Study On Rolling Bearing Fault Diagnosis Based On Deep Learning

With the progress of science technology and the continuous development of production mode,the operation speed,precision,scale and integration of mechanical equipment are increasingly higher.At present,with the improvement of the mechanical health monitoring system,mechanical equipment inspection data has gradually entered the era of "Big Data".Effective fault monitoring and diagnosis of mechanical equipment by using advanced fault diagnosis algorithm/detection equipment is the key to ensure the healthy and stable operation of mechanical equipment.How to discover abnormal operation of equipment and troubleshooting timely in big data has become a hotspots research direction in the field of mechanical fault diagnosis.The traditional diagnostic methods,such as artificial fault feature extraction is increasingly demanding for professionals,and can no longer meet the analysis requirements of "Big Data".In this paper,the rolling bearing parts in mechanical equipment as the breakthrough point of fault diagnosis research,combined with deep learning technology,a rolling bearing fault diagnosis algorithm based on deep learning is proposed.This paper elaborates on the two major research directions in the field of deep learning:Convolution neural networks(CNNs)and Recurrent Neural Networks(RNNs)that represent by Long Short Term Memory networks(LSTMs).Firstly,the paper analyzed the structure and mathematical principles of Convolution neural networks and Long Short Term Memory networks;Subsequently,CNNs model and LSTMs model are established to diagnose bearing failure.In order to ensure the reliability of the diagnosis model,an adaptive deep convolutional networks(ADCNNs)model was proposed in this paper.The model does not need to extract the feature of bearing fault data manually,instead,the data to be analyzed is directly input into the diagnostic model,and it can extract the feature of fault data independently and diagnose the fault type of bearing intelligently.The diagnosis model has the characteristics of diminishing convolution kernel size and symmetrical kernel num stacking,which ensures the running speed and high diagnostic accuracy at the same time.Aiming at the actual operating conditions of bearings,the paper designs "Variable Load Experiment" and "Noise Interference Experiment" to further test the stability and reliability of the diagnosis model.The experimental results show that the adaptive deep convolutional model can achieve more than 99%diagnostic accuracy.In the last part of this paper,a new fault diagnosis model SFDCRs is explored,which combines the feature extraction ability of CNNs and the feature automatic selection ability of LSTMs.The experimental result analysis shows that the SFDCRs model can achieve more than 99%diagnostic accuracy.