Research On Rotating Machinery Fault Diagnosis Based On Improved Convolutional Neural Network

There are problems such as wear and tear in the operation of rotating machinery,which lead to various faults,and safety accidents are likely to occur.Therefore,the fault diagnosis of rotating machinery is becoming more and more attractive.Due to the nonlinear characteristics of vibration signals of rotating machinery,the weak fault characteristics are easily submerged in the variable working conditions with the coexistence of strong noise and variable load,which increases the difficulty of fault diagnosis.Based on deep learning model and improved convolutional neural network,intelligent fault diagnosis for complex fault modes of rotating machinery is realized in this thesis.The main contents of the thesis are as follows:Firstly,a convolutional neural network(CNN)and particle swarm optimization support vector machine(PSO-SVM)method was proposed to solve the problem of unsatisfactory fault diagnosis effect in the operation of rotating machinery.Firstly,the time-frequency characteristic statistics of the signal are obtained by using the characteristic parameters,and then the convolutional neural network is used to extract the secondary features of the time-frequency characteristic statistics.Finally,particle swarm optimization support vector machine is used to classify the signals.Experimental results show that the accuracy of this method is not only higher than other classical network models,but also the shortest training time.Secondly,a one-dimensional residual convolution neural network method was proposed to solve the problem that the fault diagnosis effect of rotating machinery under different operating conditions was not ideal due to the interference of variable load and variable noise at the same time.The normalized original vibration signals of rotating machinery were input into the network model,and features were extracted by multiple one-dimensional convolutional layers with residual connections.After multiple convolutional pooling,the signals were input to Softmax layer for classification,and the fault types of vibration signals of rotating machinery were output.Experimental results show that the proposed method is robust to noise and generalization.Finally,in order to further enhance the ability of model extraction and recognition,a rotating machinery fault diagnosis method combining attention mechanism and multi-scale residual network was proposed.This method takes one-dimensional vibration signals of rotating machinery as input,and enters multi-scale network structure through first-layer convolution.By increasing the adaptability of network to the size of convolution kernel,richer feature information is extracted.Further feature extraction was carried out by multiple residual blocks,and more critical information was extracted by attention mechanism,so that model classification could make more accurate judgment.Finally,fault diagnosis was carried out by Softmax classifier.Experimental results show that the proposed method has strong robustness under complex conditions.