(663d) Fault Detection and Accommodation in Sampled Data Process Systems with Measurements and Actuation Errors

Development of systematic methods for the detection and handling of faults in chemical processes is a fundamental problem at the interface of process control and operations. The realization that malfunctions in the measurement sensors, control actuators or process equipment can cause unrecoverable losses, and may result in unacceptable performance deterioration or even instability, has led to an extensive body of research work on the development of various process monitoring and fault-tolerant control methods over the past few decades (e.g., see [1]–[4] for some results and references in this area).

While the problem is typically formulated and addressed within the conventional feedback control setting, which assumes a direct and flawless communication link between the sensors and the controller, the increasing complexity of the process/controller interface (e.g., owing to the use of network-based communication) motivates re-examining this formulation to suit the needs of the real-time operation practices. In particular, the inherent limitations on the information transmission and processing capabilities of the measurement system and the sensor-controller communication medium can erode the diagnostic and fault-tolerance capabilities of the control system if not properly accounted for. Issues such as resource constraints, data sampling and losses, processing and communication delays, measurement quantization, and real-time scheduling constraints, challenge many of the assumptions in traditional process monitoring and control methods and need to be integrated explicitly in the fault-tolerant control system design methodology.

An effort to address some of these challenges was initiated in [5] where a model-based framework for actuator fault detection and reconfiguration using discretely-sampled and delayed measurements was developed. The main idea was to develop an integrated framework for actuator handling which consists of a family of robust output feedback controllers, observer-based fault detection filters that account for the discrete sampling and delayed availability of measurements, and a switching law that orchestrates the transition from the faulty actuator configuration to a healthy fallback following fault detection.

While control system reconfiguration may be necessary for the recovery from severe failure situations, it can also be a costly strategy, especially as the realization of reconfigurable control requires the availability and use of redundant configurations, which can raise process operating costs. In cases involving only partial faults, fault accommodation by means of updating the process model and/or control design parameters based on an estimation of the fault magnitude, represents a more efficient and less costly alternative to maintaining process operation at the desired level, even under the faulty conditions. This benefit was demonstrated in [6] where a stability-based fault accommodation scheme was devised and implemented. These results were further generalized in [7] where an optimization-based approach for fault estimation was incorporated into the design methodology to facilitate the implementation of the fault accommodation strategy.

A key consideration in the previous studies is the assumption that the measurements used for monitoring, reconfiguration and/or accommodation are exact, and the assumption that the control action computed by the controller is also transmitted accurately to the control actuators. In many practical situations, however, these assumptions need to be re-examined in light of the possible degradation in the quality and precision of the transmitted sensor and actuator data over the network. Possible sources for the presence of measurement and actuation errors include interference with other signals in the field in the case of wireless networks, errors due to noisy sensor readings which are subject to drift over time, the use of discrete sensors (e.g., binary sensors) and discrete actuators (on/off valves) in the control system which cannot be varied continuously, the use of data fusion and aggregation techniques, and the need to conserve network bandwidth by reducing data precision. The end result in all these situations is that only inexact values of the state and/or input variables will be available, and the resulting errors may cause a significant deterioration in the closed-loop control and monitoring qualities, if not handled properly in the control system design.

Motivated by these considerations, we present in this contribution a methodology for the design of fault-tolerant control systems for process systems with actuator faults, discretely sampled and delayed measurements, and bounded sensor and actuator errors. A key objective of this study is to elucidate the key design parameters that can be varied to mitigate the impact of degradation in the quality of transmitted sensor and actuator data on the monitoring and fault-tolerance capability of the closed-loop system. Initially, a model-based controller that stabilizes the fault-free closed-loop system in the absence of sensor and actuator errors is synthesized. The controller includes an inter-sample model predictor to compensate for the discrete measurement sampling and a propagation unit to compensate for the delayed availability of measurements. Lyapunov techniques are them utilized to analyze the implementation of the model-based controller in the presence of bounded measurement and actuation errors. A precise characterization of the closed-loop stability region is obtained in terms of the sensor/actuator error bounds, the sensor data transmission rate, the delay size, the magnitude of fault, the size of the process-model mismatch and the choice of control configuration. The characteristic fault-free closed-loop behavior is used as the basis for deriving a time-varying residual alarm threshold for fault detection, while the stability region characterization is used to decide a suitable fault accommodation strategy. Finally, the design and implementation of the integrated monitoring and fault-tolerant control architecture are demonstrated using a chemical process example.

References:

[1] M. Blanke, M. Kinnaert, J. Lunze, and M. Staroswiecki, Diagnosis and Fault-Tolerant Control. Berlin, Germany: Springer, 2003.

[2] R. Isermann, Fault-Diagnosis Systems: An Introduction from Fault Detection to Fault Tolerance. Berlin, Germany: Springer, 2005.

[3] Y. Zhang and J. Jiang, “Bibliographical Review on Reconfigurable Fault-Tolerant Control Systems,” Annu. Rev. in Contr., vol. 32, pp. 229–252, 2008.

[4] P. Mhaskar, J. Liu, and P. D. Christofides, Fault-Tolerant Process Control: Methods and Applications. London, England: Springer-Verlag, 2013.

[5] Y. Sun and N. H. El-Farra, “Model-Based Fault Detection and Fault-Tolerant Control of Process Systems with Sampled and Delayed Measurements," Proceedings of 18th IFAC World Congress, pp. 2749-2754, Milan, Italy, 2011.

[6] T.  Napasindayao and N. H. El-Farra, “Fault Detection and Accommodation in Particulate Processes with Sampled and Delayed Measurements,” Ind. Eng. Chem. Res., 52, 12490–12499, 2013.

[7] T.  Napasindayao and N. H. El-Farra, “Model-based Fault-Tolerant Control of Uncertain Particulate Processes: Integrating Fault Detection, Estimation and Accommodation,” Proceedings of 9th IFAC Symposium on Advanced Control of Chemical Processes, to appear, Whistler, British Columbia, Canada, 2015.