Research On Fault Diagnosis Method Of Rolling Bearing Of Rotating Machinery Based On Multi-Source Information Fusion

As a critical component of rotating machinery,rolling bearings are indispensable in modern industry.Rolling bearings often operate in high-pressure and high-load environments,and damage to individual components can affect the overall operation of the equipment.This article proposes three new deep learning diagnostic models based on information fusion technology for single and composite bearing faults under different conditions,and their effectiveness is demonstrated through experiments.Firstly,two deep learning models used in this article are introduced: Convolutional Neural Networks and Visual Transformer Networks.The basic structure of CNNs and the principles and functions of each network layer are explained,and the advantages and disadvantages of commonly used parameters are discussed.Then,the relevant theoretical knowledge of the Visual Transformer architecture is introduced,and the structural characteristics and functions of each network layer based entirely on self-attention mechanisms are analyzed.Secondly,a one-dimensional Wavelet Kernel Convolutional Neural Network method for bearing fault diagnosis is proposed to address the issue of noisy input signals and the need for preprocessing of input signals in traditional deep learning networks.This method does not require manual feature extraction and uses a designed wavelet convolution kernel to integrate signal processing methods into the network model.By constructing a wavelet convolution kernel network and replacing the original convolution kernel in the one-dimensional CNN with a wavelet kernel,the network can effectively extract features from the original signal and input the obtained feature vectors into the subsequent network to achieve fault diagnosis of bearing signals containing noise.Next,a multi-source information fusion-based dual-channel Wavelet Kernel Convolutional Neural Network method for fault diagnosis of rolling mill bearings is proposed to address the problem of real-time changes in bearing fault feature frequencies under multiple working conditions and susceptibility to environmental noise.Firstly,a one-dimensional wavelet kernel network and a two-dimensional CNN model are established to respectively process the collected sound and vibration signals.The vibration signal is transformed into a wavelet time-frequency map through wavelet transform and inputted into the two-dimensional CNN.Then,the features extracted by the two networks are concatenated in the pooling layer,and the fault probability classification is outputted after comprehensive fusion of multi-sensor complementary information.Finally,the effectiveness of the proposed method is validated on a rolling mill bearing experimental platform.Finally,a multi-source information fusion-based temporal signal imaging Convolutional Visual Transformer(CVT)method for diagnosis of compound faults in bearings under multiple working conditions is proposed.Firstly,the data collected by a vibration sensor and two current sensors are imaged into a two-dimensional image with temporal correlation by Gramian Angular Summation,Difference Map,and Markov Transition Field.Then,to overcome the problem of traditional visual transformers’ strong dependence on the input data scale,a Convolutional Token Embedding and Convolutional Projection module are designed to introduce convolution into the visual transformer,and a CVT fault diagnosis model is built.Finally,experiments demonstrate that the proposed method has good diagnostic performance for the compound fault diagnosis of bearings under multiple working conditions.