Reduced kernel PCA Technique for fault

Abstract

Fault detection and diagnosis is an important problem in process engineering. It is the central component of abnormal event management (AEM) which has attracted a lot of attention recently. This thesis discuses different classes of FDD approaches for process monitoring. In addition, it presents main results of fault detection and diagnosis in a cement manufacturing plant using three monitoring techniques. The techniques are based on multivariate statistical analysis and a threshold strategy. The process is statistically modeled using Principle Component Analysis (PCA), kernel PCA and new proposed reduced KPCA to cope with the computational problem introduced by KPCA. The proposed RKPCA method consists on reducing the number of observations in a data matrix using a proposed algorithm based on fractal dimension. The Hotelling’s T², Q in addition to the new proposed index called the combined statistic f are used as fault indicators for testing PCA, KPCA and the suggested approach RKPCA carried out using the cement rotary kiln system. The three methods are compared to in terms of False Alarms Rate (FAR), Missed Alarms Rate (MDR), Detection Time Delay (DTD) and the cost function (J). The obtained results demonstrate the effectiveness of the proposed technique in reducing the number of observations from 768 to 11, leading to an 11x11 kernel matrix instead of 768x768, hence, diminishing computational time and storage requirement. Moreover, it has effectively detected the different types of faults when using statistical indices.