Fault detection algorithm of industrial process based on DW-ICA-SVM

Abstract

In order to effectively improve the fault detection and monitoring performance of support vector machine (SVM) algorithm,a new fault detection algorithm of industrial process based on DW-ICA-SVM was proposed.Firstly,the training data was normalized.The independent component analysis (ICA) was used to obtain the independent component matrix of the data and extract the hidden non-Gaussian information.Then the Durbin-Watson (DW) criterion was used to calculate the DW values of the independent components (ICs).The DW method was used to effectively extract important noise information and select the important ICs.The ICs containing important information were used as the input of the SVM model to obtain the discriminant classification function.The ICs of test data were input to the model for fault detection and monitoring.Finally,the method was applied to the nonlinear numerical example and the Tennessee-Eastman industrial process,and compared with PCA,LPP,ICA,SVM and ICA-SVM methods.The results show that the proposed method reduces the autocorrelation among samples and effectively improves the fault detection rate.The proposed method strengthens the extraction and recognition of hidden non-Gaussian information to a certain extent,and provides a reference for improving the performance of SVM algorithm in fault detection of industrial process.

Keywords

• other disciplines of automatic control technology; fault detection; durbin-watson criterion; independent component analysis; support vector machine