Research On Fault Diagnosis Method Of Rolling Bearing Based On Deep Learning

Industry is an important manifestation of comprehensive national strength,and mechanical equipment is an indispensable element in the industrial field.In the industrial field,mechanical equipment is becoming more and more systematic,complex and intelligent.Rolling bearings are an important part of mechanical equipment.Once a rolling bearing fails,it will directly cause the malfunction of mechanical equipment and even threaten people’s life and property.Therefore,research on the fault of rolling bearings has important theoretical significance and application value.The thesis researches the fault diagnosis technology of rolling bearing based on deep learning.The main work is as follows:First,the application of traditional fault diagnosis methods and deep learning methods in the field of mechanical fault diagnosis is analyzed,and the principles of two typical deep learning methods are thoroughly studied.Aiming at the problem that the deep neural network has poor ability to extract the fault features of the raw signal,the effect of different data segmentation methods on the neural network fault extraction ability is studied,and the data segmentation method is improved.This method calculates the sample length and sliding sampling interval length based on the rotation speed and sampling frequency of the sensor.Through experimental analysis,this method can significantly improve the ability of deep neural networks to extract fault features from time-domain vibration data.Then,in view of the problem that the less training data causes the generalization ability of the deep neural network to deteriorate,the characteristics of the time-domain signal are analyzed,and a data augmentation method suitable for time-domain vibration data is proposed.This method divides the samples of a single category into two parts on average,and then randomly recombines the samples of the two parts into new samples to achieve the purpose of data augmentation of this type of sample.Through experimental verification,the proposed method can effectively solve the problem of poor generalization ability of the deep neural network due to insufficient training data,and can also be used to solve the problem of faulty data imbalance.Finally,in view of the problem that most deep neural network models have many parameters and large amount of calculations,and cannot be applied to embedded devices with low computing power and real-time status monitoring,this thesis proposed a lightweight model based on wavelet time-frequency transform and deep squeeze convolutional neural network.The model uses two methods: deep squeeze convolutional network structure instead of traditional convolutional network structure and global average pooling network structure instead of fully connected network structure,both of which can greatly reduce network parameters.Through comparative experiments,this lightweight model can effectively prevent the network from overfitting with fewer parameters and maintain a higher verification accuracy.It can be applied to equipment with low computing power and real-time condition monitoring of rolling bearings.