Detect and Diagnose Unexpected Changes in the Output Probability Density Functions for Dynamic Stochastic Systems: an Identification Based Approach

摘要

In parallel to the recently developed observer based fault detection algorithm (Wang, 1998), this paper presents an identification based fault detection and diagnosis approach for the output probability density function for unknown time-invariantstochastic systems. The fault is assumed to be the unexpected changes of the system physical parameters which affect the shape of the output probability density functions of the system. In this context, the measured control input directly affects thedistribution of the system output in some probability sense. An on-line parameter estimation algorithm is constructed using the measured output probability density functions of the system. Based upon the same type of dynamic model proposed by Wang (1997,1998), a functional weighted integration of the output probability density function is employed in the evaluation of the residual signals for both fault detection and diagnosis purposes. An applicability study of the proposed algorithm to the detection of the unexpected changes of the solid flocculation in paper making is included, where a simulated example is employed to illustrate the use of the developed algorithm and encourage results have been obtained.