Research On Fault Detection And Diagnosis Method Of Industrial Process Based On Multivariate Statistical Analysis

The automation,complexity,and intelligence of industrial systems have all improved as a result of the deep integration of informatization and industrialization.At the same time,industrial system fault detection and diagnosis research has advanced significantly.The fault detection and diagnosis method based on multivariate statistical analysis can achieve the process monitoring by relying on the system process data,so this method has gradually become the research direction of many scholars.This paper focuses on the process monitoring problem in an industrial environment with nonlinear,multi-model,and complex characteristics.The main research contents of this paper are summarized as follows:(1)Aiming at the problem of multi-model fault detection in chemical processes,a fault detection method based on local prediction principal component analysis(LP-PCA)is proposed.Firstly,local prediction is performed on the training samples,and PCA is used to project the local prediction samples and training samples into the prediction space(PS).Then,the difference information between the training samples and the prediction samples are calculated in PS.Next,the new statistics are established in two prediction subspaces to monitor the status of samples.Finally,a diagnostic analysis of the detected fault samples is performed.The LP-PCA method can eliminate the influence of outliers on PCA dimensionality reduction,the difference information can amplify weak faults,and detection results of new statistics are superior to traditional statistics.The LP-PCA method and the traditional PCA method are both tested in numerical simulation examples and chemical processes.The experimental results verified the effectiveness of the proposed method.(2)Aiming at the problem of reducing process fault detection rate because of residual autocorrelation in dynamic principal component analysis,a novel fault detection and diagnosis based on residual dissimilarity index in dynamic principal component analysis is proposed in this paper.Firstly,dynamic principal component analysis(DPCA)is used to calculate the residual score of a dynamic process.Next,moving window technology and dissimilarity index are utilized to monitor the status of this process in residual score.Finally,a fault diagnosis method based on contribution chart of monitored variables is used for discovering the reason caused abnormal change of this process.The proposed method to capture dynamic characteristics of a process through DPCA,meanwhile,the proposed dissimilarity index,which is different from the conventional squared prediction error(SPE),can effectively monitor the process in which the residual scores contain significant autocorrelation.The effectiveness of DPCA-Diss is tested in a numerical case and Tennessee Eastman(TE)process.(3)Aiming at the problem of fault detection for nonlinear processes,a fault detection method for nonlinear processes based on neighborhood preserve embedding is proposed in this paper.Firstly,it calculated the score matrix of the training data set using the NPE method,which was named real scores.Then,it calculated the K-nearest neighbor mean of each sample in the training data set and projected the mean into the low-dimensional space to obtain the estimated score of the sample.Finally,it established a new statistic to monitor the status of samples in difference subspaces(DS).Compared with traditional methods such as principal component analysis(PCA),K nearest neighbors(KNN)and NPE,the results indicate the effectiveness of the proposed method in a numerical case and TE processes.(4)Aiming at the problem of multi-model online fault detection in batch processes,a fault detection method based on multiway neighborhood preserving embedding and Gaussian mixture model is proposed.Firstly,the window data are obtained by the default window width.Next,the score of the current window data set are calculated by MNPE.After that,some Gaussian components of the score are determined by GMM.Finally,a quantification index is proposed to monitor process status.MNPE-GMM can not only maintain more local structure information of the current window data set in the feature subspace,but also reduce the computational complexity of GMM in fault detection processes.By introducing a new statistic,MNPE-GMM can effectively improve the fault detection rate of some multi-model batch processes.The effectiveness of the proposed method is verified in a numerical case and the semiconductor etching process.