Dynamic process fault prediction using canonical variable trend analysis

Abstract

Fault prediction technology is important to avoid serious process failure. This paper is concerned with the fault prediction of dynamic industrial process with incipient faults and proposes a canonical variable trend analysis (CVTA) based fault prediction method. In the proposed method, canonical variate analysis (CVA) algorithm is firstly applied to analyze the process dynamics and extract the uncorrelated latent features, called canonical variables. Furthermore, support vector machine is adopted to model the relationship between the historical and future values of the canonical variables, which leads to the time series prediction model for the canonical variables. Based on the predicted canonical variables, an overall monitoring statistic is used to forecast the change of the process status. Simulations on a continuous stirred tank reactor (CSTR) system demonstrate that the proposed method can indicate the trend of the incipient faults effectively.