Study On Fault Diagnosis Method Of Rolling Bearing Based On Parametric VMD And Deep Learning

Rolling bearing as a key component of various rotating machinery equipment,its failure often affects the operation of equipment,leading to economic losses and even casualties.With powerful data processing and information fusion capabilities,data-driven deep learning models have become a research hotspot for modern rolling bearing condition monitoring and fault diagnosis.This paper analyzes three key problems that need to be solved by the current data-driven fault diagnosis model.The three problems are reducing the dependence on a priori knowledge and the influence of human factors,improving the training speed as well as fault diagnosis accuracy of the network model,and effectively extracting multi-scale features under varying operating conditions.In order to overcome the shortcomings of the existing fault diagnosis models,this paper focuses on the study of rolling bearings based on deep learning.The main research content is as follows:(1)A parameterized variational mode decomposition algorithm(IGVMD)is designed to solve the difficulty of feature extraction from bearing signals under noise interference.The algorithm firstly optimizes the basic Grey Wolf optimizer(GWO)algorithm through the nonlinear convergence factor adjustment strategy,Grey Wolf adaptive position update strategy as well as Levy flight strategy,and then adaptively search the optimal parameters of the VMD model by the minimum envelope entropy moderate function,to eliminate the influence of artificially set parameters.The parametric variational mode decomposition method filters IMF components as signal samples for preprocessing bearing signals to achieve effective fault feature extraction.(2)A one-dimensional convolutional neural network model with an adaptive PRe LU activation layer(IGVMD-1DCNN)is proposed for the one-dimensional time sequence of bearing vibration signals.Firstly,signals processed by IGVMD are input into a neural network to ensure fault diagnosis accuracy.Then,the PRe LU activation function is introduced in the neural network to overcome the limitations of the Re LU activation function,which can avoid the weak robustness of network learning caused by excessive "dead" nodes in the training process of the network.The experiment shows that the proposed model has shortened training time and achieved optimal fault diagnosis accuracy,demonstrating its excellent identification performance in fault diagnosis..(3)The fusion model of a Multi-scale convolution and improved Deep forest(Ms CNN-Igc Forest)is proposed for fault diagnosis of rolling bearings,aiming at solving the difficulty of detecting multi-scale information with rich vibration signals.The model designed the multi-scale convolution feature fusion framework inspired by the Xception structure to obtain multi-scale features.And it introduces the ECA attention mechanism to embed the multi-scale feature extraction module to suppress interference features.Then,combined with the improved deep forest to further extract the hidden features and finally realize the classification diagnosis of rolling bearing faults.The experimental results show that the model has a good fault diagnosis effect and generalization ability under different working conditions of datasets.So the model has a reference value in the field of fault diagnosis.