Research On Bearing Intelligent Fault Diagnosis Method Based On Compression Acquisition

As the main component of the rotating mechanical structure,rolling bearings are widely used in the transmission system of the equipment.In order to ensure the reliable operation of the equipment,accurate and timely diagnosis for bearing faults have important research significance.At present,the use of signal processing technology for intelligent fault diagnosis of time-domain vibration signals is a research hotspot for bearing condition monitoring.Based on the research results of compressed sensing(CS)and deep learning,two intelligent bearing fault diagnosis methods combining compression acquisition with stacked sparse auto-encoding and sparse filtering respectively are studied in this paper.Firstly,for the traditional bearing fault condition monitoring system facing the problem of compressing storage of massive data,signal processing calculation cost,and increasing communication channel bandwidth,the method of bearing vibration signal compression acquisition based on the CS theory is studied.The amount of collected data is reduced from the source,and the change from data acquisition to information acquisition is realized.By analyzing the effectively failure information acquisition of the compression observations,it is concluded that the compression acquisition can obtain and retain the fault information of the original signal,which provides theoretical support for the subsequent to directly processing the bearing vibration compression acquisition data.Secondly,for traditional bearing fault diagnosis methods,there are problems such as low efficiency of feature extraction,strong subjectivity,and excessive reliance on prior knowledge and diagnostic experience.This paper proposes a bearing fault diagnosis method with a deep neural network(DNN)which constructed based on stacked sparse auto-encoding without reconstructing the original signal.Using the powerful feature mining capabilities of the DNN model,the self-adaptive feature extraction and fault diagnosis are directly performed on compressive observations with high information content.The experimental results show that compared with the traditional bearing fault diagnosis method,the proposed method can achieve accurate and effective health status identification for different bearing fault types.Finally,for the DNN model with the problems of too many adjustable parameters and high training time complexity,a bearing fault diagnosis method combining sparse filtering and feature dimensionality reduction is proposed.This method firstly uses the sparse filtering model to achieve unsupervised feature extraction from compression acquisition data.Then it adopts Nearest Neighbor Preserving Embedding algorithm to complete feature dimensionality reduction.Finally,it uses classifier to realize fault type identification and diagnosis.Experimental results show that the combination of sparse filtering and feature dimensionality reduction can effectively improve the final recognition accuracy,and compared with the method of combining DNN,can significantly reduce the model training time and period.