Data-Based Fault Detection And Isolation For Nonlinear Process Systems Using Feedback Control.

Handling abnormal situations, like control actuator, measurement sensor and control system faults, is a subject of great importance in the chemical and process industries since abnormal situations account annually for at least $20 billion in lost revenue in the US alone. Modern chemical plants that rely on highly automated processes to maintain robust operation and efficient production are particularly vulnerable to these faults. Loss of control in a chemical process can lead to the waste of raw materials and energy resources, as well as downtime and production losses but more importantly it may lead to personnel injury or death. However, the existing paradigm to plant operations deals with the key tasks of designing control systems and monitoring schemes for detecting faults separately, thereby liming its applicability since it does not take advantage of an efficient integration of these tasks. This dissertation will present a paradigm shift to the existing approach of designing control systems and monitoring schemes in that it proposes to design control systems that are stabilizing, robust and optimal yet, they also lead to closed-loop system structures that facilitate fault isolation. To present our new method of controller-enhanced isolation, we will focus on a broad class of nonlinear process systems subject to control actuator faults and disturbances. The method allows isolating faults in the closed-loop system by designing nonlinear model-based control laws that decouple the dependency between certain process state variables in the closed-loop system. Fault detection is done using a purely data-based approach and fault isolation is achieved using the structure of the closed-loop system as induced by an appropriately designed controller. We will discuss extensions of the basic framework to deal with the issues of limited state measurements, controller optimality and networked implementation. We will present examples of large-scale process systems to demonstrate the effectiveness and benefits of the proposed method.