A class of FTC methods that has been widely studied is active FTC methods where a fault diagnosis scheme that detects and isolates the faults is combined with a fault accommodation strategy that dynamically adjusts control actions in real-time to maintain the desired closed-loop system stability and performance properties. While extensive research has been dedicated to developing fault diagnosis methods (see, e.g., [1], [2], [3]), fault estimation and accommodation are equally critical to active FTC. Accurate fault estimates guide the selection of appropriate accommodation strategies to mitigate the adverse effects of faults on system performance.
In this light, methods for the estimation and accommodation of actuator and sensor faults have been developed in previous research works (see, e.g., [4],[5],[6],[7]). While these studies have focused primarily on fault handling within an individual process unit, large-scale process networks, which feature distributed arrangements of interconnected process units, present additional challenges that must be addressed. Due to the interconnections and dynamic coupling between constituent subsystems, local faults in one subsystem may propagate throughout the plant, potentially leading to system-wide failures. Timely identification and mitigation of faults is therefore of paramount importance.
These challenges have motivated several efforts towards the development of active FTC methods for large-scale process systems (see [8] for a survey). However, existing approaches have predominantly concentrated on addressing the diagnosis, estimation and accommodation problems separately, showing limited emphasis on a cohesive framework for handling simultaneous faults distributed across multiple interconnected units. Notable exceptions include the framework presented in [9] for actuator fault identification and controller reconfiguration in linear systems. Additionally, a distributed fault estimation observer and FTC framework, based on static output feedback, was developed for linear systems with actuator faults in [10]. At this stage, a comprehensive framework for the detection, estimation, and accommodation of simultaneous sensor faults, prevalent and pivotal for the performance and stability of closed-loop systems, remains an ongoing challenge for large-scale process networks with nonlinear dynamics.
Motivated by these considerations, we present in this work an integrated approach for the estimation and accommodation of simultaneous sensor faults distributed across multiple subsystems in a process network. We focus on process networks comprised of interconnected nonlinear subsystems with local control systems that exchange (possibly faulty) sensor measurements over a shared communication medium at discrete times. The proposed approach brings together tools from supervised machine learning for classification-based fault detection and estimation and techniques from model-based control for accommodation of the faults.
Initially, a quasi-decentralized model-based controller that optimizes the rate of data transfer between the distributed control subsystems is designed to meet network resource constraints. A rigorous closed-loop stability analysis is conducted to obtain an explicit relationship linking the closed-loop stability region to the fault magnitudes, the communication rate, and the controller design parameters. The stability region provides the basis for the development of the fault accommodation strategies.
Both centralized and decentralized strategies for fault estimation are developed and investigated within the proposed framework. In the decentralized implementation, a single-output feedforward neural network is deployed within each control subsystem to classify and estimate local sensor faults, whereas in the centralized implementation, a multi-output neural network model is employed at the supervisory level for plant-wide fault estimation. The neural networks are trained on data obtained from process simulations under nominal (fault-free) conditions and varying fault magnitudes. Fault accommodation is implemented based on the stability region characterization, either by adjusting controller parameters or modifying the communication rate in the quasi-decentralized control system. The effectiveness of the proposed approach is demonstrated through a simulated process network. A comparative assessment of the advantages and limitations of the centralized and decentralized strategies in the context of the case study is presented.
References:
[1] H. Shahnazari and P. Mhaskar, “Actuator and sensor fault detection and isolation for nonlinear systems subject to uncertainty,” International Journal of Robust and Nonlinear Control 28 (2018) 1996–2013.
[2] H. Shahnazari, “Fault diagnosis of nonlinear systems using recurrent neural networks,” Chemical Engineering Research and Design 153 (2020) 233–245.
[3] S. Venkateswaran, M. Sheriff, B. Wilhite and C. Kravaris, “Design of functional observers for fault detection and isolation in nonlinear systems in the presence of noises,” Journal of Process Control 108 (2021) 68–85.
[4] Z. Gao, X. Liu and M. Chen, “Unknown input observer-based robust fault estimation for systems corrupted by partially decoupled disturbances,” IEEE Transactions on Industrial Electronics 63 (2015) 2537–2547.
[5] J. Allen and N. H. El-Farra, “A model-based framework for fault estimation and accommodation applied to distributed energy resources,” Renewable Energy 100 (2017) 35–43.
[6] S. Venkateswaran and C. Kravaris, “Design of linear unknown input observers for sensor fault estimation in nonlinear systems,” Automatica 155 (2023) 111152.
[7] A. Gajjar and N. H. El-Farra, “Machine learning-based estimation and accommodation of multiple sensor faults in sampled-data process systems,” Proceedings of the American Control Conference, Toronto, Canada, pp. 3043–3048, 2024.
[8] Y. J. Park, S. K. S. Fan and C. Y. Hsu, “A review on fault detection and process diagnostics in industrial processes,” Processes 8 (2020) 1123.
[9] D. Peng, N. H. El-Farra, Z. Geng and Q. Zhu, “Distributed data-based fault identification and accommodation in networked process systems,” Chemical Engineering Science 136 (2015) 88–105.
[10] K. Zhang, B. Jiang, M. Chen and X.-G. Yan, “Distributed fault estimation and fault-tolerant control of interconnected systems,” IEEE Transactions on Cybernetics 51 (2021) 1230–1240.