Intelligent Fault Diagnosis Method For Rolling Bearings Based On Image Coding And Deep Learning

As one of the key components of rotating machinery equipment,rolling bearings are prone to failure due to their complex and changeable working conditions.When they are partially damaged,the vibration of the mechanical equipment will be abnormal,and the mechanical equipment will be damaged.Therefore,it is of great significance to carry out research on online monitoring and fault diagnosis methods of rolling bearings to ensure the safety and reliability of equipment operation.Rolling bearing is taken as the research object of this dissertation,and comprehensively uses image coding and deep learning to carry out research on intelligent fault diagnosis methods for rolling bearings under constant working conditions and time-varying speed conditions,mainly including the following three aspects:(1)Aiming at the problems of "overfitting" in the application of traditional convolutional neural network(CNN)for fault diagnosis and the loss of time information in traditional grayscale image coding,a new method based on Gramian angle fields(GAF)is proposed.A rolling bearing fault diagnosis method combining image coding method and transfer deep residual neural network(TRN).Based on the uniqueness of the time series encoding mapping by the GAF method,the original vibration signals are encoded into Gramian angle difference fields(GADF)map and Gramian angle summation fields map(GASF),and the parameters of the Res Net18 model pre-trained on Image Net are transferred to Res Net18 with GAF as input for feature extraction of GAF under different fault modes and classification,and then achieve the purpose of fault diagnosis.The proposed method is validated by conducting experiments on rolling bearing pitting faults,and the analysis results show that the proposed method can highlight the intrinsic features of different fault modes better than the traditional grayscale map coding,and the proposed method has higher recognition accuracy of 99.30%,and faster convergence speed and stronger robustness compared with the CNN model.(2)Aiming at the problem that the parameters of the TRN model proposed above are large,a fault diagnosis method for rolling bearings based on the combination of Ghost-Res Net and RAdam is proposed.Use the Ghost convolution module in Ghost Net to replace the original standard convolution module in the Res Net model to form the Ghost-Res Net model.Taking the GADF image as input,the Ghost-Res Net model is trained by the RAdam optimization algorithm,and the feature extraction and classification of the GADF image are performed,and finally the fault diagnosis is realized.Through comparison and verification with different optimization algorithms and different diagnosis methods,it shows that the Ghost-Res Net model trained with RAdam optimization algorithm has higher diagnostic accuracy,faster convergence,better robustness and good noise resistance.The proposed method is applied to 10 classifications of rolling bearing crack faults,19 classifications of crack and pitting faults,and4 classifications of cylindrical roller bearing composite faults for verification,and high diagnostic accuracy is obtained.Again,it is verified that the Ghost-Res Net model has good generalization ability.(3)In view of the difficulty of rolling bearing fault diagnosis under time-varying speed and the high hardware requirements of diagnosis model,a fault diagnosis method based on angle domain resampling,RP image coding and EMobilenet-v3 integration is proposed.Firstly,through the angle domain resampling preprocessing method,the non-stationary time-domain vibration signal under the condition of time-varying speed is transformed into a relatively stable angle domain signal containing angle domain fault information.Secondly,the RP image is used to characterize the fault features.Finally,the RP image features are extracted and classified by the lightweight network EMobilenet-v3 model,and then the fault diagnosis is realized.The experimental results show that the RP image coding method is more suitable to characterize the corresponding fault characteristics of different fault modes under variable speed conditions than GAF and MTF image coding methods,and the highest diagnostic accuracy is 99.99% when the distance threshold percentage is 54.Compared with other diagnostic methods,the proposed EMobilenet-v3 model has the highest diagnostic accuracy and the lowest computational cost,and the GPU occupation is only 1.6G.