EWMA Kernel Generalized Likelihood Ratio Test for Fault Detection of Chemical Processes

Fault Detection (FD) is a fundamental step in process monitoring. Owning to its simplicity and effectiveness to deal with nonlinear and highly correlated process variables, kernel principal component analysis (KPCA) has been successfully used in process monitoring. However, the major drawback of this method-based kernel generalized likelihood ratio test (KGLRT) is the neglect of small faults. Inspired by the effectiveness of this detection metric and motivated by the advantages of the univariate exponentially weighted movng average (EWMA), we propose, in this paper, a KPCA-based EWMA-KGLRT FD algorithm. Hence, its performance is illustrated and compared to the conventional KPCA-based KGLRT method through continuously simulated tank reactor (CSTR). In fact, the experimental results confirmed the performance of the proposed algorithm in terms of missed detection (MD) and false alarm (FA) rates.