Research On Gearbox Fault Diagnosis Based On Improved Convolutional Neural Network

Gear box is an important part of rotating machinery,its health state plays a vital role in the safe operation of rotating machinery,so fault diagnosis is very important and necessary.In recent years,with the continuous development and progress of artificial intelligence and deep learning,data-driven deep learning intelligent diagnosis method has been widely used in gear box fault diagnosis.Based on deep learning technology,this thesis improves convolutional neural network and residual network to achieve the purpose of gear box fault diagnosis.Combined with transfer learning method,the network is further applied to the actual industry for gearbox fault diagnosis.Specific research contents are as follows:Firstly,this thesis is based on convolutional neural network and support vector machine optimized by Sparrow Search Algorithm for fault diagnosis of gearbox small sample data set.Feature extraction capability of convolutional neural network is used to extract features from original data.The Sparrow Search Algorithm was used for automatic parameter optimization of support vector machine,and the extracted feature data was input into the optimized support vector machine for fault diagnosis.By using drivetrain dynamics simulator(DDS)to collect test data,the proposed network model is verified to have higher fault identification accuracy and perform well in the fast diagnosis of small sample data.Secondly,for large sample data sets,a residual network based on mixed attention mechanism and multi-scale feature extraction is proposed.This model is based on the residual network and combines the mixed attention mechanism to increase the depth of the network.The series first layer parallel multi-scale convolutional neural network expands the width of the network and improves the capability of fault feature extraction.Add batch normalization layer to improve network stability.Transverse comparison and ablation experiments are added to prove that the network model has a high accuracy of fault recognition.The DDS test data at different signal-to-noise ratio and the public bearing data of Case Western Reserve University were used for test verification,which proved that the network has strong robustness and good anti-noise ability.Finally,based on the mixed attention mechanism and multi-scale feature extraction residual network model,and combined with the transfer learning method,the gearbox fault diagnosis problem in the actual industry is solved.Using the data from Western Reserve University as the source domain and the DDS test data as the target domain to simulate the actual industry,the transfer learning method is used to obtain a high accuracy of fault identification.Improve network generalization capability by data enhancement,adding L2 regularization formula,and adding Dropout layer.The cross-test with the variable load data of Western Reserve University proves that the proposed network model has strong adaptability and generalization ability to the variable load condition.It has certain engineering practical value and can be widely used in gearbox fault diagnosis in industrial production.