(190c) Advances in Data-Driven Fault Detection and Identification and Application to a Bench Scale Fischer-Tropsch Process

Data-driven modeling is a topic of great interest in current literature, where the core motivation is to find robust black-box models capable of replacing complex mathematical formulas in representing the dynamics of a multivariate system. Black-box models are particularly useful for situations where the mathematical model of the system is unavailable or difficult to accurately estimate. Robust data-driven modeling techniques can then be reliably applied to fault detection and identification applications, where the objective of the former is to identify when the system has deviated outside of the expected normal range of operation. On the other hand, the objective of fault identification is to identify the exact nature of the fault that the system is currently experiencing, if any. Both applications are closely interlinked and highly beneficial for the sake of critical early maintenance and improved safety standards in time-sensitive and/or expensive operations [1-3]. In this work, several linear and non-linear data-driven modeling techniques are going to be applied to real data samples gathered from a bench-scale Fischer-Tropsch process, where the ultimate objective is to compare the accuracy of the models in terms of their robustness in the presence of noise and their ability to detect faults. Linear models include Principal Component Analysis (PCA) and some of its extensions, such as Bayesian PCA and/or Multiscale PCA [4]. Non-linear models include Kernel PCA and Neural Networks [3,5].

Keywords: Fault Detection, Fischer-Tropsch, Data-Driven Modeling.

References

[1] Basha, N., Sheriff, M. Z., Kravaris, C., Nounou, H., Nounou, M., Multiclass data classification using fault detection-based techniques, Computers & Chemical Engineering, 136, 1-11 (2020).

[2] Basha, N., Kravaris, C., Nounou, H., Nounou, M., Bayesian-optimized gaussian process-based fault classification in industrial processes, Computers & Chemical Engineering, 170, 1-11 (2023).

[3] Basha, N., Ibrahim, G., Choudhury, H.A., Challiwala, M.S., Fezai, R., Malluhi, B., Nounou, H., Elbashir, N., Nounou, M., Bayesian-optimized Neural Networks and their application to model gas-to-liquid plants, Gas Science and Engineering, 113, 1-11 (2023).

[4] Malluhi, B., Basha, N., Fezai, R., Ibrahim, G., Choudhury, H.A., Challiwala, M., Nounou, H., Elbashir, N., Nounou, M., Multiscale Bayesian PCA for robust process modeling of a Fischer–Tropsch bench scale process, Chemometrics and Intelligent Laboratory Systems, 240, 2023, 104921.

[5] Fezai, R., Malluhi, B., Basha, N., Ibrahim, G., Choudhury, H.A., Challiwala, M.S., Nounou, H., Elbashir, N., Nounou, M., Bayesian optimization of multiscale kernel principal component analysis and its application to model Gas-to-liquid (GTL) process data, Energy, 284, 2023, 129221.