Research On Rolling Bearing Fault Diagnosis Based On IntResNet Deep Transfer Learning

As an essential key component in most machinery and equipment,the operating condition of rolling bearings is directly related to the safety and stability of machinery and equipment.Therefore,reliable rolling bearing fault diagnosis technology is very important.There are some problems in fault diagnosis of rolling bearings,such as difficulty in obtaining fault data,variable operating conditions and difficulty in fault feature extraction due to strong background noise interference.Therefore,this thesis proposes a rolling bearing fault diagnosis method based on Int ResNet deep transfer learning.The ImageNet image dataset is used as the source domain dataset,and the original vibration signal preprocessed intrinsic mode time-frequency map samples are used as the target domain dataset for fine-tuning to obtain a rolling bearing fault diagnosis model.Due to the fact that the input of the ResNet fault diagnosis model is an intrinsic mode time-frequency diagram converted based on intrinsic mode functions,it is referred to as Int ResNet for short.The main research contents of this thesis can be specifically divided into the following three aspects:(1)Aiming at the problem that the fault characteristics in the original vibration signal are not obvious due to strong noise and other interference components,a parameter optimized variational mode decomposition algorithm is used to decompose the original vibration data to obtain a series of intrinsic mode functions(IMFs).Due to the ability of IMF to characterize the dynamic information of fault vibration signals,the kurtosis method is used to optimize them,so that the selected signals have obvious fault impact characteristics,which is convenient for fault feature extraction.(2)Aiming at the problem of insufficient fault data in rolling bearing fault diagnosis,a deep transfer learning method was used to fine-tune the ResNet model to achieve rolling bearing fault diagnosis.A ResNet fault diagnosis model is built,the distance between the model output value and the label is calculated using the cross-entropy loss function,and the model overfitting is prevented by dropout technology.The sensitive feature component IMFmax selected from the preprocessing is subjected to continuous wavelet transform to obtain intrinsic mode time-frequency map samples,which are input into the parameter transfer ResNet50 fault diagnosis model for fine tuning to achieve classification and recognition of rolling bearing faults,achieving good results.(3)Aiming at the problem of rolling bearing fault diagnosis under small samples and varying operating conditions,the existing rolling bearing data samples are geometrically transformed and color transformed to increase the number of samples in the training set to expand the data set,and the public data set of rolling bearings from the Case Western Reserve University is used for experimental validation.Finally,under different sample sizes,a comparative analysis of the test results of the model before and after data enhancement under variable operating conditions is conducted to verify the effectiveness of data enhancement in variable operating condition fault diagnosis.