the first case. Concerning the dependency, when using it, the eddy viscosity is much higher than to the correspondent one computed without dependency (β=1) of order one molecular viscosity. In the other hand, the dependency gives more mean velocity profile and more accurate statistic of the flow. Hence, it seems from the above computations the case of α=√4 with dependency gives accurate solution with adequate statistics of the flow. In order to eliminate the instabilities present in the numerical solution for the other cases we have tried to introduce a special treatment that could refine the solution. 7

0.4

6 0.2

νt /ν

0 0

5 4

0.2

0.4

0.6

0.8

1

y/h

3

Fig. 4. Profiles of the total shear stress for different values of α and β at Reτ=950. Without filtering and with 24*32*24 grid points

α=6, β=1 α=6, β=F(y) α=4, β=1 α=4, β=F(y)

2 1

The profiles of total shear stress normalized by the shear stress at the wall is shown in figure 4 for different values of α and β. From the plot we observe that only the combination (α=√4 and β=F(y)) fit the analytical solution from the core until the location y/h∼0.20 (or y+∼190). The high modulation that appears in the first 4 grid points near the wall for all calculations could ²be due to the presence of certain instabilities created by imposing the Log profile at the first grid point. The modulations that appear from the core to the near wall region of the other solutions especially for β=1 could be due to the presence of certain numerical instabilities that should treated numerically (see the below section) The influence of both the ratio filter and the dependency on the eddy viscosity is shown in figure 5. The plot displays the eddy viscosity normalized by the molecular viscosity of the fluid along the normal-wall direction. We observe from the plot that using α=√6 gives low value of the eddy viscosity compared to the cases using α=√4. In the other hand when using the dependency (β=F(y)) the eddy viscosity is much higher than to the correspondent one computed without dependency (β=1) about plus one molecular viscosity. However, in order to predict high Reynolds number channel flow is better to use the combination of the coefficients α=√4 and β=F(y). As conclusion of the above computations, using α=√6 gives an under predicted mean velocity and an eddy viscosity of order 4 times the molecular viscosity. In the other hand, using α=√4 predicts more accurately the


0 0

0.2

0.4

y/h

0.6

0.8

1

Fig. 5. The eddy viscosity for different α and β at Reτ=950. Without filtering and with 24*32*24 grid points

IV.

Conclusion

In the present paper the work done in modeling the wall stress for LES of high Reynolds number is described. Since we are interested complex geometry flow field, and complex physics related to the presence of unsteady driving pressure gradient, we have decided to use simple and well established wall model and to extend it for the prediction of complex geometry. Considering the flexibility of the above method, and the fact that the cost of the above simulations is one order of magnitude lower than that required by fully resolved simulations, LES can b considered a promising tool in prediction flows of engineering importance.

References [1]

[2]

[3]

Deardorff JW (1970) “A numerical study of three-dimensional turbulent channel flow at large Reynolds numbers” Journal of Fluid Mechanic, 41, 453-480. Armenio V. and Piomilli U. (2000) “A Lagrangian mixed subgridscale model in generalized coordinates” Flow, Turbulence and Combustion 65, 51-81. Germano M., Piomelli U., Moin P., Cabot W. (1991) “A dynamic subgrid-scale eddy viscosity model” Physics of Fluids A 3, 1760-


1009


[4] [5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13] [14]

[15]

1765. Lilly D. K. (1992) “A proposed modification of the Germano subgrid-scale closure method” Physics of Fluids A 4, 633-635 Bou-Zeid E., Meneveau C. and Parlange M. (2000) “Large-eddy simulation of neutral atmospheric boundary layer flow over heterogeneous surfaces: Blending height and effective surface roughness” Water Resources Research, 40, 1-18. Schumann U. (1975) “Subgrid-scale model for finite difference simulation of turbulent flows in plane channels and annuli” Journal Computational Physics, 18, 376-404. Piomelli U., Moin P., Ferziger JH, Kim J. (1989) “New approximate boundary conditions for large eddy simulations of wall bounded flows” Physics of Fluids A 1, 1061-1068. Bou-Zeid E., Meneveau C. and Parlange M. (2005) “A scaledependent Lagrangian dynamic model for large eddy simulation of complex turbulent flows” Physics of Fluids 17, 1-18. Juan C. del Alamo, Javier Jimenez, Paulo Zandonade and Robert D. Moser (2004) "Scaling of the energy spectra of turbulent channels", J. Fluid Mech. 500, 135-144. Juan C. del Alamo and Javier Jimenez (2003) "Spectra of the very large anisotropic scales in turbulent channels", Phys. Fluids 15 No. 6, L41-L44. Lele, S. K. (1992) ‘Compact finite difference schemes with spectral-like resolution’, Journal of Computational Physics, 103(1), 16–42. Porté-Agel F., Menveau C. and Parlange M. B. (2000) "A scaledependent dynamic modl for larg eddy simulation/ application to neutral atmospheric boundary layer", J. Fluid Mech. 415, 261284. Piomelli U. and Balaras E. (2002) “Wall-layer models for large eddy simulations” Annual Review of Fluid Mechanic 34, 349-374 Wang M., Moin P. (2002) “Dynamic wall modeling for large eddy simulation of complex turbulent flows” Physics of Fluids 14, 2043-2051. Amoresano, A., Niola, V., Langella, F., Experimental analysis of the behavior of the droplets of high viscous fluids impacting on a flat heated surface, (2011) 8th WSEAS Intl. Conf. on Fluid Mechanics, FM'11, 8th WSEAS Intl. Conf. on Heat and Mass Transfer, HMT'11, 8th WSEAS Intl. Conf. on Mathematical Biology and Ecology, MABE'11, pp. 105-110.

Lotfi Tefiani born at the 10/09/1979 in Tlemcen, Algeria. He graduated with Bachelor studies in Naval Engineering, University of Science and Technology in Oran (USTO), Algeria in 2005, then a Master Degree in Mechanical Engineering, USTO, Oran, Algeria in 2010. At the present, he is a PhD student Naval Engineering. He works on fluid dynamics. Driss Nehari born at the 16/02/1968 in Ain Temouchent, Algeria. He is currently Professor and researcher at the Institute of Sciences and Technology, University of Ain Temouchent, Algeria. He has refereed many paper journal and conference papers.

Authors’ information Taieb Nehari, PhD student. Laboratory of smart structures, Centre Universitaire d’Ain Temouchent, Ain Temouchent, 46000, Algeria. E-mail: [email protected]. Lotfi Tefiani, PhD student. Laboratory of smart structures, Centre Universitaire d’Ain Temouchent, Ain Temouchent, 46000, Algeria. E-mail : [email protected] Driss. Nehari, Professor Laboratory of smart structures, Centre Universitaire d’Ain Temouchent, Ain Temouchent, 46000, Algeria. E-mail : [email protected]. Taieb Nehari born at the 16/05/1976 in Ain Temouchent, Algeria. He graduated with Bachelor studies in Mechanical Engineering, Technical Normal School of Oran, Algeria in 200X, then a Master Degree in Mechanical Engineering, University of Mostaganem, Mostaganem, Algeria in 2008. At the present, he is a PhD student Mechanical Engineering. He works on heat transfer.



1010


Modelling of Centrifugal Compressor Impellers Using Adaptive Neuro- Fuzzy Inference Systems (ANFIS) Layth H. Jawad, S. Abdullah, R. Zulkifli, W. M. F. W. Mahmood Abstract – In order to model the design parameters of centrifugal impeller it is necessary to find new method for design parameters prediction. Adaptive neuro- Fuzzy Inference System (ANFIS) was used to predict the design parameters. The geometry parameter definition system of radial impellers based on Bezier curves, a typical centrifugal impeller with an inlet and a radial outlet used in this paper comprises a series of three Bezier patches or segments. The hub and shroud blade sections are defined as distributions of camber line; Data were generated from Bezier curves equations. In this paper, is the use adaptive neuro-fuzzy inference systems (ANFIS) approach is developed a design prediction method for radial impellers, to use it as design parameters predictor of radial impeller angles. The results were compared by statistical criterion (RMSE). Considering the results, it is obvious that our proposed modeling by (ANFIS) is efficient and valid and it can be expanded for more general states. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: ANFIS, Centrifugal Compressor Impeller, Design Parameters

I.

Introduction

In the past decades interest has been progressively more devoted to the development of turbochargers. Because of their compact size, large capacity, high performance, and ability to improve volumetric efficiency, turbochargers are broadly used in many applications, such as marine diesel engines, automobile engines, and small gas turbines for aircraft engines [1][17]. The improvement of turbocharger compressor performance and the extension of the stable operating ranges are becoming critical for the viable future of low emission diesel engines. In the case of the centrifugal compressor, it is known that unsteady behaviour becomes apparent when the air mass flow through the compressor is lower than the critical level. This unstable phenomenon is denoted as a surge and corresponds to a backflow of compressed fluid throughout the compressor into its inlet. In general, the performance of a centrifugal compressor is expressed as a relationship between the mass flow rate and the pressure ratio on a line with a constant number of revolutions. Centrifugal compressor impeller for turbocharger is increasingly pushing the limits of efficiency, weight, inertia, compactness and cost effectiveness, these technologies require efficient impellers. Furthermore, the influences of the different diffuser meridian channel width ratios on the compressor performance under design conditions show a remarkable significance in terms of improving the efficiency of the whole machine in a micro gas turbine (MGT) centrifugal compressor [1], [17].


The effect of pulsating flow inside a centrifugal compressor and the corresponding pressure pulses on the compressor surge line can be very important because the pulsating flow is in the 40-67 Hz range (corresponding to characteristic pulsation when boosting an internal combustion engine) which increases the surge margin[2]. The application of CFD to turbocharger compressor characteristic predictions over a range of speeds between 100,000 and 200,000RPM, to develop an efficient methodology for analysing the turbocharger compressor performance, Also to compare the computation versus rig measurements [3]. In addition the stall flow phenomenon inside a turbocharger centrifugal compressor with a vaneless diffuser simulated numerical and the amplitude of the static pressure oscillation at this frequency in the diffuser is increased with reduction in compressor mass flow, the results show that there is a distinct stall frequency at the given compressor speed [4]. An analytical model for the centrifugal compressor was proposed to predict the compressor performance such as outlet pressure, efficiency and losses. The model provides a valuable tool for evaluating the system performance as a function of various operating parameters [6]. The compressor performance map is described experimentally for characterization of the automotive turbocharger and a mathematical tool has been developed for marking out surge operation points from stable compressor points [5]. The contribution to the design methodology and performance assessment of LSVD (low solidity vaned diffusers), to understand the pressure recovery phenomena in each of the three types of diffusers, and the effect of design parameters on performance was Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

1011

Layth H. Jawad, S. Abdullah, R. Zulkifli, W. M. F. W. Mahmood

studied by [7]. Narrowing the impeller flow passage depth (impeller exit width trimming), On the basis of experimental data for two impellers, one with a low flow coefficient and the other with a high flow coefficient, the effect of impeller exit width trimming were studied and discussed and effect on overall performance, blade loading and impeller diffusion was examined by [8]. The effect of the piping systems on the surge characteristics to the design of the compressor was studied and test several centrifugal compressors for turbochargers combined with the different piping systems and investigate the changes of surge characteristics, surge lines which connect surge points on the performance map by [9]. Stable working conditions and surge phenomena were simulated and boundary uses the Method of Characteristics to determine the flow conditions at compressor inlet and outlet. To downsize the engine displacement also to increase the power output and reducing fuel consumption [10].The complex shock waves within the diffuser throat and impeller inlet, respectively, within high-speed compressors. These flow phenomena do not occur in low speed compressors and are very significant in the design of these compressors [11], [12]. Many researchers have indicated that suitable treatments can extend the stable operating range of a turbocharger centrifugal compressor, but the performance is still insufficient under the majority of conditions. The flow inside blade to blade passage is very affected by the angles at the inlet of the impellers. Recently, Artificial Neural Networks (ANN) [16] and Adaptive Network-based Fuzzy Inference System (ANFIS) [14], [15] have become increasingly prevalent in solving engineering problems with a considerable reduction in computational time. ANFIS proves to be a robust approach as it has the ANN superior capabilities as well as the Neuro-Fuzzy architectures. The goal of this work was to develop a design method for radial impellers by using adaptive network-based fuzzy inference systems (ANFIS) instead of Bezier polynomial curve to predict the impellers angles at the leading edge and trailing edge as well as the angle along camber line of the hub and shroud curves camber lines curves.

II.

Meridional Flow Channel Design

the impeller are coincident with the patch boundaries on the hub and the shroud, and the third represents an exit channel.

Fig. 1. Meridional Plane View of a Full Impeller Blade

The impeller blade is defined as a ruled surface of straight lines joining points on the hub and the shroud which are equidistant along the meridional channel of the impeller, between the leading edge and the trailing edge. The hub and shroud blade sections are defined as distributions of camber line and thickness specified as Bezier functions along the normalized meridional length, whereby the leading edge and trailing edge ellipses are defined as separate parameters. The angle distribution along the hub (and a similar equation is used for the shroud) is defined as a Bezier polynomial with three internal points as in (1), as follows: 1

4 6 4

1 1 1

(1)

is the blade angle at the where the Bezier parameter hub curve and (m) in this distribution is the normalized meridional length, which varies linearly from the leading to the trailing edge along the meridional walls of the impeller [13].

III. ANFIS Model

The meridional geometry of the impeller is defined by a hub and shroud camber lines curves (based on a single subroutine in the ANSYS BladeModeler) which can be adapted to represent the different types of meridional channels that can be found, such as a radial impeller with axial inlet, radial impeller with radial inlet. The hub and shroud camber lines curves for a typical centrifugal impeller with an axial inlet and a radial outlet used in this paper comprises a series of three Bezier patches or segments as shown in Fig. 1. The first patch is the inlet channel of the impeller, the second is the impeller itself, where the leading and the trailing edges of


The ANFIS structure combines both the artificial neural network and the fuzzy inference system methodologies. One of the most notable characteristic of ANFIS is its ability to learn complicated relationships based on the pattern data. In order to model nonlinear systems, the ANFIS divides the input space into many local regions. For each local region, a simple local model is developed based on linear functions or even adjustable coefficients. The ANFIS then employs fuzzy Membership Functions (MFs) to divide each input dimension. Several local regions can be activated simultaneously in case of


1012


having the input space covered by overlapping MFs. The resolution of partitioning of the input space plays a very important role in the determination of ANFIS approximation ability. This is determined by the number of MFs in the ANFIS and the number of layers. Selecting the MFs in the ANFIS architecture affects the behaviour of the model. In this study, the Gaussian-Shaped MF Eq. (4) is tested and selected for case as it is associated with the minimum value for the Root Mean Square of Error. Fig. 2 depicts the ANFIS diagram for the three input parameters of blade radius location (R), blade meridional length (M),and blade axial location (Z) to predict the blade angle ( ) determined by the ANFIS. The use of the Root Mean Squared Error (RMSE) between pattern outputs and predicted outputs (with same inputs) is one of the conventional criterions for evaluating the performance of the ANFIS model. Selecting a suitable and complete pattern set for training of the ANFIS is very important. If an incomplete set (without all possible conditions) is selected as the training set for the ANFIS, the ability of the network when it encounters an unknown (not used in the training set) condition will be reduced. For increasing this ability, the training set of the ANFIS must be expanded as much as possible in the whole input-output data set space. In this study, the inputs database are generated as part of this study and used for training the models. In all cases, the input-output data set is divided into two (training and evaluating) subsets randomly. For each case, two thirds of the data is selected as the training subset and one-third as the evaluating subset. Training of the ANFIS is accomplished with the first subset in 1000 epochs (training stage) with hybrid (back propagation of error for nonlinear parameters and the least square Errors for linear parameters) procedure. All the ANFIS simulations are carried out using the ANFIS Code model developed for prediction the blade angle.

The Adaptive-Network-Based Fuzzy Inference System is a fuzzy Sugeno model put in the framework of adaptive systems to assist learning and adaptation. Framework makes the ANFIS modelling more systematic and less reliant on human expert knowledge. For a first order Sugeno fuzzy model, a typical rule set with base fuzzy if-then rules can be expressed as in (2): ,

(2)

The ANFIS’s architecture is shown in Fig. 3 that it has three inputs and one output. This architecture is formed by using five layer and 27 if–then rules as follows: Π

M

w

N

Π

N

Π

N

A2

Π

N

A3

Π

N

A1 R

w1

B1

N

B2

N

B3

N

w

1

f1

∑

f

C1 Z C2 C3

Layer1

Π

Π

N

Π

N

Π

N

Layer2

Layer3

w

27

Layer4

Layer5

Fig. 3. ANFIS Model Structure

Layer 1: Every node i in this layer is a square node with a node function, as in (3):

Blade radius Angle (R) (3)

(3)

anfis (sugeno)

1

where x is the input to node i, and Ai is the linguistic label associated with this node function. In other Words, is the membership function of Ai and it specifies the degree to which the given x satisfies the quantifier Ai. to be GaussianUsually has been choosing shaped with maximum equal to 1 and minimum equal to 0 as in (4):

f(u)

Blade Meridional Angle (M) (3) 27 rules Blade Angle (27)

Blade Axial Location (Z) (3) System anfis: 3 inputs, 1 outputs, 27 rules

Fig. 2. ANFIS Diagram

x

IV.

c 2σ

Structure of ANFIS

The configuration of ANFIS structure used in the present work as shown in Figure 3, which is a first-order Takagi–Sugeno type. The network analyses the system’s output for given input data set through fuzzy if-then rules. The optimal model parameter is determined by hybrid-learning algorithms.


(4)

In the above equations, x is the input of the MFs; c and are the MFs parameters to be learnt, these are the premise parameters. Layer 2: Every node in this layer is a circle node labeled П


1013


which multiplies the incoming signals and sends the product out as in (5): (5) Every node in this layer is fixed. Other T-norm operators that achieve generalized AND can be used as the node function in this layer. Layer 3: It contains fixed nodes, which calculate the ratio of the firing strengths of the rules, labeled N. The ith node analyses the ratio of the ith rule’s firing strength (degree of fulfillment) to the sum of all rule firing strengths. Outputs of this layer will be described normalized firing strengths as in (6): (6) Layer 4: The nodes in this layer are adaptive and perform the consequent of the rules as in (7): (7) where wi is the output of layer 3, and { ji , ki ,mi , ri } is the parameter set. Parameter in this layer to be determined and are referred to as the consequent parameters. Layer 5: The single node in this layer is computes the overall output as the summation of all incoming signals as in (8): ∑ ∑

used in this study to predict blade angles from generated data to save effort and computation time on studies. The prediction results were compared with actual data. The actual data was generated using ANSYS BladeModeler (based on a single subroutine computer code). Fig. 4 displays the flow chart of trained ANFIS model while Fig. 5 shows the initial (before training) and final (after training) membership functions of the same inputs respectively. Checking on the initial and final membership functions shows that there are reasonable changes worthy of being effected in the final membership functions. The initial setup of the ANFIS network model is shown in Table I. Upon getting to the 1000th training epoch, the root mean square error (RMSE) convergence curve of ANFIS exercise was determined as displayed in Fig. 6. The training error got stabilised just after epoch 200. TABLE I CHARACTERISTICS OF THE ANFIS NETWORK The number of inputs and output Input: 3,Output: 1 Activation functions Log-sigmoid Type Of Membership Function Gaussian The number of layers 5 Number of input nodes (n) 3 Number of fuzzy partitions of each variable 3 (p) Total number of membership functions 9 Number of fuzzy rules (pn) 27 Total number of nodes 78 Total number of parameters 126 Max Number of epochs 1000 Initial step size for parameter adaptation 0.01

(8)

Membership functions in the proposed ANFIS topology associated with each of the three inputs, respectively. So the input space is partitioned into three sub-domains, each of which is governed by fuzzy ‘IFTHEN ‘rules. The premise part of a rule (layer 1) defines a fuzzy sub-domain, while the following part (layer 4) specifies the output within this sub-domain. The performances of the model during training and testing were evaluated using a range of standard statistical performance assessment criteria such as, root mean square error (RMSE), as in (9). Root Mean Square of Error: RMSE

∑

x n

y 1

(9)

where x is the target value, y is the output value and n is the number of data in the data set.

V.

Results and Discussions

Adaptive neuro -Fuzzy Inference System (ANFIS) has


Fig. 4. Flow Chart of ANFIS Model


1014


1

mf2

mf1

mf3

mf3

0.8

Degree of membership


0.8

mf2

mf1

1

0.6

0.4

0.6

0.4

0.2

0.2

0

0

10

15

20

25

30

35

40

0

45

5

Blade Radius Location (R) Before Training

mf2

mf1

1

10

15

20

25

30

35

40

45

50

Blade Meridional Length (M) Before Training

mf3


0.8

0.6

0.4

0.2

0 0

5

10

15

20

25

30

Blade Axial Location ( Z) Before Training

(a) Initial mf2

mf3

1

mf2

mf1

1

mf3

mf1 0.8



0.8

0.6

0.4

0.6

0.4

0.2

0.2

0

0 10

15

20

25

30

35

40

0

45

5

Blade Radius Location (R) After Training

10

15

20

25

30

35

40

45

50

Blade Meridional Length (M) After Training

mf2 1

mf3


0.8

0.6

0.4

0.2 mf1 0 0

5

10

15

20

25

30

Blade Axial Location ( Z) After Training

(b) Final Fig. 5(a). Initial and (b) Final Gaussian Membership Function Shape for Three Inputs.



1015


9

x 10

-3

exciting practical use of modelling techniques design of centrifugal impellers. The learning rates and the overall ability of ANFIS become better through these preliminary mechanisms, hence ensuring more merits than the classical approach. Generally, in the construction of ANFIS, the use of data that had been initially worked upon gives a better precision and results compared to data that were not preprocessed. It is evident that the ANFIS can be used to predict new values for Parameters blade design with reduced computation time and without compromising the accuracy.

Root Mean Square Error (RMSE)

8

7

6

5

4

3

0

200

400 600 Epoch number

800

1000

References

Fig. 6. ANFIS Network Error Convergence Curve [1]

Figure 7 shows matching the actual, prediction and validation. Also Fig. 8 shows the actual Blade angle values compared with predicted values. The ANFIS system results show a good matching with the actual data with increased error at a higher number of cycles in the entire data domain. The ANFIS model results are in good agreement with the actual output data.

[2]

[3]

Predicted & Actual and Validated values

28 Predicted Actual Validated

26

[4]

24

[5] 22

20

[6]

18

[7] 16

0

5

10

15 20 Data Set Index

25

30

35

[8]

Fig. 7. Shows the Predicted, Actual and Validated with respect to Data set Index 30

[9]

Predicted values

Predected Actual

[10]

25

[11] 20

[12] 15 15

20

25

30

Actual value

Fig. 8. Shows the Predicted Values Versus Actual Values

VI.

[13]

Conclusion

[14]

The adaptive neuro-fuzzy inference system (ANFIS), a new and innovative soft-computing application improvised method was utilized in prediction blade angles. The method described in this paper provides an Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

Y. Yang,RongXie,Lu-yuan Gong and Yang Hai , Study of Influence of Diffuser Meridian Channel Shape on Performance of Micro-Gas Turbine Centrifugal Compressor, Power and Energy Engineering Conference (APPEEC),Asia-Pacific 978-1-42446255-1/11, 2011. J. Galindo,H. Climent, C. Guardiol, A. Tiseira, On the effect of pulsating flow on surge margin of small centrifugal compressors for automotive engines, Experimental Thermal and Fluid Science,33:1163–1171, 2009. O. Baris, Automotive turbocharger compressor CFD and extension towards incorporating installation effects, Proceedings of ASME Turbo Expo: Power for Land, Sea and Air GT2011, 2011. Q Guo,H Chen, X-C Zhu, Z-H Du, and Y Zhao, Numerical simulations of stall inside a centrifugal compressor, Power and Energy IMechE Vol. 221 Part A: J, 2007. J. Galindo,J.R. Serrano , C. Guardiola, and C. Cervello, Surge limit definition in a specific test bench for the characterization of automotive turbochargers, Experimental Thermal and Fluid Science 30 (2006) 449–462, 2006. W. Jiang,Jamil Khan, and Roger A. Dougal, Dynamic centrifugal compressor model for system simulation, Journal of Power Sources 158 (2006) 1333–1343, 2006. A. Engeda , Experimental and numerical investigation of the performance of a 240 kW centrifugal compressor with different diffusers, Experimental Thermal and Fluid Science,28:55–72, 2003. A. Engeda, Effect of Impeller Exit Width Trimming on Compressor Performance, Proceedings of the 8th International Symposium on Experimental and Computational Aerothermodynamics of Internal Flows, ISAIF8- 00135, 2007. H. Tamaki, Effect of piping systems on surge in centrifugal compressors, Journal of Mechanical Science and Technology 22 (2008) 1857~1863, 2008. J. Galindo, F.J. Arnau, A. Tiseira and P. Piqueras, Solution of the turbocompressor boundary condition for one-dimensional gasdynamic codes, Mathematical and Computer Modelling 52 (2010) 1288_1297, 2010. B. Cukurel, P.B. Lawless, and S. Fleeter, Particle Image Velocity Investigation of a High Speed Centrifugal Compressor Diffuser, Spanwise and Loading Variations, Journal of Turbomachinery, vol. 132, pp. 1-9, 2010. H. Higashimori, K. Hasagawa, K. Sumida, and T. Suita, Detailed Flow Study of Mach Number 1.6 High Transonic Flow With a Shock Wave in a Pressure Ratio 11 Centrifugal Compressor Impeller, Journal of Turbomachinery, vol. 126, pp. 473-481, 2004. M. Casey, Frank Gersbach, and Chris Robinson, An Optimization Technique for Radial Compressor Impellers, Proceedings of ASME Turbo Expo: Power for Land, Sea and Air, 2008. Y. Varol, E. Avci, A. Koca, H.F. Oztop, Prediction of flow fields and temperature distributions due to natural convection in a triangular enclosure using Adaptive-Network-Based Fuzzy Inference System (ANFIS) and Artificial Neural Network (ANN), International Communication in Heat and Mass Transfer 34 (7) 887-896, 2007.


1016


[15] Y. Varol, A. Koca, H.F. Oztop, E. Avci, Analysis of adaptivenetwork-based fuzzy inference system (ANFIS) to estimate buoyancy-induced flow field in partially heated triangular enclosures, Expert Systems with Applications 35 (4) 1989-1997, 2008. [16] A. Ozsunar, E. ArcaklIoglu, F. Nusret Dur, The prediction of maximum temperature for single chips’ cooling using artificial neural networks, Heat and Mass Transfer 45 (4) 443-450, 2009. [17] Allouis, C., Beretta, F., Amoresano, A., Experimental study of lean premixed prevaporized combustion fluctuations in a gas turbine burner, (2008) Combustion Science and Technology, 180 (5), pp. 900-909. [18] Niola, V., Quaremba, G., Amoresano, A., A study on infrared thermography processed trough the wavelet transform, (2009) Proceedings of the 8th WSEAS International Conference on System Science and Simulation in Engineering, ICOSSSE '09, pp. 57-62.

Authors’ information Department of Mechanical and Materials Engineering, Faculty of Engineering and Built Environment, The National University of Malaysia (UKM), 43600 UKM Bangi, Selangor, MALAYSIA. Layth Hasan Jawad studied Mechanical Engineering in Babylon University, Babylon IRAQ, graduated in 2000, and Masters in Mechanical Engineering Specialized in Thermo fluid Engineering-University of Babylon, IRAQ, 2002. His research interest is in the area of computational fluid dynamic, turbo machinery, internal combustion engine. Mr. Layth has some publication about internal combustion engine properties inside Iraq. He is a lecturer in the Ministry of Higher Education and Scientific Research-Baghdad IRAQ, Foundation of Technical Education. Currently he is a PhD candidate in the area of turbo machinery, National University of Malaysia, Mechanical Engineering Department. Mr. Layth is a member of the Iraqi Unions of Engineers, Federation of Arab Engineers, and Iraqi Society of Iraqi Engineers.

Associate Professor Dr Rozli Zulkifli is a member of SAE, Institution of Engineers Malaysia (IEM) and Institute of Materials Malaysia (IMM).He is the current head of Centre for Automotive Research (CAR), Faculty of Engineering and Built Environment, Universiti Kebangsaan Malaysia.

Wan Mohd Faizal Wan Mahmood was born in the State of Terengganu, Malaysia in 1972. In 1995, he graduated from Columbia University, U.S.A. with a Bachelor of Science in Mechanical Engineering (BSME) followed by a Master of Science (MS) in Power Systems Engineering from Kobe University of Mercantile Marine, Japan, in 2002. In 2011, he completed his PhD in Mechanical Engineering from The University of Nottingham, United Kingdom. He is now a senior lecturer in the Department of Mechanical and Materials Engineering, Faculty of Engineering and Built Environment, UKM. His current research subjects include computational fluid dynamics (CFD), combustion and particle tracking especially in the field of automotive engineering. His past experience as a mechanical engineer in a petroleum refinery nicely complements his current research interests. Dr. Wan Mahmood is a member in Board of Engineers, Malaysia.

Shahrir Abdullah was born in 1969 in the state of Terengganu, Malaysia. He received his B.Eng. degree in Mechanical Engineering from Universiti Kebangsaan Malaysia (UKM) in 1992, his M.Sc. degree in Design and Economic Manufacture in 1994 and his Ph.D. degree in Mechanical Engineering in 1997, both from the University of Wales Swansea. He is currently a Professor at the Faculty of Engineering and Built Environment, UKM and the Director of the Centre for Quality Assurance UKM. He has authored many technical papers in the field of computational fluid dynamics, internal combustion and powertrain engineering, machine design and numerical computation. His current research focusses on applied sciences and technologies in automotive and thermal engineering, microhydrodynamics and nanofluids. Shahrir Abdullah is a registered Profesional Engineer of the Board of Engineers, Malaysia (BEM), a corporate member of the Institution of Engineers, Malaysia (IEM) and a member of the Society of Automotive Engineers (SAE). Rozli Zulkifli was born in Selangor, Malaysia on the 31st of May 1971. He obtained a bachelor degree (Honours) in Mechanical Engineering from the University of Liverpool, UK in 1994, a master degree from the same university in 1996 and then a PhD degree from UKM. Dr. Rozli Zulkifli is currently an associate professor at the Universiti Kebangsaan Malaysia. He has published many articles in international proceedings and journals. His research interests are in heat transfer and composite materials. Currently his research thrust is in the area of jet impingement heat transfer.



1017


Numerical Study of Heat and Mass Transfer in Membrane Distillation for Desalination of Seawater A. Rachdi, R. Sebai*, F. Bouslama, R.Chouikh Abstract – The present work investigates the heat and mass transfer mechanisms in a cross-flow parallel plate membrane distillation module. The three dimensional model which accounts simultaneously for heat and water transfer in the membrane and the channels. The results are compared to the available data and the agreement is satisfactory. The effect of operating parameters such aspect ratios of the rectangular ducts, volume flow rates, the temperature differences, the heat flux and the mass flux of water vapor are also investigated. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Heat Transfer, Mass Transfer, Membrane Distillation, Cross-Flow

Subscripts h c i m o v w

Nomenclature C C

Feeding concentration (g·cm-3) Dimensionless feeding concentration C = ( c − cci ) ( chi − cci )

D h

Diffusivity (m2/s) Convective heat transfer coefficient (kW m-2 K-1) Convective mass transfer coefficient (m/s) Canal width (m) Canal length (m) Mass flow rate (kg.s−1) Reynolds number Re = uL ν Prandtl number Pr = ν α Schmidt number Sc = ν Dwv Temperature (K) Velocity (m/s) Dimensionless velocity U = Lu α ; V = Av α Water uptake (kg/kg) Coordinats (m) Dimensionnels coordinats X = x L ;Y = y A ; Z = z δ

k A L m Re Pr Sc T u, v U,V w x, y, z X,Y,Z

I.

Introduction

Several water desalination technologies have been found in the literature, based on the use of porous hydrophobic membranes and known as membrane distillation process. It is a membrane technique involving transport of water vapor through the pores of hydrophobic membranes[1] and [2]. In the present study, the direct contact membrane distillation process (DCMD) is considered. In this configuration, the both sides of the membrane are in direct contact with two liquid phases, the feed solution (warm water) and the permeate (cold water), kept at different temperatures. The vapor diffusion path is limited to the thickness of the membrane, thereby reducing mass and heat transfer resistances [3]. The membrane distillation is a very promising technology to reduce the production costs of water from fresh water sources [4] (lakes, solar pond, etc.). One of the early studies is of L. Basini et al. [5]. They have presented an experimental study on the desalination process through membrane distillation module. They studied experimentally the influence on the evaporation efficiency of the relevant process parameters such as inlet temperatures and flow rates. They have compared the mathematical model with the experimental results.

Greek symbols Membrane thickness (m) δ Height of hot water canal (m) δh Height of cold water canal (m) δc µ Dynamic viscosity (Pa s) ρ Density of water (kg/m3) α Thermal diffusivity (m2/s) Dimensionless temperature θ θ = (T − Tci ) (Thi − Tci )

λ

Heat, hot Cold Inlet Mass, membrane Outlet Vapor Water

Thermal conductivity (kW m-1 K-1)



1018

A. Rachdi, R. Sebai, F. Bouslama, R. Chouikh

R. W. Schofield et al [6] have developed the equations for heat and mass transfer in membrane distillation (MD). They tested experimentally the vapors transport for two type of membrane distillation fiber and tubular and reasonably described by combined Knudsen and molecular diffusion. Caroliene M. Guijt et al [7] have studied the heat and mass transfer membrane evaporation module for desalination of seawater. They described with molecular diffusion through stagnant air the mass transfer of the water vapour through the membrane and the air gap. The energy transports, the variation of temperature and membrane fiber diameter have developed. They indicate numerically that the highest productivity will be obtained with high temperature, small membrane fibers and a small air gap. L. Martunez-Diez and M.I. Vasquez Gonsalez [8] have investigated the method to evaluate the membrane mass transfer coefficient, the membrane heat transfer coefficient and the boundary layer heat transfer coefficient in a membrane distillation system. The results are obtained from different measurements of mass flux evaporation efficiencies with porous hydrophobic membrane. J. I. Mengual et al [9] have developed the difference between the mechanism of heat transfer in MD systems, which is coupled with transmembrane mass transfer, and the mechanism of heat transfer in pure heat exchangers. They studied experimentally the vacuum membrane distillation in a capillary membrane module and evaluated the heat transfer coefficients in both the lumen and the shell side of the membrane module. M. Qtaishat et al [10] have provided a detailed analysis of the heat transfer in direct contact membrane distillation (DCMD). A mathematical model was proposed to evaluate the experimental values of the thermal boundary layers heat transfer coefficients, the membrane-liquid interface temperatures, the temperature polarization coefficient, the membrane mass transfer coefficient and the evaporation efficiency. In this study, the obtained results showed that the mass transfer contribution to the overall heat transferred was significant only in the membrane region while it was negligible in both feed and permeate boundary layers. M. Khayet et al [11] have studied a heat and mass transfer through hydrophobic membranes applying direct contact membrane distillation (DCMD) process. They described by a three-dimensional network model of interconnected cylindrical pores with distributive pore size in membrane pore space. The influence of temperature polarization phenomenon, membrane physical properties including pores interconnectivity and the DCMD fluid dynamic conditions are developed and taken into consideration in the MC model. The present work investigates numerically the heat and mass transfer mechanisms in a membrane distillation


process. This work that enters within the framework of general study dealing with membrane distillation used in distillation process states the problem in Cartesian coordinates system, involves the use of a control-volume method and solves the set of governing equations in the air streams and in the membrane core. The numerical approach allows us to deepen the understanding of the heat and mass transfer mechanisms present in the process. The mathematical complexities are much more involved than in the sensible heat exchanger because of the interaction between the mass and temperatures fields.

II.

Theoretical Model

The hydrophobic membrane heat exchanger consists of alternate layers in a cross flow arrangements, separated by thin porous membranes. These arrangements form the hot and cold water stream passages. A schematic of a hydrophobic membrane integrated in the membrane distillation process is shown in Fig. 1. The hot and cold water streams are with different concentration.

Fig. 1. Schematic of a cross-flow water vapor exchanger with membrane hydrophobic

The dimensionless governing equations, presented in this section, use the nomenclature and coordinate system shown in the same figure. The mass flow rate at the inlet is prescribed in conjunction with a fully developed laminar flow profile. Exact solutions of such profiles are available for a variety of cross- sectional areas (Fig. 2), but we found it computationally more effective to use in this study an approximation which provides within 1% of the exact solution [12].

Fig. 2. Rectangular duct


1019


⎡ ⎛ y ⎞ n ⎤ ⎡ ⎛ z ⎞ n' ⎤ u = umax ⎢1 − ⎜ ⎟ ⎥ ⎢1 − ⎜ ⎟ ⎥ ⎣⎢ ⎝ b ⎠ ⎦⎥ ⎣⎢ ⎝ a ⎠ ⎦⎥ umax

2

V

(1)

⎛ n' + 1 ⎞ ⎛ n + 1 ⎞ = um ⎜ ⎟⎜ ⎟ ⎝ n' ⎠ ⎝ n ⎠

∂ 2θ m ∂Z m2

a 1 ≤ b 3 a 1 ≥ b 3

θ=

=0

(5)

T − Tci T −T ; θ m = m ci Thi − Tci Thi − Tci

(6)

where Thi is the inlet temperature of the hot water in, and Tci is the inlet temperature of the cold water in. Dimensionless feeding concentration is:

0.26 1.20

0.26

∂Z m2

0.12

1.11

0.8

∂ 2 Cm

=0

The dimensionless temperature is:

0.60

0.60

(4)

(2)

1.0 0.12

2

For the membrane: Due to small hydrophobic membrane thickness, heat and mass water vapor transfer in the x and y directions are considered negligible. Then heat and mass water vapor transfer in membranes can be simplified to onedimensional equation:

−1.4

⎧ if ⎪2 ⎪ n=⎨ ⎪2 + 0.3 ⎡⎛ a ⎞ − 1 ⎤ if ⎢⎜ b ⎟ 3 ⎥ ⎣⎝ ⎠ ⎦ ⎩⎪

2

∂C Pr ⎛ A ⎞ ∂ C Pr ⎛ A ⎞ ∂ 2 C V = + ⎜ ⎟ ⎜ ⎟ ∂Y Sc ⎝ L ⎠ ∂X 2 Sc ⎝ δ c ⎠ ∂Z c2 2

where umax is the maximum velocity (m s-1), a is the half-width of a rectangular duct (m), b is the half-height of a non-circular duct (m) and um is the average velocity (m s-1). Relations for the values n’ and n are calculated as follows [12]: ⎛a⎞ n' = 1.7 + 0.5 ⎜ ⎟ ⎝b⎠

2 ∂θ ⎛ A ⎞ ∂ 2θ ⎛ A ⎞ ∂ 2θ =⎜ ⎟ +⎜ ⎟ ∂Y ⎝ L ⎠ ∂X 2 ⎝ δ c ⎠ ∂Z c2

0.96

1.20

0.6 Z

1.49 0.42

1.46

1.38

C=

0.4 1.29

0.83

c − cei c −c ; Cm = m ei csi − cei csi − cei

(7)

0.2 0.60

0.83

0.96

where csi is the inlet concentration of the hot water in, and cei is the inlet concentration of the cold water in. The dimensionless coordinates:

0.0 0.0

0.2

0.4

0.6

0.8

1.0

Y

Fig. 3. The (Y–Z) u component velocity contours on channel cross-section

The u component velocity contours on supply channel cross-section is plotted in Fig. 3. The dynamic fields in both supply and exhaust channels are identical. For cross flow hydrophobic membrane distillation, the governing equations can be written in dimensionless form as follows: For hot water (distilland): In the flow region:

V=

;

δh

; X =

δc

(8)

δ

The boundary conditions are as follows. Inlet conditions for hot water: X = 0

→ θi ( X = 0 ,Y ,Z ) = 1

and Ci ( X = 0 ,Y ,Z ) = 1

(9)

Outlet conditions for hot water:

2

2 ∂θ ⎛ L ⎞ ∂ 2θ ⎛ L ⎞ ∂ 2θ U =⎜ ⎟ +⎜ ⎟ ∂X ⎝ A ⎠ ∂Y 2 ⎝ δ h ⎠ ∂Z h2 2

U

Lu

Av

x y ; Y= ; A L α α z z z Zh = ; Zc = ; Z m =

U=

2 ∂C Pr ⎛ L ⎞ ∂ 2 C Pr ⎛ L ⎞ ∂ 2 C = + ⎜ ⎟ ⎜ ⎟ ∂X Sc ⎝ A ⎠ ∂Y 2 Sc ⎝ δ h ⎠ ∂Z h2

(3)

X = 1

→

∂θ ∂C = =0 ∂X ∂X

(10)

Adiabatic boundary conditions for hot water: Y = 0 or Y = 1 ,

For cold water (distillate):


∂θ ∂C = = 0 ∂Y ∂Y

(11)


1020


On membrane surface for hot water: Z = 0 or

Z = 1 , θ ( X ,Y ) = θ mh

and C ( X ,Y ) = Cmh

(12)

Inlet conditions for cold water: → θi ( X ,Y = 0 ,Z ) = 0

Y = 0

and Ci ( X ,Y = 0 ,Z ) = 0

(13)

and 9 for cold water canal. We remark that in the channel center, the temperature and humidity levels are maximal in the hot canal and minimal in the cold canal side. This indicates that both heat and mass are transferred between the two canal streams when they flow through the unit. For weak Reynolds number, the variations in temperature are almost parallel to the diagonal line for both channels. When we increase the number of Reynolds, we remark that the contours of temperatures are elliptical in shape and symmetrical relatively to the flow direction.

Outlet conditions for cold water: Y =1 →

∂θ ∂C = =0 ∂Y ∂Y

TABLE I USED PARAMETERS OF THE MEMBRANE DISTILLATION FOR MODEL VALIDATION Symbol Value Unit Cpw 4.180 (kJ/kg K) 50.0E-02 (m) A L 50.0E-02 (m) 1.5 (m s-1) Um δ 2.0E-04 (m)

(14)

Adiabatic boundary conditions for cold water: X = 0 or

X =1 ,

∂θ ∂C = =0 ∂X ∂X

(15)

On membrane surface for cold water: Z = 0 or

Z = 1, θ ( X ,Y ) = θ mc

and C ( X ,Y ) = Cmc

(16)

δe

1.0E-02

(m)

δs

1.0E-02

(m)

λm

0.05

(W/m K)

λwater

0,6071

(kW/m K)

ρm

215

(kg/m3)

1.0

III. Solution Procedure

39.66

0.8 40.00

31.68

21.80

26.97 19.12 21.80

0.4

17.05 15.59

18.00

15.11

0.2 15.28

15.03 15.00

15.01 0.0

0.2

0.4

15.00 0.6

0.8

Fig. 4. Temperature profile in (x–y) plane at Z = δ / 2 for hot water canal, Re=102 1.0

27.70

40.00

32.82 37.10 36.00

39.75

0.8 39.20 38.88

y(m)

36.00

39.95 39.95

38.37

35.47 34.88

27.09 28.35 30.86 29.03 30.49 32.82 29.21 30.00 28.35 30.86 29.61 27.09

39.87 28.35


1.0

x(m)

0.4

The geometric and physical parameters of the hydrophobic membrane based energy recovery water vapour listed in Table I. We note that, the dimensionless temperature and concentration profiles are identical. The dimensionless temperature and water vapor profiles for different number of Reynolds in (X–Y) plane at Z = δ / 2 are plotted in Fig. 4, 5 and 6 for hot water canal and Fig. 7,8

25.45

26.39 24.00

38.32

0.6

IV.1. Temperature and Concentration Profiles

27.38 30.00 27.38

0.6

39.99

Numerical Results

29.14

36.00 37.02

38.94 36.00

39.99 y(m)

The equations of mass transfer are solved numerically using the control volumes method [13]. The discretized equations, one for each control volume, are solved by the simultaneous over relaxation method (S.O.R). One uses the Chebyshev algorithm acceleration of convergence which consists in seeking the optimal relaxation factor during the iterating process [14]. The solution is considered convergent when the relative error ranging between the new and the old value of velocity or concentration is lower or equal to 10-4. The mesh is constructed so that the boundaries of the area are located midway between the extreme nodes. There are no nodes on the border, but is considered a node on the inner side and outer side of the fictitious node.

IV.

39.33

39.98

40.00

24.95

37.79

22.70 19.76

24.95

0.2 27.70

18.00

15.49

16.36 15.16

0.0 0.0

0.2

0.4

0.6

0.8

1.0

x(m)

Fig. 5. Temperature profile in (x–y) plane at Z = δ / 2 for hot water canal, Re=103


1021


1.0

39.98 40.00

39.92 39.90

39.99

38.75

0.2

0.4

0.6

16.12

20.00 15.81 18.69 21.93

15.49 15.33

15.28 15.22

20.00

15.17

15.13

15.05 15.03

15.04 15.02

15.07

34.74

0.8

24.00 20.00

15.10

15.00 15.00

15.00

24.00

0.0

16.47 15.81 15.64

15.72 15.56 15.41

24.00

0.2 36.85

16.12

16.47

0.4

37.29 32.00 0.0

0.6 28.00

38.46 38.05

39.37

39.82

20.00

39.03

16.33 16.00

0.8

38.64

17.20

16.33

18.69 16.81

38.64

39.15

39.54

39.86 0.2

39.23

39.60

40.00 0.4

39.31

39.66

40.00

y(m)

0.6

39.95

39.77 39.72

17.72

32.00

y(m)

0.8

1.0

32.00 36.00 37.71 38.27 38.85

34.74 37.71 38.75

0.0 1.0

0.0

0.2

0.4

x(m)

0.6

0.8

1.0

x(m)

Fig. 6. Temperature profile in (x–y) plane at Z = δ / 2 for hot water canal, Re=104

Fig. 9. Temperature profile in (x–y) plane at Z = δ / 2 for cold water canal, Re=104

The temperature and water vapor fields on membrane surface in (X–Y) plane at Z = δ / 2 for hot water canal and cold water canal side are plotted in Fig. 10, 11 and 12.

The numerical results show that the shapes of the concentration contours are similar to those of temperature. The temperature and concentration differences across the membrane are resulting from relatively weak thermal resistance and small thickness of the membrane.

1.0 39.78 37.15 39.99 38.97 39.97 34.66 0.8 40.00 39.52 36.00 39.91 38.16

27.18 26.48 25.57 24.00

29.05 28.14 24.88

30.79

1.0

20.75

0.8

15.03

15.44

36.00 34.59

39.97 0.4

0.2

0.4

0.6

0.8

15.05

1.0

15.02

x(m)

0.0 0.0


30.00

0.8

30.98

24.00

y(m)

0.6

26.88 24.46 24.00 22.87 21.87 22.42

0.8

1.0

0.4

19.52

17.41

18.00 0.2 15.57

0.6 16.90 15.90

16.37 15.29

15.07

15.00

0.0 0.6

0.4

15.15

15.02 0.8

26.55 26.99 28.13 28.13 27.49 37.33 39.67 28.48 31.33 27.49 36.00 28.88 27.83 39.10 27.33 32.25 29.43 26.55 38.17 26.01 25.37 34.53 24.00 25.37 22.37 23.16 20.21 21.50 18.00 19.07 16.61 15.58 15.12 34.53

39.95

20.25 18.77

0.4

1.0

0.8

20.85

0.2

0.6

24.84

37.95

0.0

0.4

15.00

Fig. 10. Temperature distribution in membrane in (x–y) plane at Z = δ / 2 and Re=102

27.82

25.74

27.82

Y(m)

33.67

0.2

15.00

x(m)

25.74 39.78 39.10

15.75 15.19 15.08

15.00

1.0

16.93

15.44

0.2 0.0

20.41 19.16

18.00 16.31

39.89 39.72

15.00

0.0

26.72 25.91 24.83 22.87 24.00

21.85

15.00

15.01

15.06

39.99

y(m)

15.21

16.49

37.11

27.75

0.6

15.77

18.00

31.20 30.00 29.12

38.03

40.00

0.4

0.2

40.00

17.32

27.53 27.24

32.72

39.40 38.85

21.61 19.49

22.48

33.42

y(m)

0.6

40.00 40.00

27.62

30.00

0.2

1.0

x(m)

0.0

0.0


28.48

30.63

0.2

0.4

0.6

0.8

1.0

x(m)

Due to small membrane thickness, the temperature differences in membrane thickness are small. The temperature distributions on membrane surfaces are very two-dimensional. The contour lines are nearly parallel to the diagonal line. So, the thermal gradients are steep in the inlets regions of the hot water and cold water flows.


Fig. 11. Temperature distribution in membrane in (x–y) plane at Z = δ / 2 and Re=103

We remark that the heat exchange is ideal for weak numbers of Reynolds for the three compartments of the exchanger: hot water canal, cold water canal and membrane.


1022


30.00

36.21

28.69 31.96

29.47

0.8 31.96

36.21

33.15

0.6 y(m)

30.00 29.12

35.00

26.94

0.4

0.2

0.4

0.01

0.01

0.00 0.01

0.6 0.03 0.03

0.00

0.02

0.01

0.04

0.04 0.2

0.01

0.03

0.03

0.4

23.54

0.0

0.00

0.02

25.00

25.65

0.00

0.00 0.00

0.8

26.10

24.45

0.0

0.00

26.70

26.42

0.2

1.0

25.65 25.65 27.34 26.94 27.34 27.91 29.47 27.52 28.37 27.69 27.69 27.45 28.12 27.91 27.20 27.45 28.69 Y

1.0

0.00

0.02 0.00

0.04

22.73

21.71 18.90

20.00

0.6

0.8

0.00

0.00 0.00

0.0 1.0

0.0

0.2

0.4

0.6

x(m)

0.8

1.0

X

(a) Fig. 12. Temperature distribution in membrane in (X–Y) plane at Z = δ / 2 and Re=104

1.0 0.00 0.07 0.08

0.04 0.8

1.0 0.00

0.03

0.18

0.18

0.08

0.10 0.13 0.14

Y

0.29 0.53

0.16

0.4

0.6 Y

0.94

0.12

0.68

0.4

0.17 0.05

0.08 0.02

0.18

0.2

0.04 0.00

0.0

0.01

0.03 0.94

0.07

0.14

0.18 0.18 0.19

0.2

0.29

0.80

1.12 1.31

0.17

0.40

1.03

0.12

0.15

0.80

1.03

0.04

0.11

0.05 0.6

0.01

0.00

0.09

0.00

0.40

0.8

0.05

0.0

0.2

0.4

0.6

0.8

1.0

0.00

0.00

0.0

X

0.0

0.2

0.4

0.6

0.8

(b)

1.0

X

1.0

(a)

0.10

0.20

0.23

0.20

0.17 0.25

0.29

0.10 0.23

0.8

1.0

0.29

0.75 2.50

1.25

2.02

0.75

3.47

0.36

2.50 2.78

4.35

Y

5.00 4.75 5.34 5.14 4.07 5.56 2.78 5.56 3.75 6.12 2.50 3.75 6.25 5.88 6.50 0.75 1.25 0.31 0.07

0.2

0.17 0.20 0.23

0.30 0.32

4.07 4.53

0.4

0.27

0.35

Y

0.31 0.6

0.31 0.33

0.6

0.8 3.13

0.32

0.4

0.38

0.2

0.41 0.40 0.42 0.43 0.39 0.45 0.43

0.34

0.38 0.37

0.27 0.25 0.27

0.0 0.0

0.2

0.4

0.0

0.25

0.6

0.17 0.8

0.10 1.0

X

0.0

0.2

0.4

0.6

0.8

(c)

1.0

X

Fig. 14. Mass flux of water vapour

(b) 1.0

6.81

6.81

5.72

3.75 0.8

IV.2. Heat and Mass Fluxes

3.75 9.22

8.80 9.22

8.80 8.26

10.01

The dimensionless emission rates (mass fluxes water vapour (10-4 kg m-2 s-1)) through the membrane are shown in Fig. 13 and the dimensionless contours of heat flux (10-4 W/m2) through the membrane are shown in Fig. 14. We remark that the contour of temperature and water vapour values decrease along the diagonal line. The maximum heat and mass exchange is produced in the entry regions near the corner.

5.72

9.65 10.64 11.25

0.6

10.36

6.81

Y

11.90

8.26 12.18

0.4

12.45 13.26 13.54 13.97 13.76

0.2

11.55

7.50

12.73 7.50 12.99

14.69

8.80

7.50

5.72

0.0 0.0

0.2

0.4

0.6

0.8

1.0

V.

X

(c)

Conclusion

A numerical model was developed to simulate the heat

Fig. 13. Heat flux



1023


and mass transfer for the membrane distillation. The influence of some operating parameters was analyzed. The temperature and concentration profiles of hot canal, cold canal and membrane show that there is a transfer of heat and water vapor from one canal to another. For cross flow arrangement, the directions for heat and mass fluxes of water vapor decreasing are parallel to the diagonal line of the membrane. The heat exchange is ideal for weak numbers of Reynolds. We supplement our work by the variation of other operating parameters like the concentration of the feed solution and the porosity of the membrane.

Reference [1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13] [14]

I. Boukholda, S. El May, A. Bellagi“Effects of Incorporated Heat Exchangers on the Thermo Economic Analysis of a Water/LiBr Absorption Refrigeration. System Part I: Formulations” International Review of Mechanical Engineering, May 2009 Vol. 3. n. 3, pp. 306-311. I. Boukholda, S. El May, A. Bellagi“Effects of Incorporated Heat Exchangers on the Thermo Economic Analysis of a Water/LiBr Absorption Refrigeration. System Part II: Applications” International Review of Mechanical Engineering, July 2009 Vol. 3. n. 4, pp. 410-416. D. Gassara, P.Schmitz, A.Ayadi“Numerical Simulation of Particle Accumulation at the Membrane Surface in Microfiltration” International Review of Mechanical Engineering, March 2012 Vol. 6, n. 3 pp. 485-495. F. Laganà, G. Barbieri, E.Dorioli. “Direct contact membrane distillation: modeling and concentration experiments” Journal of Membrane Science 166 (2000) 1-11. L. BAsini, G. D’Angelo, M. Gobbi, G.C. Sarti, C. Gostoli. “A desalination process through sweeping gas membrane distillation” Desalination, 64 (1987) 245-257. R.W. Schofield, A.G. Fane, C.J.D. Fell. “Gas and vapour Transport through microporous membranes. I.Knudsen-Poiseuille transition” Journal of Membrane Science, 53 (1990) 159-171. Caroliene M. Guijt, Imre G.Racz, Jan Willem van Heuven, Tom Reith, André B. de Haan. “Modelling of a transmembrane evaporation module for desalination of seawater” Desalination 126 (1999) 119-125. L.Martinez-Diez, M.I. Vazquez-Gonzalez. “ A method to evaluate coefficients affecting flux in membrane distillation” Jornal of Membrane Science 173 (2000) 225-234. J.I. Mengual, M. Khayet *, M.P. Godino. “Heat and mass transfer in vacuum membrane distillation” International Journal of Heat and Mass Transfer 47 (2004) 865–875. M. Qtaishat, T. Matsuura, B. Kruczek, M. Khayet. “Heat and mass transfer analysis in direct contact membrane distillation” Desalination 219 (2008) 272–292. M. Khayet, A.O. Imdakm, T. Matsuura. “Monte Carlo simulation and experimental heat and mass transfer in direct contact membrane distillation” International Journal of Heat and Mass Transfer 53 (2010) 1249–1259. T.R. Irvine, Jr. and J. Karni. “Non-Newtonian fluid flow and heat transfer”. In: (5th Ed. ed.), Handbook of Single-Phase Convective Heat Transfer, John Wiley and Sons, New York (1987), pp. 20-1– 20-57. S.V. Patankar, Numerical Methods for Heat and Fluid Flow, 1980, p. 107. W.H. Press, B.P. Flannery, S.A. Teukolsky, W.T. Vetterling, Numerical Recipes: The Art of Scientific Computing, Cambridge University Press, 1986.



High Institute Of Industrial Systems of Gabes, Slaheddine El Ayoubi Street- 6032 Gabes , Tunisia. 2 Corresponding author. Center for Energy Research and Technology (CRTEn), BP 95 Hamamlif 2050,Tunisia. 3 High Institute of Sciences and Technology, rue de Lamine Abassi 4011 H. Sousse,Tunisia 4 Center for Energy Research and Technology (CRTEn), BP 95 Hamamlif 2050,Tunisia. E-mails : [email protected] [email protected] [email protected] [email protected]

A. Rachdi (Tunisia, 1974) got his PhD in Fluid Mechanic and Heat and Mass Transfer at the University of Tunisia in 2009. She is an Assistant Professor in High Institute of Industrial Systems of Gabes. Her research activities concern mainly renewable energies and heat and mass transfer simulation.

R. Sebai (Tunisia, 1979) got his Master in Fluid Mechanic and Heat and Mass Transfer at the University of Tunisia in 2007; and she prepares a Ph. D in Physic at the University of Tunisia in 2008. Her research activities concern mainly renewable energies and heat and mass transfer simulation in membrane processes.

R. Chouikh born in Tunis in 1961, is an Assistant Professor in the Institute of Engineering studies in Nabeul, Tunisia. PhD in Mechanical Engineering from the University of Tunis since 1998. His research activities are mainly: transport phenomena in membrane, porous media and developpement of CFD. He published several papers in international journals related to these subjects and is involved in national and international projects concerning Fuel Cell Technology and porous media. Dr. Chouikh is a member of Tunisian Society of Physics. F. Bouslama born in Tunis in 1962, is an Assistant Professor in High Institute of Sciences and Technology in Sousse, Tunisia. PhD in Mechanical Engineering from the University of Strasbourg, France since 1997. His research activities are mainly: transport phenomena, solid physics and developpement of CFD.


1024


Supervision and Control Architecture Proposal for Automation and Robotics Training on Platform Ricardo A. Castillo1, João M. Rosário2, Oscar F. Aviles3 Abstract – This work proposes a generic supervisory and command architecture for an experimentation modular automated platform equipped with remote access capacities which is conceived with the aim of improve training and research processes on Automation and Robotics, this study describes the platform´s design, dynamic modeling and implementation stages. The technologic and industrial devices integration (Programmable Logic Controllers - PLC, several types of sensors and actuators, image processing, supervisory systems and robotic manipulation devices) in a single platform which is implemented following a modular Collaborative Automatic Production System (CAPS/ADACOR) architecture allows students and researchers to interact with it by means of doing practices in order to successfully automate, supervise and manage a complete production process. Therefore, class acquired theoretical concepts are supported so improving user´s professional skills. A platform developed using the here proposed generic structure allows users to work within an educational environment coping with most of the encountered aspects in a real Manufacturing Automation System, such as Technologic Integration, Communication Networks, Process Control and Production Management. Furthermore it is possible to command the entire assembly process taking place at the platform by a remote network connection using the internet – WEBLAB (Remote Laboratory), enabling individual users and groups in different places in order to use the platform and quickly interchange information. In addition it is important to outstand that both the Modularity and Flexibility of the platform can allow readily any further hardware or software enhancement. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved. Keywords: Collaborative Automation, Integration Architecture, Process Modeling, Robotics

I.

Introduction

The integration of both industrial devices and modern Information and Communication Technologies (ICTs) motivates the development of new tools focused on supporting the educational formation, allowing the development of new approaches to training on high emphasis technological areas such as industrial automation and robotics [1],[2]. Within this context, a complete educational and training process focused on these disciplines as well as the theoretical foundations gained in the classroom must be developed in didactic and pedagogic laboratories, so enabling students to develop knowledge, skills and attitudes based on interactive and motivated interaction taking into account related areas of technological [3] in such a way as preparing competent professionals for the job market. This constructivist development of learning activities on Automation and Robotics also offers a conducive environment for researchers in these areas to formulate and assess their hypotheses [4], [5].


Currently there are several manufacturers of automated devices aimed to support training and research in technological areas; these devices provide some functionality in their control systems. However, most of these platforms are expensive, do not have open architecture for developing new experimental practices nor cover the needs required by researchers. These researchers often avoid the previous proprietary and mechanical the limitations by investing a lot of time in adaptations that can harm their research results. It can be pointed out situations in which because of these devices or platforms do not usually incorporate more than two technologies, the range of experimental applications that can be performed on them becomes limited to only one knowledge area. Consequently these activities do not reflect conditions normally experienced in the current industrial context reality, where increasingly the automated manufacturing facilities require the technology integration from conception to implementation and maintenance. Similarly, both the proposal and research activities results are limited to very specific academic fields.


1025

R. A. Castillo, J. M. Rosário, O. F. Avilés

Also, nowadays, many platforms on the market do not provide supervision nor remote control (LAN / Internet) capabilities, therefore they only provide possibilities to perform practices with fixed schedules and with a preestablished duration and organization. Thus, the use and exploitation of these devices become restricted to a small number of people [6], [7]. Due to the change from a local economic perspective to a global economy, since the last decade of the twentieth century manufacturing environments have evolved in its economic, technical and organizational dimensions causing trends in industrial automation have to adapt to a need for small and medium production [15], with an increased differentiation of families of parts and / or products, which are almost customized to the needs of individual clients and also add value to its intangible component (software, support services including, online support line, etc). Thus new paradigms such as Collaborative Manufacturing Management (CMM) and consequently the Collaborative Automated Production System (CAPS) appear as an evolution and response to new needs that are not met completely by the original concept of (Computer Integrated Manufacturing) CIM, which provided Automation Production System (APS) with a very small degree of hardware and software flexibility and integration, based on a strongly hierarchical and centralized control architecture and in a sequential planning framework which did not allow these systems to adapt quickly to environmental changes. Besides improving the previous features (flexibility and possibility of integration) and agility, the CAPS add modularity, fault tolerance, reusability, and interaction between production management components within production systems [19]. Thus the CAPS can achieve both the global and local manufacturing objectives, based on a structure no longer hierarchical but heterarchical. This new approach is based on development and integration of emerging technologies such as object-oriented control (decentralized), Intelligent Manufacturing Systems (IMS) and Mechatronics (see Fig. 1). The distributed and decentralized characteristics of CAPS modular control architecture require the implementation of local and remote supervision and control capacities [8]. Within the IMS, the Holonic Manufacturing Systems (HMS) have a holonic or holarchical structure that causes the CAPS have a high modularity (hardware software) that allows both the autonomy and the cooperation between different holons [21], [22], which are controlled in a distributed way [9]. Therefore, the HMS enable auto-configuration capabilities and ease of expansion and modification. One of the architectures proposed in the literature in order to exemplify the approach HMS / CAPS is the collaborative architecture ADACOR (ADAptive holonic Control Architecture for distributed manufacturing systems) [10] which is described below.


Fig. 1. Collaborative Industrial Automation [8]

The reliability, flexibility and agility of a CAPS not only are conditioned by the reliability, speed and flexibility of their individual mechatronic components, but depends crucially on the reliability and flexibility of the control and automation architecture (hardware). A CAPS with an ADACOR architecture organizes the use of their modules (holons), synchronizing the resource utilization, being dynamically reconfigurable, and thus capable of producing a large number of products and / or families of parts with minimal effort to change their components physical (flexibility). Thus a CAPS / ADACOR can easily adapt to a stochastic manufacturing environment, characterized by frequent occurrence of unexpected disturbances [11]. The collaborative control architecture ADACOR was proposed, developed and implemented at the Polytechnic Institute of Bragança (Portugal), taking into account a set of holons (definition given below). Currently in order to complement the theoretical instruction, the Higher Education Institutions (HEIs) implement various strategies in order to provide students with the integration of the technologies used in industrial environment (Fig. 2). The design and implementation of automation platforms focused on education is mainly developed within the HEI and research centers, an example of such applications is the PIPEFA platform (Platform for Industrial Training, Research and Training on Automation) implemented in the Laboratory of Automation Integrated and Robotics (LAIR) at the Campinas State University- UNICAMP [12] in order to demonstrate the products assembly and disassembly processes, in the LAIR laboratory has also been developed and implemented a Production System (PS) architecture learning platform for colors mixing [13]. These two applications, composed of different workstations comprise PLCs, sensors and actuators used frequently in the industrial environment. In [14] is described the design and implementation of a collaborative platform formed by two manipulator robots and a belt conveyor used as a transfer and union International Review of Mechanical Engineering, Vol. 6, N. 5

1026


element so constituting a PS that allows experimental practices in order to provide training in industrial automation. In [3] are presented three projects developed with the aim of creating integrated learning environments in automation using development kits, LEGO Mindstorms, and ROBIXTM: system for transport products in a production line, automated production and a solder machine.

development of an automated experimentation platform that will offer the possibility to implement practical applications due to the modularity and flexibility of the proposed architecture. These practical activities will use the integration of different technologies, rather than individual technologies normally used in conventional courses of Automation and Robotics. The holonic architecture of Collaborative Automated Production System - CAPS / ADACOR proposed in this work will enable users to do complete hands-on experiments within a learning environment that exemplifies routine tasks in modern Industrial Manufacturing Systems (Transport, Classification, Assembly, Inspection, Quality Control and Supervision).

II.

Fig. 2. Ways to Support the Formation Process and Automation Research

These two applications, composed of different workstations comprise PLCs, sensors and actuators used frequently in the industrial environment. In [14] is described the design and implementation of a collaborative platform formed by two manipulator robots and a belt conveyor used as a transfer and union element so constituting a PS that allows experimental practices in order to provide training in industrial automation. In [3] are presented three projects developed with the aim of creating integrated learning environments in automation using development kits, LEGO Mindstorms, and ROBIXTM: system for transport products in a production line, automated production and a solder machine. At the Santa Catarina Federal University was developed a didactic platform for working with electrohydraulic actuators to allow students to carry out experimental and theoretical analysis, this platform includes PLCs, a hydraulic power unit, a platform for circuit assembly and system acquisition and a control VXI / LabVIEWTM [16]. The Sao Joao del Rei Federal University in partnership with Junior enterprise EJEL developed academic support platforms in industrial automation field [17], the Paraná Federal University of Technology has implemented a programmable control system for distributed systems based on a FESTO ™ modular production system [18] in order to automate a process and apply concepts of tasks distribution to avoid centralized control. The proposal of technology integration in a single generic architecture application presented in this work is validated through the modeling, analysis and Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

Conception of the Platform Architecture

The educational foundation that motivates the interaction with industrial technologies through the implementation of practical activities in order to effectively support the processes of training and research on Automation is based on Jean Piaget's cognitive theory of CONSTRUCTIVISM, which proposes that knowledge is not something that can be transmitted, but is constructed in a personal way through experimentation and manipulation of individual objects of study, building new concepts based on previous [20]. From this latter approach, an effective strategy to provide a practical complement to the training process on Automation and Robotics is the interaction of students with industrial technologies through integrated learning platforms that enable the realization of practical experiments with various industrial devices. This strategy is taken into account in this work requires of these platforms having a hardware and control architecture that can represent the current APSs in order to faithfully exemplify the industrial problematic. Thus, these activities will illustrate practical situations usually found in professional practice, developing in students not only theoretical knowledge, but also technological skills to deal with the job market. The completion of these activities may be in hands-on or via the Internet [1]. In the case of collaborative platforms WebLab integration is an ever-growing area, and there is no a standardized architecture that directs the implementation and control of these devices [2], in many cases this architecture depends both on the type of technology to be integrated and on the type of experiments to be performed. Taking into account the current direction of industrial automation development and aiming to design an experimental tool in Automation and Robotics, it can be identified the problem and the scope of study where the platform will be used. After studying industrial technologies that can be integrated on Automation and


1027


Robotics it is proposed a hardware - software ADACOR holonic type architecture. The selection of technologic integration architecture is considered from two perspectives: a technological and a pedagogic points of view. From a technological perspective, the integration of the five above described technologies (Fig. 3) aim to achieve a faithful representation of the CAPS architecture that orient current manufacturing systems implementations. It should offer not only the possibility of centralized control, but also allow some level of autonomy between sub-system components without completely losing the collaboration and cooperation between them (distributed or decentralized control).

Fig. 3. Technologies Integrated in the Platform

Provided that, it is looks for a generic structure in its operational and command parts having a high degree of flexibility, modularity and reconfigurability, being open (hardware - software) to any future improvement or modification of both the operative or command part (structure and programming logic) and easily scalable. It also designed a module for online command and monitor the functioning of the platform, so providing for the user feedback and video data, thus allowing the processes configuration. The flexibility of the proposed structure will be reflected in: the possibility of reprogramming the structured logic used in implementation, the possibility of assembling different products using the same platform, and the reconfigurability of the physical structure. The integration scheme proposed for the platform can be seen in Fig. 4. The proposed integration is designed through an architecture in which each of the five industrial technologies described in Fig. 3 is exemplified through a module, and these in turn are integrated through two levels of modularity, in a first level (macro) these are associated to form an Collaborative Automated Production System CAPS, and still looking for a greater consideration of modularity in a second level (micro) this concept is also brought to the interior of the constituent parts of the CAPS. This is achieved by defining fully functional agents which will be integrated into a holonic type heterarchical architecture - ADACOR [10]. Inside the platform, these agents called holons are composed of a physical resource and a Logic Control Device (LCD) that can be programmed in order the holon to execute processes independently but collaboratively achieving shared goals, then the integration is managed in a higher level by supervision and control system (Fig. 5).

Fig. 4. Proposed Integration – Collaborative Automated Production System with Holonic type (CAPS/ADACOR) Architecture



1028


III. Implementation of Proposed Platform In order to validate the proposed ADACOR integration and the modeling of sub-processes designed to be performed using this architecture it is implemented both hardware (PC) and software (PO) parts of an automated platform (Fig. 5) in which each sub-process is specified and implemented in PLCs. In this sense, to evaluate not only the modeling of each holon, but also the integration of data, we propose and implement a complete process comprising classification, assembly, inspection and products transfer, each one with an associate sub-process. This integration is specified through a GRAFCET graph in order to implement it later using ladder logic in industrial PLCs in the same way as were performed the Petri Nets (PN) modeling. III.1. Operative Part (OP) – Description and Implementation

Classification, Assembly, Inspection and Rejection. In the proposed architecture presented in this work only will be used Classification, Assembly and Inspection the stations. III.1.2. Robotic Transfer Syetem The robotic transfer system consists of a cartesian manipulator (Fig. 6(b)) with two degrees of freedom (DF) each one driven by a stepper motor allowing translational movements: lateral (X direction) and vertical (Y direction). This robot is designed for feeding and transfer parts or assembled products, after it pass through the quality control module. The products are taken from the belt conveyor and transport towards a storage area if they get correct, otherwise they are removed by the manipulator which move the individual parts to the supply chain on which they can be inserted so starting again the process.

III.1.1. Manufacturing Cell The manufacturing cell integrated into the designed platform as validation of the proposed structure corresponds to the ICT-3 model of the BYTRONIC ™ company (Fig. 6(a)). This are composed of sensors (4 inductive, 4 photoelectric and 1 capacitive) and actuators (2 DC motors and 2 solenoids, ON/OFF and emergency switches). This operative equipment, properly programmed allows the realization of assembly and inspection of a product composed of two types of parts (a metallic base made of aluminum and a plastic ring that is put over this base). In the manufacturing cell, originally the pieces have to pass along several consecutive stations:

III.2. Command Part (PC) – Specification and Implementation For the complete PC implementation of the PC was performed as first step the substitution of the electronic control board in the manufacturing cell, this board was originally supplied by the BYTRONIC™ company. This interface allowed as user interaction merely the activation or deactivation of some sensors, therefore it was replaced by a physical command interface based in structured reprogrammable electronic in order to develop a fully reconfigurable architecture, open to future modifications.

Fig. 5. Proposed Platform - Integration Diagram



1029


III.2.1. Command Interface Implementations – CAPS / ADACOR Architecture Aiming a holonic implementation of PC, 2 Programmable Logic Controllers PLC are used, each one focusing on the control of an operational holon. Each PLC implements completely the holon’s LCD (Logic Control Device) with its decision, communication and virtual resource part following an ADACOR architecture.

So, in addition it means in FLEXIBILITY based on the ability to propose and implement various activities that can easily be made for training (on automation using the manufacturing cell, on robotics using the robotic manipulator or on mechatronics using the integrated system). Taking this holonic architecture into account it is possible to test different network communication models between the proposed holons, further extending the platform usefulness. Fig. 7 shows in detail the scheme implemented to integrate each holon following the ADACOR proposed architecture. III.2.2. Implementation of the Supervision and Command Module This module corresponds to the HS and HT holons. The supervisory system provides information about the state of significant variables of the process through indicators on a schematically screen, thus informing the user about what is happening in the platform operative part and allowing him to command the OP by means of controls to obtain a desired functioning, this system emulates a Virtual Instrument (VI). Fig. 8 shows the Supervision and Control module in LABVIEW ™ environment. The supervision and control software was structured through basic functions blocks, which are created in a modular way taking into consideration the following phases.

(a) Manufacturing Cell

III.2.3. Quality Control and Positioning Module – Design and Implementation

(b) Robotic Transfer System Fig. 6. Platform Operative Part

Taking in mind this division of responsibilities it can be achieved a high level of modularity and therefore the possibility to reconfigure and work faster and easier with any of the holons of the proposed architecture.


This module is associated to the supervision and control system to enable the integration between the two operational holons, so offering the possibility for detection of products that are differentiated by color of the plastic ring above the metal base. Therefore, the use of this module associated within an integrated process allows the development of various activities in mechatronics, adding flexibility to the proposed activities with the platform. The image processing algorithm developed for the quality control system allows the detection of different materials while they are moving through the belt conveyor as follows: a metallic base without a ring above, a correctly assembled product (the color of the ring above the metallic base determines whether the product is correct or not) and an incorrect assembled product (the color of the ring is not the determined by the user). Besides the images processing to classify materials, this system detects the product's position with regards to the first DOF of the manipulator, this position is then compared with the manipulator current position (determined by its kinematic model), for this robot minimizes the error and get the part.


1030


Fig. 7. Scheme for ADACOR Operational Holon Implementation

Thus when the system detects any material on the belt conveyor it will inform that to the supervision module in order to it stops the movement of the conveyor belt until the manipulator takes the material, carry it to the appropriate area and then return to its home position. The quality control system check the information delivered by the inspection station, and when necessary can also function without depending on this. The quality control system output information could also be used in a statistical way in a future work in order to test and explore production management models.

IV.

Use of the Proposed Architecture for Formation and Research

The possibility for propose practical activities in these areas using the platform integrated following an

ADACOR architecture is based on two main features: flexibility and modularity. Based in these characteristics, each operational holons provides the possibility to propose practices focused on one goal area depending on the technologies comprising it: the manufacturing cell holon in automation and the robotics manipulator holon in robotics. The platform proposed and implemented following a holonic hardware - software architecture allows the realization of experimental activities exemplifying tasks encountered in modern CAPSs. This work is validated in two ways: through the modeling and dynamic analysis of proposed operation to ensure achievability of its sub-processes, and through a hardware - software implementation of a real platform that follows an ADACOR architecture specifying and then implementing the proposed sub-processes in structured reprogrammable logic.

Fig. 8. Virtual Instrument developed for Supervision and Command



1031


As a suggestion for future works, it would be consider the use of the proposed modular, flexible and fully functional architecture, and from this point further study new or improved training methodologies using this tool in order to develop in the students skills on technologic areas, it is important to emphasize that a profound pedagogical study is beyond the scope of this work. However, below are presented the first approximations of how to use the platform, describing some examples of activities that can be proposed and implemented within courses on automation and / or robotics. Taking into account the modeling modularity and the hardware-software implementation, further to the possibility of experimentation with each holon separately it is also possible to define activities within each of them considering the basic operation units or sub-processes, which represent tasks similar to those found in CAPS and whose implementation has been already validated as feasible through the PN modeling. These sub-processes can occur within the smaller hardware units called stations (Table I). TABLE I SPECIFICATIONS ADOPTED FOR THE SIMULATED INVERTER Tasks with the Operative Part Tasks with the Command Part • Individual processes • Supply of raw material monitoring • Classification of raw material • Counting of pieces and assembled products • Products assembly • Manufacturing Scheduling • Products inspection of more than one product • Quality control • Monitoring and control of • Transport of finished products integrated processes

Thus, with the proposed architecture it is possible to propose practical activities with three levels of complexity (experimentation and validation) in order to apply the theoretical knowledge acquired in the classroom: a) Level A: Focused on knowledge about the operation of sensors and actuators and their application in individual tasks, using physical hardware, sensors or isolated actuators; b) Level B: Focused on the sub-processes implementation, this can develop individuals or integrated tasks using individual or associated stations to form a holon, so composing complex operating procedures; c) Level C: Focused on complete processes implementation, integrating the holons in a CAPS that performs defined production activities in order to meet a production demand. The aim of Level B practices that can be proposed is based on the initial and final markings in a station PN (Petri Net) model. In the case of practical activities integrating various stations, even from different holons, is it defined as general objective of the activity to attain the state corresponding to last station desired marking, starting from the first station initial marking.


In order to select the stations to be integrated, it should be considerate that the outputs of a station are connected to the inputs of the next, so the number of outputs from the first station should be equal to the number of inputs in the second. To propose Level C practical activities it must be integrated two complete holons in a CAPS. IV.I.

Proposed Activities within the Automation Course at Campinas State University

As a first approximation for using the platform were proposed two activities that were realized by the postgraduate students of the discipline IM333 – Logic Programmable Controllers at the Campinas State University in the first semester of 2010, these activities developed in this work include an active experimental component to complement the concepts presented about tasks programming using structured reprogrammable logic. Below are presented the two activities. IV.1.1. Activity Implemented with Classification Station of the Manufacturing Cell The activity developed in the course was active type PBL (Problem Based Learning) and is classified according to their complexity as level B activity. This activity is based on the specification and subsequent real implementation of a possible sub-process, using the hardware equipment of the manufacturing cell and the implemented command interface according to the proposed architecture. The sub-process at the station starts when the metallic and plastic pieces are moving through the supply chain towards classification station. a) If a metallic part (base) is detected at the classification station, there will be no action until it can cross completely the station and reach the ramp to the belt conveyor; b) If a plastic piece (ring) is detected at the station, it should be pushed to move by the corresponding ramp. IV.1.2. Activity Implemented within the Robotic Transfer System The activity developed in the course was active type PBL (Problem Based Learning) and is classified according to their complexity as level B activity. This activity is based on the specification and subsequent implementation of a real task, using the robot manipulator and the command interface together implemented under the proposed ADACOR architecture. The task be programmed by the students is stated as follows: a) Moving first and second stepper motor to advance and retreat the two DOF of the manipulator in a


1032


sequence that causes the end effector describes a square; b) Repeat the first sequence for the end effector in order to it describes a smaller square inside the first.

[5]

[6]

[7]

V.

Conclusions

In this research it was proposed a supervision and command generic architecture for an automated modular experimentation platform with capability to be used in remote form, designed to support and complement formal education and research on Industrial Automation and Robotics by designing, modeling and implementation a holonic- ADACOR type architecture to integrate technology into an automated platform, in which can be proposed activities that will illustrate CAPS routine tasks used in the modern industrial environment. The flexibility and modularity characteristics were considered in the conception of this architecture taking into account two perspectives: technological and pedagogical. The possibility of distributed functioning and control of this architecture, the hardware - software reconfigurability and scalability, and the use of structured reprogrammable logic promote a variety of practical activities, which enhance its complexity level as it increases the degree of integration within the platform. In addition, the possibility of remote laboratories promotes practical activities from a distance, so enhancing the use range of the platform. Two practical activities were propose to illustrate the use of the platform, which were developed by students of a postgraduate course. Accordingly, groups of students had to specify and then implement a task in each holon using the corresponding PLC. The student groups expressed an interest in activities development by reviewing and deepening the concepts studied in theoretical sessions, motivated by visualization of operating hardware functioning. This was made possible due to practical applications were not isolated, but immersed in a truly functional and accessible for students platform.

References [1]

[2]

[3]

[4]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

Dominguez, M.; Reguera, P.; Fuertes, J.J. Laboratorio Remoto Para la Enseñanza dela Automatica en la Universidad de Leon (España). Revista Iberoamericana de Automatica e Informatica Industrial, Valencia, v.2, n.2, p. 36-45, Abr. 2005. Tzafestas, Costas S.; Palaiologou, Nektaria; Alifragis, Manthos. Virtual and Remote Robotic Laboratory: Comparative Experimental Evaluation. IEEE Transactions On Education, v. 49, n. 3, p. 360-369, Aug. 2006. D'abreu, João Vilhete. Integração de Dispositivos Mecatronicos para Ensino-Aprendizagem de Conceitos na Area de Automação. Campinas, 309 p. Tese (Doutorado), Universidade Estadual de Campinas, Faculdade de Engenharia Mecanica, 2002. Chella, Marco Túlio. Ambiente de Robótica Educacional com Logo. XXII Congresso da Sociedade Brasileira de Computação – SBC2002, Florianopolis - Brasil, p. 1-8, 2002.


[20]

[21]

[22]

Yu, Xudong; Weinberg, J.B. Robotics in education: new platforms and environments. IEEE Robotics & Automation Magazine, v. 10, n. 3, p. 3, Sep. 2003. Saire, Alfredo; Gómez, Henry. Plataforma De Aprendizaje A Distancia En Automatización Industrial Empleando Laboratorios Remotos. Invest Apl Innov, v. 2, n. 2, p. 109-116, 2008. Ariza, Carlos Fernando; Amaya, Dario. Laboratorio Remoto Aplicado a la Educacion a Distancia. Ciencia e Ingenieria Neogranadina, Bogotá - Colombia, v. 18, n. 2, p. 131-145, Dic. 2008. Colombo, Armando et al. A Collaborative Automation Approach to Distributed Production Systems. Proceedings of 2nd IEEE International Conference on Industrial Informatics (INDIN´04), v.1, p. 1-6, Jun. 2004. Van Brussel, Hendrik et al. Reference Architecture for Holonic Manufacturing Systems: PROSA. Computers In Industry, special issue on intelligent manufacturing systems:, Elsevier Science Publishers B. V., v.37, n.3, p. 255-276, Dec. 1998. Leitão, Paulo; Colombo, Armando; Restivo, Francisco. Lecture Notes on Computer Science: An Approach to the Formal Specification of Holonic Control Systems. 27442004th ed.: Springer, 2004. 1090 p. ISBN: 9783540407515. Leitão, Paulo; Colombo, Armando; Restivo, Francisco. Formal Specification of ADACOR Holonic Control System: Coordination Models. Proceedings of the 44th IEEE Conference on Decision and Control, and the European Control Conference 2005. Iorio, Luis Carlos. Redes de Comunicação em Automação Industrial, Solução Tecnologica da Plataforma PIPEFA. Campinas, 152 p. Dissertação (Mestrado), Universidade Estadual de Campinas, Faculdade de Engenharia Mecanica, 2002. Aihara, Cintia Kimie. Projeto e Implantação de Plataforma Didática Aplicada ao Ensino e Pesquisa em Automação. Campinas, 234 p. Dissertação (Mestrado), Universidade Estadual de Campinas, Faculdade de Engenharia Mecanica, 2000. Baffi, Antonio Carlos. Projeto e Implementação de uma Plataforma Didatica de Manipulação Utilizando Conceitos de Automação e Integração. Campinas, 124 p. Dissertação (Mestrado), Universidade Estadual de Campinas - Faculdade de Engenharia Mecanica, 2001. Fathullah. M. et al., Investigation on Nylon PA66 Side Arms Using Taguchi and ANOVA Analysis in Reducing Cost of Producing Urinary Catheters, International Review of Mechanical Engineering (IREME), V.5, N. 7, p. 1278-1286, Nov. 2011. De Souza, Alisson. Projeto do sistema de controle de uma bancada didática para posicionadores Eletro-Hidráulicos Proporcionais. Florianopolis, 102 p. Tese (Doutorado), Universidade Federal de Santa Catarina, Universidade Federal de Santa Catarina, 2007. Barbosa, Alípio Monteiro et al. As Empresas Juniores E A Confecção De Bancadas Didáticas Como Atividades Inovadoras No Ensino De Engenharia. Anais do XXXIV COBENGE, Passo Fundo, p. 361-370, set. 2006. Seleski, Fernando; Oliveira de Araujo, Lindolpho. Projeto de Sistemas de Controle Programável de Sistemas Produtivos Distribuídos. Synergismus scyentifica UTFPR, Pato Branco, v. 1, p. 729-735, Jan. 2006. Ezral. M. et al. An Approach to Mechanization and Automation of Manual Construction Activity. International Review of Mechanical Engineering (IREME), V.5, N. 7, p. 1266-1271, Nov. 2011. Seitzinger, Joyce. Be Constructive: Blogs, Podcasts and Wikis as Constructivist Learning Tools. Learning Solutions e-Magazine: Practical Applications of Technology for Learning, p. 1-14, Jul. 2006. Van Brussel, Hendrik et al. Reference Architecture for Holonic Manufacturing Systems: PROSA. Computers In Industry, special issue on intelligent manufacturing systems:, Elsevier Science Publishers B. V., v.37, n.3, p. 255-276, Dec. 1998. Ramazan Bayindir, Selami Bulut, Design and Implementation of an Electrical Furnace for Firing Ceramic Products, International Review of Mechanical Engineering (IREME), V. 5, N. 7, p. 13071313, Nov. 2011.


1033



Universidad Militar Nueva Granada, Bogotá-Colombia, E-mail: [email protected] 2 Campinas State University, Campinas-Brasil E-mail: [email protected] 3 Universidad Militar Nueva Granada, Bogotá-Colombia E-mail: [email protected]

Ricardo A. Castillo was born in Colombia in 1980, and received his B.Sc. degree in Mechatronics Engineering by the Military University, Bogota, Colombia in 2004. Currently he is completing a M.Sc. in Mechanical Engineering at Campinas State University, Sao Paulo, Brazil working on Collaborative Automation Systems modeling and educational oriented applications. He has worked as a professor and researcher at the Military University, Colombia since 2005 been involved in Robotics, Mechatronics and Automation areas. In addition Prof. Castillo has developed projects related to Artificial Intelligence and mobile autonomous robotics. João M. Rosário, was educated at Campinas State University, São Paulo, Brazil receiving the B Sc. degree in mechanical engineering in 1981 and the M.Sc. degree in systems and control in 1983 and Specialization Degree in Production and Automation Systems in 1986 at Nancy University, France. He was awarded the Ph.D. degree in 1990 by Ecole Centrale – Paris, France, for research into Automation and Robotics. He worked briefly as a control engineer and robotics in the Hispano Suiza, France and underwater robotics at GKSS, Germany. Actually, he’s invited professor at Automation and Control department, at SUPELEC, France. Currently he is an associated professor at Faculty of Mechanical Engineering at the University of Campinas, UNICAMP, responsible of the Automation and Robotics Laboratory, and coordinator of Robotics and Automation in the Brazilian Manufacturing Network. From 1998-2002 was the head of the graduate course of Automation and Control Engineering (Mechatronics) Department. Actually develops various industrial research projects at national and international level in different areas such as: Industrial Automation, Control Design of Mechatronics Systems and Biomechanics.



1034


Injector Nozzle Spray on Compressed Natural Gas Engines: A Technical Review Semin

Abstract – In this article, previous work on the application of injector on CNG engines is reviewed. In the review of injector, the spray characteristics, the gas jet structure, the effect of injector nozzle geometry on fuel-air mixing, injector nozzle coefficients of discharge, injector nozzle spray tip penetration and cone angle as well as injector nozzle orifice shapes are outlined. Fuel-air mixing increases as the orifice diameter decreases. This can be a significant advantage for small orifice nozzles. However, multiple orifices are required to meet the desired mass flow rate as orifice diameter decreases. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved. Keywords: Fuel-Air Mixing, Gas Jet Structure, Injector Nozzle, Nozzle Orifice Shapes, Spray Characteristic

µa ∆P θ

Nomenclature Ca Cv d D do k l MWamb MWjet Pa Pf Pt Pamb r r1/2 Re t u uc uo ucl x X Xc Xcl Y Ycl Yo z µ

Area concentration ratio Velocity coefficient Nozzle orifice diameter Particle diameter Jet orifice diameter Boltzmann’s constant (1.38 x 10-23J/K), proportionality constant Orifice length Molecular weight of the ambient fluid Molecular weight of the jet fluid Ambient pressure Injection pressure Pressure total Pressure ambient Radial position Radial position at which concentration of velocity has decayed to half Reynold number thickness velocity Velocity of center of vortex Velocity at jet origin (sonic) Centreline velocity Axial position Jet fluid mole fraction Kleinstaein dimensionless jet core length Jet fluid mole fraction at jet centerline Rorational energy transition branch, jet fluid mass fraction Jet fluid mass fraction at jet centerline Jet fluid mass fraction at jet origin Penetration distance per nozzle orifice Experimentally fit constant

α αa αd ρa ρamb ρf ρo κ η

Ambient air viscosity Pressure difference across the orifice length Cone angle in the measured spray (nonuniform velocity) Cone angle Correction factor for spark ignition engines Correction factor for compression ignition engines Air density Density of ambient fluid Fuel density Density of jet fluid at orifice Kleinstein decay constant for velocity or concentration Dimensionless radial distance

I.

Introduction

In principle, the utilisation of an optimal fuel-air mixture should yield the required power output with the lowest fuel consumption that is consistent with smooth and reliable operation [1]. Over past decades, the SPI system has evolved into an electronic, pulse-widthmodulated system which utilises sequentially-timed individual injections at each intake port. According to Zhao [1], these transient sprays of 2.5 to 18.0 ms in duration and in constant phase relative to the intake valve event, either for the start of injection or the end of injection, offer significant advantages in terms of engine transient response and hydrocarbon (HC) emissions. It is important to note, however, that the meteoric expansion in the use of such systems has generally out-paced the basic knowledge and understanding of the complex, transient fuel sprays that they produce [1]-[43].



1035

Semin

According to Shiga [2], improvement of CNG injector nozzle hole geometry and understanding the processes in engine combustion is a challenge because the compression-ignition combustion process is unsteady, heterogeneous, turbulent, three-dimensional and complex. In SPI CNG engines, NG is injected by the fuel nozzle injector via the intake port into the combustion chamber, and mixing with air must occur before ignition of the gas fuel [43]. To improve the mixing process of CNG fuel and air in the combustion chamber, the geometry of nozzle holes, nozzle spray pressure, piston head modifications, the arrangement of piston top clearance, turbulent flow of air intake and variations in the CNG fuel angle of spray can be investigated [3]. The CNG fuel spraying nozzle can be modified experimentally and by computational investigations of engine power, cylinder pressure and specific fuel consumption. Czerwinski [4], [5] has investigated the sequential injection of CNG, which offers several advantages to increase CNG engine performance. The injector’s multi-hole geometry was developed to produce optimum fuel-air mixing in the engine to promote comparable engine performance [6]. According to Czerwinski [5], CNG SPI has advantages, especially greater efficiency. The power, fuel consumption and thermal efficiency of the engine are higher than in carburettor and single point systems. In the port injection CNG engine, every cylinder has at least one injector and the fuel is injected from the intake manifold into the engine cylinder when the intake valve is opened.

II.

Spray Characteristics

In contrast to the significant amount of work undertaken on the fuel-air mixing process, less information has been reported on the structure of the fuel spray from SPI injectors. The fuel spray process has long been viewed as a key element in engine performance and emissions [1]. However, it is only in recent years that the SPI process in spark ignition engines has begun to receive similar attention. Unlike the high-pressure fuel spray for which numerous standard correlations exist, low-pressure gasoline spray is relatively uncharacterised. Even though a relatively complete knowledge base has been established for fuel spray [7], much of the information cannot be easily translated to correlate the characteristics of SPI gas fuel sprays. The implications of gas fuel injection and its behaviour on combustion chamber design have received considerable attention in recent years. Most present day direct injection engines have been optimised for use with liquid fuels. The optimal configuration may be different from what is required for gaseous fuels. Abraham and Bracco [8] examined the relative effects of combustion on liquid and gaseous fuel direct injection jets. Results showed that fuel-air mixing and burning rates were initially slower for gas injection jets than for liquid sprays.


However, burning and mixing rates of the gaseous direct injection in the subsequent stages of combustion increased in comparison with liquid fuel sprays [44]. Initially, the liquid jet was more effective at entraining air, and hence was better at producing flammable mixture regions within the combustion chamber. In the end, however, the liquid spray did not burn as completely as the gas jet. Beyond the initial stage, the gas direct injection jet exhibited a higher combustion rate than that of the liquid fuel [9]. The relative decrease in burning rate for the liquid spray resulted from vapourisation of the remainder of the liquid fuel in the mixture. This caused an increase in the richness of the mixture behind the flame, leading to a smaller energy release than that exhibited by the gas jet [9]. These significant contributions have led to a better understanding of transient gaseous jet behaviour. Research on the jet model, based on experimental results, has focused on four main regions. Figure 1 illustrates a transient gas jet showing these regions: the potential core region, the main jet region, the mixing flow region, and the dilution region [10], [11]. In the potential core region shown, the gas is injected; in the main jet region; in the mixing flow region, the gas fuel is mixed with air; and in the dilution region, the gas fuel and air mixture is in the combustion process. The gaseous jet plume is characterised by a high velocity, low temperature core of rich unmixed fuel confined to the jet axis [12]. This core region is referred to as the main jet region and contains the bulk of the unmixed fuel. Turbulent vortices are generated on the periphery of the jet core as a result of shear forces exerted by the ambient air in the chamber [10], [12]. In the region close to the nozzle exit, however, less turbulence is generated. This region, the potential core region, is very stable and is characterised by very low mixing rates. It extends from the nozzle exit to a distance corresponding to a penetration distance (z) per nozzle orifice diameter (d) ratio of approximately 12.5 [11], [13]. Beyond this region, large scale vortices are generated by shear forces as air is entrained into the jet, resulting in vigorous mixing along the jet periphery. This region of enhanced mixing surrounds the main jet and is referred to as the mixing flow region. As the fuel loses its momentum, it is pushed aside by fuel flowing from upstream. The tip of the jet expands radially, forming the dilution region of the jet. This region corresponds to the jet tip and is characterised by low velocities and a high fuel concentration. An experimental study by Tanabe and Sato [14] showed that the flammable region exists only within the thin layer around the periphery of the jet. In the vicinity of the nozzle, the temperature is low and the jet velocity is high, thus resulting in poor mixing conditions. Such conditions do not provide a favourable environment for chemical reactions to take place. Downstream of the nozzle, in the mixing flow region, the temperature in the


1036

Semin

periphery of the jet is relatively high, and the mixture is roughly stoichiometric. The flow has stabilised in this area as result of momentum change between the jet and the entrained air, which provides sufficient residence time for a chemical reaction to occur. In summary, this is the most likely location for ignition to take place and, correspondingly, it is the ideal place to install a glow plug (in the periphery of the jet downstream of the potential core region). Results from Aesoy and Valland [15] further support these findings.

Fig. 1. Transient gas jet structure [11]

The core of the jet is fuel-rich and very cold. The temperature of the core is, in fact, lower than the fuel temperature before injection. This is due to the rapid expansion of the jet at the nozzle exit and the high velocity of the jet [9].The outer edge of the jet is higher in temperature because of heat transfer that occurs as the hot air is entrained into the jet. It would be very difficult to achieve ignition if an ignition assist device were to be placed the core of the jet, because a great deal of thermal energy input would be required to elevate the core to the autoignition temperature. Furthermore, the high temperature gradient between the glow plug and the jet core would result in short plug service life. These conclusions provide some explanation for the experimental findings of Aesoy and Valland [15] and Thring and Leet [16].

III. Gas Jet Structure Fig. 2 indicates the structure of a steady underexpanded gas jet with a pressure ratio Pt/Pamb>2 [17].


Initially, the gas jet expands sharply to match the reduced ambient pressure, and, as a result, the region directly downstream of the nozzle becomes supersonic. The initial expansion waves quickly reflect back towards the centreline as strong compression waves and coalesce to the oblique shocks remains supersonic. Near the centreline, the initial expansion is the highest, and the compression to ambient pressure is subsonic, thus, a slip line must exist between the supersonic and subsonic regions. Along the shear layer that is formed at the slip line, the supersonic flow accelerates the subsonic flow. If this acceleration is sufficient, the shock structure may be repeated. There has been some disagreement regarding the pressure ratio required for a second Mach disk. Donaldson and Snedeker [18] reported that a second normal shock may occur at pressure ratios as low as Pt/Pamb = 4. Ewan and Moodie [19] reported only one Mach disk in their experiments for pressure ratios up to Pt/Pamb = 14. Further downstream, the flow decays as a series of oblique shocks that follow the same structure as a moderately underexpanded jet. The length of this core region increases with Pt/Pamb, and again there is some disagreement regarding the typical length can be observed. Ramskill [17] report that the core region is typically 4 to 6 diameters long, but Ewan and Moodie [19] contend that the core region can extend for 10 to 30 diameters beyond the orifice. Rubas [20] reported that surrounding the core region of the jet is a turbulent shear layer. Upstream of the first normal shock, this turbulent mixing region is very thin, and the initial widening of the jet is due to the expansion to ambient pressure, so very little entrainment occurs in this initial region. As the jet progresses, turbulent mixing with the surroundings decelerates the jet and begins to increase the width by mass entrainment. The core eventually diminishes until an entirely subsonic jet remains. After a short transition region, a fully-developed region with a Gaussian velocity profile evolves. It is this fullydeveloped region back towards the nozzle which leads to a virtual origin called the jet pole. Normally, the jet pole lies a few diameters upstream of the exit to the orifice, but its position varies somewhat with the pressure ratio. In the fully-developed region, entrainment of the surrounding gas increases the mass flow rate at a given cross-section, but also decreases the velocity such that the momentum flux is constant. Most models of underexpanded sonic jets include two sections: the initial region, which contains both the core and transition zones, indicated in Fig. 2 and the far field or fully-developed region. The basic structure of the initial region is well characterised, but the details of this region are often unimportant in predicting the overall penetration and mixing characteristics of the jet because it is relatively short and comparatively little mixing occurs within it.


1037

Semin

The structure in the fully-developed region is independent of the initial conditions except for its starting location and initial diameter. From experimental data, Antsupov [21] has shown that the spreading angle of the fully developed region in an air jet is a constant 10.2 degrees from the jet centreline, regardless of the initial conditions.

between the mass fraction and the volume fraction for a binary mixture, which is derived from Dayton’s law of partial pressure: X =

1 (1 − Y ) MW jet 1+ Y MWamb

(5)

A curve fit of Eq. (4) and (5) in the form Eq. (2) introduces less than 3.8% error in predicting the radial mole fraction decay with: 2 X = e −53.0η X cl

(6)

Rubas [20] reported that µ in Equation (2) is a linear function of the molecular weight of the jet fluid based on experimental data from N2 and CO2 jets: Fig. 2. Underexpanded gas jet [17]

µ = 0.008125MW + 0.4925 Birch [22] hypothesised that the jet spreading angle is independent of the density ratio between the jet and ambient fluids based on comparisons between methane into air and air into air jets. The mean properties of the jet are predictable using a number of experimentally verified models [17]. The difficulty in predicting the behaviour of gas jets is in determining the length of the initial region and in finding a starting point for fullydeveloped region models. Many references provide a similar relationship for the radial velocity and the concentration-volume fraction in the fully-developed region of the jet that take the form given by Equations (1) and (2), respectively [17], [19] and [20]: 2 u = e − kη (1) ucl 2 X = e − µ kη X cl

(2)

(7)

with the assumption of Eq. (7) and the product µk = 53.0 from Eq. (6), k in Equation (1) for methane is estimated to be 85.1 and the radial velocity profile for a methane jet becomes: 2 u = e −83.1η (8) ucl with different values for µ and k, Eq. (1), (2) and (4) also hold using an alternative definition of η:

η, =

r r1 / 2

(9)

Eq. (9) is valid because Birch et al. (1978) have shown that r1/2 (defined in terms of the mass fraction) is proportional to x, where: r1 / 2 = 0.097 x

(10)

here, the dimensionless radial distance (η) is defined as: r η= x

(3)

The constants µ and k are experimentally determined and are dependent up on the species present in the gas jet and the surroundings. Birch [22] developed a similar relationship for the radial concentration (mass fraction) decay in a round free jet of methane in air: 2 Y = e−73.6η Ycl

(4)

Equation (4) can be written as a volume fraction relation in the form of Equation (2) using the conversion Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

The shapes of the velocity and concentration profiles given by Eq. (1), (2) and (4) follow the experimentally observed Gaussian profile. Kleinstein [23] proposed a widely accepted axial velocity and concentration (mass fraction) decay formula for axial symmetric, turbulent, compressible free jets based on a linearisation of the momentum equations. ucl Y −1 ⎡ ⎤ , = 1 − exp ⎢ ⎥ 0.5 uo Yo ⎢ κ 2 x ⎛ ρ amb ⎞ − X ⎥ c⎥ ⎢ d ⎜ ρ ⎟ ⎣ o⎝ o ⎠ ⎦

(11)

Kleinstein [23] assumed constant values for the eddy viscosity, κ, of 0.074 for velocity decay and 0.104 for


1038

Semin

concentration decay based on fits with previously published experimental data. Kleinstein [23] asserted that the dimensionless core length, Xc, equals 0.7 for all cases. Other researchers have modified the Kleinstein decay formulation using an alternate definition for κ to obtain slightly better fits with experimental data [17], [19]. Once the denominator of the exponential term becomes large (>10 yield a maximum error of 5%), Equation (11) can be approximated by: ucl Y , = uo Yo

1 2x ⎛ ρ ⎞ κ ⎜ amb ⎟ do ⎝ ρo ⎠

0.5

(12)

Minor variations of Eq. (12), in which velocity and concentration decay inversely with x, are widely used with various experimentally-fit values of κ [8], [24], [25]. Birch [22] developed a variation of Equation (12) for the concentration decay of a methane jet in air:

κ ' do ⎛ ρo ⎞ Y = Yo ( x + xo ) ⎜⎝ ρ amb ⎟⎠

0.5

(13)

For which κ’ = 4.0 and xo=-5.8d give good agreement with experimental data in the far-field region (defined as x ≥ 25do). Typical values for κ’ range from 4 to 6, with the higher values fitting better for near-field data. Equations (11) and (13) agree well for x/do ≥ 20.

Fig. 3. Transient gas jet [10]

Figure 3 illustrates a model of the internal structure of an incompressible transient gas jet based on particle image velocimetry (PIV) images by Fujimoto [10]. Although the gas jet exiting the fuel injector in injection NG is sonic (requiring a compressible analysis), entrainment with the surroundings decelerates the jet, and it can eventually be treated as incompressible. Thus, the leading portion of the gas jet from the fuel injector will eventually follow structure described by Fujimoto [10]. The potential core region for the transient gas jet is analogous to the potential core region in the steady-state jet.


Similarly, little entrainment occurs in this region. The region that immediately surrounds the jet’s centreline is called the main jet region, and the flow here is nearly axial. Immediately downstream of the potential core is the mixing flow region, which is divided into two parts: the first is the sub-mixing flow region and the second is the main mixing region. The sub-mixing flow region is a transition between the potential core and the main mixing flow region and it initiates the growth of the vortices that begin to entrain the surrounding fuel. The bulk of entrainment and mixing occur in the main mixing flow region because the vortices here are very large. At the head of the transient jet is a stagnant region where a large-scale recirculation flow called the tip vortex exits. The fuel at the leading edge of the jet loses axial momentum to the surroundings and gets pushed aside radially by the fuel immediately upstream. Little mixing with the surroundings occurs in the stagnation region because the velocity fluctuations and the vorticity are both too small. Because there is little mixing in this region, an annulus of relatively high jet fuel concentration exits around the head of the jet.

IV.

Injector Nozzle Geometries Effect

Numerous studies have suggested that decreasing the injector nozzle orifice diameter is an effective method of increasing fuel air mixing during injection [9], [26]. Smaller nozzle holes are found to be most efficient at fuel-air mixing, primarily because the fuel-rich core of the jet is smaller [12], [26]. In addition, decreasing the nozzle orifice diameter would reduce the length of the potential core region (defined as z/d = 12-51), and hence increase the size of the mixing flow. Unfortunately, decreasing the nozzle hole size causes a reduction in the turbulent energy generated by the jet. Since fuel-air mixing is controlled by turbulence generated at the jet boundary layer, this offsets the benefits of the reduced jet core size. Furthermore, jets emerging from smaller nozzle orifices have been shown not to penetrate as far as those emerging from larger orifices. This decrease in penetration means that the fuel is not exposed to all of the available air in the chamber. For an excessively small nozzle size, the improvements in mixing related to decreased plume size may be negated by a reduction in radial penetration [12]. Another significant feature of injection flow that requires attention is the tendency of the fuel plume to attach to the cylinder head. This behaviour is undesirable because it restricts penetration to the chamber’s extremities where a large portion of the air mass resides. Furthermore, it hampers air entrainment from the head side of the plume because the exposed surface area of the plume is reduced. The phenomenon responsible for jet deflection is known as the Coanda effect [9]. The effect arises as a


1039

Semin

consequence of the velocity and pressure fields surrounding the jet. Low pressure areas are formed above and below the jet, due to the entrainment of air mass into the jet from the local surrounding volume. Below the jet, there is a significant air mass in the volume between the piston and the injector. Above the jet, space is limited, and the air must be entrained from progressively further downstream. As the jet develops, the air gets entrained into areas which the jet would penetrate. The entrainment flow is strong enough to deflect the fuel jet upwards, causing it to attach to the cylinder head. This phenomenon must be carefully avoided in the design of a NG engine combustion chamber. The mixing was maximised when the nozzle tip was placed equidistant from the piston and cylinder head [9]. A nozzle containing many small holes would provide better mixing than a nozzle consisting of a single large hole [12]. This hypothesis has been tested by studying injectors with varying numbers of nozzle holes. The diameters of the holes were adjusted such that each nozzle delivered the same overall fuel mass flow. Computational analysis examining eight hole, four hole, two hole and one-hole nozzles revealed that the mixing rate improved with the number of nozzle holes [9], [12]. Jennings and Jeske [12] carried out a similar analysis for 8, 12 and 16-hole nozzles. Contrary to the trend, the 16hole injector performed poorly due to plume merging. Plume merging has an adverse effect on mixing because the total plume surface area available for mixing is decreased.

V.

Injector Nozzle Coefficient of Discharge

The coefficient of discharge (COD) for micro-nozzles for compressible gas flow has been measured by Snyder [27]. In later research by Siebers [28], the similar COD values were obtained again for different orifices. The COD was defined as Aeffective per Ageometry, where Aeffective was calculated from the standard isentropic compressible mass flow relation. There is a possibility for a further increase in COD values with a further increase in the Reynold number. The small orifice diameter and large l/d ratio has a strong influence on the values of COD for compressible gas flow [13]. Equations (14) – (17) are proposed for the estimation of COD [11], where d is orifice diameter, l is orifice length, COD is the coefficient of discharge, CODmax is the maximum coefficient of discharge and Re is the Reynold number. The COD decreases when cavitation occurs; indeed, the COD decreased from 0.7 to 0.6 when cavitations occurred in tests reported by [13]. Arai [29] measured COD value using diesel injection systems, where cavitations are believed to occur. The l/d ratio varied from 1 to 50, and the orifice diameter varied from 0.1 to 0.4 mm. Nearly the same coefficient of discharge (0.7) Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

for different diameter orifices were obtained. Naber and Siebers [30] obtained COD values of about 0.6 for different diameter orifices using a diesel injection system. Each orifice was tapered in the diameter, increasing by approximately 0.015 mm from the minimum diameter near the inlet to the maximum diameter at the outlet. In a later report by Siebers[28], similar COD values were obtained again for different diameter orifices. The COD values were about 0.8, which are greater than the values obtained in a previous report. The reason for the difference in COD values is not reported:

COD =

2
XT

(2)

In this equation, XT is the spray tip penetration, XL is the distance from spray tail to the nozzle hole and α is a time dependent parameter which is determined from the principle of fuel mass conservation. This conservation emphasize that the mass of the injected fuel should be equal to the mass of fuel contained in spray at any crank angle position. The principle could be shown as [20]: International Review of Mechanical Engineering, Vol. 6, N. 5

1058

R. Akbarpour Ghiasi, Y. Ajabshirchi

t

XT

b

∫0 m f dt = 2π ∫X ∫0 cρ ydxdy

(3)

L

where ρ is the mixture density. In the case of spray tip penetration, the proposed formula by Chiu et al. [19] makes divergence in numerical solution of equations. To resolve this problem, another formula is used to determine this penetration in which the diesel spray artificial break-up time has been considered [22]. This formula could be indicated as: ⎛ Pinj − P ⎞ X T = 3.72 ⎜ ⎟ ⎝ ρa ⎠

((

)

0.25

( d (t − t )) inj

tanh t − tinj / tbr

)

0.5

inj

m f zone k = 2 π ∫

(4)

0.6

in which Pinj is the injection pressure, P is the in-cylinder pressure, ρa is the density of air, dinj is the nozzle hole diameter, tinj is the injection during time and tbr is the mentioned artificial break-up time. This parameter could be presented as:

( (

tbr = 28.65 ρ f dinj / ρ a Pinj − P

))

(5)

in which ρf is the fuel density. Finally, the distance from spray tail to the nozzle hole in equation (2) could be calculated as:

before the end of injection ⎧XL = 0 ⎨ ⎩ X L = 0.5 X T after the end of injection II.2.

(C zone), premixed combustion zones (Bi zones), diffusion combustion zones (Bj zones) and air zone (A zone). The diffusion combustion zones are formed along the skin of core zone since air continues to entrain into the spray. In addition, premixed and diffusion zones have desired quantities of the fuel. The burning of premixed zones is instantaneously and the rates of this burning are high. But, mixtures in diffusion combustion zones are burned with low burning rates. It must be noted that all of the zonal boundaries have constant equivalence ratios at each time step. Fuel and air masses of the zones are given by the following formula [19], [20]: c ρ ydxdy

(9)

k +1

ϕk

∫ϕ (1 − c ) ρ ydxdy

(10)

k +1

Ns ⎛ −m u y − pVk + ⎞ ⎜ k ∑ i =1 k i,k ⎟ 1 ⎜ k k ⎟ Tk = ⎜ +Qk + β m f h f + χ me he ⎟ mk cvk ⎜ ⎟ ⎜ +ψ m m hm + ξ m a ha ⎟ ⎝ ⎠

(11)

where:

Fuel spray undergoes a delay period before the start of ignition. For calculating this period, there are a large number of correlations which are based on the theoretical and/or experimental investigations. But in current study, the following relation is selected [23]:

τ = 4.3 × 10−3 P −2.5 φ tot −1.04

XT XL

XL

ϕk

∫ϕ

In this formula, φk and φk+1 are equivalence ratios of zone boundaries. In addition of the mentioned equations, a series of the time-derivate energy conservation equations are solved in current research work. Number of these equations is equal to the number of cylinder zones (NZC) at any crank angle. These equations are introduced as:

(6)

Ignition Delay of Fuel Spray

⎛ 5000 ⎞ exp ⎜ ⎟ ⎝ T ⎠

mazone k = 2 π ∫

XT

⎧1 ⎩0

β =⎨

in core zone ⎧0 in air zone χ =⎨ in other zones ⎩1 in other zones ,

⎧−1 in air zone ⎩0 in other zones

ξ =⎨ (7)

where φ tot is the total equivalence ratio, P is the incylinder pressure and T is in-cylinder temperature. It should be mentioned that ignition starts when the following condition is satisfied:

⎧0 ⎪

ψ = ⎨ −1 ⎪1 ⎩

in Bi, previously formed Bj and air zones in core zone at the time of Bj formation in new formed Bj zone

(8)

yi,k is the mass fraction of the ith species in the kth zone, uk is the internal energy of the kth zone, mk is mass of the kth zone, p is in-cylinder pressure, Vk is the rate of volume change in kth zone, Q is the rate of heat transfer

Governing Equations and Their Solution Methods

from/to kth zone, m f , m a , m e , m m are the mass rates of

As shown in Fig.1, in current model, the vapor jet is divided into different zones such as: rich fuel core zone

fuel, air, air entrainment and mixture, respectively. Also, h is the specific enthalpy.

t

dt

∫0 τ

≥1.0

k

II.3.



1059


In addition, the ideal gas state equations of different zones are participated in calculations. These equations could be shown as: PVk = nk R Tk

k = 1,...,NZC

(12)

in which nk is total mole number of kth zone. Also, the rate of cylinder volume change is utilized as a constraint of problem. This rate could be indicated as [6]: 1 ⎧ ⎡ − ⎤⎫ sin θ − 1 R 2 − sin 2 θ 2 ⎥ ⎪ ⎪ θ ⎢ 1 C V = VC ⎨ ( rc − 1) ⎢ 2 ⎥ ⎬ (13) 2 ⎪ ⎢( − sin ( 2θ ) ) θ ⎥⎪ ⎣ ⎦⎭ ⎩

(

)

In this equation, VC is the clearance volume, rc is compression ratio, RC is the ratio of connecting rod length to crank radius and θ is crank angle. Equations (11), (12) and (13) are solved with AdamsBashforth predictor-corrector method [24]. Then the pressure is obtained from the ideal gas state equation of cylinder charge. It should be mentioned that the composition of burned products is calculated by using Olikara-Borman's method [25]. Also, it should be noted that the thermodynamic properties are determined based on the local temperatures and zonal fuel-air ratios. II.4.

0.7 ⎧ K ⎫ ⎪a B ( Re ) (Tb − TWall ) + ⎪ ⎨ ⎬ 4 ⎪+bA Tb4 − TWall ⎪ ⎩ ⎭

(

)

(14)

where a and bA are the coefficients with constant quantities, Re is Reynolds number based on the piston speed and cylinder bore, B is the cylinder bore, Tb is the bulk temperature of cylinder charge, TWall is the wall temperature, Stot is the total surface of combustion chamber at any crank angle and K is the thermal conductivity of the zone. The parameters Stot and K could be calculated by the following correlations:

Stot

Cp µ 0.7

(16)

where Sh is the cylinder heat surface area, Sp is the piston crown surface area, Ls is stroke, Cp is the specific heat capacity and µ is the dynamic viscosity. As mentioned before, total heat transfer is apportioned to the various zones in relation to their mass and temperature [18]. In current research work, heat transfer between the next zones is considered as conduction form. Also depend on the zone number, convection or radiation formula is used to calculate heat transfer between the zone and the cylinder wall. Fig. 2 shows heat transfer calculations procedure in more details. It must be noted that the air swirl is not shown in Figs 3 and 4, but, this important phenomena is considered in calculations.

Heat Transfer Equations

It is obvious that the calculation of transferred heat from each zone to the next zones is an important stage in improvement of the simulation model. This transferred energy has important effect in zonal temperature and consequently in concentrations of the species. It is worth to know that the Annand's correlation [21] is the most commonly used model of heat transfer in scientific articles about multi-zone combustion models. This correlation could be shown as:

Q = Stot

K=

⎡ RC + 1 − cosθ + ⎤ 1⎥ ⎢ ⎥ 2 2 − RC − sin θ 2 ⎥ ⎣⎢ ⎦

π B Ls ⎢ = Sh + S p + 2

(

)

(15)


Fig. 2. Some important parameters of heat transfer calculations in spray different zones (k is the zone number)

The associated correlations for calculations of heat transfer to/from the different zones of the model are shown as: Fuel core zone: ⎧ ⎛ TC − Tk −1 ⎞ ⎫ k ⎪QC = KC SC ⎜ ⎟⎪ ⎝ X C′ − X k′ −1 ⎠ ⎪ ⎪ ⎪ ⎪ K + K k −1 ⎪ KCk −1 = C ⎪ , 2 ⎪⎪ ⎪⎪ ⎨ ⎬ XC + X L , ⎪ X C′ = ⎪ 2 ⎪ ⎪ X C + X k −1 ⎪ ⎪ ⎪ X k′ = ⎪ 2 ⎪ ⎪ ⎩⎪Tk −1 > TC ⎭⎪

(17)

Burning zones:


1060


⎧ ⎛ Tk − Tk −1 ⎞ k −1 ⎪Q k = − K k S k −1 ⎜ ⎟+ ⎝ X k′ − X k′ −1 ⎠ ⎪ ⎪ ⎪+ K k S ⎛⎜ Tk +1 − Tk ⎞⎟ − b S T 4 − T 4 Wall ⎪ k +1 k ⎝ X k′ +1 − X k′ ⎠ A k k ⎪ ⎪ K k −1 = K k + K k −1 ,K k = K k +1 + K k k +1 ⎨ k 2 2 ⎪ X k +1 + X k X + X k −1 ⎪ ′ , X k′ = k , ⎪ X k +1 = 2 2 ⎪ ⎪ X ′ = X k −1 + X k − 2 ⎪ k −1 2 ⎪ T T T > > k k +1 ⎩ k −1

(

)

⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ (18) ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎭

Air zone: KA ⎧ ⎫ 0.7 ⎪QA = − Stot a B ( Re ) (TA − TWall ) + ⎪ ⎪ ⎪ ⎛ Tk +1 − TA ⎞ ⎪ ⎪ k +1 − K A Sk +1 ⎜ ⎟ ⎪ ⎪ ′ ′ ⎝ X k +1 − X A ⎠ ⎪⎪ ⎪⎪ ⎨ k +1 K A + K k +1 ⎬ ⎪K A = ⎪ 2 ⎪ ⎪ X k +1 + X k X T + X k +1 ⎪ ⎪ , X A′ = ⎪ X k′ +1 = ⎪ 2 2 ⎪ ⎪ ⎩⎪Tk +1 > TA ⎭⎪

NZC

NZC

k =1

k =1

k =1

∑ Q k = ∑ PVk + ∑ U k

(21)

burning rate and lower heating value of the fuel, respectively. Emissions Formation

II.6.

In the case of NO emission study, formation kinetics is based on the extended Zeldovich mechanism. Associated reactions of this mechanism and their rate constants are shown in Table I.

K1

N + O2

K ZZZ X YZZ Z NO + O 2

13

9

K3

K 3 = 4.1 × 10

13

It should be mentioned that the rate of NO formation is obtained after some simplification processes [6,17] and this rate is shown as:

d [ NO ] dt

According to equation (1), y is a function of x, b and t for a constant value of equivalence ratio. Also parameter b is the radius of spray cross-section and is a function of x. In reality, this cross-section is elliptic because of the air swirl existence, but for the aim of simplification, it is considered as equivalent circular form [19]. In addition, Xk in this equation is the length of the zone with number k in spray axial direction and this parameter is a function of zonal boundary equivalence ratio. The lower limit of integral (XL) is the distance of spray tail from the nozzle hole. In equations (18) and (19), coefficients a, bA and also cylinder wall temperature are considered as 3.28×1011 W/m2K4, 0.35 and 550 K, respectively. These quantities are selected same as the quantities in Annand's formula for the purpose of preparing the same conditions in comparing of the results.

⎛ 3150 ⎞ ⎟ ⎝ Tk ⎠

K 2 = 6.4 × 10 T exp ⎜ −

ZZZ X N + OH YZZ Z NO + H

(20)

⎛ 38000 ⎞ ⎟ ⎝ Tk ⎠

K1 = 7.6 × 10 exp ⎜ −

(19)

XL

II.5.

NZC

It is necessary to mention that m fb and LHV represent

ZZX Z N 2 + O YZZ Z NO + N

Xk

∫ 2 π yk dx

H .R.R. = m fb × LHV −

TABLE I NO FORMATION REACTIONS AND RELATED ARRHENIUS RATES Reaction Arrhenius rate equation

For first time, in Cummins engine multi-zone models, surface area of each zone at each crank angle is calculated by some mathematical processes. The base formula of these processes is an integral which is introduced as: Sk =

net heat release rate as [6]:

(

2 K1 [O ]e [ N 2 ]e =

)

⎛ ⎛ NO ] ⎜1 − ⎜ [ ⎜ ⎜ [ NO ] e ⎝ ⎝

⎞ ⎟ ⎟ ⎠

2

⎞ ⎟ ⎟ ⎠

⎛ ⎛ [ NO ] ⎞ ⎜ ⎜ ⎟ K O N ⎜ [ NO ] ⎟ 1 [ ]e [ 2 ]e ⎜ e ⎠ ⎝ ⎜1 + ⎜ K 2 [ N ]e [O2 ]e + K 3 [ N ]e [OH ]e ⎝

(

(

)

)

⎞ ⎟ ⎟ ⎟ ⎟ ⎠

(22)

in which the brackets with subscription e show the concentrations in equivalent state. In the case of CO emission study, formation and oxidation processes are controlled kinetically. This formation is assumed to be closely linked to the complex fuel decomposition process and can be written as: d CO f dt

1

⎛ −12000 ⎞ = A f m fv P 2 exp ⎜ ⎟ ⎝ Tk ⎠

(23)

in which Af is the total surface area of formatted CO and mfv is the mass of fuel vapor. Also, the rate of CO oxidation is calculated by considering kinetics of the main oxidation reaction. This reaction and its rate calculations are as follows [17]:

Heat Release Rate K

Energy conservation equations are combined to give


1 ZZZ X CO + OH YZZ Z CO2 + H

K2

(24)


1061


{

dCOox = K1 [OH ]e [CO ] − [CO ]e dt

}

⎛ T ⎞ K1 = 6.76 × 1010 exp ⎜ k ⎟ ⎝ 1102 ⎠

(25)

(26)

Finally, the net production rate of CO is computed as: d COnet d CO f dCOox = − dt dt dt

(27)

In the case of soot emission study, various relations have been introduced by the researchers. One important correlation is Nishida's formula [26]. This formula contains calculations of soot formation and oxidation rate and also total mass rate (Eqs. (28)-(30)). These correlations are suitable to use in MZCM of diesel engines [17]. In equation (28) and (29), Asf and Asc are the total surface area of formatted and combusted soot, respectively. Also in equation (30), ms is mass of soot and PPO2 is the partial pressure of O2: ⎛ −6313 ⎞ = Asf m fv P 0.5 exp ⎜ ⎟ dt ⎝ Tk ⎠

(28)

⎛ −7070 ⎞ dmsc = Asc ms PPO2 P1.8 exp ⎜ ⎟ dt ⎝ Tk ⎠

(29)

dmsnet dmsf dmsc = − dt dt dt

(30)

dmsf

II.7.

Error Percentage Calculation

Error percentage is an important creation in evaluating of the developed model’s prediction accuracy. This parameter could be calculated as: ⎛ X − X exp Error (% ) = 100×⎜ ⎜ X exp ⎝

⎞ ⎟ ⎟ ⎠

(31)

In this formula, X is the engine characteristic parameter (such as in-cylinder pressure) or emission concentration and also Xexp is the experimental value of X.

III. Results and Discussion The first parameter which is analyzed in current study is the in-cylinder pressure. It should be noted that experimental data of this parameter have been obtained by Pirouzpanah et al. in different operating conditions of OM-355 diesel engine [27], [28]. Specifications of this engine are given in Table II.


In addition, for illustration of the developed MZCM ability in prediction of the engine combustion parameters and emissions, three cases of study are selected. Specifications of the selected cases are shown in Table III. TABLE II SPECIFICATIONS OF OM-355 ENGINE Engine specification Cylinder bore (m) Connecting rod (m) Compression ratio Number of cylinders Cylinder stroke (m) Nozzle hole diameter (mm) Nozzle holes number Injection pressure (bar) IVC (CA-BTDC) EVO (CA-ATDC)

Quantity 0.128 0.28 16.1 6 0.15 0.31 4 175 120 118

TABLE III SPECIFICATIONS OF THE SELECTED CASES IN CURRENT STUDY Case number 1 2 3 Engine speed (rpm) 1200 1400 1600 Rate of fuel mass (g/s) 5.47 4.81 8.13 Rate of air mass (g/s) 120.38 141.79 148.48 Total equivalence ratio 0.68 0.51 0.82 Trapping temperature (K) 314.2 313.2 300.2 Trapping pressure (bar) 1.13 1.14 0.98 Start of injection (CA-BTDC) 13 14 15 End of injection (CA-ATDC) 2.5 3 4

For the selected cases of study, the in-cylinder pressure variations and also the calculated levels of errors are shown in Figs. 3 and 4, respectively. Also in these figures, result of the developed Cummins engine MZCM and the older model, which applies Annand's correlation in heat transfer calculations, are compared. As shown in Fig. 4, predicted values by the developed model has maximum error levels of 10.4% for case No. 1, 9.6% for case No. 2 and 10.2% for case No. 3. At the same time, in older Cummins engine model, these percentages are about 15.5 % for case No. 1, 17.3% for case No. 2 and 20.3% for case No. 3. These percentages show significant prediction improvement of the developed model over the traditional model. The second characteristic parameter which is considered in this study is the net heat release rate. As known, this parameter is an important factor in engine performance evaluations. Also analysis of this parameter gives more realistic view about of the model accuracy to the researcher. So, variations of the net heat release rate for the selected cases are given in Fig. 5. It should be mentioned that the experimental data of heat release rate analysis in this figure is obtained according to the method which has been described previously by Fathi et al. [29]. Same as the in-cylinder pressure analyses, improvement in predictions of the new model is clearly considered in heat release rate figures. The in-cylinder temperature is another important parameter that is seriously affected from transferred heat quantity.


1062


(a)

(a)

(b)

(b)

(c)

(c)

Fig. 3. The in-cylinder pressure variation for: (a) case No. 1, (b) case No. 2 and (c) case No. 3

Fig. 4. Error percentages of the pressure for: (a) case No. 1, (b) case No. 2 and (c) case No. 3



1063


In Fig. 6, variations of this parameter are shown for the selected cases and also for both considered multizone models. As shown in this Figs. 5 and 6, both temperature and net heat release rate of the case No. 3 have higher quantities in comparison with two other cases. This is because of the longer ignition delay period [5], [6] and also because of the higher load of engine in this case. The longer ignition delay period causes greater quantity of premixed mixtures of fuel and air which burn with higher rates of heat release. At the time of premixed charge burning, air entrains into spray and prepares combustible mixture which burns with lower rates of heat release. This type of burning is nominated as diffusion burning.

(a)

Fig. 6. The in-cylinder temperature variations for the selected cases of study

In the section of emission analyses, NO, CO and soot concentrations are calculated and also compared with the associated experimental data [27], [28]. The NO concentration level is calculated kinetically [17] by using Eq. (22). The variations of this emission for the selected cases of study are indicated in Fig. 7. As seen in this figure, the quantity of NO rises after the start of ignition and continues to a peak value. Finally, it freezes when the temperature is decreased during expansion process. It must be noted that, because of the higher level of incylinder temperature in case No. 3, NO concentrations have higher quantities in this case. It is essential to compare the obtained theoretical values of emissions with the associated experimental data. So, this comparison in the case of NO concentrations (in mass bases) is shown in Table IV. Some higher error percentages are because of this point that the experimental data have been obtained for NOX (not only for NO) concentrations. As seen in this table, the developed MZCM can predict the quantities better than the older model.

(b)

(c) Fig. 5. Net heat release rate variations for: (a) case No. 1, (b) case No. 2 and (c) case No. 3



1064


Fig. 8. The in-cylinder temperature variations for the selected cases of study

Fig. 7. The in-cylinder temperature variations for the selected cases of study TABLE IV COMPARISON BETWEEN EXPERIMENTAL AND THEORETICAL NO Case 1 Case 2 Case 3 NO Experimental (g/kW.hr) 8.05 7.76 13.25 Present model (g/kW.hr) 7.02 6.88 12.48 Present model error (%) 12.8 11.34 5.81 Annad’s correlation model (g/kW.hr) 6.79 9.25 15.54 Annad’s correlation model error (%) 15.65 19.2 17.28

In Figs 8 and 9, variations of the CO and soot concentrations are indicated for the selected cases. As shown in these figures, concentrations of both species follow same trends. This is because of the similarity in mechanisms of productions. As shown in Fig. 8, from the start of ignition, carbon monoxide concentrations are increased rapidly for all of the considered cases. Then most part of these generated CO is oxidized rapidly. Because of this oxidation process, the concentrations are decreased sharply and approached to the lower quantities at exhaust valve opening. In Table V comparisons between experimental and numerical data of CO are shown in mass bases. It can be seen that the error percentages are in acceptable ranges for the developed model in comparison with the older one. Also, with considering soot production trends in Fig. 9, it can be seen that the oxidation of this species is done with a blunt slope for all of the selected cases. In addition, considerable quantities of this emission (nearly 1 to 2 (g/m3)) could escape from the oxidation process at the latter part of expansion step. Same as the other emissions, comparisons between the experimental and theoretical results for soot are given in Table VI. As shown in this table, the error percentages are in acceptable ranges for the developed model in cases No. 2 and No. 3. But this is not true in case No. 1. It must be mentioned that the error levels for engine characteristic parameters and exhaust emissions will be better in the developed simulation model if the spray wall impingement, spray primary and secondary break ups, droplets collisions, coalescence and evaporation subprocesses and the effect of residual gases are considered.


Fig. 9. The in-cylinder temperature variations for the selected cases of study TABLE V COMPARISON BETWEEN EXPERIMENTAL AND THEORETICAL CO Case 1 Case 2 Case 3 CO Experimental (g/kW.hr) 1.86 3.73 3.46 Present model (g/kW.hr) 1.53 3.36 3.11 Present model error (%) 17.74 10 10.11 Annad’s correlation model 2.42 3.05 2.95 (g/kW.hr) Annad’s correlation model error 30.1 18.23 14.74 (%) TABLE VI COMPARISON BETWEEN EXPERIMENTAL AND THEORETICAL SOOT Case 1 Case 2 Case 3 Soot Experimental (g/kW.hr) 0.47 0.62 0.32 Present model (g/kW.hr) 0.53 0.79 0.38 Present model error (%) 12.76 23.39 18.75 Annad’s correlation model (g/kW.hr) 0.42 0.84 0.23 Annad’s correlation model error (%) 10.63 35.48 28.12

IV.

Conclusion

In this article, Cummins engine MZCM is developed for theoretical calculations of combustion parameters in


1065


direct injection diesel engines. Also this model is used to determine important emissions concentrations in different operating conditions. The development is done by reforming in heat transfer calculations. Then the results of the new model are compared with experimental data and also with the outputs of the older model. The older model use the most common Annand's correlation in heat transfer calculations. Results of analysis reveal that the developed MZCM can predict the in-cylinder pressure, the net heat release rate and the main emissions (CO, NO and soot) better than the older Cummins model. It could be seen that the maximum error percentage for in-cylinder pressure prediction is 10.2 % for the developed model but, this quantity is 20.3% for the older one. Also, in prediction of engine emissions, error percentages are in acceptable levels for the developed MZCM. Maximum error percentages of emissions which are predicted by this model are 12.8 %, 17.74 % and 23.39 % for NO, CO and soot, respectively. But these percentages are 19.2 % for NO, 30.1 % for CO and 35.48 % for soot in older model.

Acknowledgements The authors would like to write in memoriam of Professor Vahab Pirouzpanah, who devoted his life to enlightening a myriad of students as well as making noteworthy contributions to several aspects of internal combustion engines. In addition, the authors wish to thank gratefully Mr. Morteza Fathi, MSc. of mechanical engineering, due to his ongoing scientific support in heat release analysis. Also, the authors would like to thank Iranian Fuel Conservation Organization (IFCO) for financial support of this work.

References [1] [2]

[3]

[4] [5] [6] [7]

[8]

[9]

C. Baumgarten, Mixture Formation in Internal Combustion Engines (1st edition, Springer, 2006). H. Hiroyasu, T. Kadota, M. Arai, Development and Use of a Spray Combustion Modeling, Bulletin of the JSME, Vol. 26, n. 214, pp. 569-575, 1983. S. M. Shahed, W. T. Flynn, W. T. Lyn, A model for the formation of emissions in a direct injection diesel engine, In J. N. Mattavi, C. A. Amann, Combustion Modeling in Reciprocating Engines, (New York: Plenum Press, 1980, 345-368). J. H. Weaving, Internal Combustion Engineering: Science and Technology (Elsevier Applied Science, 1990). J. I. Ramos, Internal Combustion Engine Modeling (Hemisphere Publishing Corporation, 1989). J. B. Heywood, Internal Combustion Engine Fundamentals (McGraw Hill International Editions, 1989). T. M. Belal, E. M. Marzouk, M. M. Osman, Numerical Investigation on the Effect of Different Carbon Dioxide and Hydrogen Concentrations on Closed Cycle Diesel Engine Performance and Pollutant, IREME, Vol. 4, n. 6, pp. 733-747, 2010. H. Ge, Y. Shi, R. Reitz, D. Wickman, W. Willems, Development Using Multi-dimensional CFD and Computer Optimization, SAE Technical Paper Series, No. 2010-01-0360, 2010. L. Shenghua, J. W. Hwang, J. K. Park, M. H. Kim, J. O. Chae, Multizone Model for DI Diesel Engine Combustion and Emissions, SAE Technical Paper Series, No. 1999-01-2926, 1999.


[10] P. Ottikkutti, J. Vangerpen, K. R. Cui, Multi-Zone Modeling of a Fumigated Diesel Engine, SAE Technical Paper Series, No. 910076, 1991. [11] O. Durgun, Z. Şahin, Theoretical Investigation of Heat Balance in Direct Injection (DI) Diesel Engine for Neat Diesel Fuel and Gasoline Fumigation, Energy Conversion and Management, Vol. 50, n. 1, pp. 43-51, 2009. [12] P. Zhou, S. Zhou, D. Clelland, A Modified Quasi-Dimensional Multi-Zone Combustion Model for Direct Injection Diesels, Int. J. of Engine Research, Vol. 7, n. 4, pp. 335-346, 2006. [13] D. A. Kouremenos, C. D. Rakopoulos, D. Hountalas, Multi-Zone Combustion Modeling for the Prediction of Pollutants Emissions and Performance of Direct Injection Engines, SAE Technical Paper Series, No. 970635, 1997. [14] P. Q. Tan, Z. Y. Hu, K. Y. Deng, J. X. Lu, D. M. Lou, G. Wan, Particulate Matter Emission Modeling Based on Soot and SOF from Direct Injection Diesel Engines, Energy Conversion and Management, Vol. 48, n. 2, pp. 510-518, 2007. [15] G. Stiesch, G. P. Marker, A Phenomenological Model for Accurate and Time Efficient Prediction of Heat Release and Exhaust Emissions in Direct Injection Diesel Engines, SAE Technical Paper Series, No. 1999-01-1535, 1999. [16] Z. Gao, W. Schreiber, A multi-zone analysis of soot and NOx emission in a D.I. diesel engine as a function of engine load, wall temperature and intake air O2 content. Proc. ASME Fall Technical Conf., ICE division, Peoria, Illinois, 2000. [17] Z. Bazari, A DI Diesel Combustion and Emission Predictive Capability for Use in Cycle Simulation, SAE Technical Paper Series, No. 920462, 1992. [18] S. M. Shahed, A. Mathematical Model of Diesel Combustion, Proc. IMechE, C94/75, pp. 119-128, 1975. [19] W. S. Chiu, S. M. Shahed, W. T. Lyn, A Transient Spray Mixing Model for Diesel Combustion, SAE Technical Paper Series, No. 760128, 1976. [20] Z. Şahin, O. Durgun, Multi-Zone Combustion Modeling for the Prediction of Diesel Engine Cycles and Engine Performance Parameters, Applied Thermal Engineering, Vol. 28, n. 17-18, pp. 2245-2256, 2008. [21] W. J. D. Annand, Heat Transfer in the Cylinders of Reciprocating Internal Combustion Engines, Proc. IMechE, Vol. 177, n. 3, pp. 973-990, 1963. [22] A. J. Yule, L. Filipovic, On the Break-up Times and Lengths of Diesel Sprays, Int. J. of Heat and Fluid Flow, Vol. 13, n. 2, pp. 197-206, 1992. [23] V. Pirouzpanah , R. Khoshbakhti Saray, Enhancement of the Combustion Process in Dual-fuel Engines at Part Loads Using Exhaust Gas Recirculation, Proc. ImechE Part D, Vol. 221, n. 7, pp. 1-12, 2007. [24] J. H. Mathews, Numerical Methods for Computer Science, Engineering and Mathematics, (Prentice-Hall Int. Editions, 1987). [25] C. Olikara, G. L. Borman, A Computer Program for Calculating Properties of Equilibrium Combustion Products with Some Applications to IC Engines, SAE Technical Paper Series, No. 750468, 1975. [26] K. Nishida, H. Hiroyasu, Simplified Three-Dimensional Modeling of Mixture Formation and Combustion in a Direct Injection Diesel Engines, SAE Technical Paper Series, No. 890269, 1989. [27] V. Pirouzpanah, B. O. Kashani, Prediction of Major Pollutants Emission in Direct Injection Dual Fuel Diesel and Natural Gas Engines, SAE Technical Ppaer Series, No. 1999-01-0841, 1999. [28] V. Pirouzpanah, S. M. Mirsalim, Y. Jeyhouni, M. Afghahi, Reduction of pollutants emissions of OM-355 diesel engine to Euro2 by converting dual fuel engine (diesel+gas), Proc. 1st Conf. Conversion of Automotive Fuel to CNG, Tehran, 2003, pp. 84-94. [29] M. Fathi, R. Khoshbakhti Saray, M. David Checkel, Detailed Approach for Apparent Heat Release Analysis in HCCI Engines, Fuel, Vol. 89, n. 9, pp. 2323-2330, 2010. [30] Amoresano, A., Cameretti, M.C., Tuccillo, R., Combined experimental - Numerical approach for the fuel jet study in a LPP combustor, (2011) Proceedings of the ASME Turbo Expo, 2 (Parts A and B), pp. 1097-1108.


1066



Internal Combustion Engines Laboratory, Department of Mechanical Engineering, University of Tabriz, Tabriz, Iran 2

Department of Agricultural Machinery, University of Tabriz, Tabriz, Iran

(Corresponding Author), Building No. 8, Department of Mechanical Engineering, University of Tabriz, 29 Bahman Avenue, P.O.Box: 51666-14766, Tabriz, Iran. Tel: +984113392461 E-mail: [email protected] Reza Akbarpour Ghiasi Date of birth: 1978 BSc: Mechanical Engineering, Solid Mechanics, University of Tabriz, Tabriz, Iran, 1997-2001. MSc: Mechanical Engineering, Energy Conversion, University of Tabriz, Tabriz, Iran, 2002-2005. He published one journal paper and three

conference papers. His research interests are: 1- Internal Combustion Engines 2- Combustion and Emission Modeling 3- Thermodynamic Analysis of the Thermal Systems Yahya Ajabshirchi Date of birth: 1949 BSc: Agricultural Engineering, Mechanics of Agricultural Machinery, University of Tabriz, Tabriz, Iran, 1956-1960 MSc: Mechanical Engineering, Internal Combustion Engines, French Institute of Petroleum, France, 1974-1976. PhD: Mechanical Engineering, Energy Conversion, University of Tabriz, Tabriz, Iran, 1994-2000. Associate Professor in department of Agricultural Machinery, University of Tabriz, Tabriz, Iran. He published 15 journal papers and 8 conference papers. His research interests are: 1- Combustion 2- Emission Control in Internal Combustion Engines 3- Renewable Energies Dr Ajabshirchi is a member of Iranian Society of Combustion.



1067


Prediction of Machining Parameters of Surface Roughness of GFRP Composite By Applying ANN and RSM S. Ranganathan1, T. Senthilvelan2

Abstract – Glass fibre-reinforced polymer (GFRP) composites are alternative to engineering materials because of economic, light weight, corrosive resistance and superior properties. The experimental research under taken by the scholars is to study the impact of machining parameters on surface roughness of composite material by applying artificial neural network (ANN) and response surface method (RSM). The orthogonal turning operations were carried out on the composite material using tungsten carbide (WC) insert. During machining, the cutting speed (Vc), feed rate (fs) and depth of cut (ap) were varied. Turning experiments were designed based on the statistical three level full factorial experimental design techniques. An artificial neural network and response surface method have been developed, which can predict the surface roughness of the machined workpiece. The experimental results concur well with the results obtained from predictive model. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Composite Material, Turning, Surface Roughness, Artificial Neural Network (ANN), Response Surface Method (RSM)

I.

Introduction

Glass fiber-reinforced polymer (GFRP) composites are finds many applications such as aerospace, aircraft, automobile, robots and machine tools because of their excellent properties such as high specific strength, low thermal expansion, light weight, corrosive resistance and good dimensional stability. For making hole on the GFRP plate, conventional drilling with twist drill remains one of the most economical and efficient machining process for riveting and fastening structural assemblies in the automotive and aerospace industries. Turning of glass fiber-reinforced polymer (GFRP) composite materials are facing difficulties while machining because of fibre pull-out, matrix bonding and fibre fuzzing. Sarma et al studied the evaluation of surface roughness in machining of GFRP composite using digital image processing [1]. Birhan Ishk investigated the surface roughness in orthogonal cutting of unidirectional GFRP composite [2]. Wang et al developed a regression model to predict the cutting force while orthogonal cutting of GFRP composite material by using different rake angle, clearance angle, depth of cut and cutting speed [3]. Lasri et al modeled the chip separation in machining of FRP composite by using stiffness degradation concept [4]. Naveen Sait et al optimized the machining parameters of glass fiber reinforced composite using desirability function [5]. Palanikumar et al analyzed the surface roughness parameters in turning of FRP composite material using PCD cutting tool [6].


Seeman et al studied the tool wear and surface roughness in machining of particulate aluminum metal matrix composite using response surface methodology [7]. Erol Kikickap modlled and optimized the burr height in drilling of Al-7075 using response surface methodology [8]. Noordin et al described the performance of coated carbide tools when turning of AISI 1045 steel using response surface methodology [10]. The wear behaviour of glass fibre-polyester composite was studied by Amar Patnaik et al and the results indicate that erodent size, fiber loading, impingement angle and impact velocity are the significant factors in a declining sequence affecting the erosion wear rate [11]. Lee presented the precision machining of GFRP with respect to tool characteristics and stated that the surface roughness is not changed according to the depth of cut and cutting speed for the various tools and the surface roughness and surface quality is good in the GFRP cutting at decreased feed rates [12]. Paulo Davim and Mata studied the GFRP material fabricated by hand laid up method and turned by PCD and cemented carbide cutting tools to investigate the surface roughness and specific cutting pressure [13]-[14]. The proposed methodology of combining artificial neural network (ANN) and response surface methodology (RSM) has wide applications like prediction of machining parameters in turning of GFRP composites WC insert. In the present work, composite GFRP rod was turned by using WC insert under different cutting parameters, viz., cutting speed (Vc), feed rate (fs) and depsth of cut (ap) at three different levels of cutting Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

1068

S. Ranganathan, T. Senthilvelan

parameters. Experiment were conducted using the Taguchi method of parametric design have been employed for prediction of surface roughness Therefore, the artificial neural network (ANN) and response surface methodology (RSM) has been considered for the prediction of surface roughness of the turned GFRP composites. Finally, analysis of variance (ANOVA) test have been conducted to find the contribution of machining parameters to predict the surface roughness.

II. II.1.

Experimental Procedure Fabrication of GFRP Composite Rod

The work piece material used for the present investigation was glass fiber reinforced plastic having a 30 mm diameter and length of 250 mm fabricated by hand-lay-up method. The roving are set an angle of 300 while fabricating the GFRP rod and shown in Fig. 1. In the present work, the artificial neural network (ANN) and response surface method (RSM) were developed to determine the surface roughness and the main and interaction effects of process parameters viz cutting speed (Vc), feed rate (fs) and depth of cut (ap) were studied through analysis of variance (ANOVA).

Fig. 1. Fabricated GFRP rod and WC inserts

II.2.

Machining Conditions

Turning experiments were designed based on the statistical three level full factorial experimental design techniques. Design of experiments (DOE) methods can be an important part of a thorough system optimization, yielding definitive system design or redesign recommendations. In this present study, factorial design creates 3n training data, where n is the number of variables. In these studies, three independent variables, such as the cutting speed (Vc), feed rate (fs) and depth of cut (ap) had total of 33 = 27 experimental runs. The range of process parameters are shown in table. I. The turning of GFRP composite work was carried out on an ALL GEAR LATHE machine with a maximum speed of 1200 RPM and a 5 KW drive motor. Fig. 1 gives the detail of the machine with workpiece. TABLE I RANGE OF PROCESS PARAMETERS AND THEIR LEVELS Cutting speed (Vc) Feed rate (fs) in Depth of cut Level in m/min m/rev (ap) in mm 1 29.68 0.25 0.4 2 73.04 0.376 0.8 3 113.1 0.381 1


III. Results and Discussion III.1. Analysis of Variance (ANOVA) Analysis of variance (ANOVA) test was performed to identify the significance of the process parameters which affect the surface roughness of GFRP composite turning operations. The result of ANOVA for the cutting performance is presented in Tables II. Statistically, the Ftest provides a decision at a 95% confidence level to estimates the significant parameters on deciding the performance characteristics. Low experimental error of 0.680% is found for turning operations reveals the accuracy of the applied procedure. From this study, the percentage contribution of the cutting parameters on the surface roughness (Ra) is shown in Table II and it can be observed that the feed rate is found to be that it is decide the surface roughness (Ra) and contribute 31.08 %, combination of cutting speed and feed rate decides the surface roughness by 31.8 % and the combination of cutting speed, feed rate and depth of cut contribute to predict the surface roughness by 21.74 %. The cutting speed and depth of cut (ap) have less significant factor on the surface roughness of the turned surface. TABLE II ANOVA TABLE FOR THE CUTTING PERFORMANCE Source of Df SS Mean SS % Contribution Variation Cutting Speed 1 0.6480 0.6480 3.28 A Feed rate B 1 6.1256 6.1256 31.08 Depth of cut 1 0.7921 0.7921 4.01 C AB 1 6.276 6.276 31.8 AC 1 0.1089 0.1089 0.55 BC 1 1.464 1.464 7.43 ABC 1 4.2849 4.2849 21.74 Error 8 0.00287 0.000357 0.014

III.2. Artificial Neural Network (ANN) for Prediction of Surface Roughness in Turning of GFRP Composite Artificial Neural Networks (ANN) has been established as a tool for effortless computation. Artificial neural network (ANN) have been successfully employed in solving problems in areas such as fault diagnosis, process identification, property estimation, data smoothing and error filtering, product design and development, optimization, and estimation of activity coefficients. A neuron is the basic element of neural networks, and its shape and size may vary depending on its duties. An Artificial neural network (ANN) may be seen as a black box that contains hierarchical sets of neurons (i.e. processing elements) producing outputs for certain inputs and the mechanism of the different steps involved in the training process have been discussed in the published literatures [9]. In this study, the effects of cutting speed, feed rate and depth of cut on workpiece


1069


surface roughness were statistically investigated in turning of GFRP composite. The best artificial neural network (ANN) architecture was designed by Neurointelligence evaluation software package. Neuron in the input layer corresponds to cutting speed (Vc), feed rate (fs) and depth of cut (ap). The output layer corresponds with surface roughness (Ra). In this model, the inputs are fully connected to the hidden layer and the hidden layer neurons are fully connected to the outputs. The input and output layer have nine neurons and three neurons, respectively. Artificial neural network (ANN) models have one hidden layer with four neuron and one output neuron as shown in Fig. 2.

Fig. 3. Surface roughness by experimental and ANN prediction

Fig. 2. Artificial Neural Network predictive model Fig. 4. Error improvement with respect to the number of iteration

For training the network, the TRAIN function of Neurointelligence software package was used [16]. The function works on batch back propagation algorithm. By batch back propagation method in the Neurointelligence software package the learning rate is given as 0.1 and a momentum 0.9 was chosen. The training process takes about 14 seconds on an IBM-P4 processor PC for about 50,000 training iterations. Mean square error for the training data was computed to be 0.008608. Training of the neural network model was performed using 27 experimental data set. The results predicted from the artificial neural network (ANN) and response surface method (RSM) model are compared with those obtained from the experimental test in Table. 4. It is seen from Table. IV show that the (ANN) prediction has a good agreement with the (RSM) predictions and experimental results. Fig. 3 shows the experimental results and the artificial neural network prediction of surface roughness of the turned GFRP material. Fig. 4 shows the error improvement with respect to the number of iteration of ANN prediction with experimental results of turning of GFRP. From the results, it is found that the developed ANN prediction model has a good interpolation capability and therefore it can be used as an efficient tool for the prediction of surface roughness of the turned parts. The scatter plot presented in Fig. 5 is reveals that the residuals fall on a straight line implying that the errors are distributed normally with respect to the predicted values. The residuals do not show any obvious pattern and are distributed in both positive and a negative direction implies that the model is adequate.


Fig. 5. Scatter plot of target and experimental output

III.3. Response Surface Method to Predict the Surface Roughness (Ra) Response surface methodology (RSM) is an advanced tool, now a day’s commonly applied involves three factorial designs giving number of input (independent) factors and their corresponding relationship between one or more measured dependent responses [17]-[18]. A simpler and more efficient statistical model using RSM designed in Design-Expert .8 evaluation software package. The method used to predict surface roughness by Response surface methods (RSM) is central composite design, 3-factor approach. RSM creates polynomial models for the available data set as must rate in the following equation.


1070


n

n

n

∑ βi xi + ∑ βi x 2i + ∑∑ βij xi x j + ε i =1

i =1

(1)

i =1 j =1

Surface roughness = +6.14 - 0.39 * A - 0.23 * B + 0.21 * C + 0.56 * A * B -0.090 * A * C + 0.65 * B * C + 0.55 * A2 + 0.036 * B2 - 0.081 * C2 where β0, βi and βij are turning parameters at different cutting parameters level and n is the number of model parameters. In creating RSM models 27 experimental data measurements obtained from the cutting conditions i.e. cutting speed, feed rate and depth of cut are shown in Table III. TABLE III COMPARISON OF THE SURFACE ROUGHNESS (RA) OF EXPERIMENT RESULT WITH ARTIFICIAL NEURAL NETWORK PREDICTION AND RESPONSE SURFACE MODEL PREDICTIONS Ex Cutting Speed Feed rate Depth of cut Surface roughness (Ra) in No (Vs) m/min (fs) mm/rev (ap) mm µm Experiment ANN RSM 1 29.68 0.25 0.4 7.42 7.3962 7.2498 2 29.68 0.25 0.8 6.31 7.6152 6.6542 3 29.68 0.25 1 8.5 5.5198 5.2134 4 29.68 0.376 0.4 8.42 8.4897 8.3214 5 29.68 0.376 0.8 7.51 7.4670 7.4625 6 29.68 0.376 1 6.92 6.9425 6.9348 7 29.68 0.381 0.4 5.36 5.6327 5.5324 8 29.68 0.381 0.8 6 5.8893 5.9874 9 29.68 0.381 1 5.54 5.5255 5.4587 10 73.04 0.25 0.4 7.1 7.0973 7.0687 11 73.04 0.25 0.8 6.22 6.0396 6.1257 12 73.04 0.25 1 7.38 7.3945 7.3754 13 73.04 0.376 0.4 6.42 6.4856 6.3124 14 73.04 0.376 0.8 6.46 6.3935 6.2147 15 73.04 0.376 1 8.92 8.8925 8.7487 16 73.04 0.381 0.4 9.46 9.1345 9.2478 17 73.04 0.381 0.8 8.88 8.9499 8.9478 18 73.04 0.381 1 7.56 6.6931 6.8745 19 113.1 0.25 0.4 6.97 6.6087 6.7492 20 113.1 0.25 0.8 6.45 6.4428 6.3548 21 113.1 0.25 1 5.59 5.5097 5.4789 22 113.1 0.376 0.4 7.76 7.7313 7.6483 23 113.1 0.376 0.8 6.7 6.7257 6.8261 24 113.1 0.376 1 8.5 8.4395 8.6124 25 113.1 0.381 0.4 5.34 6.3416 6.2879 26 113.1 0.381 0.8 6.81 6.8691 6.7425 27 113.1 0.381 1 7.26 5.4708 6.9875

Fig. 9 shows a SEM micrograph of the machined surface of GFRP composite bar machined at 113.1 m/s, feed rate of 0.381 mm/rev and depth of cut of 1 mm. It was observed that, the workpiece has less rupture of fibre due to high machining conditions. Fig. 10 shows a SEM micrograph of the machined surface of GFRP composite bar machined at 29.68 m/s , feed rate of 0.25 mm/rev and depth of cut of 0.4 mm. It was observed that, the workpiece has more rupture of fibre due to low machining conditions. III.4. Comparison of Surface Roughness (Ra) Values Between ANN and RSM An attempt was made to compare the artificial neural network (ANN) with response surface method (RSM) for the study of turning of GFRP composite. The predicted output values of artificial neural network (ANN) with response surface method (RSM) are shown in Table IV. Though both the models performed well and offered stable responses in predicting the combined interactions of the independent variables with respect to the response, yet the ANN based approach was better in fitting to the measured response in comparison to the RSM method. Both construction of an artificial neural network model and response surface method requires less computational time and low cost by Neurointelligence and Design-Expert.8 evaluation package respectively. Normal Plot of Residuals

Design-Expert® Software Surface roughness Color points by value of Surface roughness: 8.5

99

95

Normal % Probability

5.34

90 80 70 50 30 20 10 5

1

-2.47

-1.26

-0.04

1.17

2.39

Internally Studentized Residuals

Fig. 6. Normal probability plot Vs residual

The adequacy of the model has been carried out using residual analysis, predicted values are presented in Fig. 7. The normal probability plots is presented in Fig. 6 and reveals that the residuals fall on a straight line implying that the errors are distributed normally with respect to the predicted values. The residuals do not show any obvious pattern and are distributed in both positive and a negative direction implies that the model is adequate. The relation between predicted values and experimental values are shown in Fig. 7. The experimental values are very close to the predicted values and it has been seen that most of the points are close to the centre line and hence the empirical model provides reliable prediction.

Predicted vs. Actual

Design-Expert® Software Surface roughness 8.50

Color points by value of Surface roughness: 8.5 5.34

7.70

Predicted

n

Y = β0 +

6.90

6.10

5.30

5.32

6.11

6.91

7.70

8.50

Actual

Fig. 7. Predicted and experimental values of the surface roughness



1071


Surface roughness

Design-Expert® Software 0.38

Surface roughness Design Points 8.5 5.34

0.35

Actual Factor C: Depth of cut = 0.63

B: Feed rate

6.25726

X1 = A: Cutting speed X2 = B: Feed rate

6.25726

6

0.32

6.5868

6.91635 0.28

7.24589

7.57543 0.25 29.68

50.54

71.39

92.25

113.10

A: Cutting speed

Fig. 8(a) Fig. 10. Typical SEM image of the machined surface of the GFRP composite at lower cutting speed (29.68 m/s), feed rate (0.25 mm/rev) and depth of cut (0.25 mm)

Surface roughness


Surface roughness Design Points 8.5 5.34

Actual Factor B: Feed rate = 0.32

0.81

C: Depth of cut

X1 = A: Cutting speed X2 = C: Depth of cut

The ANN model requires a several iterative computations and selection of learning rate, momentum and weight randomization range will decide the accuracy of the results. Comparison of experimental measurements with predicted results from ANN and RSM is shown in Fig. 11.

7.04845

6.79999 0.63

6.55153 6.30307

6

6.05461

0.44

0.25 29.68

50.54

71.39

92.25

113.10

A: Cutting speed

Fig. 8(b) Surface roughness


Surface roughness Design Points 8.5

6.46936 6.17619

5.34

Actual Factor A: Cutting speed = 71.39

0.81

C: Depth of cut

X1 = B: Feed rate X2 = C: Depth of cut

6.17619

6

0.63

5.88301

Fig. 11. Comparison of experimental measurement with predicted results from ANN and RSM

6.46936

0.44

5.58984 5.29667 0.25 0.25

0.28

0.32

0.35

0.38

IV.

Conclusion

B: Feed rate

Fig. 8(c) Fig. 8 (a), (b) and (c) Estimated contour plots for surface roughness in turning of GFRP composites

Fig. 9. Typical SEM image of the machined surface of the GFRP composite at higher cutting speed (113.1 m/s), feed rate (0.381 mm/rev) and depth of cut (1 mm)


The experimental observations of the surface roughness (Ra) of the turned part were incorporated into ANN based Neurointelligence software package and Design Expert 8.0 evaluation package for turning of GFRP composite and the findings are follows: ¾ From the ANOVA tables, the feed rate (fs) is the single process parameter to contribute 31.08 % for deciding the surface roughness (Ra) of the turned surface. ¾ The combination of cutting speed (Vc) and feed rate (fs) are the most significant parameter to decide the surface roughness (Ra), which contributed 31.8 % turning of GFRP composites and combination of all cutting parameters are contribute to 21.74 % for surface roughness of the turning of GFRP composites. ¾ The depth of cut is of least significant parameter to predict the surface roughness of the machined surface.


1072


¾ An artificial neural network (ANN) model and response surface method (RSM) model were developed to predict the surface roughness within the wide range of cutting parameters, found to be capable of better prediction of surface roughness within the range that they had been trained. Good agreement has been observed between the predictive models and experimental results. ¾ The results of the artificial neural network (ANN) and response surface method (RSM) model are robust and accurate to estimate the surface roughness of the turned part in turning of GFRP composites. ¾ Using the ANN and RSM models, saving in time and cost can be achieved for finding surface roughness (Ra) in turning of GFRP composites.

References [1]

P.M.M.S. Sarma, L. Karunamoorthy, K. Palanikumar, Surface roughness parameters evaluation in machining GFRP composite by PCD tool using digital image processing, Journal of Reinforced Plastics and Composites, Vol. 28,No 13, 1567-1585, 2009. [2] Birhan Isik, Experimental investigations of surface roughness in orthogonal turning of unidirectional glass-fiber reinforced plastic composite, International Journal of Advanced Manufacturing Technology, Vol. 32, No 1-2, 42-48, 2008. [3] DH. Wang, M. Ramulu, D. Arola, Orthogonal cutting mechanisms of graphite epoxy composite. Part I: unidirectional laminate, International Journal of Machine Tools and Manufacture, Vol. 35, 1623–1638, 1995. [4] L. Lasri, M. Nouari, M. El Mansori, Modelling of chip separation in machining unidirectional FRP composites by stiffness degradation concept, Composites Science and Technology, Vol. 69, 684-692, 2009. [5] A. Naven Sait, S. Aravindan, A. Noorul Haq, Optimisation of machining parameters of glass-fibre-reinforced plastic (GFRP) pipes by desirability function analysis using Taguchi technique, International Journal of Advanced Manufacturing Technology, Vol. 43, No 5-6, 581-589, DOI: 10.1007/s00170-008-1731-y. [6] K. Palanikumar, F. Mata, J. Paulo Davim, Analysis of surface roughness parameters in turning of FRP tubes by PCD tools, Journal of Materials Processing Technology, Vol. 204, pg. 469474. 2008. [7] M. Seeman, G, Ganesan, R. Karthikeyan, A. Velayudham, Study on tool wear and surface roughness in machining of particulate aluminum metal marteix composite-response surface methodology approach, International Journal of Advanced Manufacturing Technology, Vol. 48, No 5-8, 613-624, DOI: 10.1007/s00170-0092297-z. [8] Erol Kilickap. Modeling and optimization of burr height in drilling of Al-7075 using Taguchi method and response surface methodology”, International Journal of Advanced Manufacturing Technology, Vol. 49, No. 9-12, 911-923, DOI: 10.1007/s00170009-2469-x. [9] Dilbag Singh, P. Venkateswara Rao, A surface roughness prediction model for hard turning process, International Journal of Advanced Manufacturing Technology, Vol. 32, No. 11-12, 1115-1124, DOI: 10.1007/s00170-006-0429-2. [10] M.Y. Noordin, V.C. Venkatesh, S. Sharif, S. Elting, A. Abdullah, Application of response surface methodology in describing the performance of coated carbide tools when turning AISI 1045 steel”, Journal of Materials Processing Technology, Vol. 145, 46-58, 2004. [11] Amar Patnaik. Alok Satapathy. S.S. Mahapatra, R.R. Dash, A Taguchi Approach for Investigation of Erosion of Glass FiberPolyester Composites, Journal of Reinforced Plastics and Composites , Vol. 1, 1-18. 2008.


[12] E.S. Lee, Precision machining of Glass Fibre Reinforced Plastic with respect to Tool Characteristics, International Journal of Advanced Manufacturing Technology, Vol. 17, 791-798, 2001. [13] J.P. Davim, F. Mata, Influence of cutting parameters on surface roughness in turning glass-fiber-reinforced-plastics using statistical analysis. International Journal of Lubrication and Tribology, Vol. 56/5, 270–274, 2004. [14] J.P. Davim, F. Mata, Optimisation of surface roughness on turning fiber reinforced plastics (FRPs) with diamond cutting tool, International Journal of Advanced Manufacturing Technology, Vol. 26, 319–323, 2005. [15] M. R. Khoshravan, F. Azimpoor, Modeling of Delamination in Woven Composites Based on a Unit Cell, International Review of Mechanical Engineering, Vol.3, No.4, 473-480, 2010. [16] Arindam Majumder, A Comparative Study of the Ann with RSM for Predicting Bead Geometry of Gas Tungsten Arc Welded aa7039 aluminium Alloy Joints, International Review of Mechanical Engineering, Vol.7, No.4, 833-839, 2010. [17] Ibrahim M. Deiab, Hany A. El Kadi, Artificial Neural Networks Based Prediction of Tool Wear Progression, International Review of Mechanical Engineering, Vol.4, No.4, 410-416, 2009. [18] D.C. Montgomery, Design and Analysis of Experiments, (John Wiley and Sons, 1991) [19] Design – Experts software. Trial version 8.0, User’s Technical Manual; Start Ease Inc, Minneapolis, MN, 2010. [20] Niola, V., Quaremba, G., Amoresano, A., A study on infrared thermography processed trough the wavelet transform, (2009) Proceedings of the 8th WSEAS International Conference on System Science and Simulation in Engineering, ICOSSSE '09, pp. 57-62.

Author’s information 1

Department of Mechanical Engineering, Saveetha School of Engineering, Saveetha University, Saveetha Nagar, Chennai-602105, India. Phone: +91 97874 15094 E -mail: [email protected] 2 Department of Mechanical Engineering, Pondicherry Engineering college, Pondicherry – 605104, India. Phone: + 91 9442066544. E-mail: [email protected].

Dr. S Ranganathan is working as Professor, Department of Mechanical Engineering, Saveetha School of Engineering, Saveetha University, Chennai, India. He has obtained his Bachelor degree in Mechanical Engineering from CIT Coimbatore, Master degree in Engineering Design from Bharathiar University and Doctor of Philosophy from SCSVMV University, Kanchipuram. He has teaching and research activities since the last 14 years. His field of specialization is Hot Turning, Hard Turning and Machining of Composite Materials. He has published 2 national, 5 international journals and 4 International conferences. Dr. T. Senthilvelan, is working as a Professor & Head in Mechanical Engineering Department, Pondicherry Engineering College, Pondicherry, India He has obtained his Bachelor degree in Mechanical Engineering, Master degree in Production Engineering and Doctor of Philosophy from Annamalai University. His research area includes in the field of Powder Metallurgy, Machining of Composite Materials and sintering. He has published 30 papers on various research areas such as P/M, Hot Turning, EDM and Composite Materials in national and international journals.


1073


Characterization of Two Sprays Interacting A. Amoresano1, C. De Nicola2, F. De Domenico2

Abstract – In order to obtain an appreciable effect of extinction of the heat from a heat source (eg a fire) it is necessary that the water flows are high (20- 30 liters per minute) and the droplet diameters are between 20 and 100 microns. To achieve both of these two features the type of nozzle to be used has to be multi hole and supply pressures have to exceed 80 bars. In these conditions there are strong interactions between the individual holes - that generate particular phenomena such as the air entrainment - and the generation of areas where the drops tend to stagnate. To understand the physics of this second phenomenon, the interaction of two sprays is studied in this paper. The article focuses on the region of interaction between the two sprays, which is characterized by using a PDPA system. Copyright © 2012 Praise Worthy Prize S.r.l. All rights reserved.

Keywords: Spray Characterization, PDPA Measurements, Two Phase Fluid

I.

Introduction

Pressure swirl atomizers [1] use the conversion of pressure into kinetic energy to achieve a high relative velocity of the liquid compared to the surrounding gas. In the pressure swirl atomizer, a swirling motion is also given to the liquid so that, under the action of centrifugal force, it spreads as a conical sheet as soon as it leaves the orifice. There are two basic types of pressure nozzles: the solid-cone pressure atomizer and the hollow-cone pressure atomizer. In the former, the spray consists of drops that are distributed fairly uniformly throughout its volume (Fig. 1). The latter produces a hollow-cone spray, in which most of the droplets are concentrated at the outer edge of a conical spray pattern. The main drawback of solid-cone nozzles is a coarse atomization, the drops at the center of the spray being larger than those near the periphery. Hollow-cone nozzles provide better atomization and their radial liquid distribution is also preferred for many industrial processes, especially for combustion applications. The simplest design of the hollow-cone nozzle is the simplex atomizer. Liquid is fed into a swirl chamber through tangential ports that give it a high angular velocity, thereby creating an air-cored vortex. The outlet from the swirl chamber is the final orifice, and the rotating liquid flows through this orifice under both axial and radial forces to emerge from the atomizer with the shape of a hollow conical sheet, the actual cone angle being determined by the relative magnitude of the tangential and axial components of exit velocity. As it deals with surface area, Sauter mean diameter is a good way to describe a spray that is to be used for processes involving evaporation. Manuscript received and revised June 2012, accepted July 2012

Because of the complexity of the physical phenomena involved in atomization by pressure swirl nozzles, the study of atomization has been performed principally by empirical methods, resulting in correlations for SMD or D32 of the following general form: SMD=2,25σ 0,25 µ0,25mL0,25 ρ-0,25 ∆PL-0,5 where: σ is the surface tension. µ is the kinematics viscosity. ρ is the density m L is the mass flow rate. ∆PL is the injection pressure differential across the nozzle. This formula takes into account the spray angle, which is known to affect mean drop size. To overcome some of the deficiencies of this relation, and to explain some of the apparent anomalies that careful measurements often reveal, Lefebvre has proposed an alternative form equation for the mean drop sizes produced by pressure-swirl atomizers.

Fig. 1. Spray produced by pressure-swirl atomizers: hollow cone, solid cone


1074

A. Amoresano, C. De Nicola, F. De Domenico

This equation is not the result of a mathematical treatment of the subject, but is based on considerations of the physical process involved in pressure swirl atomization. Several different mechanisms have been proposed to describe the atomization process. It is generally agreed that the disintegration of a liquid jet or sheet issuing from a nozzle is not caused solely by aerodynamics forces, but must be result of turbulence or other disruptive forces within the liquid itself. These disturbances are very important for the sheet disintegration, especially in the first stage of atomization. Subsequently, the relative velocity between the liquid and the surrounding air has a profound effect on atomization through its influence on the development of waves on the initially smooth surface. Any increase in this velocity causes a reduction in the size of the ligaments that disintegrate and become smaller drops. The process of atomization in pressure swirl atomizers is complex so it is useful to divide it into two main stages. The first stage represents the generation of surface instabilities due to the combined effects of hydrodynamic and aerodynamic forces. The second stage is the conversion of surface protuberances into ligaments and then into drops. The present work focuses on the analysis of the interaction between the spray jet and the surrounding gas on the boundary layer. This research includes the investigation of the distribution of the spray droplet size the analysis of the behaviour of the droplets outside the boundary layer in the area between the liquid phase and the quiescent air. This approach takes an important role when a multi-hole atomizer is used or, alternatively, nozzles working close to one another. These configurations have several industrial applications in fields such as gas turbines, internal combustion engines, agricultural sprays [1]. In Fig. 2 [2] a typical multi-hole atomizer for firefighting is shown. Picture represents the flux density distribution. The lines between the central hole and each peripheral hole show the presence of regions where the distritibution of the droplets depends on the dynamic field created by the interaction of the holes.

Fig. 3. Liquid interaction in the liquid break-up region of a twin-fluid atomizer

Figures 3 and 4 [3] respectively show the theoretical and experimental dynamic mechanisms that arise when a single hole pressure swirl nozzle is used. Fig. 3 is obtained by a CCD camera and highlights the instabilty on the boundary layer between the two phases [4]. The liquid film causes a drop of static pressure close to the boundary layer of the a liquid phase and causes also the obscillations of the cone angle[5]. In this region it is possible that some droplets detach themselves from the liquid boundary layer and are called back from the external region. As a consequence, if a secondary single-hole exists, an interaction between the two sprayswill arise. This interaction depends on the the supply pressure of the atomizers, on the distance of the holes and above all on the characteristics of the flow field [6].

Fig. 4. Experimental image of the contour of gas liquid interaction in the liquid break-up region of a single hollow cone nozzle

II.

Experimental

Velocity and droplet size measurements have been carried out at ambient temperature and atmospheric pressure by means of the Phase Doppler Anemometry (PDA) technique (Fig. 5).

Fig. 2. Flux density distribution in a multi-hole watermist pressure swirl atomizer



1075


Both droplet sizes and the axial and the radial components of the droplet velocities have been measured along the axis of the intersection of the two jet in backscattering position [6]. The PDA system includes an argon-ion laser operating at 514.5 and 488 nm, a 310 mm focal length transmitting optics with a beam separation of 65.0 mm and modular collecting optics working in forward scattering mode at 30° off-axis [7]. A single aperture of 0.025 mm has been set into the receiver to limit the scattered light to the detectors. The transmitter and collecting optics have been mounted on a y-z translation frame that has allowed the positioning of the probe volume at different locations within the spray.

Input gas pressure ranging from 0.01 MPa to 1.0 MPa produces a linear output pressure of the fuel from 2.5 to 35MPa. A pressure tank of 1.0 dm3, located between the injection pump exit and the injectors, recovers the pressure oscillations due to the dynamic air recharge. Test points have been located on the vertical axis of the jet, as shown in Fig. 6, which depicts the layout of the measurement grid. The investigated points have been 7.5, 10, 12.5, 15, 17.5, 20 and 30 mm from the nozzle along z axis

x H/P PUM RESERVOI

FUEL INJECTO

z

TRANSMITTE

30° EXTERNA INPUT RECEIVE

Fig. 7. Measurement axis SPECTRU ANALYZE

PHOTOMULTIPIERS

ARGO -IO LASER

Fig. 5. PDA test bench

Experiments have been conducted at ambient temperature and atmospheric backpressure, injecting the water into a vessel under quiescent conditions with the bottom of the vessel connected to an exhaust blower in order to extract water droplets. The cycle-resolved data have been analyzed off-line by applying the ensemble averaging technique in order to estimate the axial and radial mean velocity and the mean diameter (D10). In order to spray water through two Danfoss Nozzles, an injection apparatus - making use of a hydro-pneumatic pump activated by pressured gas - has been used (Fig. 6).

A remote control panel has been designed in order to acquire the correct burst of data for each acquisition. This panel is able to control, at the same time, the supply pressure of the pump (air line), the supply pressure of the spray and the water mass flow rate. Furthermore, it controls the pressure drop of the two supply lines so as to be sure that the pressure is in the correct fixed range. Fig. 8. shows the control panel designed in Labview language. So it is possible to correlate this quantities with the data recorded from the PDPA and analyse correctly the mean statistical distribution of the diameters and velocity of the droplets [9][10][11].

Fig. 8. Remote control panel and data acquisition system

Fig. 9 schematically shows the layout of the interaction of the two sprays. As a result, when a liquid jet.

Fig. 6. Spray test-bench



1076


The figure shows that the interaction of two sprays, or two or more holes of a spray head, creates three regions. The first is situated between the two tips of the nozzles and it may be called depressure region or empty region. The second region is also called “free zone”. In this part of space the nozzle can be analyzed like a single nozzle because the fluxes do not interact with each other. The third region is called “intersection zone” because of the overlapping of the two sprays. Fig. 8 also shows that the measurement grid crosses all the three regions.

Free zone: It is the region where the behavior of the spray is the same as the behaviour of a single spray.

Fig. 10. Break-up regimes in the parameter space Rel - We

III. Results and Conclusions

Fig. 9. Schematic flow field generated by the interaction of two nozzles

The depressure zone is about 12 mm far from the tip of the nozzle, so it is situated at the same height where the break-up of the droplets occurs. This region develops exactly at the same height where the liquid film breaks up as shown in Fig. 8. In this space two phenomena take place. The first phenomenon is due to the interaction of the two liquid and gaseous phases. Close to the tip of the nozzles, the energy of pressure is quickly transformed into kinetic energy, so the injected liquid interacts with the surrounding quiescent gas (air) and the breakup of the ligaments begins. The interaction of the two phases creates a shear stress on the their boundary layers [8]. Consequently, some droplets leave the liquid phase and are called back by the depressure generated between the two nozzles. As a result, when a liquid jet of diameter Dl and velocity Ul issues from a nozzle and discharges into a stagnant gas, it becomes unstable and breaks into droplets. In the presence of a high-speed coaxial gas stream, with a momentum flux greater than the flux of the liquid, the breakup of the jet is caused by the transfer of kinetic energy from the high speed gas to the liquid. The cause of the break-up between the liquid jet and the quiescent gas depends strongly on the supply pressure of the nozzle [12] as shown in Fig. 10. Intersection zone: In the intersection zone the droplets generated by the two nozzles interact among them and can coalesce. In this region we will expect droplets with large diameter due to the interaction of the droplets in a low velocity flow field. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

The measurements start from the centre of the two nozzles and move in horizontal and vertical directions. The first valid line of measurement is the one where the geometrical and spherical validation [11] of the PDA is realized. The first analysis sets 5 minutes as validation limit. Fig. 11 and fig. 12 show the contour of the counts of the droplets and the D10 in the plane measurement grid.

Fig. 11. Representation of droplet count in the plane measurement grid

The second analysis focuses on the mean velocity and Sauter mean diameter profiles across the radial span of the interacting droplets sprays at a given axial position. They are constructed using the average values computed through the validation of over 30.000 signals. On the x axis the zero point represents the origin of the coordinates as shown in fig.7. It is possible to divide the diagram into three parts representing the three regions: the empty zone between the two sprays, the intersection zone and the free zone of the single spray.


1077


Fig. 14. Droplet radial mean velocity for several axial coordinates Fig. 12. Representation of D10 in the plane measurement grid

III.1

Droplet Mean Velocity Components

The axial and radial mean droplet velocity component profiles across the radial span of the interacting sprays are presented in Figs. 13-16. Fig. 13 and Fig. 14 represent the axial and radial mean droplet velocity along the entire measurement region. Figs. 15 and 16 represent the characterization of the empty zone only. It shows that some droplets coming from the boundary layer of the two atomizers are present in the empty zone. The axial mean velocity component profiles are characterized by the typical trend of a single nozzle for all the axial distances except at Z=15.6mm, where the profile begins to have an almost constant trend. The lateral mean velocity components profiles have a positive mean component magnitude, which is maximum near the center and decreases in the radial direction. In addition, the magnitudes of the maximum decrease and the axial location increases [14]. Only for Z=15.6 mm a sign change occurs because the radial velocity components move in opposite directions and in this zone the effects of the interaction of two sprays strongly influence the evolution of the particles.

Fig. 15. Droplet axial mean velocity for several axial coordinates in the empty zone

Fig. 16. Droplet radial mean velocity for several axial in the empty zone

For a better understanding of the droplet flow field the mean velocity vectors for each grid point are shown in Fig. 17. III.2. Droplet Mean Diameters The D10 and the Sauter mean diameter profiles across the radial span of the interacting sprays are presented in Fig. 18-21.

Fig. 13. Droplet axial mean velocity for several axial coordinates



1078


In the empty zone, the trend initially increases and then decreases. Particles with larger diameter have a higher speed, because they depend on a greater momentum and kinetic energy. The value of the common mean diameter is low close to the boundary layer, increases along the horizontal direction and achieves a minimum on the 0 point of the x axis. This trend is the same for each distance from the y axis.

Fig. 20. Droplet D10 mean diameter for several axial coordinates inside the empty region

Fig. 17. Droplet mean velocity vector for the entire flow field

Fig. 21. Droplet D10 Sauter mean diameter for several axial coordinates inside the empty region

The SMD diameters increase moving from the gas liquid interface and achieve a maximum before of the x=0. In this point all the D32 distributions achieve a minimum and then have another maximum going towards the other gas liquid interface. The measurements recorded and the related diagrams highlight the following conclusions: the empty zone interacts with the two sprays and a “fog” arises in the region between them. This “fog” is due to the callback of the droplets from the gas liquid interface. These droplets float thanks to the depressure inside the empty zone.To further study the phenomenology described so far, the number of Rel and We are calculated from the values of the mean diameter and the axial velocity for each point of acquisition, where Re l = U l D l/ν l and We = ρ g U 2lD l/σ .

Fig. 18. Droplet D10 mean diameter for several axial coordinates

The graph below shows (Fig. 22 ) how each points in the regime of the Rayleigh instability break-up. In this condition the liquid viscosity has a stabilizing effect that lowers the breakup rate and increases the size of the observed droplets. However, it is possible to consider that the results for the empty area are almost at the

Fig. 19. Droplet Sauter (SMD) mean diameter for several axial coordinates



1079


Rayleigh threshold and there for the phenomenon of break-up may occur anyway.

Fig. 24. Volume fraction distribution in the free region

Fig. 22. Parameter space Rel – We for thedouble nozzle

III.3

Volume Fraction Distribution

These statements are supported by the following diagrams, where the volume fraction distribution is represented. Fig. 23-25 shows the volume fraction distribution of the droplets in the empty region being measured in the three points, x = -1.5 mm, x=0mm, x=1.5mm and 7.6 mm far from the tip of the nozzles. This distribution shows that most of the droplets have a diameter between 55 and 100 µm and only few small droplets have 1015 µmdiameter. It is interesting to compare the distribution of the drops of the empty region with that of the other two regions defined in this paper. Fig. 8 shows that the MVD inside the free region is the same distribution of a single spray [4]. It is possible to notice that in this case the volume fraction distribution is quite independent of the radial coordinate. Most of the droplets have a diameter between 50 and 85 µm. This is due mainly to the exchange of momentum and to the phenomenon of coalescence among the droplets.

Fig. 25. Volume fraction distribution in the intersection region

IV.

References

[1]

Nasr, G. G., Yule, A. J., Bendig, L., Industrial Sprays and Atomization, Springer, 2002 [2] Lefebvre, A. H., Atomization and Sprays, Taylor & Francis, 1988 [3] U. Shavit, Gas liquid interaction in the liquid break-up region of Twin-fluid atomization, “Experiment in fluids 31 (2001), p. 550570 [4] T.Marchione, C. Allouis, A. Amoresano, F. Beretta, Experimental Investigation of a Pressure Swirl Atomizer Spray, Journal of Propulsion and Power vol. 23 (2007), p.1096-1101. [5] C. Allouis, A. Amoresano, F. Beretta, Experimental study of lean premixed prevaporized combustion fluctuations in a gas turbine burner, Combustion Science and Technology vol. 180/5 (2008), p.900-909. [6] N. Damaschke, H. Nobach, N. Semidetnov, C.Tropea, Optical Particle Sizing in Backscatter. Applied Optics, Vol. 41, pp. 57135727 (2002) [7] B. Esposito, M. Marrazzo, “Application of PDPA System with Different Optical Configuration to the IWT Calibration”, 45th AIAA Aerospace Sciences Meeting 2007-01-08 [8] J. C. Lasheras and E. J. Hopfinger, Liquid Jet Instability and Atomization in a Coaxial Gas Stream, Annu. Rev. Fluid Mech. 2000. 32:p. 275–308 [9] Wang, X., Lefebvre, A. H., Mean Drop Sizes from Pressure-Swirl Nozzles, Journal of Propulsion and Power , vol. 3, Jan.-Feb. 1987, p. 11-18 [10] W.D. Bachalo, M.J. Houser, Phase Doppler spray analyzer for simultaneous measurements of drop size and velocity

Fig. 23. Volume fraction distribution in the empty region



1080


[11] [12]

[13]

[14]

distributions, Journal of Optical Engineering 1984, vol. 23, no5, pp. 583-590 H.-E. Albrecht, M.BorysN. Damaschke, C.Tropea - Doppler and Phase Doppler Measurement Techniques. Springer 2003. Amoresano, A., Cameretti, M.C., Tuccillo, R., Combined experimental - Numerical approach for the fuel jet study in a LPP combustor, (2011) Proceedings of the ASME Turbo Expo, 2 (Parts A and B), pp. 1097-1108. Li G. Zheng, Min G.Yu, Shui J. Yu, Zhi C. Liu, Characteristics Determination of Water Mist for Suppressing Pool Fire, AIP Conf. Proc. 914, pp. 530-536 J.Heinleinß, U. Fritsching , Droplet clustering in sprays, Experiments in Fluids (2006) 40: 464–472.

Authors’ information DiME, Naples University Federico II Mechanical and Energetic Department ,Via Claudio 21 80125 Naples, Italy. E-mail: [email protected] Amedeo Amoresano was born in Naples on October 27, 1963. He took his degree in Mechanical engineering at University of Naples Federico II in 1991 by discussing a thesis concerning the analogic to digital conversion of data of a 3D PDA. In 1994 he took his PhD in Thermomechanical and Energetic Systems discussing a thesis on the fluidodynamic of two phase systems. In 1997 he became researcher of the University of Naples “Federico II” at DiME (Mechanical and Energetic Department). From 2001 he is Assistant Professor of Fluid Machinery and is an adviser for the italian government of the Innovative Power Plant. In 2007 he was responsible of PRIN (National Research Program) “Analysis and experimental characterization of fire suppression spray”. From 2009 he is Aggregate Professor of “Innovative Power Plant”. His principal research fields are: - Spray and atomization systems - Mild and diluted combustion and gasification systems - Optical diagnostics and thermal images processing - Aircraft Deicing System During his career he tutored several graduated and PhD students and gave lessons in the Italian Accademy Air Force where is responsible of the experimental activity on the Wind Tunnelmework of the combustion courses for chemical engineers at the University of Naples. He is author of about sixty works among the ones published on international journals, on the proceedings of international and national meeting in reduced or extended form.



1081


A Sensitive Methodology for the EGR Optimization: a Perspective Study A. Amoresano, V. Niola, A. Quaremba Abstract – The policy of reducing emissions in the energy sector is one of the scientific community’s main research topics. There have been significant developments in automotive and business fields, particularly in diesel engines. Nowadays, the method, which is widely used in reducing emissions of a Diesel engine, is the dilution factor called EGR through which it is possible to lower NOx levels. With the reduction of NOx also the performance of a diesel engine (e.g. the stable combustion and engine power) should be evaluated. This paper presents a useful methodology for analysing the change of the dynamic of the entire engine system, which is due to small changes in EGR dilution ratio. The adoption of systems with sophisticated signal processing such as the one presented in this paper, allows easier control of complex systems, such as diesel systems HCCI. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: EGR control, NOx reduction, Discriminating Analysis

I.

Introduction

When combustion temperatures exceed 1700 °C, atmospheric nitrogen begins to react with oxygen during combustion. The result is the creation of several compounds called nitrogen oxides (NOx), which play a major role in urban air pollution. To reduce the formation of NOx, combustion temperatures must be kept below the NOx threshold. This is done by recirculating a small amount of exhaust gases through the "exhaust gas recirculation" (EGR) valve. The EGR valve controls a small passageway between the intake and exhaust manifolds. When the valve opens, intake vacuum draws exhaust through it. The exhaust dilutes the incoming air/fuel mixture and has a quenching effect on combustion temperatures, which keeps NOx within acceptable limits. In particular, the mass flow rate of the exhausts is given by two elements. The first element depends on the exhaust grabbed in the combustion chamber while the second element is due to the recirculation ratio. So, the combustion of the fuel heavily depends on the EGR ratio [1]. To a lesser quantity of fuel/air ratio less oxygen is available for the oxidation of soot, which consequently then grows. It follows that it is sensible to assume that the evolution of the mixing ratio affects the reducing emissions but also the different propagation of the flame front. The recirculation of the exhaust reduces the combustion temperatures and can have significant effect on power since it lowers the peak of combustion and the related overpressure. Reaching a high value of the EGR ratio, the combustion can become unstable (Jinyoung Kwon et al. [2]) and to keep it stable it is necessary to change the timing of the spark injection. Manuscript received and revised June 2012, accepted July 2012

The goal of Zero Emissions will be reached with technological solutions, like electrical vehicles, fuelcells, hydrogen utilization, etc. These aims can be achieved developing precise control of the recirculation of exhaust gases (EGR). In this scenario, the actual market trend is to produce engines with high specific power, characterized by a lower displacement and an higher mechanical power. This “downsizing” is determined by development of modern supercharging systems (mechanical [3] or turbo [4]). The coupling of a turbocharger group to a diesel engine is nowadays a very common solution but it is more and more adopted also in the case of the SI engine. In general, it shown that the Brake Specific NOx (BSNOx) decreases when the EGR increases under load conditions. Unstable combustion conditions will arise, for a fixed regime, when the EGR ratio increases. This behavior is emphasized [5] under low load because the front of the flame decreases its velocity and partial burning or misfire conditions occur. A correlation analysis between combustion performances and EGR dilution ratio has been carried out by Johan Molin [6] and studied by Saheed et al. [7]. They have shown the EGR can increase if the turbulence of the flame front intensifies, otherwise instability conditions arise. This paper analyzes the engine behavior for each small variation of the EGR ratio by using mathematical tools. This analysis is carried out studying the changes of the vibrational morpho dynamics of the engine. This approach can be useful if applied to each engine (Diesel Engine PCI, DI) but it may be more suitable if used in the analysis of HCCI engines. This paper introduces a new sensitive technology applied to a Diesel engine, but it could be useful, in


1082

A. Amoresano, V. Niola, A. Quaremba

general, for the control engine unit and simulation [8] for all types of engines.

II.

Experimental Set-Up

The experimental set-up is presented in Fig. 1; the engine test bench has been equipped to test a 1910 cm3 Common Rail Diesel engine. Engine main design characteristics are specified in Table I, and the main measurement equipments consist of an electric brake (Schenk WS 260), a data acquisition system, an exhaust gas analyzer and a special ECU, open to control data change. The main engine control signals acquired are: - Torque - Speed - Air flow rate (Hot wire Anemometer) - Fuel flow rate (AVL 7030 - A05) - Controlled intake air condition (humidity 50% and temperature 25 °C) - 14 points of temperatures (turbocharger, intercooler, intake and exhaust manifold, oil, 6 points on exhaust line, atmospheric, fuel and cooling liquid) - 8 Mean Pressures (atmospheric, turbocharger, intake and exhaust manifold, oil, exhaust line, intercooler and fuel) - Emissions NO - NO2 – NOX (Chemiluminescence measurement - ECO PHYSICS) CO - CO2 - O2 – HC (Infrared Photometer - ABB) - ECU parameter (pilot and main injection parameter, EGR level, torque parameter, etc.)

Two NOx and Lambda sensors - one upstream and one downstream - are installed on the engine exhaust line (Fig. 2). These sensors are used for catalyst management in vehicles with gasoline or Diesel engine and consist of a ceramic sensor element and an electronic control unit. The ceramic sensor measures the oxygen concentration carried by the exhaust gas through a diffusion barrier into a first cavity. The oxygen concentration inside the cavity is controlled with respect to the constant concentration of a few ppm NOx. Other components of the exhaust gas (also entering the cavity as HC, CO and H2) are oxidized on the Pt electrode. The gas to be analysed, which contains NOx and few ppm O2, passes into a second cavity, where gaseous oxygen is totally removed by an auxiliary pump. At the measuring electrode, the equilibrium: 2NO N2 + O2 is changed by removing the oxygen generated by the reduction of NO. The amperometric measurement of this generated oxygen represents the NOx concentration of the exhaust gas. An electronic control unit provides the power control for heating the sensor in order to manage the temperature. The measured values are transmitted by a CAN line and are acquired by a National Instruments device (NI-USB 8473). The "regularity" of the engine is evaluated through the identification and classification of the clusters generated by its vibrational conditions; these are created by increasing inert gas recirculation in the combustion chamber (i.e., combustion gas) and by setting the solenoid valve of the Exhaust Gas Recirculation (EGR) circuit. All of this is necessary to acquire the vibrational dynamics generated by the engine. To achieve this purpose several accelerometer signals are acquired along the x direction, i.e., parallel to the axis of the motor shaft as well as perpendicular to the y axis of the crankshaft and, at the same time, parallel to the crankcase. In particular, an accelerometer is placed in order to acquire the vibration along the z axis perpendicular to the crankcase (hence perpendicular to x and y axes) (Fig. 3).

Fig. 1. Engine test equipment TABLE I MAIN ENGINE CHARACTERISTICS Displacement 1910 cm3 Compression ratio 18,45 : 1 Bore x stroke 90,4 x 92,0 mm Max power 100 kW at 4000 RPM Max torque 302,7 Nm at 2500 RPM Alimentation Garrett turbocharger Injection Common Rail Fig. 2. Engine



1083


Fig. 5. Details (0.25s) of the signals reported in the Fig. 4

Fig. 3. Accelerometer locations

The main two methodological conditions applied are as follows: • stationary running of the engine during all the test-run • working points set at 2000rpm and 2bar regarding the mean effective pressure (MEP) Firstly, the test-run is developed by acquiring the accelerometer signals from the engine for roughly 2 minutes. The engine is set on standard running conditions (i.e., amount of recirculated gas equal to that normally used by the standard mapping of the motor). These data are used to construct the so-called “reference signal” (baseline). In other words, the data collected in such operating conditions represent the vibrational “signature” related to the engine as a mechanical system.

Fig. 4. Examples of accelerometer signals (2 minutes)

Subsequent acquisitions are obtained by varying the set-up of the solenoid regulating the EGR in order to gradually increase the amount of exhaust gas in the combustion chamber. Examples of the accelerometer signals referred to the z-axis are reported in Fig. 4. It shows 6 sequences of the original signals, in the time domain, each of them equivalent to 121s (∼5×106 points). The first signal refers to the baseline (top) and the other five operating conditions, obtained by changing the EGR, are referred henceforth as EGR1, EGR2,..., EGR5. The Fig. 5 shows a detail of the above signals corresponding to roughly 0.25s (equivalent to 8 laps of the crankshaft).


From the combustion point of view the increase of EGR (performed by means of 5 very small steps, i.e., with the EGR ratio starting from 2.0 and increasing of 1%, until a final value of 2.2) causes a less abrupt combustion with lower local peak of temperature and consequently this leads to less formation of nitrogen oxides (NOx). Such tests (i.e., obtained by varying the EGR), as mentioned above, have been referred to as EGR1, EGR2, ..., EGR5. The different points of the operating engine differ simply because they are characterized by a slightly different combustion affecting almost exclusively the formation of pollutants (NOx reduction and increase of carbon monoxide CO) without influencing the engine performance or power. As already pointed out, the purpose of the test was to quantify the "degree of regularity" of the engine for the several set-ups of the EGR by processing only the accelerometer signals acquired during the test-run. The main aim of the proposed method of signal processing is the verification of the following basic idea: an engine fueled with a greater amount of inert (i.e., gas recirculation) should exhibit, if it is compared to a standard power supply, a more regular operating condition. The measurements are referred to the morphodynamical vibration shown by the engine when it is powered with a gradual increase of the percentage of inert. To obtain this weak information from each cluster (baseline, EGR1, etc.) the accelerometer signals are "synchronized" with the tachometer signal acquired during the running-test. It is possible to notice that the morphodynamics regularity of the vibrational signal gradually increases up to a limit beyond which a further inflow of exhaust gas tends to invert the phenomenon. The explanation lies in the fact that the combustion becomes less sharp and therefore more regular by increasing the percentage of EGR. It is well known that the EGR has mainly the aim of reducing the emission of NOx. Moreover, the production of NOx increases when the combustion is faster and high local peaks of temperature arise. In order to evaluate such a phenomenon, a signal named “surrogate” has been reconstructed for each cluster. It is representative of an averaged class of


1084


accelerometer signals in terms of spectrum (i.e., frequencies and amplitudes) obtained during the test-run (baseline and EGRs). Therefore, the following procedure for the construction of a “surrogate” has been developed for each family of signals: a) Determination of the number of points for obtaining detailed vibrational description of a complete lap. This means, 2000rpm (speed of the engine) = 33.33roots/s → 0.03s/root → 0.03 × 40960 (sampling rate) ≅ 1228-1229points/root. b) Extraction, from each cluster of accelerometric signals, of 51 randomized sequences (record) synchronized with the corresponding tachometer signal. c) Extraction from those records of the frequency spectra by applying the Fast Fourier Transform (FFT) with subsequent determination of an averaged spectrum (geometric or arithmetic). d) Reconstruction of the averaged signals by applying the Inverse Fast Fourier Transform (IFFT) hereinafter referred to as "surrogate". e) Convolution between the surrogate signal and each of the 51 records randomly selected within each test-run (cluster). f) Calculation of several corresponding parameters of regularity, including the correlation coefficient and the error of cross-correlation. All the above parameters have been used for the subsequent multivariate classification (i.e., discriminant analysis). The following Figs. respectively show the comparison between the baseline frequencies spectra with EGR1 (Fig. 6) and baseline frequencies spectra with EGR5 (Fig. 7). However, as it can be easily observed, the comparison of the baseline spectra with reference to EGR1 shows little differences. These slight differences give little or no information about the more or less regular vibration of the engine. A similar reasoning can be made when comparing the spectra extracted from the baseline and the EGR5 cluster.

Fig. 7. Comparison of Baseline # EGR5 spectra

Fig. 8 shows the comparison between the surrogates extracted from each cluster. It is quite clear that the higher the percentage of EGR, the greater its phase shift and amplitude. In particular, the change of the surrogate signals can be observed in terms of amplitude and frequency, depending on the cluster from which they were extracted. It was possible to apply to such data the multivariate discriminant analysis to verify the assumptions on which the proposed method is based. Briefly, it is to be noticed that discriminant analysis aims to statistically distinguish two or more groups of events obtained from previous observations.

Fig. 8. Comparison of 6 surrogates

For example, the events are the operating conditions of the engine which generate outputs that, if appropriately parameterized and coded, allow the identification and classification of similar clusters. In order to distinguish the clusters, a collection of variables able “to measure” the characteristics for discriminating among several groups is usually employed. For example, with reference to an internal combustion engine, these variables must be sensitive to speed changing or to applied torque or to minimum variations of EGR. From a mathematical point of view, the goal of the discriminant analysis is “to weigh” and linearly combine

Fig. 6. Comparison of Baseline # EGR1 spectra



1085


the set of variables selected for study so as to significantly differentiate the clusters from one another. In other words, it is possible in this way to discriminate among groups in order to distinguish them. For example, if a certain “maneuver” can lead to an important and significant response of the engine, variables play a crucial role in attaining the target: to discriminate among the groups. Therefore, to achieve this goal one or more linear combinations of variables must be built in order to discriminate the clusters in the following form:

Di = wi1Z1 + wi 2 Z 2 + … + wip Z p

(1)

reduction of the correlation at EGR5 set-up, beyond which any further supply of inert gas may affect the smooth vibrational running of the engine. The same reasoning, but reversed, can be made for the other discriminating parameter: the error of crosscorrelation, which measures the symmetry of the records with reference to their respective surrogates of each cluster. Here the minimum corresponds to the cluster named EGR4, confirming the behavior of the first parameter. The statistical significance of the "discriminating power" of each function is evaluated on the basis of the value assumed by the respective eigenvalues (Table II). The first two functions used for discrimination alone "explain" about 99% of the variance.

where Di is the score of the ith discriminant function, w are the weights given by each variable, Z are the standardized values of the p variables used in the analysis. Fig. 9 shows how these parameters, defined above, help the classification of the 306 accelerometer signals.

Fig. 10. Correlation coefficient (306 records)

Fig. 9. Classification map (306 records)

Fig. 9 shows the presence of 6 well “discriminated” clusters. It is apparent that a significative “distance” between the "centroids" exists. It refers to the baseline and to the cluster named EGR1, generated by the minimum change of the EGR ratio, (in terms of percentage from EGR ratio 2.0 of baseline to 2.02 of EGR1). It is also possible to deduce the sensitivity of the method employed. It can be noticed that only two discriminant functions were used for the classification of all records: the linear composition of the two parameters and the correlation coefficient with the error of cross-correlation. Fig. 10 and Fig. 11 show the diagrams with reference to each cluster. The 95% confidence interval (CI) is also reported for each cluster. In particular, the correlation coefficient shows a rather rapid increase from the baseline condition up to the set-up of EGR1, reaching the maximum under the set-up referred to as EGR4. Fig 10 also shows a


Fig. 11. Cross-correlation error (306 records) TABLE II EIGENVALUES (306 record) % of cumulative canonical Function eigenvalue variance % correlation 1 179,356a 91,6 91,6 ,997 2 14,142a 7,2 98,8 ,996 ,6 99,4 ,733 3 1,161a 4 ,784a ,4 99,8 ,663 a ,2 100,0 ,528 5 ,386 a. First 5 canonical discriminant functions were used in the analysis.


1086


Subsequently, it was carried out further extensive statistical assessment composed of 612 records randomly chosen from each test-run. By repeating the method from b to f items, the result of classification is shown in Fig. 12. The relative position of the cluster referred to as baseline is clear if it is compared to the other five clusters. Multivariate analysis aggregates the clusters according to their characteristic properties, which are expressed indirectly in a quantization of the degree of regularity of the engine, depending on the mass flow rate of exhaust gas recirculation. In practice, the combinations allow the correct identification of the first two clusters, and help the achievement of a good smooth running. Vice versa, much attention should be paid when reaching the amount of gas recirculation which would alter the surrogate as shown in Fig. 12.

becomes much more irregular: the best value is concentrated around the fourth cluster (Fig. 13), where the correlation function reaches the maximum value and conversely the error of cross-correlation reaches the minimum value (Fig. 14). Thanks to its response and sensitivity, the proposed method could be used to optimize, for example, the curve of EGR for each engine family, using the response of the mechanical system in terms of vibration.

Function 1 2 3 4 5

TABLE III EIGENVALUES (612 record) % of cumulative Eigenvalue variance % 87,465a 90,9 90,9 7,457a 7,7 98,6 ,7 99,3 ,668a ,6 100,0 ,597a ,0 100,0 ,043a

canonical correlation ,994 ,939 ,633 ,612 ,204

Fig. 13. Correlation coefficients (612 records) Fig. 12. Classification map (612 records)

Even with this sample, the eigenvalues confirm the validity of the analysis, explaining, with the first two discriminat functions, almost 99% of the cumulative variance (Table II). Finally, it is interesting to note that the trend of the two parameters with greater discriminatory power is very similar to the one achieved by a smaller sample size. Moreover, the interval of confidence shows less variability confirming the validity of the analysis. The vibrational dynamics in terms of regularity of the engine can be analysed by studying the variables statistically selected for DAM (Discriminating Analysis Model): the correlation coefficient and error of crosscorrelation. The correlation coefficient confirms that an engine powered with exhaust gas recirculation produces a vibrational signature very similar to its surrogate and therefore very repetitive (about +40%). Conversely, the engine not powered by exhaust gas has a more irregular trend (about -25 %). It should be noticed that there is a limit to the gas recirculation beyond which the engine Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

Fig. 14. Cross-correlation error (612 records)

By acting on the map of the engine, the right balance between fuel and inert gas recirculation can be optimized reaching the following dual purpose:


1087


1) The reduction of NOx ; 2) More regular operating with lower vibrations, functional benefit and duration of all the mechanical parts subjected to less thermal stress.

[2]

[3]

III. Conclusion In this paper a new technique to measure the effect of the EGR on a four stroke and direct injection Diesel engine is developed. From the diagram and the Fig.s above analyzed it is possible to verify that the little change of the set-up of the EGR map modifies the vibrational “signature” of the entire engine. The results carried out show that little variations well define the repeatability classes of the EGR. These results demonstrate that the morphodynamics analysis “feels” the system changing and can verify the instability conditions before they occur. This behavior is analyzed for little variations of the EGR ratio starting from 2.0 up to 2.2 increasing it by 1%. From the analysis of Fig.s 6 and 7 the frequency of the vibrational signature occurs around 5000Hz (multiple harmonic of 33.33Hz) corresponding to 2000 rpm. So it is possible to exclude the fundamental harmonics of the engine and to state that this behavior is due to the combustion phase. Assuming that the crankle angle does not change it is possible to maintain that this frequency is due to the combustion phase. The change of the EGR ratio can affect the propagation of the flame front and the output is a variation of the morphodynamic behavior. Taking into account that the used engine is a direct injection four stroked diesel engine, a little variation of the EGR does not affect the engine performances. However, it is possible to say that the engine system modifies its vibrational map. In a DI Diesel engine the front of flame follows the corresponding injection law and it is difficult to say if the measurement done can be operative and can lead to an optimization of the EGR ratio. This methodology is very sensible when analyzing the vibration check of the engine and it is possible to say that this technique is highly suitable if applied to HCCI Diesel engine. In a HCCI engine the front of flame is not well defined because it depends on the distribution of the pilot flame and on the stochastic ignition point inside the injected fuel. The analysis of the behavior by means of this methodology can be used for analyzing how the engine changes its morphdynamics signature and for verifying how the combustion occurs. It defines a functional map of the engine vibration for each sequence of parameters such us cranck angle, EGR ratio and the characteristics of combustion. This kind of study is still in progress.

References [1]

[4]

[5]

[6]

[7]

[8]

Jinyoung Cha, Junhong Kwon, Youngjin Cho and Simsoo Park “The effect of Exhaust Gas Recirculation (EGR) on Combustion Satbility, Engine Performance and Exhaust Emissions in a Gasoline Engine” KSME Journal Vol. 15 No. 10. Pp 1442-1450, 2001. Gimelli A., Rapicano A., Barba F., Pennacchia O., (2012), “Reciprocating Compressor 1D Thermofluid Dynamic Simulation: Problems and Comparison with Experimental Data”. International Journal of Rotating Machinery, , vol. 2012, Article ID 564275, 13 pages, 2012. doi:10.1155/2012/564275. http://www.hindawi.com/journals/ijrm/contents/. Bozza F., Gimelli A..(2009). “Unsteady 1D Simulation of a Turbocharger Compressor”. SAE 2009World Congress & Exhibition. 20-23 April 2009. Detroit, MI, USA. Session: Modeling of SI and Diesel Engines. SAE paper 2009-01-0308.On SP-2244,Modeling of SI and Diesel Engines, 2009, pp. 67-76, ISBN 978-0-7680-2140-0.Also in SAE International Journal of Engines, vol. 2, pp. 189-198, E199717 - Print ISSN: 1946-3936, E200637 - Online ISSN: 1946-3944, October 2009. Haiyong PENG, Yi CUI, Lei SHI, Kangyao DENG “Effects of EGR on combustion process of DI diesel engine during cold start”, Front. Energy Power Eng. China 2008, 2(2): 202–210 Johan Molin” Investigation of Correlations Between COV of Ion Integral and COV of IMEP in a Port-Injected Natural-Gas Engine”, Master thesis, Linköping, 12 December, 2008 Shaeed O. Wasiu, ShaharinA.Sulaiman and A.Rashid A.Aziz “ An Experimental Study of Different Effects of EGR Rates on the Performance and Exhaust Emissions of the Stratified Bozza, F., Fontanesi, S., Gimelli, A., Severi, E., Siano, D., (2012). “Numerical and Experimental Investigation of Fuel Effects on Knock Occurrence and Combustion Noise in a 2-Stroke Engine”. SAE 2012World Congress&Exhibition. 20-23 April 2012. Detroit, MI, USA. SAE Paper 2012-01-0827, also in SAE International Journal of Fuels and Lubricants 5(2):2012, doi:10.4271/2012-01-0827.

Authors’ information DiME, Naples University Federico II Mechanical and Energetic Department ,Via Claudio 21 80125 Naples, Italy. E-mail: [email protected] Amedeo Amoresano was born in Naples on October 27, 1963. He took his degree in Mechanical engineering at University of Naples Federico II in 1991 by discussing a thesis concerning the analogic to digital conversion of data of a 3D PDA. In 1994 he took his PhD in Thermomechanical and Energetic Systems discussing a thesis on the fluidodynamic of two phase systems. In 1997 he became researcher of the University of Naples “Federico II” at DiME (Mechanical and Energetic Department). From 2001 he is Assistant Professor of Fluid Machinery and is an adviser for the italian government of the Innovative Power Plant. In 2007 he was responsible of PRIN (National Research Program) “Analysis and experimental characterization of fire suppression spray”. From 2009 he is Aggregate Professor of “Innovative Power Plant”. His principal research fields are: - Spray and atomization systems - Mild and diluted combustion and gasification systems - Optical diagnostics and thermal images processing - Aircraft Deicing System During his career he tutored several graduated and PhD students and gave lessons in the Italian Accademy Air Force where is responsible of the experimental activity on the Wind Tunnelmework of the combustion courses for chemical engineers at the University of Naples. He is author of about sixty works among the ones published on international journals, on the proceedings of international and national meeting in reduced or extended form.

Hailin Li, GhaziA, Karim “Modeling the performance of TurboCharged Spark Ignition Natural Gas Engine With Cooled Exhaust Gas Recirculation” Journal of Engineering for Gas Turbine and Power, May 2008 Vol 130.



1088


Numerical Analysis of Anti-Icing and De-Icing Thermal Systems F. De Domenico1,*, A. Amoresano2, C. De Nicola3 Abstract – An aircraft flying at low-moderate altitudes can experience ice formation on its forward surfaces. The impact with super cooled droplets can generate a water film or beads and rivulets on a solid wall which in turn can freeze and cause the ice accretion. In order to protect the aircraft surfaces, an accurate knowledge of local and total impingement characteristics and afterwards of the ice shapes on a real aircraft component is the first prerequisite in designing a proper ice protection system. The present paper proposes a three-dimensional method to estimate the ice accretion on finite wings from its single section using a two-dimensional method. The method is based on the Lagrangian formulation which considers the gas phase as a continuum and calculates the trajectory of each particle in the flow as a result of convection action of various forces (drag, gravity) agents on the particle itself. Integral methods based on the Prandtl liftingline theory have been implemented to calculate the effective angle along the wingspan and successively to calculate the thermal power necessary to avoid ice formation and accretion. It is possible to give the wing characteristics as a table of geometric variables or as a CAD. The comparison with experimental data drawn from literature points out a significant potential towards accurate prediction of ice accretion characteristics on real finite wings. Further investigations on up-to-date test cases are scheduled to fully validate the present method. Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved. Keywords: Ice Accretion, Ice Protection Systems, Thermal Resistances

C k/c m s

Nomenclature A Clα Cd Cp Dp IGES LWC M MVD NURBS P1 Prt Re Sp SLD Stk T1 Tr Ue V XLE

Γ

A

Projection Matrix Lift curve slope Spherical particle drag coefficient Caloric air capacity at constant pressure, J/Kg/K Droplet diameter, µm Initial Graphics Exchange Specification Liquid water content, kg/m3 Free stream Mach Number Mean Volumetric Diameter, µ Non Uniform Rational B-Splines Free stream pressure, Pa Turbulent Prandtl number (Prt ≈ 0, 9) Reynolds number Droplet frontal area, m2 Super cooled Large Droplets Parameter defined by the boundary layer thickness δ (s) Free stream temperature, K Recovering temperature in front of each node, K Flow field velocity out of the boundary layer, m/s Phase velocity, m/s Position of the leading edge, Circulation Phase acceleration, m/s2

Airfoil chord, m Ice horn height-to-chord ratio Phase mass, kg Curvilinear abscissa along the airfoil (from the stagnation point), m Ice horn location-to-chord ratio Time, s Angle of attack, degrees Zero lift angle, degrees Collection efficiency Angle of twist refers to the chord, degrees Dimensionless wing span Dynamic viscosity, kg/(m s) Phase mass density, kg/m3

s/c t

α α 0l β ε η µ Subscript g p

Carrier gas phase Droplet particle phase

I.

Introduction

Aircraft flying at subsonic speeds through clouds at low-moderate altitudes can be subject to ice formation on critical aerodynamic surfaces. Ice accretion results from very small super cooled droplets (droplets cooled below freezing point), usually 5 to 50 microns in diameter, which can freeze upon impact with the aircraft exposed surfaces, or from super cooled large droplets (SLD), like so-called drizzle droplets up to



1089

F. De Domenico, A. Amoresano, C. De Nicola

400 micron diameter or rain droplets beyond 400 micron. The ice accretion usually occurs near the leading edge zone which implies geometry alteration damaging the aerodynamic performance and handling characteristics intended as loss of lift, increase drag and stall speed, and reduced stability and controllability of the aircraft. This can be enough to cause flow separation, decrease lift and, in addition, increase drag enough to cause flight instability. This problem was known since the early aviation steps: even Charles Lindbergh risked to fail the first transatlantic flight in 1927 because of the ice accretion on his “Spirit of St. Louis”. Thus, aircraft must be designed with the necessary equipment required for ice removal or prevention. Recently this problem has been the subject of many theoretical, experimental and numerical studies aimed at defining a baggage of knowledge to fully understand the phenomenon and develop prediction methodologies more and more accurate and refined. In fact, the problems associated with the detection and location areas where to make simulation tests, led the certification authorities to contemplate the possibility of granting certification through the use of estimated ice shapes. Hence, given its complexity, the problem of ice on surfaces of aircraft requires a high use of human and technical resources. The role of the computational aerodynamics is fundamental: the development and validation of efficient algorithm for ice accretion prediction could help to reduce the experimental and certification work load, which is di cult and costly, and could also provide a valuable support for designing anti/de-icing systems. This is particularly true for CFD-based approaches in aircraft icing. Thus, an accurate knowledge of ice accretion characteristics on a real aircraft components is the first prerequisite in designing a proper ice protection system. To date, there are no methods or numerical models yet that can predict correctly, and especially in short computational time, ice shapes. Hence, with the aim to minimize the computational time, the development and validation of a code which implements a quasi bidimensional ice accretion method on finite wings is presented. The development activity represents the extension of the two-dimensional code “ICE2D” through the analysis of the single sections of a finite wing, then proceeding with the CAD generation and successively to calculate the thermal power necessary to avoid ice formation and accretion. The bi-dimensional code has been written by the ONERA aerodynamic department and is included in the ONERA Icing Code. For now, the present paper gives some details about the bi-dimensional Lagrangian method for ice accretion calculation and successively is focused on the philosophy of the three dimensional approach from the type of wing geometry input to the calculation of the effective angle of the sections along the wingspan through two finite wing methods: Multhopp and Weissinger.


II. II.1

The Ice 2D Code

The Bi-Dimensional Approach

The modeling of the icing growth is made by using twice four subroutine to: • compute the flow field around the airfoil; • compute the droplet trajectories and the local catch efficiency coefficient; • compute the heat transfer coefficient on a rough airfoil; • compute the thermodynamic balance which gives the ice growth rate and the ice shape after a given icing time. Classically the models are time-dependent, that means that the flow field is calculated one more time as soon as the ice deposit is scheduled to modify it. The ONERA model uses a different method, similar to a predictorcorrector method, which gives relatively good results: the ice shape is first estimated for a given icing time in one step. Then the flow field, the trajectories and the heat transfer coefficient are calculated on this “estimated shape”. Assuming that the values of the local efficiency and the heat transfer coefficient are varying linearly from their values on the clean airfoil to their values on the airfoil covered by the estimated shape, the thermodynamic balance is made and the ice shape calculated. II.2

Flow Field Computation

The potential equation for the flow field is solved for a “C” grid by using a finite element method [2] the code takes into account the compressibility effects but does not take the viscous effects. Moreover, it cannot determine and compute a boundary layer separation.

Fig. 1. Airfoil computational mesh grid


1090


II.3

0,5Cf(s)=0,168/[ln[846δ(s)]/k(s)]

Trajectories Computation

The trajectories are computed considering that the only forces to take into account are the drag forces, produced by the velocity between the droplets and the flow field around it. The motion equation of a droplet of mass mp is. mp dVp/dp=D+mg

(1)

The drag force on a spherical droplet is equal to: D=0,5 rgVr2SpCd=mpap

(2)

II.5.

(8)

Thermodynamic Balance Computation

The code uses the Messinger equation [4] to determine the local freezing fraction. Starting at the stagnation point, this code solves the steady state heat balance equation for each successive facet. It calculates the local equilibrium surface temperature, the freezing fraction and the local rate of icing. The unfrozen fraction is then allowed to run back to the next face.

where Vr is the relative velocity between the droplet and the surrounding flow field. The drag coefficient Cd is function of the local droplet Reynolds number (Rep): Rep=Dpρa Vr /mg

(3)

The trajectories are computed with an explicit method: knowing the droplet velocity and its position in a given t time, it is possible to compute the drag force applied to the droplet and the its velocity and position at t + dt time (though the integration of the acceleration along x and y direction). For the resolution of 1, in general, the integration starts at a sufficient distance from the body so as to assume that particles, suspended in air, have the same asymptotic velocity of the flow. The catch efficiency is equal to the ratio of the mass flow of water impinging in a particular point to its value away from the airfoil (referred as at infinite distance, equal to 4 chords).

β =ṁi/ ṁ∞

(4)

In a bi-dimensional calculation, the mass flow in a tube defined by two trajectories stay constant:

ṁ∞ ∆S∞= ṁi ∆Si

(5)

where ∆S represents the area delimited by two trajectories. So the catch efficiency, for two trajectories close to each other, is equal to:

β= ∆S∞/ ∆Si=dS∞/ dSi II.4.

(6)

Heat Transfer Coefficient Computation

The determination of the heat transfer coefficient taking the roughness of the wall into account uses the Makkonen correlation [3]. The roughness of the airfoil has an effect only for a turbulent boundary layer. In that case the heat transfer coefficient is equal to: h(s)=[ 0,5 ρg CpUe Cf(s)]/ [Prt+[0,5 Cf(s)]1/2 [Stk(s)-1]]

(7)


Fig. 2. Definition of catch efficiency in the Lagrangian approach

III. The Ice 2d-Evo Method The evolution of the code “ICE2D” presented above, provides the analysis on finite wings from the ice accretion of several sections along the wingspan. The method requires four steps: 1. Assignment of the wing geometric characteristics in tabular format or CAD; 2. Calculation of the effective working angle for every sections; 3. Icing prediction on the single sections by using the “Ice2D” code; 4. Calculation of the thermal power necessary to avoid ice formation and accretion. III.1. The Data Input The first step provides the assignment of the wing geometric characteristics such as: •η the dimensionless half wingspan; • c, chord in meters; • X.L.E. position of the leading edge in meters; • ε, angle of twist refers to the chord in degree; • x – y dimensionless coordinates of an airfoil. The figure below shows the adopted convention. A panel method will calculate for each airfoil the variables Clα and αl0 necessary to the effective angle calculation. It is possible to provide this variables by two approach. International Review of Mechanical Engineering, Vol. 6, N. 5

1091


For each grid point barycentric coordinates, with regard to the current (projected) triangle, are calculated from which the z-value is obtained: /

(10) 1

Z = w1 V1z + w2 V2z + w3 V3z

After obtaining the airfoil coordinates, the variables c, X.L.E. ε, Clα and αl0 will be calculated successively through a geometric and aerodynamic study of the sections extrapolated from the CAD.

Fig. 3. Definition of the wing geometric characteristics

The first one is a simple assignment of these variables in tabular format. The airfoil is selected through a text file.

η 0.0 0.25 0.5 0.75 1.0

TABLE I WING SECTIONS DATA c(m) X.L.E(m) 1.162 0.0 1.0525 0.16975 0.9385 0.3395 0.82675 0.50925 0.715 0.679

(11)

ε 0.0 0.0 0.0 0.0 0.0

III.2

Finite Wings Aerodynamic Analysis

Assigned the wing geometric characteristics, an integral method proceeds with the calculation of the aerodynamic characteristics of a finite wing applying the Prandtl lifting-line theory. The lift distribution, along the wing span, can be calculated from the Prandtl integral equation: Cl(y)= Clα[α(y)- αι(y)] where:

The second approach needs, instead, the import of a CAD from which the required variables are obtained. Care must be taken with this approach, especially with regard to the extrapolation of the airfoil coordinates based on the projection in barycentric coordinates [5]. The CAD has to be composed of triangular panels and, fixed the h section, grid points are generated. The method consider every triangle projected to the x - y plane and determine whether grid points lie inside.

Fig. 4. Projection in barycentric coordinate

For a given triangle generated by (V1, V2, V3) points, the non-singular projection matrix is calculated:

αι(y)=0,25πV∞


Γ

(13)

The lift coefficient of a wing generic section is a linear function of the effective working an:

αeff=α- αι

(14)

For this purpose, Multhopp method has been implemented for the analysis on straight wings. In order to make reliable the analysis even on swept wings, Weissinger method, an extension of the Prandtl theory known as approximated lifting-surface theory, has been implemented. The model provides that 1. The wing is replaced by a single adherent vortex, placed along the line of fire (swept), with variable circulation; 2. The swept adherent vortex and the free vortices contribute to the down wash; 3. The condition to the limit of the tangential flow is realized within the two-dimensional flow limits of the thin airfoil theory. N.B. The effective working angle is considered constant during the analysis of ice accretion throughout the icing time. This hypothesis does not affect the accuracy of the obtained ice shapes under linearity conditions of the lift curve because the Clα and αl0 of the clean airfoil and the iced airfoil are almost identical[11]. III.3

(9)

⁄ ⁄

(12)

Icing and Thermal Power calculation

The last step sets up the ice accretion calculation on each airfoil. Fixed the airfoil coordinates, the chord value


1092


and the effective working angle, the ice shape on each section is evaluated imposing the Mach number M∞, the free stream temperature T∞ (K), the free stream pressure P∞ (Pa), the droplet diameter Dp (µm), the liquid water content LWC (Kg=m3) and the icing time ti (s). After that, a function constructs the “Non Uniform Rational BSplines” (NURBS) structure given the control points and the knots, from the clean and iced airfoil coordinates, and convert it to an IGES file (Initial Graphics Exchange Specification). The ice horn location-to-chord ratio (s/c) and the ice horn height-to-chord ratio (k/c), useful for a successively stall path analysis, are geometrically extracted from the ice shapes. Finally, after the analysis of ice accretion, the thermal power, necessary to avoid the ice formation and accretion, is calculated from the heat transfer coefficient through the equation: Qs(s) = h(s)[Tr(s) - T0]

consisting of the outboard portion of a general aviation business jet tail. The tail tip consists of a semi-cylindrical cap. The tail airfoil is a symmetric 8% thick NACA 64A008 section and it is kept constant from root to tip. The location of maximum thickness for this airfoil section is at x=c = 0:39. The Mean Aerodynamic Chord is 0:956m and is located approximately 0:559m above the tunnel floor.

(15)

This equation is based on the assumption to get, through a reverse process, the effective thermodynamic state to a recovering temperature on the body Tr from the reference thermodynamic state by the subtraction of heat at constant pressure. T0 is the reference thermodynamic state temperature where, if water, T0 = 273,16 K. The numerical results obtained are an excellent starting point for future experimental campaigns necessary for the design of anti/de-icing protection systems. So, with the assumption that for each section is installed an electrical resistance is placed in stream-wise direction, the aim is to calculate the energy required by the system in a fixed period by adding the contributions of all sections. To control the aerodynamic performance degradation of each section, a key parameter for the activation of the electrical resistances is set: the thickness of the ice horn dimensionless respect to the chord. Set this parameter, for each section the loss of lift, indicated as a percentage of the lift coefficient in clean configuration, and the increase of drag force are fixed. It is possible to control the maximum thickness of the ice shape using the electrical resistance as a sensor other than as an ice protection system. Hence, It is possible to calculate, from equation 15, the energy required for a time t of activation for each section from the formula:

Fig. 5. Horizontal swept tail

The wing characteristics are listed in Table I. The results of the airfoil placed in the middle of the wing will be presented (c= 0:9385 m). The test conditions are listed below. Table II : Test condition for the swept wing.

Run

M∞

A B C D

0.3 0.3 0.3 0.3

TABLE II TEST CONDITION FOR THE SWEPT WING MKD LWC AOA(deg) T∞(K) (µm) (Kg/m3) 5 248.15 50 0.0013 5 248.15 20 0.0013 5 263.15 50 0.0013 5 248.15 50 0.0026

t(s) 300 300 300 300

(16)

where the subscript i indicates the section and E = [Joule=cm2] (ds is dimensionless).

IV.

Results

In this section some results obtained with the Ice2DEvo method for ice accretion calculation are presented. The test case is a full-scale reflection plane tail model Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved

Fig. 6. Thermal flux necessary to avoid ice formation and accretion Run A

The first run shows a typical trend of the power flow required to avoid ice formation and accretion. In the case of clean wing, the curve has a more regular trend with values much lower compared to the case of iced wing. Varying the value of the MV D, it is possible to note that


1093


only the curve of the iced wing shows a considerable reduction of the thermal power required due to the decrease of the average diameter of the drop. Increasing the upstream temperature, the typology of ice shape changes and the curves of the clean and iced wing show values much smaller than case A.

This is substantially due to the upstream condition much closer to the reference. thermodynamic state. Finally, increasing the Liquid Water Content, the ice shape is much larger because of the increased presence of water. There are not so many differences regarding the thermal power curves respect the Run A. The taper ratio of the wing indicates an ice growth rate higher for sections with smaller chord. Then, each section will have a different cycle of activation/deactivation of resistance .Fixed percentage of the horn a thickness equal to T=C = 0:02, the following activation times of the resistances were obtained from the curves of the maximum thickness percentage versus time for each section:

Section 1 238s

TABLE III ELECTRICAL RESISTANCE ACTIVATION TIME Section 2 Section 3 Section 4 222s 195s 165s

Section 5 125s

Fig. 7. Thermal flux necessary to avoid ice formation and accretion Run B

Fig. 7. Maximum thickness percentage versus time

Then, for each section an average of one hour cycle is calculated with a time of activation equal to t = 60 s and, considering all the sections, a maximum value of energy is obtained from Eq. (17):

Fig. 8. Thermal flux necessary to avoid ice formation and accretion Run C

1

1

1,7

0,3 (17)

V.

Conclusion

A quasi-bidimensional Lagrangian method to solve the ice accretion problem on finite wings has been presented. The method provides ice accretion shapes and thermal power flux on several wing sections along the wing span. Moreover a method for the electrical resistances design and activation has been developed. A general estimate of the total energy spent per hour was calculated

Fig. 9. Thermal flux necessary to avoid ice formation and accretion Run D



1094


considering five sections on wing. Further research will be focused on experimental test cases to be analyzed towards a full validation useful for designing more effective ice protection system on real aircraft components.

References [1] [2] [3] [4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

Giuseppe Mingione, Massimo Barocco, Il volo in condizioni favorevoli alla formazione di ghiaccio, IBN Editore, 2005. M. Bredif, A fast finite element method for transonic potential flow calculation, AIAA paper 83-0507. L. Makkonen, Heat transfer and icing of a rough cylinder, Cold Regions Science and Technology NCR n° 23312, 1984. Messinger B.L., Equilibrium temperature of an unheated icing surface as a function of airspeed, Journal of the aeronautical sciences, Vol. 20, pp. 29-42, n° 1 January 1953. Svatopluk Zakarias, Projection in Barycentric Coordinates, Department of Mathematics, Faculty of applied sciences, University of West Bohemia, Pilsen, Czech republic, December 3°, 1995. James R.M., On the Remarkable Accuracy of the Vortex Lattice Method, Computer Methods in Applied Mechanics and engineering, 1972. Andy p. Broeren, Michael B. Bragg, Effect of Residual and Inter cycle Ice Accretions on Airfoil Performance, DOT/FAA/AR02/68, May 2002. Andy p. Broeren, Sam Lee, Christopher M. LaMarre, Michael B. Bragg, Effect of Airfoil Geometry on Performance With Simulated Ice Accretions, Volume 1: Experimental Investigation, DOT/FAA/AR-03/64, August 2003. Jianping Pan, Eric Loth, Effect of Airfoil Geometry on Performance With Simulated Ice Accretions Volume 2: Numerical Investigation, DOT/FAA/AR-03/65, August 2003. Yihua Cao, Kungang Yuan, Guozhi Li, Effects of ice geometry on airfoil performance using neural networks prediction, Aircraft Engineering and Aerospace Technology: An International Journal 83/5 266-274, 2011. Frank T. Lynch, Abdollah Khodadoust, Effects of ice accretions on aircraft aerodynamics, Progress in Aerospace Science 37, 669797, 2001. Farooq Saeed, State-of-the-Art Aircraft Icing and Anti-Icing Simulation, 25th Annual ARA Congress, Cleveland, OH, July 2000. O. Meier, D. Scholz, A handbook method for the estimation of power requirements for electrical de-icing systems, DLRK, Hamburg, 31 August - 3 September, 2010.

Amedeo Amoresano was born in Naples on October 27, 1963. Hhe took his degree in Mechanical engineering at University of Naples Federico II in 1991 by discussing a thesis concerning the analogic to digital conversion of data of a 3D PDA. In 1994 he took his PhD in Thermomechanical and Energetic Systems discussing a thesis on the fluidodynamic of two phase systems. In 1997 he became researcher of the University of Naples “Federico II” at DiME (Mechanical and Energetic Department). From 2001 he is Assistant Professor of Fluid Machinery and is an adviser for the italian government of the Innovative Power Plant. In 2007 he was responsible of PRIN (National Research Program) “Analysis and experimental characterization of fire suppression spray”. From 2009 he is Aggregate Professor of “Innovative Power Plant”. His principal research fields are: - Spray and atomization systems - Mild and diluted combustion and gasification systems - Optical diagnostics and thermal images processing - Aircraft Deicing System During his career he tutored several graduated and PhD students and gave lessons in the Italian Accademy Air Force where is responsible of the experimental activity on the Wind Tunnelmework of the combustion courses for chemical engineers at the University of Naples. He is author of about sixty works among the ones published on international journals, on the proceedings of international and national meeting in reduced or extended form.


Ph.D. Candidate, Aerospace Engineering Department, Naples, Italy E-mail: [email protected] 2 Professor, Mechanical and Energetic Engineering Department (DiME), Naples University Federico II Mechanical and Energetic Department, Via Claudio 21 80125 Naples, Italy. E-mail: [email protected] 3 Professor, Aerospace Engineering Department, Naples, Italy. E-mail: [email protected]



1095


ERRATA CORRIGE In the article “Tooth Shape Optimization of the Timing Belt with Tangent Grooves Using FEM”, International Review of Mechanical Engineering (I.RE.M.E.), Vol. 6, N. 3, March 2012, pp. 468-478, for a print mistake the name of the second author is incorrect. The correct name of the second author of the paper is Young-Doo Kwon instead of Young-Doo Kwo. Many apologies to the authors and to our readers for this mistake.



1096

1970-8742(201207)6:5;1-P Copyright © 2012 Praise Worthy Prize S.r.l. - All rights reserved