Research On Fault Diagnosis Method Of Rolling Bearing Based On Time-Frequency Analysis And CNN

Rolling bearings are widely used in industrial production.As one of the core parts of mechanical equipment,their health conditions have an important impact on the safe operation of equipment.Rolling bearing health monitoring and fault diagnosis are of great significance for the safe operation and maintenance of equipment.Rolling bearing operation conditions and fault types are variable including changes in equipment operating load,changes in spindle speed and so on.Their vibration signals exhibit non-stationary characteristics and are often accompanied by a lot of noise.Studying the fault diagnosis under variable conditions of rolling bearings in a noisy environment is of great significance for maintaining the safe operation of equipment.This paper focuses on the problem of bearing fault diagnosis and diagnosis under variable conditions in noisy environments,and time-frequency denoising of bearing noise signal,deep learning diagnosis model and deep transfer learning method are studied.The main work is as follows:(1)In order to solve the problem that the spectrum fault feature is not significant when the signal contains noise,a threshold denoising synchrosqueezing transform(TDSST)method based on STFT time-frequency correlation coefficient threshold optimization is proposed.The optimal denoising threshold is determined by the STFT time-spectrum correlation coefficient curve,and the time-frequency denoising is performed according to the threshold.On this basis,the time-frequency spectrum after denoising and compression is obtained by synchrosqueezing transform.Fault simulation signal analysis verifies its outstanding performance in timefrequency resolution and noise reduction.(2)Aiming at the problem of information redundancy and single convolution kernel size in dense block of DenseNet,a Multipath Residual Dense Convolutional Neural Network(MRDenseNet)is proposed.In the dense block,the residual learning is introduced,and the convolution kernel of Multipath is designed.By increasing the scale of convolution kernel to increase the width of the network while ensuring the integrity of information flow,the multiscale parallel feature extraction is realized.The fault simulation signal of the inner ring of rolling bearing shows that the proposed model is effective.(3)Aiming at the problem of network performance bottleneck caused by the single distribution difference measurement index in transfer learning,a joint maximum mean discrepancy(JMMD)indicator integrated with CORAL is proposed.By optimizing the loss function of the convolutional neural network by JMMD,a model framework of bearing fault diagnosis under variable conditions based on TL-MRDenseNet is constructed.The analysis results of rolling bearing fault simulation data under different working conditions show that the JMMD index has good applicability,and the constructed model can effectively improve the fault diagnosis ability under variable conditions.(4)TDSST is used as data preprocessing method,the rolling bearing fault diagnosis method based on MRDenseNet model and the rolling bearing fault diagnosis method under variable conditions based on TL-MRDenseNet model are applied to practical applications.The advantages of the MRDenseNet model in single bearing and compound fault diagnosis were verified by bearing failure experiment data and 6205 bearing composite failure experiment data,and the advantages of the TL-MRDenseNet model in fault diagnosis under variable conditions was verified by using bearing failure data under different operating conditions.