Research On Gearbox Fault Diagnosis Method Based On Improved VMD And Deep Learning

Gearbox is widely used in automobiles,machine tools,wind turbines,aerospace equipment and so on.However,in the process of using the gearbox,varying degrees of damage and failure will inevitably occur,which will cause a lot of human and financial losses.In order to avoid these losses and ensure the safe operation of the gearbox,it is of great significance to study the fault diagnosis method of the gearbox.Gearbox fault diagnosis methods based on data drive are popular to use widely,but signal noise and signal feature extraction have always been the bottleneck to improve the accuracy of fault diagnosis.Therefore,this paper studies the variational mode decomposition(VMD)signal processing method and to decrease the signal noise and studies the one-dimensional cavity convolution neural network model to solve the problem of singal feature extraction.A new fault diagnosis method of gearbox is formed by combining these two methods.Firstly,the related theory of VMD method is introduced,which needs to artificially set the number of decomposition and central frequency and other related parameters.But these key paraments are important to the decomposition effect of VMD method,a new fitness funmction and a particle swarm mutation optimization algorithm are proposed for adaptive optimization of VMD parameters.Aiming at the problem of one dimensional cavity convolution neural network(1DCNN)with the weak feature understanding ability,and no memory,the long-short memory neural network(LSTM)is introduced into 1DCNN,and a new structural model of one-dimensional cavity convolution-long-short memory(DCNN-LSTM)depth nettwork is constructed.The method has been effectively verified in the ECG heartbeat classification data set on kaggle.In order to solve the problems of high noise,complex components and difficult signal feature extraction in gearbox fault diagnosis,adaptive VMD method and DCNN-LSTM are applied to gearbox fault diagnosis in this paper.The adaptive VMD method is used to decompose and reconstruct the gearbox fault signal to remove the signal noise,and the DCNN-LSTM method is used to identify the fault of the reconstructed signal.This method is applied to PHM2009 gearbox data and gearbox data obtained from QPZZ-II experimental platform.The results show that this method can effectively identify gearbox faults.Compared with other machine learning methods,the method proposed in this paper has higher diagnosis accuracy.