Fault Diagnosis Of Rotating Machinery Parts Based On Deep Learning

With the development of industrial intelligence,all kinds of rotating machinery are becoming more and more complex.Due to its poor working environment,it is very easy to produce faults.How to find faults in time and ensure the normal operation of equipment is very necessary.Therefore,based on the deep learning method,this paper studies and analyzes the bearing and gear fault diagnosis,and specifically makes the following research:(1)Summarize the research status of fault detection of rotating parts at home and abroad,analyze and summarize the failure forms of rolling bearings and gears in depth,and finally start with the detection accuracy and model training speed,determine to use convolution neural network as the main network to carry out fault diagnosis of rotating machinery.(2)Aiming at the problem that the single neural network structure is not sufficient for feature extraction and the diagnosis efficiency is low in bearing fault diagnosis,a onedimensional convolutional neural network with LSTM is proposed,which effectively combines the advantages of long and short time memory neural network that is good at processing time series data and the advantages of convolutional neural network that is good at local feature extraction to achieve the goal of combining global features with local features.Finally,through experimental verification,the accuracy of the proposed model is significantly higher than other methods,and the diagnostic rate is as high as 100%.(3)Aiming at the problem of low accuracy of model diagnosis under high load and variable working conditions,a two-dimensional CNN variable working condition diagnosis model with attention mechanism is proposed.Based on the concept,a multifeature extraction module is designed,the convolution attention mechanism is fused,and the short-time Fourier transform is used to process the original vibration signal.Finally,through the experimental verification,the accuracy of the model reaches 99.6% under single working condition and more than 98% under variable working condition.(4)Aiming at the problem that some sample data are difficult to obtain,which makes it difficult to collect fault diagnosis signals,a reconstructed data migration learning algorithm based on improved residual network is proposed;The SDAE algorithm is constructed to remove the noise of the input data and concentrate the useful information hidden in the original data;Two loss functions are designed to reduce the distribution difference between the source domain data and the target domain data in the feature space.Finally,the experimental results show that compared with the traditional network model,the proposed model not only reduces the training time,but also greatly improves the classification accuracy;Compared with the mainstream transfer learning model,the accuracy of transfer is improved by at least 4% under different working conditions.Figure 45 Table 17 Reference 71...