(583b) Sensor Fault Estimation and Accommodation in Networked Distributed Processes

The problems of fault identification and fault accommodation have been the subject of considerable interest in the control research community in response to the ever increasing requirements on the reliable and safe operation of automatic control systems. At present, however, only relatively few studies have been devoted to the development of systematic methods for the diagnosis and handling of faults in spatially distributed processes. This is an important problem given that many industrial processes are characterized by spatial variations and are modeled by Partial Differential Equations (PDEs)  (e.g., [1], [2]). The focus of existing results in this direction has been on the detection and compensation of either actuator (e.g., [3]–[6]) or component faults (e.g., see [7]). By contrast, the problem of sensor fault-tolerant control has not attracted as much attention despite the common occurrence of sensor faults in practice and the detrimental effect they can have on the monitoring capability and overall control quality of the closed-loop system.

An effort to tackle this problem was made in [8] where a sensor fault detection and reconfiguration strategy was developed to reduce the performance deterioration associated with sensor faults in highly dissipative distributed processes. Fault detection was achieved by means of a residual generation approach in which the residual was compared against a suitably designed fault-free alarm threshold. In this study, however, the faults considered were assumed not to jeopardize closed-loop stability. The detection approach was stability-based and thus could leave partial faults that do not compromise closed-loop stability go undetected. Furthermore, the problems of identifying the location and the magnitude of the sensor faults were not addressed. It was assumed instead that a suitable mechanism for fault identification was already built into the system and able to determine the size and location of the fault. Finally, and similar to other existing methods for fault-tolerant control of distributed parameter systems, the fault-tolerant control structure was designed within a conventional feedback control setting with continuous and flawless sensor-controller communication.With the increased integration of communication networks in control systems, the impact of sensor-controller communication constraints (e.g., resource limitations, communication delays) on closed-loop stability can no longer be ignored and needs to be explicitly accounted for (e.g., see [9]).

Motivated by these considerations, we present in this work an integrated approach for data-based fault identification and accommodation of sensor faults in a class of spatially distributed processes with low-order dynamics, and sensor-controller communication constraints. The main idea is to synthesize a model-based networked output feedback controller based on an appropriate reduced-order system obtained via model order reduction of the highly-dissipative PDE. By analyzing the fault-free behavior of the networked closed-loop state, the closed-loop stability region is characterized in terms of the sensor-controller communication rate, the communication delay size, the model uncertainty, the control actuator placement and the controller and estimator design parameters. The fault identification mechanism is then implemented by solving a moving-horizon optimization problem, which is embedded in the sensors, to estimate the size of the faults using the historical sampled output and input data. Once the faults are identified, estimates of the fault magnitudes are sent to the controller at the next sensor transmission time. These estimates, together with the characterization of the networked closed-loop stability region, are used to determine the suitable measures for fault accommodation, which include updating the control model parameters, adjusting the control actuator placement, and/or switching to a new stabilizing feedback gain. The proposed methodology is illustrated using a simulated diffusionreaction process subject to both abrupt and gradual sensor faults.

References:

[1] P. D. Christofides, Nonlinear and Robust Control of PDE Systems: Methods and Applications to Transport-Reaction Processes. Boston: Birkhauser, 2001.

[2] A. Smyshlyaev and M. Krstic, Adaptive Control of Parabolic PDEs. Princeton University Press, 2010.

[3] M. Demetriou, “A model-based fault detection and diagnosis scheme for distributed parameter systems: A learning systems approach,” ESAIM-Control Optimisation and Calculus of Variations, vol. 7, pp. 43–67, 2002.

[4] S. Ghantasala and N. H. El-Farra, “Robust actuator fault isolation and management in constrained uncertain parabolic PDE systems,” Automatica, vol. 45, pp. 2368–2373, 2009.

[5] Z. Yao and N. H. El-Farra, “Robust fault detection and reconfigurable control of uncertain sampled-data distributed processes,” in Proceedings of 50th IEEE Conference on Decision and Control, Orlando, FL, 2011, pp. 4925–4930.

[6] A. Baniamerian and K. Khorasani, “Fault detection and isolation of dissipative parabolic PDEs: Finite-dimensional geometric approach,” in Proceedings of American Control Conference, Montreal, Canada, 2012, pp. 5894–5899.

[7] A. Armaou and M. Demetriou, “Robust detection and accommodation of incipient component faults in nonlinear distributed processes,” AIChE J., vol. 54, pp. 2651–2662, 2008.

[8] Z. Yao and N. H. El-Farra, “Performance-based sensor reconfiguration for fault-tolerant control of uncertain spatially distributed processes,” in Proceedings of 19th IFAC World Congress, Cape Town, South Africa, 2014, pp. 5193–5198.

[9] Y. Sun, S. Ghantasala, and N. H. El-Farra, “Networked control of spatially distributed processes with sensor-controller communication constraints,” in Proceedings of American Control Conference, St. Louis, MO, 2009, pp. 2489–2494.