Research On Method Of Bearing Mechanical Vibration Signal Analysis Based On Machine Learning

Bearing is an indispensable component in the production system.Timely bearing fault diagnosis can minimize the economic loss,because the early detection of new problems can save valuable time and cost,so the classification and location of its faults are of great significance in industrial production.Fault diagnosis is an important system in manufacturing industry.With the development of intelligent manufacturing,data-driven fault diagnosis has become a hot topic.However,the traditional data-driven fault diagnosis method experts based on feature extraction.The feature extraction process is a difficult process and greatly affects the final result.Deep learning(DL)provides an effective method to automatically extract the features of source data,and convolutional neural network(CNN)is one of them.In this paper,CNN is applied to bearing fault classification and location1.Bearing signal classification: bearing vibration signal is one-dimensional time series signal.We use a conversion method to convert one-dimensional time series signal into two-dimensional matrix,and then use the convolutional neural network(CNN)to classify the image.We use the global average pooling(GAP)layer instead of the traditional full connection layer.This will transform the problem of signal analysis into the problem of image processing,through the classification of images,indirect classification of source vibration signals.The experimental results clearly verify the effectiveness of this method.We use the bearing vibration data set of Case Western Reserve University to verify our experiment.The results show that the scheme can classify the bearing fault signals accurately.2.Bearing fault location: on the basis of fault classification,this paper adds the algorithm of feature visualization,that is,gradient weighted class activation mapping(Grad-CAM),which can find the feature area in the image,that is,the area of convolution neural network as the classification basis.Then we map these points back to the source sequence signal,and we can find the special sequence points in the source sequence signal Finally,we use these points to locate the fault.We use the bearing vibration data set of Case Western Reserve University to verify our experiment.The result shows that the bearing fault location method proposed in this paper can roughly locate the bearing fault location.