E. Ayesa*,**, A. De la Sota***, P. Grau*,**, J.M. Sagarna****, A. Salterain*,** and J. Suescun***** *CEIT, P de Manuel Lardizabal 15, 20018 San Sebastian, Spain (E-mail:
[email protected]) **TECNUN, Engineering School of San Sebastian, University of Navarra, Spain ***Consorcio de Aguas Bilbao –Bizkaia, 48910 Sestao, Bizkaia, Spain ****MSI S. Coop, Ama Kandida 21, 20140 Andoain, Gipuzkoa, Spain *****CADAGUA S.A., Gran Vı´a 45, 48011 Bilbao, Spain Abstract This paper presents the theoretical basis and the main results obtained during the development and full-scale experimental validation of the new supervisory control strategy designed for the Galindo-Bilbao wastewater treatment plant (WWTP). The different phases of the project have been carried out over the last 8 years, combining model simulations, pilot-plant experimentation and full-scale validation. The final control strategy combines three complementary control loops to optimise the nitrogen removal in pre-denitrifying activated sludge plants. The first controller was designed to maintain the average concentration of the ammonia in the effluent via the automatic selection of the most appropriate DO set point in the aerobic reactors. The second control loop optimises the use of the denitrification potential and finally, the third control loop maintains the selected amount of biomass in the biological reactors by automatic manipulation of the wastage rate. Mobile-averaged windows have been implemented to incorporate commonly used averaged values in the control objectives. The performance of the controllers has been successfully assessed through the full-scale experimental validation in one of the lines of the WWTP. Keywords Automation; control; nitrogen removal; PID controllers; practical experience; WWTP operation
Water Science & Technology Vol 53 No 4–5 pp 193–201 Q IWA Publishing 2006
Supervisory control strategies for the new WWTP of Galindo-Bilbao: the long run from the conceptual design to the full-scale experimental validation
Introduction
Over the last few years it has been recognised that a more extensive use of instrumentation, control and automation (ICA) in the wastewater treatment plants (WWTP) could significantly improve both the stability of the process and the quality of the effluent and at the same time achieve a simultaneous reduction in running costs. There has been a combination of factors that have made the progressive implementation of ICA technology to WWTPs possible: improvements in monitoring systems, more actuators, and process control systems, low prices of computers, advances in plant modelling and better trained and qualified operators (Olsson, 2002). However, although many new systems for ICA have often been proposed (Vrecko et al., 2003), only a few of them have been verified in full-scale applications (Ingildsen et al., 2002). Probably, the main technical limitation has been the lack of reliable measuring devices for control purposes but, at present, new online sensors are available and this is no longer the main bottleneck to set in practice ICA proposals (Jeppson et al., 2004). During the second half of 1996, a consortium whose members are a research centre (CEIT), a control engineering company (MSI S. Coop), a water engineering company (CADAGUA) and a water authority (Consorcio Aguas Bilbao-Bizkaia), with the financial support of the Basque Government started a new R&D project focused on the development of control strategies to optimise the operation of the new Galindo-Bilbao WWTP. The flexibility of the plant and the monitoring equipment brought great possibilities for doi: 10.2166/wst.2006.124
193
operation and control. Moreover, there was a calibrated model of the plant available from previous studies (De la Sota et al., 1994). In spite of all these stimulating factors, the whole project has needed several phases and the full-scale implementation of the controllers was not initiated until 2003. This paper describes the supervisory control strategy designed for the Galindo WWTP, the tasks carried out during the different phases of the project and the discussion of the main results obtained during the final validation stage. E. Ayesa et al. 194
Description of the supervisory control strategy
The secondary treatment of the Galindo-Bilbao WWTP was designed for the carbon and nitrogen removal of a wastewater flow of 345,600 m3/day using a predenitrification – nitrification configuration (Concha and Henze, 1996). The main objective of the supervisory control strategy designed for the plant was the optimisation of the existing plant capacity, improving process stability and effluent quality with minimum cost. Therefore, the automatic supervision was conceived as a way to select, at any moment, the most appropriate set-point for the lower-level controllers, leaving the human decision for the top of the hierarchical plant control structure. It is well known that in a conventional or ‘rigid’ operational strategy of a predenitrificatrion activated sludge plant, the process operator has to select (based on their knowledge and experience) several parameters such as the DO concentration in the aerated volumes, the recirculation flow of nitrates to the anoxic zone and the wasting of the excess of sludge. The concentration of DO in the aerated zones should be maintained at a sufficiently high level to support the growth of the adequate organisms, to guarantee the required nitrification rate and to maintain the required level of mixing, but sufficiently low to save energy, avoid the excess mixing and to reduce, as much as possible, the oxygenation in the anoxic volumes induced by the nitrate recirculation. Nitrification is improved by high DO values; however, lowering the DO set-point not only favours the denitrification in the anoxic zone but also some simultaneous nitrification –denitrification in the aerated zone (Olsson and Newell, 1999). It is important to point out that for a conventional urban WWTP, this DO set-point is not able to eliminate completely the hourly fluctuations in the effluent ammonia. Therefore, it is suggested that a variable DO set-point should be selected to only reach the required (not the minimum) averaged value of the ammonia concentration at the end of the aerated volume. The first controller designed for the Galindo-Bilbao WWTP (Figure 1 loop a) had the aim of maintaining the selected 24 h-averaged concentration of the ammonia in the effluent by automatically selecting the most appropriate DO set point in the aerobic reactors (Suescun et al., 2001). Logically, the measurement of effluent ammonia concentration should be made at the outflow of the biological reactors, in order to eliminate the delay and the equalisation effects produced by the volume of the clarifiers. The 24 h mobileaveraged window is a simple mathematical concept that filters the typical variation in the effluent concentration generated by the daily load profile. Consequently, the DO set-point selected by the controller moves smoothly according to medium- and long-term disturbances (for example dry or wet weather, unexpected load variations, changes in water temperature, weekly and seasonal load variations, etc.). The width of the mobile average window can also be used to ‘distribute’ these cyclical perturbations between the effluent ammonia and the DO set-point. For example, a narrower mobile average filter (8 or 12 h) would reduce both the controller’s time-response and the height of the peaks in the instantaneous effluent ammonia, but at the expense of increasing the short-term fluctuations in the DO set-point selected by the controller. Therefore, the plant operator can select the most appropriate size of the mobile average window for each particular plant. However, it should be clearly understood that narrowing of the mobile-averaged windows
Ammonia set-point
Ammonia DO set-point controller
DO controller
WWTP DO
Effluent ammonia
(loop a)
Effluent ammonia 24-hour mobile-average
Nitrates Internal recycle controller Internal recycle controller flow set-point
WWTP (loop b)
E. Ayesa et al.
Nitrates set-point
Nitrates in the last anoxic reactor Mass set-point
Mass controller
Mass of solids
Sludge waste flow set-point
Sludge waste controller
WWTP (loop c) Mass of solids
24-hour mobile-average
Figure 1 Schematic description of the three control loops designed for the Galindo-Bilbao WWTP
for effluent ammonia will cause higher short-term fluctuations in the DO set-point and therefore, a net increase in the total air consumption for a similar mass of ammonia that is removed. This negative effect is due to the nonlinear relationship between the DO concentration and the nitrification rate combined with the decrease in oxygen transfer efficiency at higher DO levels. The second control strategy aims at the optimum use of the denitrification potential of the plant, selecting the most appropriate recycle flow of nitrates from the exit of the aerated volume to the anoxic zone. This means that there is maximum nitrates removal, maximum carbon consumption without oxygen and maximum recovery of the water alkalinity. For this purpose, the second control loop (loop b) designed for the GalindoBilbao WWTP is continuously selecting the required recycle flow to maintain a low and stable concentration of nitrates at the end of the anoxic zone. A low nitrates set-point ensures the complete denitrification using the minimum recycle flow, with a minimum amount of recirculated oxygen and a maximum hydraulic retention time in the anoxic volume. It should be noted that this control is based on the instantaneous value of the nitrates in the anoxic zone and does not need any kind of mobile averaged filter. The selection of an appropriate value for the total amount of sludge in the biological reactors also has a significant influence on the process behaviour. Operating the plant with a high mass of solids (low sludge wastage flow) increases the oxygen demand (because of endogenous respiration) and the solids flux to the secondary settler but, on the contrary, it reduces the sludge production, increases the global biological activity and improves the stability of the process. Then, the third control loop (loop c) has the aim of maintaining a selected averaged value of the total mass of suspended solids in the biological reactors via the automatic manipulation of the wastage rate. The use of the total mass of solids (instead of the direct MLSS measurement), makes the generalisation of this strategy to step-feed or contact-stabilisation plants possible. However, for selecting an appropriate value for the mass set-point, the operator should take into account the maximum solid flux that can be accepted by the settler (associated with the MLSS in the final reactor). A 24-h mobile average filter has been also introduced to reduce the disturbances produced by the daily influent load profile.
195
E. Ayesa et al.
The three supervisory loops previously described above were implemented over the conventional control loops existing in the SCADA of the WWTP, using incremental PI and PID controllers with limits of the derivative gain (Astrom and Hagglund, 1995). The full-scale implementation of the control algorithms required many additional improvement to the computing code so as to guarantee a safe operation of the process (upper and lower limits for all the manipulated variables, digital filtering and sampling of the signals, detection and elimination of outliers, automatic procedures for the initialisation and commutation of the control loops, etc.). The initial tuning of the controllers was made by optimising the trajectory of the closed-loop response models for each manipulated variable. However, it is important to point out that the experimental response-analysis required for the calibration of the SISO models is very difficult in full-scale WWTP where different cause-effects relationships are interrelated and many perturbations are affecting the process simultaneously. At this point, the predictions of the calibrated plant model were extremely useful in combination with the experimental data. The behaviour of the resulting controller was also verified by simulations (using the complete plant model) before the final implementation and, finally, some small additional tuning was also required during the experimental full-scale validation. Description of the R&D project tasks
The design, development and implementation of the advanced control strategies for the Galindo-Bilbao WWTP was carried out by a multidisciplinary working team with the research centre being responsible for the design of the WWTP control strategy. The industrial companies guaranteed the use of robust and professional tools for the development and implementation of the controllers and, finally, the technical staff at the water authority were responsible for the realistic definition of the operational plant objectives. Table 1 shows the main tasks carried out during the R&D project. The project included three complementary working lines: plant modelling and simulation, pilot-plant experimental verification and full-scale validation. The model of the biological treatment was based on the well known ASM1 (Henze et al., 1987) and most of the model parameters and influent characterisation were obtained from previous experimental studies (De la Sota et al., 1994). The simulation platforms WEST (http://www. hemmis.com) and MATLAB-Simulink (http://www.mathworks.com) were used for the numerical simulation of the full-scale and pilot-scale plants, respectively. The first phase of the work (T1 and T2) focused on the design, development and verification of the most appropriate operational strategy for the biological treatment at the Galindo-Bilbao WWTP using both mathematical modelling and pilot-plant experimentation. Once the pilot plant was modelled and calibrated, the optimum operational strategies of the process were designed using model-based optimisation algorithms (Galarza et al., 2001a). The resulting automatic controllers were developed and verified by model simulation and pilot-scale experimentation. However, although the simulation results were very positive, the experimental validation at pilot-scale of the controllers (T2) was severally limited due to an incorrect operation of the on-line analysers for ammonia and nitrates. In fact, only the sludge mass control was successfully verified at the pilot plant Table 1 Time-schedule of the tasks carried out during the whole project 1996
196
Simulation Pilot-scale Full-scale
1997
1998
1999
2000
2001
T1
2002
2003
T5 T2
T4 T3
T6
E. Ayesa et al.
during this initial phase of the project (Galarza et al., 2001b). After several months, new analysers were installed and successfully tested under full-scale conditions (T3) and lately put into operation in the pilot-plant in order to continue with the experimental validation of all the control strategies (T4). Finally, the partners of the research project decided to take on the full-scale development, implementation and validation of the control strategies at one of the six lines of the Galindo-Bilbao WWTP. With this objective in mind, the dynamic behaviour of the full-scale plant was modelled beforehand and simulated in open and closed loops (T5), and the resulting controllers were finally implemented and verified in the plant (T6). Some of the results obtained during this final experimental validation are presented and discussed below.
Results and discussion Results of the numeric simulation of the WWTP behaviour
24-h average N-NH4 (mg/l)
During the first few stages (T1), dynamic simulations of the process permitted an in depth discussion and understanding of the plant potential, facilitating the conceptual definition of the control loops and a preliminary assessment of the benefits that could be expected from the implementation of the controllers. During the final validation period (T5), the simulation of the full-scale plant model was extremely useful for estimating the dynamic response of each of the manipulated variables and for the tuning of the controllers, taking into account the difficulties for decoupling different cause-effects relationships in a multivariable, slow and highly perturbed process like a running WWTP. The main results obtained for 1-year simulation of the pilot plant with the supervisory control strategy was previously presented in Galarza et al. (2001a). As a summary, the simulations with the calibrated plant model estimated a significant increase in process stability and effluent quality (total nitrogen) with a simultaneous reduction of air flow consumption (15% approximately). Figure 2 shows some recent simulation results obtained with WEST for the full-scale plant (Task T5) that illustrate the effect of the supervisory control strategy to the effluent ammonia concentration and the air flow supply, corroborating the simulation results previously obtained for the pilot plant. 6 5
Without supervisory control
With supervisory control
Without supervisory control
With supervisory control
4 3 2 1
0 200
kLa (d–1)
150 100 50 0
May
June
July
August September October November December January February
March
April
Time (month) Figure 2 Some comparative results of the long-term simulation (during 1 year) of the 24-h averaged effluent ammonia and the air supply requirements predicted by the plant model (with and without the proposed control strategy)
197
Results of the full-scale experimental validation of the supervisory control loops
E. Ayesa et al.
The control strategy finally implemented at the plant is based on three complementary and decoupled PID control loops. Figure 3 presents some results obtained during a few weeks of the full-scale experimental validation period. Figure 3 a shows both the experimentally measured concentration of the effluent ammonia (instantaneous and 24 h averaged values at the exit of the biological reactors) and the DO concentration in the aerobic volumes. It can be observed how the supervisory loop regulates the 24-h averaged concentration near the selected set-point (1.0 mg N/L throughout this experimental period) with some occasional high points in the instantaneous concentration that would be significantly flattened in the clarification volume. The proposed feed-back loop is not able to completely eliminate the periodic oscillations around the selected reference and consequently, a safety factor should be considered by the plant operators when selecting the most appropriate set-points. This oscillating behaviour could probably be reduced by introducing some feed-forward action in the controllers or, simply by moving the on-line ammonia analyser from the exit to an intermediate point in the nitrification volume to improve the sensitivity of the controller to the continuous variations in ammonia. This possible modification is currently under study. The supervisory loop for regulating the selected mass of solids within the biological reactors (loop c) was able to maintain a stable value for the averaged mass by manipulating the waste flow (Figure 3b), in spite of process perturbations such as the fluctuations in the influent load (very frequent during the start-up of the different lines of the plant) or the transfer of mass between reactors and settlers. Finally, loop b was able to regulate the instantaneous nitrates concentration at the exit of the final anoxic reactor to a low value (0.5 mg N/L). This supervisory loop has a fast input-response and does not require any kind of mobile-averaged window. On the one hand, this fact facilitates the experimental calibration of the open-loop process model but, on the other, the resulting controller can suffer an excessive propagation of the measurement noise, especially when the derivative terms are included in the control law. Figure 4 shows the significant improvement obtained when an appropriate low-pass filtering of the nitrate signal was introduced during the validation period. The main benefits expected from the implementation of the supervisory control strategy are focused on process stability, effluent quality and reduction in cost during ‘under-loaded conditions’. Some results from the experimental assessment of the air flow consumption (a) 6
N-NH4
24-h average N-NH4
2.5
DOsp
1.5 1.0
2
DO set point (mg/l)
N-NH4 (mg/l)
2.0 4
0.5
100 90
2.0 Mass of solids
Waste flow 1.8
80 70
1.6
60
1.4
50
1.2
40 30 23/08
198
0.0
0
28/08
02/09
07/09
12/09 17/09 Time (dd/mm)
22/09
27/09
02/10
1.0 07/10
Figure 3 Some results of the full-scale experimental validation of the supervisory control loops
Waste flow (m3/min)
Total mass of solids (T)
(b)
Internal recycled flow 1.2
Introduction of low pass filter
0.8 0.4 10/12
11/12
12/12 Time (dd/mm)
13/12
14/12
Internal recycled flow (m3/s)
N-NOx (mg/l)
1.6 N-NOx
0 15/12
E. Ayesa et al.
6 5 4 3 2 1 0 09/12
Figure 4 Effect of signal filtering in the behaviour of the Nitrates loop
are presented in Figure 5. The picture compares the DO concentration and the air flow consumption within the first nitrification volume in two parallel lines during 1 week. One of the lines was operated with a constant DO set-point (1.5 mg/L), while the other had the supervisory loops implemented. The data show how the supervisory control moves the DO concentration according to the averaged effluent quality, producing a net reduction in air flow consumption during low-loaded periods. A quantitative assessment of the expected improvement is not easy to estimate, because it is clearly determined by the load and temperature at all periods of the year. From the experimental results obtained at the Galindo WWTP during the validation period, a simultaneous reduction of both 2.0 mg/L of effluent nitrates (and consequently in total nitrogen) and a fall of 15– 20% in aeration has been obtained, corroborating the results that had been estimated by simulation some years previously. The improvement in effluent quality is mainly based on the optimum use of the denitrification potential while the global reduction in air flow supply is a combination of three factors combined: first, the use of higher amount of nitrates as electron acceptor in anoxic conditions, second, the improved oxygen transfer efficiency at lower DO concentration and third, the continuous adaptation of the ammonia removal to strictly comply the effluent requirements. Other possible collateral and undesirable effects of the low DO concentration, such as the deterioration of the settling properties of the flocks or the settling of solids in the final biological reactors were not observed during the validation period. Once the performance of the controllers was validated at full-scale, the final version of the supervisory loops is now being offered as a commercial product (CIMCO) by the engineering companies involved in the R&D project.
DO (mg/l)
3.0
With supervisory control
Without supervisory control
With supervisory control
2.0 1.0
0.0 3000
Air flow rate (m3/h)
Without supervisory control
2000 1000 0 08/09
09/09
10/09
11/09 12/09 Time (dd/mm)
13/09
14/09
15/09
Figure 5 Comparative assessment of the air flow consumption with and without the supervisory control loops
199
Conclusions and further research
E. Ayesa et al.
The application of supervisory control strategies for the automatic selection of the main operational parameters in WWTPs can be a very useful tool for optimising the exploitation of the process. At the Galindo-Bilbao WWTP, it has been demonstrated that the synergetic combination of three supervisory control loops facilitates a stable operation of the biological process, improves the effluent quality and lowers simultaneously the air flow needed. The first control loop has been able to maintain the average concentration of the effluent ammonia, the second loop has optimised the use of the denitrification potential and the third loop has regulated the mass of solids in the reactors. The controllers have been based on simple and easy-to-implement feed-back control loops. The possible incorporation of more complex control structures to the plant can be studied but the practical experience acquired during the development of the R&D project suggests a careful analysis of all the difficulties associated with the practical application of the controllers to the full-scale process. The researchers ought to pay special attention to supervising the correct procedures for this final implementation because it will determine the success or failure of the whole project. The close relationship among researchers, engineers and water authorities has been a significant contribution to its success. In fact, this positive experience has promoted new collaborations focused on the design and development of automatic controllers for SBR and alternating processes. Current work at the Galindo-Bilbao WWTP is focused on completing the experimental full-scale validation of the control strategies, including longer seasonal variations in influent load and water temperature, and on the implementation of the supervisory control to the six lines at the plant. The possible modification of the control strategy to incorporate biological P removal or the synchronisation of the mass control loop with the sludge treatment (‘plant-wide control’) are also being discussed as possible topics for further research. Acknowledgements
The authors wish to express their gratitude to the Basque Government (SPRI) for the financial support given for the initial research project. Thanks are also due to Miguel Lueje (Consorcio de Aguas Bilbao-Bizkaia) for their fruitful collaboration in the different phases of the R&D project.
References
200
Astrom, K. and Hagglund, T. (1995). PID Controllers: Theory, Design and Tuning, ISA, The International Society for Measurement and Control, NC, USA. Concha, L. and Henze, M. (1996). Advance design and operation of municipal wastewater treatment plants. RDN and DRDN processes. Science Research and Development Technologies for Environmental Protection, Report 10 EUR 16869, EN Environment and Climate Programme; European Commission Directorate General XII. De la Sota, A., Larrea, L., Novak, L., Grau, P. and Henze, M. (1994). Performance and model calibration of RDN process in pilot-plant. Water Sci. Technol., 30(6), 355 – 364. Galarza, A., Ayesa, E., Linaza, M.T., Rivas, A. and Salterain, A. (2001a). Application of mathematical tools to improve the design and operation of activated sludge plants. Case study: the new WWTP of GalindoBilbao. Part II: Selection of operational strategies and automatic controllers. Water Sci. Technol., 43(7), 167 – 174. Galarza, A., Ayesa, E., Salterain, A., Linaza, M.T. and Flo´rez, J. (2001b). Real time control strategies for activated sludge plants – Application to RDN and DRDN processes. Proceedings of the 1st IWA Conference on Instrumentation, Control and Automation - ICA 2001. Malmo, Sweden, June 3 –7, 2001. Published in February 2002 in Water Intelligence Online 1(2), (www.iwaponline.com/wio/2002/02/ wio2002028.htm).
E. Ayesa et al.
Henze, M., Grady, C.P.L., Jr., Gujer, W., Marais, G.v.R. and Matsuo, T. (1987). Activated sludge model No. 1, IAWPRC Scientific and Technical Reports, London, UK, IAWPRC (IAWQ). Ingildsen, P., Jeppsson, U. and Olsson, G. (2002). Dissolved oxygen controller based on on-line measurements of ammonium combining feed-forward and feedback. Water Sci. Technol., 45(4 –5), 453 –460. Jeppson, U., Alex, J. and Ayesa, E. (2004). Report of the Working Group 1 (Plan Operation). COST624 Action (2004). Olsson, G. (2002). Lessons learnt at ICA2001. Water Sci. Technol., 45(4– 5), 1 – 8. Olsson, G. and Newell, B. (1999). Wastewater Treatment Systems. Modelling, Diagnosis and Control, IWA Publishing, London, UK. Suescun, J., Ostolaza, X., Garcı´a-Sanz, M. and Ayesa, E. (2001). Real-time control strategies for predenitrification – nitrification activated sludge plants: biodegradation control. Water Sci. Technol., 43(1), 209 –216. Vrecko, D., Hvala, N. and Carlsson, B. (2003). Feedforward-feedback control of an activated sludge process: a simulation study. Water Sci. Technol., 47(12), 19 – 26.
201