Lightweight network fault diagnosis method combining temporal features with spatial features
In order to solve the problem of low model training accuracy caused by cross information and repeated features among multi-sensor data,the fault diagnosis of rolling bearing under acoustic radiation signal was studied.A lightweight network fault diagnosis method based on the fusion of spatial and temporal features(SF-TFNet)was proposed.Firstly,convolutional neural network was used to extract space features(SFs)of the original bearing acoustic array signal,and then long short-term memory(LSTM)was used to extract time features(TFs)in the acoustic array signal,and the extracted SFs and TFs were fused to generate a new feature matrix.In order to eliminate the overlapping features and information redundancy caused by fusion features,kernel principal component analysis(KPCA)method was introduced to reduce the dimensionality of the newly generated feature matrix,remove the redundant components of the features,and construct a new spatio-temporal feature data set of rolling bearings.Finally,AdaBoost algorithm was used to classify the faults of the newly generated data set,and the final fault diagnosis result of the rolling bearing was obtained.The research results show that,on the rolling bearing fault test bench in semi-anechoic chamber,the SF-TFNet method can achieve a high fault classification accuracy of 99.75%,and the clustering effect of the proposed method is obvious.Comparing with ResNet,ICNN and AlexNet in the strong background noise environment,the SF-TFNet method not only converges quickly,but also has the highest accuracy,which can reach 99.25%.The comparison results further show that the proposed method has high fault identification accuracy.It provides a theoretical basis for fault diagnosis of multi-channel acoustic radiation rolling bearing.
rolling bearingacoustic radiation signalmulti-information fusionlight weight feature fusionfault diagnosislong short-term memory(LSTM)time features(TFs)kernel principal component analysis(KPCA)
万方数据
维普
