Two-level modeling strategy based incipient fault detection for nonstationary industrial processes

Abstract

Nonstationary problem widely exists in the real industrial processes, in which the mean values and variance of the nonstationary variables change with time. The interval of fault-free data is very wide for the nonstationary variables. Thus, the fault especially the incipient fault that has a small magnitude including early changing and the slow-developing, may be buried by nonstationary trends resulting in low fault detection rate. In this paper, a two-level fault detection modeling strategy is proposed to detect the incipient fault for nonstationary industrial processes. The nonstationary variables are first distinguished from the stationary variables. Then, a two-level fault detection model is constructed to detect the incipient fault. In the lower level, cointegration analysis (CA) is utilized to investigate the relationship between nonstationary variables, while principal component analysis (PCA) is performed to extract the process characteristic of the stationary variables. A total fault detection model is constructed in the upper level to describe the relation between nonstationary variables and stationary variables. The proposed method can effectively distinguish the incipient fault from the normal trend of the nonstationary variables. To illustrate the feasibility and effectiveness, the proposed algorithm is applied to a real industrial process of the thermal power plant.