Fault Diagnosis And Prediction Of Rolling Bearings Based On Deep Learning

With the continuous development of industry,the level of mechanization is getting higher.In order to avoid losses caused by damage to mechanical equipment,people have higher requirements for equipment fault diagnosis.Rolling bearings,as the joints of mechanical equipment,are vulnerable components.Once a failure occurs,it will seriously affect the operation of the equipment.Therefore,the research on intelligent fault diagnosis of rolling bearings is very necessary and valuable.For the traditional diagnosis method of mechanical equipment,it is generally to collect equipment signals,and then use time domain analysis,frequency domain analysis and other methods,combined with experience to perform equipment fault diagnosis.This method relies heavily on experience.Intelligent fault diagnosis technology based on deep learning can overcome the shortcomings of traditional diagnosis.This paper revolves around the intelligent fault diagnosis and prediction of rolling bearings,and completes the research of the following contents:(1)Fault features and extraction methods of rolling bearings.First analyze the fault and characteristic frequency of the rolling bearing,and then for the convolutional neural network,this article introduces the SDP analysis method,which is used to convert the vibration signal into a two-dimensional image.Then analyzes the advantages and disadvantages of the SDP method.In view of the shortcomings of the SDP image conversion method,this paper proposes the original data fusion method to convert the signal into a two-dimensional image.Through experimental verification,the method proposed in this paper is effective.(2)This paper proposes a deep learning model based on residual neural network.The residual neural network,by introducing the mapping of the shallow network,allows the neural network to have a deeper depth and better training effects.For CNN,the increase in model depth means that more complex classification can be achieved.Then for the shortcomings of too many parameters of the residual neural network,the dense neural network introduced in this paper,the dense neural network is another way to realize the idea of the residual neural network.At the same depth,its training parameters are much smaller than the residual neural network,and in simple scenarios,the dense neural network can effectively suppress the problem of overfitting.Through experimental analysis,the fault diagnosis models based on the two algorithms have good performance.(3)The single pixel of the vibration characteristic map of rolling bearing is prone to overfitting In order to improve the generalization ability of the model and suppress overfitting,this paper improves the residual neural network.First,the random weight reduction method is used to improve the training speed and training effect.It is an improved algorithm of stochastic gradient descent method.Then for the phenomenon of easy overfitting,this paper uses the dropout layer to suppress the overfitting phenomenon.Experiments show that the two methods used have a certain improvement in training effect and training speed.