Application Of Multimodality Fusion And Deep Learning Algorithm In Rolling Bearing Fault Diagnosis

Rolling bearing is an important mechanical foundation.Once the failure of the rotating mechanical bearing in modern production will seriously affect the efficiency and safety.The rolling bearing in actual operation is always in variable load and noise interference.Complex working conditions will lead to the change of vibration characteristics,making fault diagnosis difficult.Therefore,it is of great significance and value to carry out fault diagnosis of rolling bearing under complex working conditions.The traditional fault diagnosis method is mainly to decompose vibration signals in multi-scale and select features manually.The process is more random,which results in information loss and cumulative error.In recent years,the fault diagnosis method using deep learning to extract features is gradually emerging.This method is mainly based on the analysis of vibration signal in time domain,frequency domain or a domain in wavelet domain,No Federation extraction of other domain features is available.The experimental results show that the above two methods are not accurate and generalization for rolling bearing fault diagnosis under complex conditions.In view of the above problems,this paper combines the deep learning convolutional neural network(CNN)and the multimodal fusion technology(MTF),and proposes a fault diagnosis method for rolling bearing under complex working conditions based on multimodal convolutional neural network.The main research work and results are as follows:(1)In order to solve the problem of fault diagnosis of rolling bearing under variable load,this study uses depth learning method to extract the features separately from time domain and frequency domain data,and integrates the features of two domains to strengthen the relationship between different domain features,supplement the information that may be missed by single domain features,and construct TFMCNN integrated learning model with 1D-CNN.The results of many experiments on bearing test bench show that the average accuracy of TF-MCNN under various load conditions is more than 96%,which is nearly 20%higher than traditional SVM diagnosis method,11%higher than that of single frequency domain CNN,and 6.67%higher than WDCNN in single time domain.(2)In this study,the original time domain signal is analyzed by DB4 wavelet,and the fourth layer high frequency component is drawn to input 2D-CNN to extract the wavelet domain features.At the same time,the time-domain and frequency-domain signals are input into 1D-CNN to extract the features respectively,and the network WTF-MCNN is constructed to integrate the three domain features to carry out the complex working condition axis Fault diagnosis.The experimental results show that the average accuracy of the proposed method is 20%higher than that of WDCNN in single time domain under weak noise and various variable loads,about 3%higher than that of traditional methods of EMD+VPMCD fusion in time and frequency domain;the single time domain method with moderate noise has no diagnostic ability,and the accuracy of the proposed method is 7.3%higher than that of traditional fusion method;the accuracy rate of the proposed method is higher than that of traditional fusion method in case of strong noise The method is more than 20%higher and has strong robustness.Therefore,WTF-MCNN has stronger anti noise ability than traditional and single domain diagnosis methods,and is more suitable for fault diagnosis under complex conditions.