Nonlinear chemical processes fault detection based on adaptive kernel principal component analysis

When kernel Principal Component Analysis (KPCA) is applied to fault detection, kernel Principal Components (KPCs) are divided into two spaces according to the size of variance for fault detection, respectively. However, it is easy to cause the mutation feature to be scattered, thereby resulting in a high missed alarm rate. For this problem, an Adaptive KPCA (AKPCA) method based on online samples is proposed for the fault detection of nonlinear chemical process. AKPCA selects the KPC with the highest mutation probability as the Dominant Variation KPC (DV-KPC) and then selects the KPCs which have strong similarity with DV-KPC as the Non-Dominant Variation KPCs (NDV-KPCs). Finally, the DV-KPC and the NDV-KPCs form the Adaptive KPCs (AKPCs) which are used to construct the statistics for detection. Tennessee Eastman (TE) process is used to verify the feasibility and effectiveness of the AKPCA method in the fault detection of nonlinear chemical processes.