Design and implementation of an intelligent grinding assistant system ...

106

Int. J. Abrasive Technology, Vol. 1, No. 1, 2007

Design and implementation of an intelligent grinding assistant system Michael N. Morgan* and Rui Cai AMTReL, General Engineering Research Institute, Liverpool John Moores University, Byrom Street, Liverpool L3 3AF, UK E-mail: [email protected] E-mail: [email protected] *Corresponding author

Andrea Guidotti Balance Systems s.r.l., Italy E-mail: [email protected]

David R. Allanson, J.L. Moruzzi and W. Brain Rowe AMTReL, General Engineering Research Institute, Liverpool John Moores University, Byrom Street, Liverpool L3 3AF, UK E-mail: [email protected] E-mail: [email protected] E-mail: [email protected] Abstract: In modern competitive manufacturing industry, machining processes are expected to deliver products with high accuracy and assured surface integrity, using shorter cycle times with reduced operator intervention and increased flexibility. To meet such demands, the trend towards increased use of machine intelligence in machining systems and operations is clear and unlikely to be revised. This paper describes the structure, content and relations employed in a fully integrated intelligent grinding system for adaptive controlled cycle optimisation, thermal damage avoidance, dressing interval optimisation and data retention. This system, termed: ‘Intelligent Grinding Assistant’ (IGA©) is a unique, robust and versatile software system with embedded links and protocols for communication with the Computer Numerical Control (CNC) and instrumentation of a grinding machine. The IGA© was evaluated and implemented on a commercially available production machine and had its international launch in September 2005. The CNC, machine tool instrumentation, process monitoring and data analysis systems, Profibus link, and control strategies all constituted the IGA© system. Grinding performance was monitored and assessed in real time. An intelligent database was also developed to support the IGA© in the provision of optimised and/or safe starting cycle data to the operator/adaptive system and in the selective

Copyright © 2007 Inderscience Enterprises Ltd.

Design and implementation of an intelligent grinding assistant system retention of data. Importantly, the IGA© has been designed and implemented in a way to ensure that an operator can readily use the system and achieve process improvement without recourse to extensive training. Keywords: intelligent grinding; artificial intelligence methods; adaptive control; thermal model; grinding kinematics. Reference to this paper should be made as follows: Morgan, M.N., Cai, R., Guidotti, A., Allanson, D.R., Moruzzi, J.L. and Rowe, W.B. (2007) ‘Design and implementation of an intelligent grinding assistant system’, Int. J. Abrasive Technology, Vol. 1, No. 1, pp.106–135. Biographical notes: Michael N. Morgan manages the Advanced Manufacturing Technology Research Laboratory (AMTReL, GERI) and is Reader in Manufacturing Technology in the Liverpool John Moores University, UK. He has 20 years experience in manufacturing research and has contributed to key developments and innovations in grinding process technology through major collaborative research contracts. His work has resulted in award winning products in the grinding machine tool industry. Currently, he is leading the AMTReL in a range of research investigations, developing his interests in high speed grinding technologies, High Efficiency Deep Grinding (HEDG), sensors/instrumentation for grinding process monitoring, AI techniques for machine tool control and fluid delivery in grinding. Rui Cai received her MPhil in Ultra Precision Engineering in PR China and joined AMTReL as a PhD student on an EPSPC project: ‘Precision grinding with CBN wheels’. She has worked as Postdoctoral Researcher in the Mechatronics Research Centre, Wolfson School of Engineering, Loughborough University, investigating vibrations in wood machining, multisensor fusion and machine part design. Since her return to AMTReL she has completed a project concerned with Adaptive Control CNC grinding. The present research concerns development of a User Guidance Manual for coolant delivery in grinding, a feature of the current EPSRC project – ‘Optimisation of coolant delivery in grinding’. Andrea Guidotti is the Product Marketing Manager of the Italian company Balance Systems s.r.l., responsible for the market segment dedicated to auxiliary measuring and control systems for machine tools. He joined the Company in 1994 and is presently Project Leader for products for grinding processes applications. His area of expertise includes development of hardware, software, sensors and actuators for real-time applications in grinding machine and machine tool. He received a Doctoral degree in Electronic and Automation Engineering from the Politechic of Milan, in 1992. He has authored or co-authored a number of technical papers and industrial patents. David R. Allanson is a Senior Lecturer in the School of Engineering, LJMU and has published on a wide range of topics in the field of abrasive technology. He has 20 years experience in manufacturing research and has undertaken work on a broad range of programmes related to the modelling and control of the grinding process. He has published papers on control systems design and areas concerned with cycle optimisation, in-process gauging and strategies for avoidance of thermal damage. The subject of his current interest is augmentation of conventional CNC systems through the application of adaptive algorithms, process models and intelligent database technologies.

107

108

M.N. Morgan et al. J.L. Moruzzi is a Consultant Engineer with Extensive Industrial and Academic Experience in Control Systems Engineering, Electronics and Sensor Applications. He was the Founder of Goodwin Electronics and has contributed to a broad range of industrial research programmes in the fields of machine tool technologies and manufacturing. He is an Honorary Professor of Liverpool John Moores University and is closely associated with AMTReL. He has a strong publications record, holds a number of patents and has contributed to key developments in the grinding machine tool industry. He was principal coordinator for the programme of research presented in this paper. W. Brian Rowe has Broad Industrial and Academic Experience in Mechanical Design and Manufacture. He is particularly known for his extensive publications in the fields of grinding research, machine tools, bearings and tribology. He is an Emeritus Professor of Liverpool John Moores University closely associated with AMTReL, the Advanced Manufacturing Technology Research Laboratory. He was a principal author of Tribology of Abrasive Machining Processes (William Andrew Publishing, 2004), A Handbook of Machining with Grinding Wheels (CRC Press, 2006) and the author of Hydrostatic and Hybrid Bearing Design (Butterworths, 1983).

1

Introduction

Improvements to production efficiency are increasingly important in grinding as the processes develop to compete with increased efficiencies in other primary material removal processes and alternative finishing processes. However, the grinding process differs from many of the other processes in the complexity of the relationship between the machining parameters and the process performance. As a consequence, in both manual and Computer Numerical Control (CNC) operations the process quality and productivity depend to a large extent on the experience of the operator and as a result many operations are undertaken at conditions far from optimal. The application of Artificial Intelligence (AI) technologies using computers and controllers is seen as a way forward to produce higher quality components more efficiently with smaller batch sizes and more frequent changeovers. Two main trends are evidenced in the development of AI technologies in grinding: desktop systems to assist tool and parameter selection and self-optimising systems integrated within the machine controller (Rowe et al., 1991a). The application of AI, summarised by Rowe et al. (1994), described how AI techniques are applied in-process models and wheel selection systems, for machining parameter selection, process monitoring, process control, AC optimisation and database creation. Five principal techniques were identified: knowledge-based expert systems, fuzzy logic, neural networks, genetic algorithms and adaptive control for optimisation. Adaptive Control Optimisation (ACO) is the integration of a Knowledge-Based System (KBS) into a controller. Any of the AI technologies described above may be used including expert systems, fuzzy systems, neural systems, model-based systems and hybrids of these techniques. Various adaptive control systems have been developed. Miler (1975) classified adaptive control into technological adaptive control and geometrical adaptive control. Geometrical adaptive control aims to control the accuracy

Design and implementation of an intelligent grinding assistant system

109

properties such as size, roundness and surface roughness by adjusting the machining parameters. Technological adaptive control aims to control productivity according to Adaptive Control Constraint (ACC) and ACO. ACC systems are based on the premise that the machining proceeds within grinding machine limits such as maximum motor load, maximum infeed or maximum spindle speed. In addition to machine limits, other constraints may be considered including force, torque, grinding temperature and residual stress. Milner pointed out that practically all adaptive systems operate on the basis of ACC and, therefore, work in a suboptimal manner. Amitay et al. (1981) developed an adaptive control system, which aimed to optimise grinding and dressing conditions for maximum removal rate subject to the constraints that 1

workpiece burn should not occur and

2

surface roughness should remain within a specified limit.

The heart of the control system was an algorithm incorporating the optimisation strategy. Based on the input, the controller generated an updated infeed. The updated infeed was immediately used to control the grinding cycle. In parallel, the infeed was also employed by a further module in the system to calculate an updated value of burn power. Malkin proposed another optimisation method based on the coupling of two optimisation strategies (Malkin, 1981). The first strategy was to minimise the cycle time for a conventional cycle as described by Amitay. The second strategy was to modify the conventional cycle employing accelerated spark-out. A modified control algorithm was then proposed to minimise the cycle time in cylindrical plunge grinding (Malkin and Koren, 1984). The previous technique was refined by incorporating infeed velocity control based on direct measurement of the radial allowance remaining on the workpiece. The optimal infeed control policy is advantageous for grinding systems having a high time constant. In-process diameter measurement is now readily available for grinding and improves the ability to hold size tolerance (Liverton et al., 1993). Accelerated spark-out is a technique where the infeed is programmed to overshoot the target size position in order to reduce the time required to bring the workpiece to size. Accelerated spark-out has been studied by several investigators including Spiwak and Kaliszer (1983), Malkin and Koren (1984) and Webster and Zhao (1990). Accelerated spark-out is noted to be particularly useful when there are large deflections in the system and long time constants. In principle, the method is more efficient than a constant infeed control. However, computation complexity is increased due to the requirement for in-process parameter estimation. The computational capability required is not so great when using adaptive dwell period to control workpiece size and roundness and the spark-out period is adjusted by updating grinding parameters (Allanson et al., 1989; Kelly et al., 1989; Webster and Zhao, 1990). He (1984) developed an adaptive control system which used different strategies in different stages of the grinding process. In coarse grinding, the control strategy is aimed to maximise grinding efficiency. As a consequence of infeed adjustment, grinding productivity is maximised. In fine grinding, the strategy is aimed to achieve size, roundness and roundness accuracy using a constant infeed. Accuracy was controlled by determining the position at which spark-out started and ensuring that the length of the spark-out period was sufficient to achieve the required size and roundness accuracy.

110

M.N. Morgan et al.

Brinksmeier and Popp (1991) developed an adaptive-control-system for the external grinding process. Because the behaviour of the grinding process was substantially influenced by disturbances a self-tuning controller was employed. A system identification procedure was used to allow a permanent adaptation of the controller to the changing process parameters. An adaptive control system for internal grinding developed by Tönshoff and Zinngrebe (1986) aimed at reducing the manufacturing time while ensuring the accuracy of bearing rings. Gap elimination was employed and infeed was varied to maintain constant average force. Kelly et al. (1989) developed a novel grinding cycle termed the ‘pecking cycle’ to allow the centreless grinding of thin-walled cylinders with a large degree of initial variance. The principle behind the ‘pecking cycle’ was to allow the system to evaluate the compliance of an individual workpiece with size measurement taking place between pecks. Although there are many problems in traverse grinding, comparatively little research has been reported. Zhao et al. (1988) investigated adaptive control for traverse grinding of long slender steel rollers used in the film and metal industry. With the adaptive control system, the deflection was compensated by adjusting the steady automatically in accordance with in-process measured workpiece diameter data. A fuzzy logic model based on the experience of the operator was employed so that the operator’s skill was replaced by the intelligence of the computer. A time-optimum adaptive strategy for plunge grinding control was proposed by Webster and Zhao (1990) and Zhao (1989). The strategy aimed at achieving the required workpiece quality in the shortest grinding cycle, by identifying the time constant and employing accelerated spark-out. Rowe et al. (1993a) developed an adaptive system for plunge grinding which included different features depending on whether diameter gauging was or was not available. A common feature for both cases was that infeed was varied to achieve a specified maximum power. An adaptive dwell cycle was employed, on the basis of the in-process identification of time constant. The time constant was identified by integrating the power signal during the infeed period. Accelerated spark-out could be employed but was not employed for close tolerances. Allanson (1993) designed an improved adaptive dwell cycle for multiplunge grinding to vary the dwell time in proportion to the time constant at different positions on the workpiece. The time constant was identified in the dwell period using the technique of weighted Least Mean Squares (LMS) and logic rules. The technique was shown to be reliable even with up to 20% noise in the power signal. König et al. (1995) presented modelling and adaptive control of plunge grinding operations using an Acoustic Emission (AE) sensor. The plunge grinding process was either planned a priori or adaptively controlled to achieve a desired surface finish. Rowe et al. (1996a) reviewed the use of intelligent control and optimisation techniques in grinding and proposed the incorporation of intelligent techniques into CNCs. Intelligent CNC with ACO has been shown to provide significant improvements in productivity. As progress is achieved with process improvements account of these improvements can be automatically accommodated within the knowledge base due to the constant feedback from the process. There is the potential to greatly reduce set-up time, process proving time and the extent of operator intervention by improved AI systems with the involvement of ACO (Rowe, 1996a). The case has been made for an initial

Design and implementation of an intelligent grinding assistant system

111

parameter selection tool which recommends grinding wheels and grinding conditions required to achieve a satisfactory performance. The selection tool should be integrated into the CNC to ensure participation in production operations. It is evident from the literature reviewed above that intelligent CNC with integrated adaptive control can provide significant improvements in workpiece quality and productivity, providing potential to greatly reduce set-up time, process proving time and the extent of operator intervention (Allanson, 1993; Allanson et al., 1989, 1992; Amitay et al., 1981; Billatos and Tseng, 1992; Brinksmeier and Popp, 1991; Chen, 1998; Chen et al., 1998; Gao et al., 1992; König and Werner, 1974; König et al., 1995; Liverton et al., 1993; Malkin, 1981; Malkin and Koren, 1984; Rowe et al., 1993b, 1996a; Tönshoff and Zinngrebe, 1986; Wada and Kodama, 1977; Xiao et al., 1992, 1993; Zhao, 1989; Zhao et al., 1988). Although the technique of ACO integrated with CNC has been investigated by many researchers and some initial experiments completed, it has not been implemented in industry in production environments. The work herein describes a system that applies the techniques which have been proven to be reliable and easily transferable for different machine tools, grinding operations and controller systems to the production machine tool.

2

System overview

The Intelligent Grinding Assistant (IGA©) is a software package that links with the CNC and instrumentation on a grinding machine and analyses the grinding performance of the machine in real time. On the basis of results produced by the various algorithms incorporated within the software the IGA© will make recommendations to the CNC to alter certain grinding parameters aimed at improving the performance of the grinding cycle. In particular it will prevent burn occurring during the cycle and it will adjust the infeed rate and dwell time to ensure the fastest possible cycle time commensurate with the part specification produced by the design engineering specifications. The key algorithms employed are known as the: •

burn algorithm

•

time constant algorithm

•

wheel dress algorithm.

These algorithms and associated strategies will each be discussed separately in detail. However, common to each of the algorithms, and indeed necessary for implementation, are the following data: •

material properties of the component/part

•

material properties of the grinding wheel

•

material properties of the coolant used

•

the grinding cycle for the component/part

•

the physical specification of the grinding machine.

These data are retained in a MS Access database which was developed as an integral feature of the IGA©. The structure of the database was so designed that all the attributes that define a grinding cycle and its components were included. Cycles for a specified part that have been optimised are saved to the database for use in the future production of

112

M.N. Morgan et al.

those parts. The grinding database is also needed to store the adaptive control algorithms data and process data required for IGA© system implementation. Optimised cycle data is sent directly to the CNC and is used to produce the next component. A decision-making system for the selection of initial grinding parameters is also integrated into the database. It is also possible to save to the database the detailed machining parameters including the maximum grinding temperature and specific energy of the cycle for each individual part. This data are referred to as the Individual Part Record (IPR) and can be used to monitor the production of ‘critical’ parts. The IGA© also provides an option for the user to influence the process performance. This is achieved by allowing manual input of the parameters required by the algorithms, via the SETUP Menu. The process is subsequently ‘optimised’ on the basis of user inputs. This effectively provides a ‘tuning’ facility for the adaptive algorithms. This is a key attribute of the system in that the IGA© engages with the operator and allows the operator some flexibility in how to employ the intelligence and the adaptive strategies and how to tune the system to satisfy/deliver the operator requirements. The IGA© communicates with the CNC and the VM20 instrumentation system attached to the machine via a Profibus field bus and ETHERNET system. A schematic diagram of the complete system is shown in Figure 1. Figure 1

A schematic diagram of the IGA© system

Design and implementation of an intelligent grinding assistant system

113

The system has been developed to be flexible and to accommodate all major grinding operations. The system described presently refers to external cylindrical plunge, centreless plunge and plane surface operations. Further process operations are the subject of current IPR discussions. The IGA© Human Machine Interface (HMI) provides for a display language selected from English, Italian, German, Spanish or French.

3

Adaptive control system: cycle definitions

The adaptive strategies and the database structures supporting the overall system are contained in a separate PC, termed the IGA©, which has communication links with the machine, and the CNC controlling the machine. The database structures were coded using MS Access. The adaptive algorithms use data from its own system database. This database obtains information from the following sources: •

the CNC cycle

•

the component/part specification database feature

•

the material property database feature

•

the wheel property database feature

•

rules devised by the system developers.

To ensure accurate cross-referencing of data used by each of the IGA©, the database and the adaptive system three cycles were defined: Active Cycle, Reference Cycle and Optimised Cycle. These three cycles may be regarded, simplistically, as original data, in-process data and optimised data, respectively. The cycle data for a component is initially provided by the CNC. This data provides the current ‘best’ machining parameters and is stored in the IGA© as the Active Cycle. The IGA© can also be instructed by the CNC to save this cycle as the Reference Cycle in the IGA database. However, if after many successful grinding cycles improved cycle parameters have been identified, then the Active Cycle can be saved as the Optimised Cycle in the IGA database.

4

Profibus and ethernet communication

There is a two-way communication link between the IGA© and the machine tool CNC, as shown in Figure 1. The CNC provides the IGA© with details of the ‘cycle’ to be executed. The ‘cycle’ details include: part and batch reference; work material and diameter; cycle; wheel type, speed and diameter; coolant type; infeed; workspeed and coarse cut. A complete ‘cycle’ package must be sent to the IGA before a cycle is started. The CNC will also provide a second package of instruction data. This data are used to instruct the IGA to the required cycle data and to set flags that describe the status for the cycle start, dwell and progress within the cycle.

114

M.N. Morgan et al.

During the grinding cycle the IGA will analyse the process data and on completion of optimised dwell will send to the CNC modified or updated grinding parameters. If the recommended infeed and dwell values identified in the updated parameter set are accepted by the CNC it will place them into the CNC cycle and retransmit the modified cycle back to the IGA. The IGA will place these values into its Active Cycle for analysis once the next cycle begins. If the CNC decides to save these optimised values for future use it can send a command to the IGA to copy the data just sent to the Optimised Cycle database. There is a Profibus link between the CNC and the IGA. The spindle power and AE signals are provided to the IGA on a regular basis. The CNC is slave to the IGA and requests signals every 50 ms during a cycle.

5

Adaptive burn control and dressing strategy

The adaptive burn control and dressing optimisation features are activated by the operator via the SETUP Menu. A thermal model is used to predict the grinding temperature of the grinding cycle and adaptive control strategies ensure that the specified critical temperature is not exceeded. The thermal model employed is based on the work of Rowe et al. (1991b, 1995, 1996b, 1997). Key aspects of the model are: use of effective contact length (Rowe et al., 1993b), use of effective wheel thermal properties (Rowe et al., 1998), a heat partitioning solution based on the work of Hahn (1962) and the triangular heat source. Parameters required for thermal analysis are described fully in references (Rowe et al., 1991b, 1995, 1996b, 1997, 1998).

5.1 Temperature tolerance and strategy for infeed increment The critical threshold (burn) temperature for a particular material will be stored in the material database of that material. The grinding temperature and specific grinding energy is computed during the coarse feed cycle. In the case where part geometry and surface finish requirements are being satisfied, a new infeed may be suggested (increased or decreased). The change will aim at improving the cycle efficiency. The algorithm will recompute temperature and specific energy based on this new value of infeed. However, the new infeed must satisfy the temperature tolerance and dressing interval strategies.

5.1.1 Temperature tolerance The temperature tolerance strategy is that the IGA© will not permit an infeed value to be used in the cycle such that the predicted temperature exceeds a preset fraction of the critical temperature, that is (Temperature tolerance) × (Burn temperature). Typically the temperature tolerance will be set at 90%. Thus if the burn temperature for the material is 300°C then the IGA will not suggest infeed values that would produce a predicted grind temperature greater than 270°C. A lower infeed will be suggested if the burn criteria has been exceeded. At the end of the current grinding cycle the new value of the infeed is sent to the CNC. It is the task of the CNC to either accept or reject this suggested value. If it accepts the value then the CNC amends its cycle and then retransmits the amended cycle back to the IGA where it will be placed in the ACTIVE Cycle.

Design and implementation of an intelligent grinding assistant system

115

5.1.2 Burn exceeded If the burn model predicts a grinding temperature greater than the burn temperature it will, in addition to suggesting a new infeed, also send an explicit signal to the CNC indicating that the burn limit has been exceeded. This allows the operator the choice to remove the part from the batch being produced, if desired.

5.1.3 Infeed increment If the burn is active then the IGA will decide, on the basis of the predicted grind temperature, whether or not it is safe to increase the infeed used during the roughing part of the cycle. If the predicted temperature is below the threshold set by the temperature tolerance and the material burn temperature the suggested infeed will be incremented by the infeed increment which is typically set at 10%. If the temperature threshold has been exceeded in the last cycle then the suggested infeed will be decremented to a safe value. This information is passed to the CNC via the DataToCNC structure. The burn algorithms are seen, therefore, to operate in conjunction with the parameters temperature tolerance and infeed increment. The burn algorithm is also used by the dress algorithm. If the recommended changes breach the dressing interval criterion a wheel dress cycle is proposed.

5.2 Dressing optimisation algorithm The burn model calculates the efficiency of the grinding process in the last cycle, using a parameter known as the cycle specific energy. The lower this value is the more efficient is the grinding process. High values of this parameter indicate that the wheel has become dull or that the grinding parameters are incorrect. This parameter is also used by the system dress algorithm to evaluate the condition of the wheel. To do this the system must first be dressed and the IGA informed that a dressing cycle has been completed. A number of (N1) grinding cycles are then carried out in order to condition the wheel. After the wheel has been newly dressed and conditioned the specific energy of the next N2 cycles are used to calculate an average specific energy for the system. During subsequent cycles the specific energy of each cycle is monitored and if the value exceeds a preset threshold (say 120% of the average value) then a dress request is generated. This can be summarised as follows: •

A dress cycle is executed.

•

N1 normal (conditioning) grinding cycles are executed and no data gathered.

•

Over the next N2 cycles the cycle specific energy is calculated and an average is found. This is referred to as eCA and is used as an indication of a ‘very sharp’ or ‘sharp’ or ‘blunt’ wheel condition.

If ec > SpEnergyTol × eCA send message to screen and a dress request to the CNC. The values of N1, N2 and the SpEnergyTol can be set in the algorithm data structure. Prior to the predicted temperature calculation, the sharpness of the wheel, force ratio and roughness factor will each be attributed to one of three values given in Table 1.

116

M.N. Morgan et al.

Table 1

Recommended values for contact length model parameters Wheel condition V. Sharp

Sharp

Blunt

Grain contact radius, mm

12

15

20

Force ratio, Ft/Fn

0.7

0.45

0.2

Roughness factor

12.5

9

5

6

Plunge time constant model (internal and external)

In order to compensate for the effects of deflection during grinding it is essential to include a dwell or spark-out period at the end of the feed cycle. The machine operator normally sets the dwell time on the basis of experience. The dwell time required depends on the infeed rates employed within the cycle and the flexibility of the part, grinding wheel and machine. As the magnitude of the grinding force changes with wheel condition it is often necessary to set a conservative dwell period. This means that most grinding cycles are not optimised for minimum cycle time and take longer time than required. The time constant model allows the dwell time to be optimised in real time and automatically compensates for increases in grinding force and changes in part flexibility such as encountered in the grinding of multidiameter or slender parts. The time constant model describes the behaviour of the grinding system (machine, wheel and part) as a first order system and identifies the time constant from the decay of grinding power during the dwell period. The time constant is then used to calculate the correct time at which to terminate the dwell period and terminates the cycle. As the dwell time is automatically controlled for each workpiece this is a real-time adaptive process. This algorithm requires the following information: •

the spindle power being used

•

cycle start signal

•

the dwell period has started.

From this information the algorithm will calculate the optimum dwell time and will issue an output to the CNC to terminate the cycle. The system will need guidance as to the number of time constants required for the appropriate surface finish required, for example, 3 time constants. This number will be determined by post-cycle examination of the part. Basing the dwell time on the power decay allows the time constant model to cope with a variety of infeed cycles including those that do not achieve the programmed feedrate. In order to determine the dwell time and trigger grinding wheel retraction at the appropriate time the control algorithm compares the present dwell time with a target dwell time based on the estimated time constant and the specified number of time constants. Estimating the system time constant from the power data allows for a more consistent control of dwell time than for example, basing the decision to retract on the power level alone. Such a simple approach is very susceptible to the effects of signal noise and

Design and implementation of an intelligent grinding assistant system

117

measurement errors. The importance of having an appropriate dwell period cannot be overstated, short dwells give rise to workpieces with poor roundness and accuracy and, on the other hand, overly long dwell periods may result in excessive dulling of the grinding wheel grits and other problems. Retracting the wheel automatically helps reduce these problems, particularly on parts of variable compliance.

6.1 Dwell control algorithm and grinding theory The grinding system, that is the machine tool, workpiece and grinding wheel can be modelled as a first order mechanical system, as illustrated in Figure 2. Figure 2

First order mechanical system representation

In Figure 2, where Xp is the programmed radial infeed position; Xi is the actual radial infeed position; X i′ is the rate of change of workpiece radius; a is the instantaneous depth of cut; Fn is the normal grinding force; dw is the workpiece diameter and Vw is the workpiece surface speed and Ks is the static grinding force coefficient. It relates the depth of cut to the normal grinding force by the relationship: Fn = K s a

(1)

where Ke is the effective spring stiffness that represents the compliance of the grinding machine, workpiece and grinding wheel. The response of the above first order system responds to the applied grinding cycle in a variety of ways. However, it must be appreciated that the above model only predicts the behaviour during material removal. Secondly, the assumption that there is a linear relationship between depth of cut and normal grinding force is valid for a large range of workpiece materials at reasonable depths of cut. There may be some non-linear behaviour during very small cuts that border on rubbing conditions or for some very hard to grind materials. The response of the grinding system to an applied feed cycle can be understood by considering two phases of a typical grinding cycle, the infeed phase and the dwell period.

6.2 Infeed behaviour During the infeed phase of a grinding cycle the grinding wheel approaches the workpiece surface, makes contact with the surface and then starts to remove material from the workpiece. The following diagram can represent this behaviour, Figure 3.

M.N. Morgan et al.

118 Figure 3

Infeed phase of a grinding cycle

From Figure 3, it can be seen that after the initial contact between workpiece and grinding wheel the actual infeed position (the actual material removed from the workpiece) (Xi) lags behind that programmed (Xp). This lag builds up during the initial stages of infeed and given a long enough infeed period eventually reaches steady state. The grinding force giving rise to deflection of the grinding wheel/workpiece/machine system causes the lag in position. During the infeed stage of the grinding cycle the actual infeed position is given by the following relationship:

(

X i = Vf t − τ + τ e(

-t/ τ )

)

(2)

where time t is measured from touch detection and τ is the system time constant given by the following expression:

τ=

π dw K s Vw K e

(3)

The above relationship shows that the time constant depends on the relationship between the static grinding force coefficient (Ks), the effective stiffness of the machine and workpiece (Ke) and the workpiece rotational speed. The effects of the above parameters are impossible to predict accurately before grinding. The overall stiffness of the grinding system is not accurately known and the static grinding force coefficient varies with changing wheel sharpness. It is, therefore, necessary to measure the time constant in real time in order to accurately control the feed cycle.

Design and implementation of an intelligent grinding assistant system

119

During the dwell period material continues to be removed from the workpiece as the deflection is relieved and the deflection and the depth of cut decay in an exponential manner. The material removal during dwell can be modelled by the following expression:

(

X d = x1 1 − e(

− ( td ) / τ )

)

(4)

where x1 is the deflection present at the start of dwell, td is the time into the dwell period measured from the start of dwell and Xd is the infeed achieved during the dwell period. At completion of the grinding cycle the effective infeed position is given by adding the feed achieved during the infeed and dwell sections of the grinding cycle. X i = X1 + X d

(5)

The residual deflection at the end of the grinding cycle can then, if required, be calculated by subtracting the actual infeed position (Xi) from the programmed infeed position.

6.3 Estimation of the system time constant During grinding it is possible to monitor the power consumed by the grinding wheel spindle. However, the power monitored during the grinding includes the no-load power and it is necessary to define the grinding power as: Grinding power = Total power − No-load power

Previous work has shown that there is an approximately linear relationship between grinding power and the metal removal rate. This can be used to predict the shape of the power decay during the dwell period. The exponential decay of depth of cut during dwell gives rise to an exponential decay in the grinding power. The grinding power during the dwell period can be shown to follow the expression: P = PD e − ( td /τ )

(6)

where PD is the grinding power at the commencement of dwell period and P is the instantaneous grinding power during the dwell period. The above relationship between grinding power and dwell time can then be used together with sampled power data to estimate the system time constant τ during the dwell period. A robust method of estimating the time constant from the power data is to use the LMS technique. In order to employ LMS in the estimation of time constant it is necessary to linearise the power data and time data. Rearranging the previous equation gives: P = e − ( td / τ ) PD

(7)

taking logs to the base e gives: ⎛ P ⎞ ⎛1⎞ loge ⎜ ⎟ = − ⎜ ⎟ td ⎝τ ⎠ ⎝ PD ⎠

(8)

120

M.N. Morgan et al.

The above equation is of the form y = ax, and it is now possible to use LMS to estimate the value of the time constant directly from the sampled power data. Close examination of the power data generated during a real grinding test has indicated that the time immediately following the start of dwell is very susceptible to the effects of power signal noise. Also at the end of the dwell period as the power level approaches the no-load power level, noise can generate significant errors. To avoid these problems it is necessary to use a Weighted Least Means Squares (WLMS) estimation procedure with the weight applied to each power sample being adjusted in real time. The selection of the individual weighting is based on the approximate power level at that instant. In order to allocate a weighting to an individual power sample the power decay is subdivided into a series of bands. Each band is allocated a weighting that is then applied to all power samples lying within that band. In this way power data from stages of the dwell period susceptible to noise can be given a low rating whilst data from stages of the dwell period known to give good results can be given a high weighting. The structure of the WLMS dwell control algorithm is illustrated in Figure 4. Figure 4

The structure of the WLMS dwell control algorithm

In order to implement the above algorithm ‘notice’ of the following events is required: •

start of cycle (CNC to IGA)

•

touch/start of grinding (CNC to IGA)

•

start of dwell (CNC to IGA)

•

end of dwell (IGA to CNC).

Notice is provided via digital flags. The flags enable control of the operation via the following sequence of events: 1

At receipt of the start of cycle flag, the no-load power is continually monitored.

2

At receipt of the touch flag, the no-load power is stored and the grinding power is continually monitored.

Design and implementation of an intelligent grinding assistant system

3

At receipt of the start of dwell flag, the last grinding power is stored and the WLMS algorithm is started.

4

The WLMS algorithm estimates the time constant continually. The estimated time constant is then used to calculate the required dwell time.

5

The dwell control algorithm compares the elapsed dwell time with that calculated and at the appropriate time signals the CNC to terminate the dwell period.

7

Experimental data

121

The typical power curve generated by a single plunge grinding operation is shown in Figure 5. It can be seen that the power during the infeed and dwell periods of the grinding cycle corresponds to the expected response of a first order system. From the typical power curve illustrated above, the effects of noise generated by the grinding wheel spindle drive can be clearly seen. Also clearly visible is the step in the grinding wheel power attributed to the drag exerted by the coolant impinging on the wheel surface. This small increase in coolant will give rise to a slightly increased no-load power level. The effect is seen more clearly in the following Figure 6, which presents the same power data subject to a simple 200 ms moving average. Figure 5

Typical power curve obtained from Balance Systems s.r.l. VM20

The power level during grinding is then monitored and the power at dwell commencement recorded. Data from the power decay is then used to estimate the time constant of the decay. In order to reduce the effects of noise only power data within a selected portion of the decay is used. This is illustrated in the following Figure 7.

122

M.N. Morgan et al.

Figure 6

Initial stages, running average 0.2 sec

Figure 7

Monitored power during dwell period

The output of the time constant estimation algorithm is illustrated below in Figure 8. It can be seen that in this example the algorithm generates an estimation of the time constant after approximately 330 ms and that the most consistent value of time constant is generated during a period corresponding to one time constant of dwell. Once the average specific energy has been calculated for a newly dressed wheel then as the wheel wears progressively the specific energy in each cycle will increase. The specific energy tolerance is the % increase allowed in the cycle specific energy

Design and implementation of an intelligent grinding assistant system

123

before a dress request is issued by the IGA to the CNC. An operator can view the specific energy of each cycle by opening the calculated data or the specific energy dialogue boxes. Note that only for the first nine cycles is the specific energy logged. Thereafter the value for the last cycle is shown along with the calculated average specific energy (see above). Figure 8

8

Values of estimated time constant

Intelligent grinding database

The database was developed in Access 2000. The structure of the database was designed in such a way that all the attributes that define a grinding cycle and its components were included. The database elements were: machine − cycle − material − wheel − dress − coolant

Machine data: this provided details of machine parameters. The data were uploaded from the CNC interfaced to the system. Cycle data: this provided details of work material, work diameter, coolant, wheel type and grinding parameters for each part. The data were uploaded from the CNC interfaced to the system. Optimised cycles for a specified part were used for the future production of those parts. There were three Tables associated with Cycle Data:

The first table stored the reference data named as Reference Cycle Data. This data remained unchanged and had limited access. The second Table provided temporary storage of current data termed: Active Cycle Data. The third Table provided storage for adapted or optimised data termed: Optimised Cycle Data.

124

M.N. Morgan et al.

The development of the system in this way was critical for system implementation. Material data: this provided details of the mechanical and thermal properties of work material. Wheel data: this provided details of wheel type, size, mechanical and thermal properties of the wheel. Dress data: this provided details of dressing parameters. The data would be uploaded from the CNC interfaced to the system. Coolant data: this provided details of physical and thermal properties of the coolant. IPR: this stored all agreed details of ground parts including batch no, part no, cycle data, calculated data, including the maximum grinding temperature, algorithm data and work post assessment data. This ‘Record’ could be used to monitor the production of ‘critical’ parts.

The database HMI was programmed in Visual Basic 6.0 and enabled the operator to read and edit data (password protected for data editing) from the database. It is illustrated schematically in Figure 9. Data validation routines were written to ensure that data input was valid and within stated limits. Figure 9

Database front-end

The records in machine data, material data, coolant data and wheel data covered the data to be used in machining. If the CNC machine could not find the data from the records, an information message box appeared and the database front-end allowed new records to be added in the database.

Design and implementation of an intelligent grinding assistant system

125

The database also allowed optimised cycle data records to be saved to reference cycle data for future use. After grinding one part, the Cycle Data and other useful data were saved in the IPR as a reference. Figure 10 shows the relationship between the IGA© and the intelligent database. The above description summarises the methodology employed in the writing and saving of records to the database. The methodology for automatic updating and retrieval of the most recent and/or optimised records from the database is now described. Figure 10

Flowchart of input cycle data to CNC

8.1 The intelligent grinding parameters selector This feature aided the selection of optimised start machining parameters for a workpiece material or part size new to the database, based on AI techniques and historical saved information. A combination of Case-Based Reasoning (CBR) and Rule-Based Reasoning (RBR) was employed to select grinding conditions and update the knowledge base of the database. A fundamental characteristic of CBR systems is the requirement for sufficient cases to be resident in the database to cover the target Cycle Specification. If insufficient cases are available for a successful search the system will default to RBR procedures. Figure 11 shows the data flowchart for the Intelligent Grinding Parameters Selector (IGPS).

126

M.N. Morgan et al.

Figure 11

Data flowchart for the IGPS

Design and implementation of an intelligent grinding assistant system

127

8.2 Principal of CBR system CBR is an approach which seeks to identify a close match between a new operation to be performed and the characteristics of a previously successful case stored in a case base. The approach solving a Cycle Specification is to remember a similar Cycle Specification solved in the past and adapt the IGPS Output to solve the new Cycle Specification. The basic approach for using CBR is to: •

recall the most appropriate Cycle Specification from IPR

•

modify the Cycle Specification to meet the new Cycle Specification

•

test and further modify

•

save to Reference Cycle Data and IPR for future reference.

The principle of the CBR system developed was to: •

Assign an index to each of the key features of the Cycle Specification. Indexation was employed to establish similarity between cases.

•

Retrieve past cases with similar indexes to the Cycle Specification. The indexes assigned to the key features of the Cycle Specification were matched with similar cases in memory.

•

Adapt the case from memory to match the new Cycle Specification. In the instance when similar cases were identified the nearest case was adapted.

•

Test the new IGPS Output. If the test was successful the output was stored. If unsuccessful, the case was further modified and the loop repeated until success or cases exhausted.

The case contained knowledge concerning workpiece data, wheel data, coolant data, dressing data, process and postassessment data. These data were obtained from the IPR and, consequently, the IPR was kept as a case base. The case was defined in three parts: Cycle Specification, Index and IGPS Output: Cycle Specifications: included workpiece data and machining requirements. The features in a Cycle Specification were defined as ‘Primary’, ‘Secondary’ or ‘Tertiary’. The ‘Primary’ indexes were used to index the basic similarity of cases. Index: based on keywords or values to identify features of the Cycle Specification. Different Cycle Specifications gave different combinations of features that became different cases. The indexes were described by a set of symbols. IGPS Output: an IGPS Output would include specifications of the grinding wheel, the coolant, the dressing conditions and other values of control parameters. An IGPS Output was assumed to be an optimal combination. However, in practice, the grinding wheel may sometimes be constrained by other manufacturing considerations. With a variety of batches of different workpieces it is often necessary to avoid wheel changes and a few general purpose wheels may be used to grind a wide variety of materials, even if at less than optimum efficiency. In addition, wheel speed and wheel diameter may be fixed or varied by the operator according to the requirements of the production

128

M.N. Morgan et al.

situation. However, the more the restrictions were imposed on the IGPS Output the fewer the similar cases in the case base. The only dressing tool currently considered was the single point diamond. If a different dressing tool was employed the dressing conditions would need to be adjusted appropriately.

8.2.1 Case collection-using the index Selection of an appropriate case was achieved when the Cycle Specification Index was matched with a case in the case base. When a user input the Cycle Specification, the system identified the features of the Cycle Specification and created the Cycle Specification Index. The system identified the features of the Cycle Specification with the Index of a case. This was achieved as a result of the following logic: the system matched the Primary Indexes. Primary–Secondary factors are listed in Table 2. The Primary Indexes of the Cycle Specification and of the case in the case library were required to be same. If the match were successful for one or more cases, a set of applicable cases would assume to have been located. If the system did not proceed to the next step, the reasoning process had failed and the executor would quit. Table 2

Primary–Secondary factors

Primary

Secondary

Cycle type

Subwork material

Work material

Subtarget roughness

Target roughness

Work diameter

Hardness

Wheel speed

Wheel type

Wheel diameter

A nearest case could be retrieved from the set of applicable cases through matching where possible the Primary Indexes. A similarity metric was proposed, to judge the similarity between a new case defined by the Cycle Specification and an old case located by the Primary Indexes. The nearest case was ruled to be the case with the highest similarity value. The similarity metric was defined as Similarity =

∑

n i =1

weight i × sim i

∑

n

(9)

weight i i =1

weighti denotes the importance of each feature and the complexity of the modification. The more important the feature is or the more difficult the modification, the higher is the value of the weight. Simi denotes the similarity of the ith feature of two cases. For the material feature, sim (material) is defined as: if the materials are the same, the sim (material) = 1, otherwise, sim (material) = 0. For other features, simi is defined as: ⎛ CycleSpi − Casei ⎞ sim i = 1 − ⎜ ⎟ Range ⎝ ⎠

2

(10)

Design and implementation of an intelligent grinding assistant system

129

8.2.2 Case modification In most situations, the case retrieved would not exactly fit the Cycle Specification definition. The case would, therefore, have to be modified to conform to the new requirements. Since the Cycle Specification case was almost similar to a stored IGPS Output, it was expected that the stored case need only a minor modification to apply to the Cycle Specification. The strategy for modification of a case was based on the following assumptions: •

Since similar cases belong to the same material and roughness groups, the wheel and fluid need not be changed.

•

The wheel speed, vs, either has a fixed value or an optimal value has been established, which is not affected by the variations of other variables. Therefore, wheel speed does not need modification.

•

There are many factors, which affect surface roughness. The most sensitive parameter is dressing lead fd. Dressing lead is, therefore, modified by the system to satisfy the required surface roughness and other parameters are unchanged.

•

For changes in workpiece diameter from dun to dwn or wheel speed from vs0 to vsn the feedrate is changed from vf0 to vfn,

vfn = vf 0 •

dw0 vsn dwn vs0

(11)

For changes in equivalent diameter from de0 to den or wheel speed from vs0 to vsn the workspeed is changed from vw0 to vwn, vwn = vw0

den vsn de0 vs0

(12)

where

de = •

ds dw ds + dw

(13)

For a different surface roughness, dressing depth ad and equivalent chip thickness heq are unchanged, a change of surface roughness from Ra0 to Ran will require a change in dressing lead from fd0 to fdn: fdn =

fd0 Ran2 2 Ra0

(14)

The modification of the value of dressing feed is not an exact process due to the uncertainty introduced by the shape of the dressing diamond which will change with the wear. However, with a little modification the method is feasible.

8.3 Principle of RBR system The RBR system is a KBS employing production rules as the main form of knowledge representation and manipulation. A rule is a conditional statement that specifies an action that is supposed to take place under a certain set of conditions. The RBR system was

130

M.N. Morgan et al.

developed on the basis of a series of coded rules. The rules, expressed in Boolean, were developed from data that is given in tables such as those found in Machining Data Handbooks (Grinding Data Book, 1992; Machinability Data Centre, 1980) and from other published sources. The basic architecture of the RBR agent for selection of grinding conditions includes: •

a knowledge base containing a set of production rules

•

an inference engine

•

a working memory containing the current state description.

An Inference ‘engine’ was developed to provide the reasoning strategies for searching the database. In many reference materials the suggested speed for a certain grinding condition was not a single value but a range. The average value of the range was used for inputting the data to Active Cycle Data.

8.4 Evaluation of the IGPS Output It was necessary to firstly evaluate the output of the IGPS to ensure that the data was fit-for-purpose. The evaluation was completed on a production machine tool at the partner plant using actual grinding data. The methodology for the evaluation was: •

operator input cycle data to CNC (data obtained from previous experience)

•

cycle data ported automatically to database

•

IGA© adaptive cycles executed

•

‘new’ machining parameters proposed from adaptive system recorded manually

•

‘new’ data input to database and to CNC

•

CNC used ‘new’ parameters to complete subsequent adaptive cycle

•

database and IGA© assessed for functionality, that is – communication protocols, writing to records and algorithms.

The database was designed to allow new outputs from the IGPS to be input to Active Cycle Data for grinding with ACO. This process provided the learning capability of the CBR system. Within an adaptive control environment the values of parameters are adapted to take account of the changes in the process performance in order to maintain optimal grinding conditions. The adaptive control modules were integral features of the IGA©. It was important, therefore, that the database provided facility for an adaptive control system to feedback optimised values. In this instance, the case based on reasoning system would store the new data for future reference. For the rare case when two optimised cycles had the same index, the system avoided the risk of overwriting good quality data with lower quality data by comparing the specific material removal rates and determining which is better:

Qw′ = π dw vf

(15)

Since the wheel diameter dw would be the same in the two cases, the feedrate vf was used for the comparison. The case having the higher vf was retained in the case base.

Design and implementation of an intelligent grinding assistant system

131

8.4.1 IGPS front-end The implementation of the CBR and RBR was separated into the Input (problem description) module and the select (solution) module. •

•

System input module: the system input module described the grinding ‘problem’. −

Process information input: process information described by the Part Name, Cycle No and Cycle Type, is shown in Figure 12.

−

Workpiece information input: all data concerning the workpiece was input via the workpiece information module.

−

Wheel information input: there were two ways to select the wheel: (1) By the system or (2) by the user. Figure 13 illustrates the input format of the wheel information module.

System select module: on completion of Inputs, the system Select function was actuated. If there were no suitable cases available, RBR was applied. The selection included the wheel, values of grinding and dressing parameters and coolant selection (refer Figure 14, note, however, that the wheel could have been selected either by the system or by the user).

Figure 12

Process information input page

132

M.N. Morgan et al.

Figure 13

Wheel information – user input page

Figure 14

Output from IGPS

Design and implementation of an intelligent grinding assistant system

9

133

Conclusions

An intelligent grinding advisory system (IGA©) and its component parts for the grinding operation has been described. The database was developed as an integral part of a system. The database elements included grinding database, the grinding rule base and the reasoning modules. The database stored: machine data, wheel data, dress data, coolant data, cycle data and individual part program data. A decision-making system, termed: IGPS was integrated in the database. This assisted in the selection of initial grinding parameters. The IGPS was evaluated in a production environment and the working principles were demonstrated successfully. Application of the database provides the potential to greatly reduce set-up time and process proving time. The Intelligent Database system thus developed had flexibility, intelligence and extendibility.

Acknowledgements Thanks are extended to the Italian government for the SIPOR project grant support and also to the collaborating partners, Balance Systems s.r.l. (Italy), Tacchella Macchine s.p.a. (Italy).

References Allanson, D.R. (1993) ‘Coping with the effects of compliance in the adaptive control of grinding processes’, PhD Thesis, Liverpool John Moores University. Allanson, D.R., Kelly, S., Terry, S., Moruzzi J.L. and Rowe, W.B. (1989) ‘Coping with compliance in the control of the grinding process’, Annals of the CIRP, Vol. 38, No. 1, pp.311–314. Allanson, D.R., Thomas, D.A., Moruzzi, J.L. and Rowe, W.B. (1992) ‘Adaptive and learning strategies for CNC grinding’, Proceedings of SERC ACME Conference, pp.84–89. Amitay, G., Malkin, S. and Koren, Y. (1981) ‘Adaptive control optimisation of grinding’, Journal of Engineering for Industry, Transaction of ASME, Vol. 103, pp.103–108. Billatos, S.B. and Tseng, P.C. (1992) ‘Knowledge based optimization for intelligent machining’, Journal of Manufacturing Systems, Vol. 10, No. 6, pp.464–474. Brinksmeier, E. and Popp, C. (1991) ‘A selftuning adaptive control system for grinding Processes’, Annals of the CIRP, Vol. 40, No. 1. Chen, X., Allanson, D.R. and Rowe, W.B. (1998) ‘Life cycle model of the grinding process’, Computers in Industry, Vol. 36, Nos. 1–2, pp.5–11. Chen, Y. (1998) ‘A generic intelligent control system for grinding’, PhD Theses, Liverpool John Moores University, UK. Gao, Y., Jones, B. and Webster, J. (1992) ‘An integrated size and roundness adaptive control system for the plunge grinding process’, International Journal of Machine Tools and Manufacture, Vol. 32, No. 3, pp.291–303. Grinding Data Book (1992) UK: Universal Grinding Wheel Co. Ltd. Hahn, R.S. (1962) ‘On the nature of the grinding process’, Third MATADOR Conference, pp.129–155. He, X.S. (1984) ‘The research of practical adaptive control system for external cylindrical grinding processes’, Proceedings of the 25th International MTDR Conference, pp.169–175. Kelly, S., Rowe, W.B. and Moruzzi, J.L. (1989) Adaptive Grinding Control, Advanced Manufacturing Engineering, Vol. 1, Butterworths, pp.287–295.

134

M.N. Morgan et al.

König, W., Altintas, Y. and Memis, F. (1995) ‘Direct adaptive control of plunge grinding process using acoustic emission (AE) sensor’, International Journal of Machine Tools and Manufacture, Vol. 35, No. 10, pp.1445–1457. König, W. and Werner, G. (1974) ‘Adaptive control optimization of high efficiency external grinding concept: technological basis and application’, Annals of CIRP, Vol. 23, No. 1, p.101. Liverton, J., Thomas, D.A., Allanson, D.R. and Rowe, W.B. (1993) ‘Adaptive control of cylindrical grinding’, Proceedings of SME 5th International Grinding Conference. Machinability Data Centre (1980) Machining Data Hand Book, 3rd edition, Vol. 2, USA: Metcut Research Associates Inc. Malkin, S. (1981) ‘Grinding cycle optimisation’, Annals of CIRP, Vol. 30, Part 1, pp.223–226. Malkin, S. and Koren, Y. (1984) ‘Off-line-grinding optimisation with a micro-computer’, Annals of the CIRP, Vol. 29, No. 1, pp.213–215. Miler, D.A. (1975) ‘An introduction to adaptive control’, Production Engineer, pp.175–180. Rowe, W.B., Allanson, D.R., Pettit, J.A., Moruzzi, J.L. and Kelly, S. (1991a) ‘ Intelligent CNC for grinding’, Proceedings of the Institute of Mechanical Engineers, Vol. 205, pp.233–239. Rowe, W.B., Black, S., Mills, B., Morgan, M.N. and Qi, H.S. (1997) ‘Grinding temperatures and energy partitioning’, Proceedings of the Royal Society, A, Vol. 453, pp.1083–1104. Rowe, W.B., Black, S.C., Morgan, M.N. and Mills, B. (1995) ‘Experimental investigation of temperature in grinding’, Annals of CIRP, Vol. 44, No. 1, pp.329–332, ISBN 3-905-277-23-9. Rowe, W.B., Li, Y. and Inasaki, I. (1994) ‘Applications of artificial intelligence in grinding’, CIRP Annals, Vol. 43, No. 2, pp.521–531. Rowe, W.B., Li, Y., Mills, B. and Allanson, D.R. (1996a) ‘Application of intelligent CNC in grinding’, Computers in Industry, Vol. 31, pp.45–60. Rowe, W.B., Morgan, M.N. and Allanson, D.R. (1991b) ‘An advance in the thermal modelling of grinding processes’, Annals of CIRP, Vol. 40, No. 1, pp.339–342. Rowe, W.B., Morgan, M.N., Black, S.C.E. and Allanson, D.R. (1998) ‘Validation of thermal properties in grinding’, Annals of CIRP, Vol. 47, No. 1, pp.275–279. Rowe, W.B., Morgan, M.N., Black, S.C.E. and Mills, B. (1996b) ‘Simplified approach to thermal damage in grinding’, Annals of the CIRP, Vol. 45, No. 1, pp.299–302, ISBN 3-905277-25-5. Rowe, W.B, Morgan, M.N. and Qi, H.S. (1993b) ‘The effect of deformation on contact length in grinding’, Annals of CIRP, Vol. 42, No. 1, pp.409–412, ISBN 3-905-277-19-0. Rowe, W.B., Thomas, D.A., Moruzzi, J.L. and Allanson, D.R. (1993a) ‘Intelligent CNC grinding’, Manufacturing Engineering, pp.238–241. Spiwak, S. and Kaliszer, H. (1983) ‘Multi-microprocessor control for plunge grinding’, Proceedings of 24th International MTDR Conference, pp.247–257. Tönshoff, H. and Zinngrebe, M. (1986) ‘Optimisation of internal grinding by microcomputer based force control’, Annals of CIRP, Vol. 35, No. 1, pp.293–296. Webster, J. and Zhao, Y.W. (1990) ‘Time-optimum adaptive control of plunge grinding’, International Journal of Machine Tools and Manufacture, Vol. 30, No. 3, pp.413–421. Wada, R. and Kodama, H. (1977) ‘Adaptive control in grinding’, Bulletin of Japan Society of Precision Engineers, Vol. 11, No. 1, p.1. Xiao, G., Malkin, S. and Danai, K. (1992) ‘Intelligent control of cylindrical plunge grinding’, Proceedings of American Control Conference, Vol. 1, Chicago, pp.391–398. Xiao, G., Malkin, S. and Danai, K. (1993) ‘An autonomous system for cylindrical plunge grinding’, Proceedings of NSF Design and Manufacturing System Conference, pp.399–403. Zhao, Y., Webster, J. and Kaliser, H. (1988) ‘In-process size and roundness measuring system for the adaptive control of cylindrical grinding’, Proceedings of 27th International MTDR Conference, pp.289–297. Zhao, Y.W. (1989) ‘In-process measurement and computer control of cylindrical grinding’, PhD Thesis, Birmingham University.

Design and implementation of an intelligent grinding assistant system

Nomenclature a

instantaneous depth of cut

ad

dressing depth

Casei

the value of ith feature in existing case

CycleSpi

the value of ith feature in Cycle Specification

de

equivalent workpiece diameter

dw

workpiece diameter

eC

cycle specific energy

eCA

average cycle specific energy

fd

dressing lead

Fn

normal grinding force

Ft

tangential grinding force

heq

equivalent chip thickness

Ke

effective spring stiffness

Ks

static grinding force coefficient

N1, N2

integers used in dressing strategy

P

instantaneous grinding power

PD

grinding power at commencement of dwell

Qw′

specific removal rate

Simi

similarity of the ith feature of two cases

t

time

td

time into the dwell

Variablerange

variable range of the value of ith feature

vf

feedrate

vs

wheel speed

vw

workspeed

weighti

importance of each feature and the complexity of the modification

x1

deflection present at the start of dwell

Xd

infeed achieved during the dwell

Xi

actual radial infeed position

X i′

rate of change of workpiece radius

Xp

programmed radial infeed position

τ

system time constant

135