Image Transformation Of Vibration Signal And Quality Level Analysis Model Construction Of Rolling Bearing

Rolling bearing is a precision mechanical part.As the main component of rotating machinery,it plays a major role in ensuring the safe operation of equipment.Therefore,it is conducive to improving the safety factor of rotating machinery by establishing a complete set of bearing quality evaluation methods.The traditional industrial model relies on manual inspection,the experience of technicians and measuring equipment to realize the quality assessment of bearings.This method cannot complete the inspection of large batches of samples,and the process is complicated and cumbersome.In view of this situation,a rolling bearing quality analysis method based on vibration image and deep learning was proposed.In the early stage of delivery,the characteristic vibration information of the bearing is weak and noise interference was accompanied,which affects the quality evaluation.The denoising method for vibration signals based on Empirical Mode Decomposition(EMD)and Fast Independent Component Analysis(Fast ICA)was proposed.The vibration signals of healthy bearings with artificial noise added were decomposed by the EMD method,and the first denoising operation was implemented through the correlation-kurtosis joint index.Then the remaining EMD components were reconstructed by Fast ICA for the second denoising,and the data after reconstruction of the remaining components were used as the final experimental data.Through the analysis of the final data by the envelope spectrum,it can be seen that the characteristic frequency points of the bearing components are effectively identified.It shows that the denoising method based on EMD combined with Fast ICA is feasible.Aiming at the insufficient ability of machine learning methods to extract the features of bearing vibration signals,a bearing quality assessment method based on multi-vibration image fusion was proposed.Vibration signal images are respectively generated by Short-time Fourier Transform(STFT),Continuous Wavelet Transform(CWT)and Fast Spectral Kurtosis analysis,and the quality assessment model of data fusion was built by convolutional neural network(CNN).By comparing 1D-SVM,1D-CNN,2D-CNN and other quality assessment models,the experimental results show that the effective features of one-dimensional vibration signals are weak and difficult to obtain.The vibration image can combine the time domain and frequency domain of the vibration signal.Comparing analysis of the respective quality assessment models of vibration images,the results show that the multi-vibration image fusion method has the advantages of fast convergence and high accuracy of quality assessment,and the final accuracy rate is 98%.Aiming at the problem that vibration image generation requires a large amount of calculation and complex parameters setting in the multi-vibration image fusion method,a rolling bearing quality evaluation method based on STFT and Auto-Encoder(AE)ensemble multi-activation function is proposed.Firstly,the deep features of the STFT images are extracted by a multi-layer AE method.secondly,different forms of the STFT image features are obtained by changing the activation function of the AE network.Then multiple activation functions are integrated using the proposed activation function weighted voting method,and finally a stable quality assessment model is constructed.In the experiment,the Back Propagation Network(BPNN)and Support Vector Machine(SVM)methods are compared,and the results show that the method has excellent quality assessment ability with an accuracy rate of99.26%.In order to verify the effectiveness of the multi-layer AE network and the integration method,the single-layer AE network evaluation model and the multi-layer AE network evaluation model are compared,and the accuracy rates are 86.67% and94.67%,respectively.Indicating that the deep AE network can effectively acquire vibration features.The accuracy of the method proposed in this paper is 99.3%,indicating that the ensemble method makes the quality evaluation ability of the deep AE network more stable and accurate,and can well achieve the purpose of bearing quality evaluation.