Research On Equipment Anomaly Recognition Based On Audio Fusion Feature

With the rapid development of industrial production automation.The normal and safe operation of mechanical equipment plays an important role in industrial production.With the complexity of industrial equipment development,it is difficult to manually confirm the occurrence of equipment failures,and when the fault develops to a certain extent,it will cause major economic losses,and even more,it will even lead to bad disasters.Therefore,it is extremely important to study the abnormal diagnosis of mechanical equipment.The sound recognition technology has broad application prospects in the abnormal recognition of mechanical equipment,and it is worthy of further research.Aiming at how to effectively identify the abnormal operation of equipment and improve the accuracy of abnormal signal recognition,this paper focuses on the feature extraction of sound signals based on the sound data of high-pressure pump.In this paper,the wavelet packet decomposition energy characteristics,MFCC features and their differential characteristics of the signal are extracted respectively,and analyze the recognition accuracy of the two methods in the abnormal sound of mechanical equipment.Secondly,based on the two,the concept of MFCC fusion feature is proposed,which is to extract the wavelet packet decomposition energy feature,MFCC feature and signal short-time zero-crossing rate characteristics of the signal,and perform linear fusion.The SVM classifier is used for classification training after the fusion feature.It is found that the fusion feature has better effect on the abnormal diagnosis of the device than the single feature,but the experimental running time has increased.In response to this problem,this paper uses the PCA method to optimize the fusion features,and performs dimensionality reduction on the fusion features in different dimensions.After dimension reduction,the best dimensional features were identified for the final experiment,which resulted in better recognition results and reduced experimental run time.In terms of classifier selection,the paper mainly chooses two kinds of classification models:support vector machine(SVM)and Gaussian mixture model(GMM),and uses the above features to carry out classification experiments.By comparing the recognition accuracy,it is concluded that the SVM classification effect has better recognition accuracy than the GMM method in the abnormal sound signal recognition of mechanical equipment.Finally,it can be concluded that the method of combining MFCC fusion characteristics after dimension reduction and SVM has better recognition effect in the aspect of abnormal sound recognition of mechanical equipment by experiments.