Research On Rolling Bearing Fault Diagnosis Method Based On Data-driven

As one key components with the most widely used in rotating machinery,rolling bearing being fully applied in machining centers,engines,wind turbines,high-speed locomotives,construction machinery and other mechanical equipment.However,the health state of rolling bearing will directly affect the overall operation of the machine and equipment,and its state change can lead to subsequent chain reaction,safety accidents and economic losses.Therefore,accurate fault diagnosis and health monitoring of rolling bearing has become one of the typical subjects with important social significance for the development of national economy.This thesis aims to solve the problem of rolling bearing fault diagnosis using deep learning theory and methods.A novel algorithm is proposed to solve the problems of low accuracy and difficulty in determining the best hyperparameters,and the local optimal problem is solved by optimizing Chebyshev chaotic mapping,adaptive inertia weight,and random walk strategy.The penalty factor and kernel parameters of support vector machine is optimized by using the improved sparrow algorithm and a new rolling bearing fault diagnosis model is bulit.Experimental results show that the fault classification accuracy,algorithm operation efficiency,effectiveness and superiority of the proposed method have been effectively improved compared with other methods.To solve the problem of extract fault features and insufficient sensitivity to noise for traditional machine learning models,stacked denoising auto-encoder is introduced to automatically extract signal features to avoid the common problem of traditional methods relying heavily on manual extraction of features.And combined with polynomial kernel function and wavelet kernel function to build a new hybrid kernel extreme learning machine,the sparrow search algorithm is employed to iteratively optimize the hyperparameters of the deep hybrid kernel extreme learning machine to construct the fault classifier,integrated to build a novel end-to-end rolling bearing fault diagnosis model.Through the verification of the two data set instances,it can be seen from that the test results that the proposed method has made significant progress in noise resistance,feature extraction and fault classification.Multi receptive field graph convolution layer,local extreme graph convolution layer,and graph attention convolution layer is integrated to build multi graph convolutional neural network,which is used to extract the bearing data structure features,and solve the difficulty of the traditional deep learning models to achieve accurate fault diagnosis under different working conditions.The domain adversarial idea is introduced to construct a cross-domain fault diagnosis model of rolling bearing under unsupervised domain adaptation,which is tested and verified by public data set,it is shown that the proposed cross-domain fault diagnosis model method can better realize the alignment of class labels,domain labels and data structure,so that the model can more effectively deal with various cross-domain fault diagnosis tasks.