(487d) Handling Multiple Simultaneous Faults in Networked Process Control Systems Using a Data-Driven Methodology

Recent years have witnessed significant growth in communication and networking capabilities, accompanied with advancements in sensor and actuator manufacturing technologies. As a result, modern process control systems are designed with an extensive network of interconnected sensors and actuators over a shared communication network. The growing dependence of process operations on networked sensors and actuators increases the likelihood of faults occurring in these components, which have the potential to cause significant degradation in the achievable closed-loop control stability and performance, leading to economic losses and safety hazards. In this light, the subject of fault-tolerant control has received considerable attention over the past few decades, with a focus on the problems of fault detection, diagnosis, estimation, and accommodation in process control systems (e.g., see [1, 2, 3]).

In the implementation of fault-tolerant control approaches, the problems of fault detection and fault accommodation are often treated separately in the literature. Fault detection requires distinguishing between normal process disturbances and faults, using combinations of past data and first-principles based models. In methods that primarily utilize process data, fault information is typically extracted by comparing the process state trajectory with historical process data [4]. Fault accommodation, on the other hand, has been studied in prior works in the context of traditional feedback control systems where it is assumed that the output measurements or states are continuously available to the controller. However, networked control systems wherein sensor-controller communication is discrete in nature pose challenges. The stability and performance characteristics of these systems exhibit a critical dependence on the sensor-controller communication rate, making fault accommodation more challenging.

Several methods have been proposed to address the problem of sensor fault accommodation. These include passive fault-tolerant control methods [5], sensor reconfiguration-based approaches [6], performance-based accommodation of multi-rate sensor faults [7] and performance-based accommodation of sensor faults under sampling delays [8]. These approaches aim to mitigate the impact of sensor faults on the stability and performance of the closed-loop system. Nonetheless, the success of these approaches is dependent on the effectiveness of the schemes developed for detecting and identifying sensor faults. Effective fault compensation strategies often require knowledge of the existence and magnitudes of the faults.

Despite recent progress on sensor fault-tolerant control, the problem of handling multiple simultaneous sensor faults still remains a challenging task. Developing robust algorithms for the detection and estimation of such faults is crucial, especially when compared to the case of systems subject to a single fault. While the problems of detection [9] and accommodation [10] in systems subject to multiple simultaneous faults have received some attention lately, the problem of estimation of multiple simultaneous sensor faults has not been rigorously addressed. Furthermore, a unified framework for the detection, estimation and accommodation of such faults remains lacking at this point.

This work presents an integrated framework for the active detection, estimation and accommodation for multiple simultaneous sensor faults in networked process control systems. The framework brings together tools for data-based characterization of closed-loop stability, neural-network based fault estimation and model-based controller reconfiguration for accommodation. The region of guaranteed sensor fault-tolerance is initially characterized based on the magnitudes of the faults, the frequency of sensor-controller communication, and the parameters of the controller design. This characterization highlights the critical factors that can be employed to effectively mitigate the impact of these faults once they are detected and identified. A multi-output neural network (MONN) model is trained off-line using state measurements obtained under fault-free conditions and under various fault modes. This classification is used for detection as well as estimation of the magnitudes of the sensor faults. Once the MONN model is trained, a framework is proposed for online implementation of the integrated detection, estimation and accommodation strategy. The effectiveness of the proposed approach is demonstrated through an application to a chemical process example.

References

[1] Mhaskar P, Liu J, Christofides PD. Fault-Tolerant Process Control: Methods and Applications. Springer Science & Business Media. 2012.

[2] Blanke M, Kinnaert M, Lunze J, Staroswiecki M. Diagnosis and Fault-Tolerant Control. Springer, 2016.

[3] Amin A, Hasan K. A review of fault-tolerant control systems: Advancements and applications. Measurement. 2019;143:58–68.

[4] Venkatasubramanian V, Rengaswamy R, Kavuri SN, Yin K. A review of process fault detection and diagnosis: Part III: Process history based methods. Computers & Chemical Engineering. 2003;27(3):327–346.

[5] Gani A, Mhaskar P, Christofides PD. Handling sensor malfunctions in control of particulate processes. Chemical Engineering Science. 2008;63(5):1217–1229.

[6] Yao Z, El-Farra NH. Performance-based sensor reconfiguration for fault-tolerant control of uncertain spatially distributed processes. IFAC Proceedings Volumes. 2014; 47(3):5193–5198.

[7] Chen S, Allen JT, El-Farra NH. Model-based strategies for sensor fault accommodation in uncertain dynamic processes with multi-rate sampled measurements. Chemical Engineering Research and Design. 2019; 142: 204–213.

[8] Allen J, El-Farra NH. Performance-based fault mitigation in sampled-data process systems with sensor faults and measurement delays. IFAC-PapersOnLine. 2020; 53(2):13757–13762. 21st IFAC World Congress.

[9] Voutsinas S, Karolidis D, Voyiatzis I, Samarakou M. Development of a multi-output feed-forward neural network for fault detection in Photovoltaic Systems. Energy Reports. 2022;8:33–42. Technologies and Materials for Renewable Energy, Environment and Sustainability.

[10] Zhao D, Polycarpou MM. Distributed fault accommodation of multiple sensor faults for a class of nonlinear interconnected systems. IEEE Transactions on Automatic Control. 2021;67(4):2092–2099.