Fault Diagnosis On Industrial Equipment Based On Data Correlations

Fault diagnosis on industrial equipment is one of the most important technologies in intelligent manufacturing.It locates the fault,and gets its information,such as happening time,duration,cause,etc.Massive diverse monitoring data from multiple sources,in which few faults are labelled,is generated in modern industrial equipment.Because of the correlation of detection data among parts,inside parts,and time,existing fault diagnosis method based on a single characteristic signal aims at specific parts or faults,and the results have low certainty and precision.Most of the fault diagnosis methods involving multi-source information perform information fusion after extracting fault features.This approach relies on a researcher’s understanding of specific types of fault and is insensitivity to sudden events and infrequent failures.To solve these problems,this thesis proposes a fault diagnosis scheme on industrial equipment based on data correlations.The specific research contents are as follows:(1)A fault diagnosis framework on industrial equipment based on data correlation is proposed.First,the multi-source information is fused based on data similarity before feature extraction.And the original sampling data is segmented using multidimensional time series segmentation method for state detection;Then,the features of data segments are extracted to describe their corresponding equipment running states;Finally,combined with the temporal correlation among different equipment running states,the outlier detection based on cluster analysis is used for fault diagnosis.(2)A time series segmentation scheme based on multi-source information fusion is proposed.Because the existing multi-dimensional sequence segmentation method has the following shortcomings: 1).The number of segments needs to be manually input;2).The structured information of the equipment cannot be used;3).The correlation between the monitoring signals inside the component is ignored;4).The optimization target and industrial monitoring Data does not match.This thesis propose a time series segmentation scheme based on multi-source information fusion.First,the segmented combination information entropy is optimized after minimizing the shape error,and the time series segmentation algorithm based on shape and information entropy(MIEn)is used to initially segment the multi-source PCA.Then the number of initial segments is used as the input parameter of the memetic segmentation algorithm based on time series information entropy(MAIEn).This method consider the information of other components which can improve the performance on industrial data.(3)A fault diagnosis scheme based on time-correlation mixture model is proposed.Outlier detection technology can be used for fault detection of industrial equipment.At present,the temporal correlation of data in clustering is ignored in most of the clusterbased outlier detection methods.Therefore,a fault diagnosis scheme based on timecorrelation mixture model is proposed in this thesis.In this method,the correlation of data in time is modeled using a hidden Markov model and the distribution in the feature space is modeled using a Gaussian mixture model.These two models are used to train a set of parameters,and clustering is carried out by the time period after the parameter converge.After clustering,the fault degree index is constructed by the outlier factor to complete the fault diagnosis.Finally,the proposed method is experimentally evaluated using two real industrial equipment monitoring datasets.Compared with the existing methods,the results show that the proposed scheme has great advantages in fault diagnosis accuracy.