J. Alex*, L. Rieger**, S. Winkler*** and H. Siegrist** * ifak, Institute for Automation and Communication, Steinfeldstrasse, 39179 Barleben, Germany (E-mail:
[email protected]) ** EAWAG, Swiss Federal Institute for Environmental Science and Technology, Ueberlandstrasse 133, PO Box 611, 8600 Duebendorf, Switzerland (E-mail:
[email protected]) *** Institute of Water Quality and Waste Management, Vienna University of Technology, A – 1040 Vienna, Karlsplatz 13/E 226, Austria (E-mail:
[email protected]) Abstract To show the impact of sensor behaviour on the control result, four strategies for aeration control are tested using different sensor characteristics. It is demonstrated, on the one hand, how an increasing response time will limit the achievable control quality and, on the other hand, how a given sensor characteristic can be taken into account for the controller design. The presented tests show that an improvement potential by control for WWTPs is available but this potential is limited compared to proper DO control with fixed set-points. To activate this control potential, sufficient control authority must be available and a careful control design is required. It can be shown that using feedback control, sensors with a small response time have significant advantages compared to conventional sensors. Using feed forward control, the improvement potential by control is considerably higher and additionally, the sensor delay can be integrated into the controller design. The presented discussion is based on simulation studies performed on a standardised benchmark case. For these tests it was necessary to include sensor models into the simulation model. It can be stated that the usage of sensor models is necessary for the application of dynamic simulation for the design and evaluation of WWTP control and in general to achieve realistic results. Keywords Benchmark simulation; control strategy evaluation; plant dynamics; process control; sensor model
Water Science and Technology Vol 47 No 2 pp 113–120 © IWA Publishing 2003
Progress in sensor technology – progress in process control? Part II: results from a simulation benchmark study
Introduction
In the last few years new types of sensors suitable for application in wastewater treatment facilities have been developed and released to the market. These new sensors ease the application of continuous measurement systems at locations (raw sewage, primary effluent), which could hardly be operated when applying conventional analyser systems. Comprehensive studies have been carried out investigating the accuracy, reliability and applicability of some of the new sensor systems (Ingildsen and Olsson, 2002; Rieger et al., 2002 ; Winkler et al., 2002). Some of the new sensor systems operate with only a very short response time (< 5 sec, Thomann et al., 2002), which is advantageous for control applications. The importance of a small response time of sensors or analysers becomes significant in cases where the time constants of the controlled process are similar or smaller than the sensor response time. It can be assumed that the availability of reliable measurements with shorter response times will allow more advanced control strategies. A proven tool for the testing and optimisation of control strategies is the dynamic simulation of activated sludge systems (i.e. with the ASM family, Henze et al., 2000). A limiting factor for the comparability to field applications is the common assumption of ideal (no delay or noise) or simplified (only delay) sensor models. The evaluation of the effect of sensor behaviour to the overall control performance by dynamic simulation requires that appropriate sensor models are used. In part I of this work (Rieger et al., 2003), models for specific sensors and a classification of typical sensor types for use with the COST simulation
113
benchmark environment are proposed. In this simulation study, four controllers are tested using an ideal sensor model and a sensor model with a response time of 40 minutes. This shall demonstrate the impact of the measuring equipment on the control result. Problem analysis J. Alex et al.
The design of control systems for wastewater treatment plants is often not a synchronous process of selection of measurement equipment and control design. Instead it can still be observed that these tasks are carried out separately without sufficient interaction. This is partly caused by a lack of interdisciplinary knowledge of the experts involved and partly due to a sometimes insufficient time budget for the planning phase. This can lead to nonoptimal control systems. Regarding only the measuring equipment, the main problems for control purposes are the response time, noise effects and different systematic errors. On the controller side, often the control authority of the control handles is not sufficient or limited by the adjusting range. For example, blowers are often designed for estimated future load scenarios and the air flow rate can subsequently not be reduced to an appropriate level during low load periods. Other limitations with respect to manipulation of the air flow rate are aerator properties, for example ceramic diffusers cannot be used for intermittent aeration. In order to investigate the required control authority for a given plant and load scenario, dynamic simulation is a suitable tool. The application of this tool will yield better results if realistic sensor models are included in the simulation model. The modelling of the control handles can be important as well; however it has not been considered in detail in this paper. The combination of biological, sensor, controller and control handle models allows one to define the requirements of the control strategy with regard to the measuring equipment or to optimise existing control systems. Based on these preconditions it becomes possible to design well suited control systems using dynamic simulation with a good chance to prove their potential in practice as well. Most of the control strategies applied in wastewater treatment plants today represent – from a control engineering point of view – a relatively low level of sophistication. Often fixed time phase controller, two-point controllers or simple PI-controllers are applied. These control strategies proved to yield satisfactory results and are very stable and robust in operation. In this respect it has to be considered that the good control results can be achieved partly due to the ample process volume available, which leads to an immediate dilution of incoming peak loads. This makes the process very robust against disturbances and subsequently does not require an immediate control action. Nevertheless, with constantly increasing demands with regard to effluent quality and optimised plant operation, more sophisticated control strategies are gaining importance. Obviously – from a control point of view – a process can be operated more successfully if a controller anticipatory drives the system as opposed to reacting on process disturbances. Feed forward control strategies trigger control actions based on information about process influences before a reaction of the process can be observed. With this, a process can be operated in a way that the influence of predictable disturbances is minimised. Simulation study on the effects of ammonium sensor properties
114
In the following simulation study the effect of ammonium sensor behaviour on different aeration control systems is compared. Today dynamic simulation is widely accepted as a tool for system analysis and optimisation. In order to compare and evaluate different control strategies for wastewater treatment plants, benchmark simulation models have been developed (Alex et al., 1999). The benchmark simulation models provide a standardised environment, i.e. the plant configuration, size and dynamics of the plant load are
variable Zone 1
Denitrification M M Influent
variable Zone 2
Nitrification
Secondary Clarification
Nitrification
ye P direct
SC
NH1
qair5 qair4 qair3
18446+385 NH4
qair2
internal recirculation
return sludge
excess sludge
P
yw
385
J. Alex et al.
Qir
Figure 1 Block diagram of a COST benchmark simulation model implementation in SIMBA
defined. Also methods for evaluation of the control results have been developed (Copp, 1999) which take into account the effluent quality and energy consumption. The COST benchmark model (Figure 1) was applied for the simulation study presented in this paper. This model describes an activated sludge plant with pre-denitrification. The plant model consists of five cascade tanks in series, where the first tank cannot be aerated and is therefore used for denitrification. In divergence from the original benchmark model description, tanks 2 and 3 are implemented as nitrification/denitrification tanks, i.e. they can alternatively be aerated or mixed only. Tanks 4 and 5 are – like in the original COST benchmark model – used for nitrification. The total tank volume is 6,000 m3, the same as the volume of the final clarification, which has a surface area of 1,500 m2. The operation temperature is defined at T = 15°C. The dry weather average flow is defined at 18,500 m3, the total COD load is 7,050 kg/d and the total nitrogen load is 908 kg/d. This corresponds to a relatively high volumetric loading of 0.6 kgBOD5/m3 and a very high N:COD-ratio of 0.13. From these numbers it can be concluded that the plant is not ideal for achieving high nitrogen removal efficiency. The dry weather influent (Figure 2) shows relatively high dynamics, with a peak factor of the flow of 1.74 and the COD load of 2.34, respectively. The dry weather case has been chosen instead of the rain weather case as from the simulation results it was obvious that the dry weather case is more critical with respect to ammonium peaks. The definition of the rain weather influent data is such that the effect of the first flush is not included. Instead immediate dilution takes place resulting in a less critical load case. The excess sludge withdrawal is defined with a fixed flow rate of 385 m3/d from the return sludge, resulting in a maximum total sludge age of approximately nine days. The 4
2
x 10
4000
2000
1
TKN load [kg N/d]
COD load [kg COD/d]
COD
TKN 0
7
7.5
8 t [d]
8.5
0 9
Figure 2 Influent load pattern of two selected days as defined in the COST benchmark model
115
return sludge flow is operated with a constant rate of approximately 100% of the mean influent flow. In order to improve the nitrogen removal performance, an internal recycle from the fifth into the first tank with a constant flow rate of 30,000 m3/d (approximately 160% of mean influent) is included. Due to the high N:COD ratio this flow rate is more than sufficient to avoid nitrate limitation in the denitrification tank, in fact denitrification is more likely to be substrate limited. J. Alex et al.
Description of the applied control strategies
Four different control strategies were applied in the simulation study which are described in the following paragraphs. The configuration of the aeration system for all strategies is such that tanks 4 and 5 and tanks 2 and 3 are supplied by a common air pipe, respectively. The air supply to tank 2 is controlled by a butterfly valve, which is opened or closed on a switching condition defined in the respective control strategies. DO controller. This control system serves as the reference case. It is a classic DO controller, which controls DO concentration in tank 4 at 1.7 mg/l, and the DO concentration in tank 3 at 2 mg/l. The air supply to tank 2 is always shut off. NH4 feedback (cascade). This control concept is a typical concept which is proposed frequently for treatment plants e.g. in Germany. On top of the DO controller described above an ammonium based controller is implemented to determine the DO set-points of the DO controller. An effluent ammonium measurement is used to detect the load situation of the plant, three cases are defined: low, medium and high. Depending on the actual load situation, DO set-points for an underlying DO controller and the state of the air flow valve to tank 2 are defined. NH4-feed forward/feedback (cascade). This controller is an extension of the NH4 feedbackcascade controller described above by a feedforward controller. The ammonium concentration in the influent and the influent flow is measured continuously and from this the ammonium influent load is calculated. Two threshold values are defined. If the influent ammonium load exceeds the lower threshold value, tank 3 is forced to be aerated. A maximum condition selects the larger of the two DO set-points (one defined by the feedback part, the other by the feedforward part) in the case where tank 3 is aerated. If the second threshold value is exceeded, the air valve to tank 2 is forced to be opened. NH4-feedback (direct). This control concept is applied at a number of treatment plants in Austria. An effluent ammonium measurement is used directly for aeration control. The effluent ammonium concentration is controlled at 1.5 mg/l by means of a PI-controller. The air distribution is defined by the following scheme. Initially, only tanks 4 and 5 are aerated. If the air flow into these two tanks exceeds 90% of the maximum air flow, than tank 3 also starts to be aerated. If the air flow into tank 3 is at its maximum, the air flow valve to tank 2 is opened. The simulation model presented in Figure 1 utilises this control option. 1.6
L
M 1.2
116
Load state
3.0
H 2.6
Low Medium High
Air valve
DO set-point
DO set-point
tank 2
tank 3 (mg/l)
tank 4 (mg/l)
OFF OFF ON
0.0 1.0 2.0
1.5 1.7 1.7
Figure 3 State transition diagram and DO set-point table of the NH4 feedback controller for change between the three load states, low(L), medium(M) and high(H) on condition of the NH4-N effluent concentration [mg/l]
Simulation results NH4 feedback controller (cascade)
J. Alex et al.
All figures showing the results from the simulation study include the simulation results applying the DO controller as the reference case (thin solid line). The sensor classes refer to the classification according to Rieger et al. (2003), where class A is a sensor with a very short response time (1 min) and class C is a sensor with a higher response time (here 40 min). Only for class C is noise considered. Figure 4 shows a comparison of the influence of the ammonium effluent sensor properties on the performance on the NH4 feedbackcascade controller. Additionally, the simulation results of the simulation applying the NH4 feedforward/feedback-cascade controller with a class C sensor are included in these figures. As a starting point for each simulation, a steady state of the plant was used as defined in the benchmark specification (long time dry weather conditions). In Figure 4 and Figure 5, the results from day 7 and 8 of a 14 day simulation period are presented, thus potential long term effects of less nitrification would already be visible. In addition, Table 1 provides information about the mean effluent concentration for NH4-N and inorganic N, the NH4-N maximum value at day 7 as well as the necessary aeration energy for the controllers considered. From the time series for the NH4-N concentration in the last tank it can be clearly noted that the benchmark plant operates in a very high load situation. Nitrification works stable; however fairly high effluent peaks occur. The denitrification capacity is limited as well, and thus the mean effluent value for inorganic N is around 14 g/m3 with peaks up to 20 g/m3. The aim of the proposed controllers is to reduce NH4-N peaks and to increase the overall N elimination rate by a better denitrification during low-load situations (night). The NH4-cascade controller has control options available to increase or decrease the nitrification potential (possible aeration of tanks 2 and 3) compared to the DO controller reference. Thus one could expect that an increased denitrification during the low-load night hours can be combined with an appropriate action to avoid NH4-N peaks during the highload hours. This can partially be observed for the case of an ideal sensor (class A). The improved denitrification capacity during the night can be achieved together with a slight reduction (5%) of the NH4-N peak for the morning flush compared to the DO controller reference. However, already for a sensor with a usual response time of 40 min (class C1, T90 = 40 min) this cannot be achieved. The peak is more than 7% higher than for constant operation, although more nitrification capacity would be available in this case. Obviously, load increases are detected too late. Subsequently the aeration is increased at times when the ammonium concentration is already at an increased level compared to the DO controller. In conclusion of these results it must be stated that a successful control of the nitrogen removal by adjustments of DO set-points and the usage of variable zones for nitrification and denitrication requires sufficient fast response times for the measurement devices used and sufficient control authority available. In this case, a relative large volume was operated as a variable zone nitrification/denitrication zone (tank 2 and 3 representing 39% of the overall volume). NH4 feedforward/feedback controller (cascade)
Analysing the performance of the NH4-cascade controller (Figure 4), it must be stated, that even if more of the tank volume can be aerated than for the reference case in the high load situation, only approximately the same peak level can be achieved. This is related to the principal limitation of feedback control to reject disturbances, especially for plug-flow type plants. The only way to improve the control behaviour is to apply feed forward control. However, this is only possible when applying the new generation of sensors, as they are able to measure even in difficult locations such as raw influent or effluent of the primary clarifier.
117
Table 2 Simulation results, mean values of the last 7 days of a 14 day dry weather simulation period in tank 5, maximum value NH4-N at day 7 Controller
NH4-N max.,
NH4-N
day 7, g N/m3 mean, g N/m3
J. Alex et al.
DO controller 9.1 8.6 NH4 feedback (cascade), class A 9.8 NH4 feedback (cascade), class C (40 min) NH4 feedforward/feedback (cascade), class C (40 min) 6.8 9.4 NH4 feedback (direct), class A NH4 feedback (direct), class C (40 min) 10.3
Aeration energy
mean, g N/m3
kWh/d
14.7 13.6 13.7 13.9 14.4 14.5
6,909 6,820 6,820 7,066 6,239 6,152
DO controller NH4-feedback (cascade), class A NH4-feedback (cascade), class C (40min) NH4-feedforward/feedback (cascade), class C (40min)
12 NH4-N (Effluent N zone) [g/m3]
2.3 2.4 2.7 1.8 2.5 2.8
NH4-N+NO3-N
10
8
6
4
2
0 7
7.2
7.4
7.6
7.8
8 t [d]
8.2
8.4
8.6
8.8
9
NH4-N+NO3-N (Effluent N zone) [g/m3]
22 20 18 16 14 DO controller NH4-feedback (cascade), class A NH4-feedback (cascade), class C (40min) NH4-feedforward/feedback (cascade), class C (40min)
12 10 8 7
7.2
7.4
7.6
7.8
8 t [d]
8.2
8.4
8.6
8.8
9
Figure 4 Simulation results for the cascade-NH4 controller compared to the DO controller
Figure 4 shows the effect of an additional feed forward action, where the ammonia peaks are reduced by approximately 26% and an improved denitrification is achieved during the night. In addition, this performance can be achieved even with a slow sensor. This behaviour is the result of the possibility to consider the sensor dynamics during the design of the feed forward action. As this controller decreases the mean NH4 effluent concentration by approximately 20% compared to the DO controller, a slightly increased aeration energy need (2%) is inevitable. NH4-feedback controller (direct)
118
The cascade controller discussed above gains already some robustness against NH4 measurement delays, as the underlying DO control loops will suppress the effect of increased NH4 load already to some extent. For the direct NH4 controller, one could expect a higher sensitivity of the controller performance to the response time of the used NH4 measurement device. In Figure 5, the performance of the direct NH4 controller is compared to the DO controller.
DO controller NH4-feedback (direct), class A NH4-feedback (direct), class C (40 min)
10
8
6
J. Alex et al.
NH4-N (Effluent N zone) [g/m3]
12
4
2
0 7
7.2
7.4
7.6
7.8
8 t [d]
7
7.2
7.4
7.6
7.8
8 t [d]
8.2
8.4
8.6
8.8
9
NH4-N+NO3-N (Effluent N zone) [g/m3]
22 20 18 16 14 12 10 8
DO controller NH4-feedback (direct), class A NH4-feedback C (40 min)9 8.2 8.4 (direct), 8.6 class 8.8
Figure 5 Simulation results for the direct-NH4 controller compared to the pure DO controller
For the application of the fast sensor (class A), a similar performance for the NH4-N effluent concentration can be achieved than by using the cascade approach. An improved denitrification during the night hours can be achieved without increased peak levels of the NH4-N concentration for the morning peak. The improvements of this controller regarding nitrogen removal are not as good as for the cascade controller but a significant reduction of the aeration energy (approximately 10%) can be noted for this control option. Thus, these properties of this controller may be favourable for low loaded plants. As assumed, this control option shows an increased sensitivity to the response time of the NH4-N sensor. The application of a sensor with a response time of 40 min will lead to considerably higher NH4 peaks than using DO control. To deal appropriately with the high load peaks of the benchmark plant, a sensor with a short response time is mandatory. Conclusions
In reality, high response times up to and more than half an hour for sensors measuring nutrients or total components are a common situation. If this is not taken into account, the results of a simulation study to develop or evaluate the control of a WWTP must be evaluated very carefully. If one expects to migrate control structure and parameter settings obtained by a simulation study and to achieve similar results in practice it becomes necessary to use realistic sensor models for a simulation study. For the evaluation of the effect of sensor dynamics to the control performance, it can be concluded from the presented results that sensors with high response times can still be suitably adapted to low load situations. However they cannot sufficiently deal with high load peaks. If the sensor dynamics can be considered within the controller design (e.g. for feed forward action), the effect of sensor delays can be compensated. Since already the set-up and tuning of the proposed controllers in a virtual environment
119
J. Alex et al.
(benchmark simulation) was not a trivial task, it must therefore be concluded for practical applications that improvements by control are not easy to achieve. Thus, the additional potential provided by sensors with small response times are very welcome and will increase the likelihood of proper control performance in practice. For example, the direct NH4 controller works only sufficiently with a sensor with a small response time. This clearly shows that there is a strong need to define the requirements of the control strategy on the measuring equipment. With regard to the possible control results, control of N-elimination should only be used as an alternative to DO control if enough control authority is available to avoid worse behaviour during high load situations. From the performance of NH4 controllers (cascade and direct) it can be concluded that assuming a sufficient control authority and a carefully designed controller, it is possible to improve the plant performance without an additional danger of violating effluent limits compared to conservative DO control. This can be achieved for realistic sensor behaviour (e.g. for the cascade controller with T90 = 40 min). However sensors with smaller response times increase the quality of control and minimise the risk of non-optimal operation. Feed forward control was found to be the best option to deal with peak load situations. This control option additionally provides the opportunity to consider the sensor dynamics explicitly. However, the additional influent measurement – in the presented case of NH4-N – requires sensors which can be operated in raw wastewater without too much maintenance effort. Experiences (Rieger et al. 2002, Winkler et al. 2002) show that this can be expected from novel sensors. Nevertheless, it has to be mentioned that all results should be considered with respect to the high loaded benchmark environment. Other, perhaps more typical, plants with a temporary or constantly low loaded influent and a less steep dynamic will allow a greater reduction of energy consumption parallel to stable nitrification during peak loads. References Alex, J., Beteau, J.-F., Copp, J.B., Hellinga, C., Jeppson, U., Marsili-Libelli, S., Pons, M.N., Spanjers, H. and Vanhooren, H. (1999). Benchmark for Evaluating Control Strategies in Wastewater Treatment Plants. Proceedings ECC’99 (CD), Karlsruhe, BM-10. VDI/VDE, Duesseldorf. Copp, J.B. (1999). The IWA simulation benchmark: background and use. IWA Scientific and Technical Report Task Group: Respirometry in Control of the Activated Sludge Process, Internal Report#6. Copp, J.B. (Editor) (2002). The COST simulation benchmark – Description and simulator manual. COST action 624, Office for official publications of the European Communities, Luxembourg. Henze, M., Gujer, W., Mino, T. and van Loosdrecht, M. (2000). Scientific and Technical Report No 9: Activated Sludge Models ASM1, ASM2, ASM2d and ASM3. IWA Task Group on Mathematical Modelling for Design and Operation of Biological Wastewater Treatment, IWA Publishing. Ingildsen, P. and Olsson, G. (2002). Exploiting online in-situ ammonium, nitrate and phosphate sensors in full-scale wastewater plant operation. Wat. Sci. Tech. 46(4–5), 139–147. Rieger, L., Siegrist, H., Winkler, S., Saracevic, E., Votava, R. and Nadler, J. (2002). In-situ measurement of ammonium and nitrate in the activated sludge process. Wat. Sci. Tech. 45(4–5), 93–100. Rieger, L., Alex, J., Winkler, S., Böhler, M., Thomann, M. and Siegrist, H. (2003). Progress in sensor technology – progress in process control? Part I: Sensor property investigation and classification. Wat. Sci. Tech. 47(2), 103–112 (preceding article). Thomann, M., Rieger, L., Frommhold, S., Siegrist, H. and Gujer, W. (2002). A monitoring concept for online sensors. Wat. Sci. Tech. 46(4–5), 107–116. Winkler, S., Rieger, L., Bornemann, C., Thomann, M. and Siegrist, H. (2002). In-line monitoring of COD and COD-fractionation: Towards new data quality for dynamic simulation of wastewater treatment plants. Proc. 3rd IWA world water congress, Melbourne (paper No. e21604a).
120
