Development of performance assessment and fault identification strategy based on kernel GDA

Statistical performance monitoring aims at improving process operation by distinguishing abnormal process conditions from common cause variations. However, for some complicated cases in industrial process, because of the nonlinear correlations between process variables, conventional linear statistical methods often have poor ability for monitoring these processes. In this paper, a novel nonlinear on-line performance monitoring and fault identification method based on kernel discriminant analysis (kernel GDA) was proposed. The basic idea of this method is to first map data in the original space into high-dimensional feature space via nonlinear kernel function and then extracts the optimal Fisher feature vector and discriminant vector to perform process monitoring. If faults occurred, it uses the similar degree between the present discriminant vector and the optimal vector of fault in historical dataset to diagnosis. The proposed method can effectively capture nonlinear relationship in process variables. It was evaluated by the application to the CSTR process and its effectiveness was demonstrated.