Research On Rolling Bearing Fault Diagnosis Method Based On Deep Learning And Multi-source Information Fusion

Industrial transformation measures based on data and computer technology continue to develop,and mechanical equipment will continue to be updated.As a key component of rolling bearings,their performance increasingly affects the operating status of the entire equipment,and intelligent fault diagnosis is of important research significance.Faced with the status quo of large-scale fault data,complex structure,redundancy,and numerous error messages,the development of traditional fault diagnosis technology is limited.The data collection of a complex system is easily affected by uncertain factors such as the actual industrial environment temperature and noise.A single sensor may cause the collected data to be incomplete and biased.How to realize intelligent fault diagnosis of rolling bearings when the equipment is becoming more and more complex,the scale of data is becoming larger and the traditional diagnosis technology is becoming weaker;how to achieve effective and reliable fault diagnosis of rolling bearings under the condition of uncertainty and ambiguity in the fault representation of a single data source is a hot issue currently studied.Based on the above background,this article takes rolling bearings as the research object,uses deep learning algorithms and multi-source information fusion technology to deeply study rolling bearing fault diagnosis methods,and realizes efficient and intelligent fault diagnosis of rolling bearings.The main research contents are as follows:Firstly,a fault diagnosis method of intelligent rolling bearing based on 1D-CNN+LSTM is studied,and the original vibration signal collected by the sensor is directly analyzed and diagnosed.According to the training task of rolling bearing fault diagnosis,a diagnosis model based on convolutional neural network(CNN)and long-short-term memory network(LSTM)is designed,and a one-dimensional convolution structure suitable for bearing input is designed,retaining the time correlation of bearing vibration signals,and connecting the LSTM structure to the internal signal data mining Timing information,select a suitable RELU activation function and Adam optimization algorithm to improve the training effect of the network model.Experimental results show that the designed 1D-CNN+LSTM intelligent diagnosis method effectively improves the accuracy of rolling bearing fault diagnosis.Secondly,research a method of fusion decision fault diagnosis based on D-S evidence theory to realize comprehensive fault diagnosis of equipment.Firstly,establish a diagnosis sub-network based on1D-CNN+LSTM,and perform preliminary diagnosis on multiple data source data through three independent diagnosis sub-networks.Then,the D-S evidence theory diagnosis model is constructed,the fault category identification framework is established,the evidence body and probability distribution are revised,and the preliminary diagnosis results of multiple data sources are analyzed by decision-level fusion processing to realize the global fault diagnosis of electrical equipment.Experimental results show that the proposed fusion decision diagnosis method is effective and fault-tolerant.After the diagnosis sub-network misjudges the fault mode,it can still correct the misjudged fault category through evidence fusion decision and improve the accuracy of fault diagnosis.