Research On Fault Detection Method Based On ICA-SVDD

With the increasing degree of industrial intelligence,it is necessary to guarantee the safety of production process and the quality of products.The process of industrial production is producing and storing a large amount of data.In recent years,the technology of process detection based on data has been greatly developed.Multivariate statistical analysis method has been widely used in the field of industrial chemical process fault detection.The traditional principal component analysis(PCA)is a process detection method based on global structure feature,which can capture the whole structure of the process data,and cannot effectively mine the local information.The PCA method needs to assume that the process data obey the Gauss distribution and linear characteristics in the process of industrial process detection.However,in the process of chemical production,there are always some unfavorable factors such as non-Gauss information and noise interference and The PCA method only eliminates the correlation of variables and does not analyze the independence.The independent component analysis(ICA)method can make up for the deficiency of the PCA method,and obtain the independent data information by the independent component of the process data,which is consistent with the Gauss distribution,The ICA method has a better processing ability to process data with non-Gauss property.However,the ICA method is not as good as the support vector data description.Support vector data description(SVDD)method,in which the original data is projected into the feature space,is constructed in the feature space.Fault detection based on SVDD model.This topic is combined with ICA method and SVDD method to achieve fault detection,as follows:In the industrial process data has the characteristics of complex distribution,and the actual industrial process data are nonlinear and non-Gauss problem,based on this,this paper combined the ICA algorithm and SVDD algorithm to complete the chemical process fault detection,put forward a kind of industrial process fault detection method based on ICASVDD.First,the data of industrial process data are extracted by ICA algorithm independent component,extracted the data not only is gaussian,and were independent of each other between the data,extract the independent components need to look for a separation matrix W,through the linear change,independent principal component can be separated from the mixed signal.Then,the extracted data using SVDD algorithm for reconstruction of data,and establish a model based on SVDD statistics for industrial process monitoring and statistics.Finally,the completion of the Tennessee-Eastman(TE)numerical simulation experiment,the experimental results verify the validity of the proposed fault detection model.