Fault Detection of Chemical Process Based on Functional Kernel Entropy Component Analysis

In order to reduce the influence of data nonlinearity and noise on industrial process fault detection, a functional kernel entropy component analysis (FKECA) is proposed for industrial process fault detection. Firstly, using the function data analysis strategy, by introducing the rough penalty term in the curve fitting process, the industrial process data is transformed into functional data, which can effectively eliminate the abnormal value and noise interference, and make up for the missing data. Secondly, the functional data is mapped to the high-dimensional linear feature space through the radial basis function. Thirdly, the entropy cumulative contribution in descending order is proposed to determine the number of principal components, and the monitoring statistics and control limit are calculated. Finally, the fault detection method of nonlinear process in noise environment is verified. The proposed method is applied to Tennessee Eastman process. The results show that compared with principal component analysis (PCA) and functional principal component analysis (FPCA), the proposed method not only can effectively filter out noise and outliers, but also can improve fault detection rate in a certain extend.