A Fault Detection Method Considering Fault Mode Information

Mainstream data-driven fault detection methods either use only normal samples for offline modeling, or use both normal samples and fault samples for offline modeling. In the actual industrial process, it is often possible to obtain some information about fault mode based on prior knowledge or historical data, which is useful for fault detection. Based on this consideration, this paper presents a fault detection method of considering fault mode information (FMI). Firstly, the mathematical description of the fault model considering FMI is given; Secondly, by analyzing the fault signal ratio (FSR), a fault decoupling subspace (FDS) projection method based on FSR maximization is proposed; Thirdly, the traditional multivariate statistics method is used to construct statistics and to carry out fault detection. Finally, the effectiveness of the proposed method is verified by a classical numerical simulation.