Research On Bearing Fault Diagnosis Method Based On Deep Learning

With industrial upgrading and technological innovation,handicraft industry has gradually been replaced by various mechanical equipment.In order to liberate more labor,machinery and equipment are moving towards automation.Rolling bearings have been widely used in various mechanical facilities.As a vulnerable part of mechanical equipment,the health of rolling bearings means whether the equipment can work stably and efficiently.If there is a problem with the rolling bearing,the service life of the mechanical equipment will be shortened.Therefore,in order to avoid economic losses caused by bearing failures,it is necessary to monitor the health of bearings.In recent years,due to breakthroughs in deep learning technology,deep learning knowledge has been continuously applied in the field of fault diagnosis.The traditional bearing fault diagnosis technology based on expert knowledge and manual extraction of fault features has gradually been replaced by data driven and deep learning based fault diagnosis methods.This dissertation mainly conducts relevant research on bearing fault diagnosis methods by using deep learning technology from three aspects: improving the diagnostic accuracy of the model,reducing the size of the model,and reducing the training time of the model.The main research contents are as follows:(1)Aiming at the problem that the features extracted from a bearing fault diagnosis model using a single time-frequency analysis technique to process signals are not comprehensive,a fault diagnosis method using a multi-scale convolutional neural network with multiple inputs is proposed.Firstly,the original vibration signal is converted into two time-frequency domain images through continuous wavelet transform(CWT)and short time Fourier transform(STFT),respectively;Then,two time-frequency images are input into convolutional networks with different convolutional kernel sizes for feature extraction;Finally,the extracted different fault feature maps are superimposed and input into the classification layer to achieve fault classification,and experimental verification and analysis are conducted on the dataset.The results show that the proposed method has stronger fault feature extraction ability and higher diagnostic accuracy compared to the bearing fault diagnosis method using a single time-frequency analysis technique to process signals.(2)A fault diagnosis method for light bearings based on deep separable convolution was proposed to solve the problem of incomplete features contained in a single position sensor and multiple parameters in the fault diagnosis model.This method can simultaneously learn different features from vibration signals at different locations to improve the stability of the diagnostic model.The lightweight unit designed reduces the size of the model and the number of parameters that need to be learned.The bearing vibration signals at different positions are converted into two-dimensional time-frequency images through STFT and input into the model.Experiments were conducted on two bearing datasets and compared with other models.The results show that this model has smaller model size,higher accuracy,and less computational burden compared to other classical deep learning models.(3)Aiming at the problems of long training time and incomplete feature extraction in current bearing fault diagnosis methods based on deep network,a bearing fault diagnosis method based on transfer learning was proposed.By fine-tuning the pre training model to achieve ideal diagnostic results and shorten training time.The effects of freezing different convolution modules and using different learning rates on the diagnostic performance of the model were analyzed on the dataset.And compared with other classic networks.Experimental results show that the proposed method has shorter training time and diagnostic accuracy.Figure 40 Table 30 Reference 79...