Diagnostic tools for multivariable model-based control systems

Abstract

In this paper we propose a set of diagnostic tools that can be used for detecting and estimating the causes for performance degradation in multivariable model-based control systems. Two simple diagnostic tests are proposed, on the basis of prediction error analysis, to detect the performance degradation and bifurcate the potential causes for the degradation. These tests are useful for narrowing down the list of potential causes for performance degradation. For detailed fault estimation, we propose to use a stochastic model based on Gaussian-sum statistics. The optimal estimation for this model requires a bank of a growing number of filters and is computationally intractable. To propose a practical solution, we combine the different suboptimal approaches in the literature. The results of the diagnosis can be used for alerting the operators about the potential faults and for adapting the controllers.