Performance monitoring and fault detection in control systems

Citation

Tyler, Matthew Lamont (1997) Performance monitoring and fault detection in control systems. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/xxrh-h734. https://resolver.caltech.edu/CaltechETD:etd-01182008-083425

Abstract

As the sophistication of systems used in chemical processing industries increases and demands for high quality products manufactured at low costs mount, the need for improved methods for automatic monitoring of processes arises. This is particularly true for systems operating under automatic control where the control system often acts to eliminate early warning signs of process changes. This thesis examines problems in the area of dynamic system monitoring, with emphasis on control systems. Problems in the areas of controller performance monitoring, estimation, and fault detection are considered.

In the area of controller performance monitoring, techniques for assessing performance in a minimum variance framework are developed. In contrast to previous methods, the current approach is applicable to general systems, including unstable and nonminimum-phase plants and systems with unstable controllers. Through two simple examples, it is shown that significant errors may be encountered when information on the unstable poles and non-invertible zeros of a system is not properly included in the performance evaluation techniques. An alternative approach to evaluating deterioration in performance of control systems is formulated using a framework in which acceptable performance is expressed as constraints on the closed loop transfer function impulse response coefficients. Likelihood methods are used to determine if the constraints are met. This second approach can be applied to more general performance criteria than the minimum variance based method.

The problem of constrained state estimation is pursued using Moving Horizon Estimation. It is shown that previous formulations of this estimation technique can be unstable when constraints on the innovations and estimated states are included. By expanding the constraint set and modifying the estimation objective, stability is guaranteed. The proposed algorithm can be implemented as a quadratic program.

Several approaches to fault detection are considered. First, the simultaneous design of linear fault detection filters and controllers is considered using the four parameter controller framework. It is shown how this framework may be considered as a special case of a more general interconnection framework for which a deep synthesis theory exists. Second, using the Moving Horizon Estimation framework, a model based fault detection scheme capable of directly incorporating a class of bounded model uncertainty is developed. The proposed method is compared to other methods employing an adaptive threshold, and is demonstrated on a simulation example of a cold tandem steel mill. Finally, a statistical framework for general change detection problems is presented. This method uses a two-model approach, where signals and parameters subject to change are modeled by Brownian motion for the faulty case and by constant values in the nominal case. A detection algorithm using likelihood ratio testing is implemented through the use of recursive dynamic filtering.

The use of qualitative modeling in detection and control problems is formulated using propositional logic. By representing literals using integer variables, qualitative features can be incorporated into control and detection problems. Symptom aided detection and multiobjective performance prioritization are among the problems which can be solved in this framework using mixed integer linear and quadratic programming.

Thesis Files