Research On Bearing Fault Diagnosis Based On EMD And CNN

Rolling bearings are an industrial component that exists in most mechanical equipment in modern industry,and are also extremely prone to failure.The working condition of rolling bearings directly affects the performance of mechanical equipment.When a fault occurs,it may cause damage to the equipment and even cause serious accidents.Achieving fault diagnosis of rolling bearings is an important research direction of current rolling bearings.In this paper,the vibration mechanism of the bearing is expounded in theory,and the common analysis methods of bearing fault signal are studied.Aiming at the problem that bearing fault signals are noisy and the characteristic frequency of faults is difficult to extract,this paper deeply studies and analyzes the Empirical Mode Decomposition(EMD)method,and proposes a method to improve EMD combined with wavelet.Obtain the characteristic frequency of the fault signal clearly.In order to further solve the problem of relying on manual judgment in the traditional EMD bearing fault diagnosis process,a BN-CNN(Batch Normalization-Convolutional Neural Networks)model is proposed to realize automatic identification of the bearing fault signal The research contents are as follows:(1)In view of the problem that the upper and lower envelope curves obtained by cubic spline interpolation in the original EMD decomposition process are prone to envelope overshoot and undershoot,an improved EMD method based on rational cubic Hermite interpolation is proposed.By analyzing the single-component signal,the objective function of the optimal interpolation parameters is obtained,and combined with the beetle whisker optimization algorithm,the best fitting envelope curve is obtained.This method effectively improves the EMD decomposition accuracy,and can suppress the modal aliasing phenomenon to a certain extent.(2)Due to the excessive noise of the actual bearing signal,in order to obtain the bearing fault characteristic frequency more accurately,the wavelet method with improved threshold and the improved EMD method are combined and applied to the bearing fault signal.First,the improved threshold wavelet denoising method is used to denoise the bearing fault signal,and then the improved EMD is used to decompose the denoising signal.Through envelope spectrum analysis,a clearer fault characteristic frequency is obtained.(3)In order to solve the problem of traditional EMD algorithm overly relying on human judgment in bearing fault diagnosis,firstly,the improved EMD combined with wavelet method to process the reconstructed bearing fault data for time-frequency domain statistical feature extraction,and then build a BN-CNN model to extract features Fault identification.Finally,the final classification results are analyzed by F1-Score and confusion matrix.The results show that,compared with traditional machine learning methods,the BN-CNN model can effectively identify bearing failure types and has a higher failure recognition rate.