TY - CHAP
T1 - Fault detection in a benchmark simulation model for wastewater treatment plants
AU - Ramin, Pedram
AU - Flores-Alsina, Xavier
AU - Topalian, Sebastian Olivier Nymann
AU - Jeppsson, Ulf
AU - Gernaey, Krist
PY - 2022
Y1 - 2022
N2 - The International Water Association (IWA) Benchmark Simulation Models (BSM1 and BSM2) have been successfully used in both industry and academia to test and verify control strategies in wastewater treatment plants (WWTPs). In this study, a new (plant- wide) benchmark simulation model, the BSM2-LT, is developed to evaluate monitoring algorithms. This platform provides opportunities to generate various sensor/actuator and process faults. To make this realistically, different Markov-chain models are used to re- create the alternation of sensor/actuator states based on predefined occurrence probability. The same principle is used to describe the occurrence of toxic/inhibitory compounds. Using this platform, one can test the performance of a monitoring algorithm such as a fault detection method. To demonstrate this in an example, a multivariate method based on adaptive dynamic principal component analysis (dPCA) was used to detect faulty events. The performance of the monitoring algorithm is evaluated with a penalization index, scoring from 0 to 100. While the tested method had a good false alarm score, it resulted in a low false acceptance. While the results could be certainly improved, the main focus of this study is the benchmark simulation model and not presenting a well optimized monitoring algorithm. The software which will be produced and freely distributed in the near future, will allow an objective evaluation of monitoring algorithms for WWTPs for any user.
AB - The International Water Association (IWA) Benchmark Simulation Models (BSM1 and BSM2) have been successfully used in both industry and academia to test and verify control strategies in wastewater treatment plants (WWTPs). In this study, a new (plant- wide) benchmark simulation model, the BSM2-LT, is developed to evaluate monitoring algorithms. This platform provides opportunities to generate various sensor/actuator and process faults. To make this realistically, different Markov-chain models are used to re- create the alternation of sensor/actuator states based on predefined occurrence probability. The same principle is used to describe the occurrence of toxic/inhibitory compounds. Using this platform, one can test the performance of a monitoring algorithm such as a fault detection method. To demonstrate this in an example, a multivariate method based on adaptive dynamic principal component analysis (dPCA) was used to detect faulty events. The performance of the monitoring algorithm is evaluated with a penalization index, scoring from 0 to 100. While the tested method had a good false alarm score, it resulted in a low false acceptance. While the results could be certainly improved, the main focus of this study is the benchmark simulation model and not presenting a well optimized monitoring algorithm. The software which will be produced and freely distributed in the near future, will allow an objective evaluation of monitoring algorithms for WWTPs for any user.
KW - Benchmark simulation
KW - Fault detection
KW - Markov chains
KW - Monitoring algorithms
KW - Wastewater treatment
U2 - 10.1016/B978-0-323-85159-6.50227-X
DO - 10.1016/B978-0-323-85159-6.50227-X
M3 - Book chapter
AN - SCOPUS:85136146989
SN - 978-0-323-85159-6
T3 - Computer Aided Chemical Engineering
SP - 1363
EP - 1368
BT - 14th International Symposium on Process Systems Engineering
PB - Elsevier Science Publishers B.V.
ER -