Research On Rolling Bearing Defect Detection Technology Using Computational Imaging

Rolling bearing is an important component in rotating machinery and equipment.As a "joint" that connects rotating components and fixed components,they are one of the most prone to failure in rotating machinery and equipment.The key to improve the reliability and safety of rotating equipment is to monitor the running status of rolling bearings continuously and to detect and identify the potential faults in time.At present,domestic fault diagnosis technology for rolling bearings has not been able to achieve automatic detection and identification of faults,and the existing methods have disadvantages such as low recognition rate and poor environmental adaptability.Therefore,in-depth research on bearing defect detection technology is required to improve the information’s analysis and processing capabilities to ensure the normal operation of rolling bearings,and reduce the incidence of accidents.Taking rolling bearing as the research object,the paper uses non-stationary random signal processing method to conduct time frequency analysis and time scale analysis on the signal.The obtained time-frequency energy distribution is used as the imaging basis to generate fault feature image,so as to complete the imaging from the bearing fault signal to the image and obtain the sample database.By the imaging algorithm and deep convolutional neural network,the problem of defect detection of rolling bearing is transformed into the problem of fault feature image recognition and classification.Paper uses deep learning to complete the defect detection and recognition,First,a convolutional neural network(CNN)was constructed.The feature image data set based on the Short-Time Fourier Transform(STFT),Wavelet Transform(WT),and Wigner-Ville Distribution(WVD)was used to convolve the CNN.The network is trained,and a proper number of samples are selected to make a training set and a test set.The designed CNN is trained on the training set,and then the test set is used to test.It proves that the CNN can recognize and classify the fault feature images to realize the defect diagnosis of rolling bearings.Simulation results show that the convolutional neural network generated by the training of fault feature images based on wavelet transform has good recognition accuracy,and the comprehensive accuracy rate is significantly better than the other two types of convolutional neural networks,reaching 98.9%.