Fault detection of nonlinear dynamic industrial processes based on improved kernel entropy component analysis

Fault detection in nonlinear dynamic processes is essential to ensure the safety of industrial production. Interactions among process data exhibit strong nonlinear and dynamic characteristics. These interdependent features have limited the performance of traditional Kernel Entropy Component Analysis (KECA) methods in fault detection within nonlinear dynamic industrial processes. Therefore, this paper proposes a fault detection method for nonlinear dynamic processes based on Moving Window Kernel Entropy Component Analysis, abbreviated as MW-KECA. This method initially utilizes KECA mapping to obtain the score vectors and nonlinear feature subspaces of the process data set, achieving effective dimensionality reduction for multidimensional feature matrices. Subsequently, a training model is established, followed by the dynamic update of the model using a moving window. Finally, the MW-KECA method is simulated in the Tennessee Eastman (TE) process and compared with kernel principal component analysis (KPAC) method and KECA method. The experimental results validate the feasibility and effectiveness of the proposed method.