A Hybrid Model Based and Statistical Fault Diagnosis System for Industrial Process

Abstract

This thesis presents a hybrid model based and statistical fault diagnosis system, which applied on the nonlinear three-tank model. The purpose of fault diagnosis is generating and analyzing the residual to find out the fault occurrence. This fault diagnosis system includes residual generator and residual processor. The fault generator is applied with the Luenberger observer, which has its own algorithm to compute the parameters. The residual processor is applied with the Shiryayev sequential probability ratio test, which calculating the posteriori probabilities to detect and isolate a change in residual in the independent measurement.

The thesis starts with introduction and literature review, and then shows the methods of residual generation and residual processing. The chapter six presents the simulation results, which operated by MATLAB Simulink. The fault diagnosis system successfully captured different kind of fault in three-tank model. The results prove the effectiveness of this fault diagnosis system.