Fault Detection, Isolation and Accommodation Using the Generalized Parity Vector Technique

In real industrial processes continuous production is required to achieve productivity and profitability requirements. As a result, stopping a production line suddenly in the middle of a process, to fix or replace a faulty sensor, may produce significant economic losses. Therefore, the current fault management strategy challenge is not only to detect and isolate faults, but also to accommodate them, to keep the safe operation in the plant while maintenance can be scheduled without significantly disturbing the process.;This research extends the generalized parity vector (GPV) approach originally proposed by Viswanadham, Taylor and Luce and continued by Omana and Taylor, to offer a complete sensor fault detection, isolation and accommodation (FDIA) technique viable for implementation in real industrial applications. Fault detection and isolation is also provided for actuators.;A new systematic approach to implement a recursive on-line transformation matrix computation block using optimization is developed. The calculation of this transformation matrix represents an important contribution to the FDI field using directional residuals because it eliminates the restriction on the number of faults that previous researches were able to isolate and significantly increases FDI robustness. The special case for sensor-actuator faults and the hyperplane intersection problem are identified and solved by extending the objective function during the optimization process to compute the transformation matrix. This modification significantly improves the isolation results by reducing the ambiguous cases produced by these inevitable special geometrical situations given by the system dynamics. This is a major contribution, because it identifies and overcomes these critical limitations of FDI using directional residuals that previous researchers were not aware of. The plant model availability issue is overcome by incorporating an on-line system identification module before executing the FDIA block. This shows that while the GPV approach is a model-based FDI technique, it is still viable for those plants where an a-priori mathematical model is not available.;A fault management strategy is implemented using a novel fault-size estimation, classification and accommodation method based on the static GPV magnitude signature. The proposed fault accommodation technique not only preserves closed-loop stability, but also compensates the actual variable affected by the faulty sensor. A initialization section is introduced to make this FDIA technique capable of handling model and operating point changes. The FDI robustness is significantly improved by incorporating an on-line threshold computation block and combining the strengths of the static and dynamic GPV implementations during the decision-making process.;In this work the FDIA technique is successfully analyzed and simulated on a gravity three-phase separation process used in oil production facilities. This model closely simulates a large scale process, which allows the GPV technique to be validated in a higher dimensional space with more complex system dynamics.