Research On Gearbox Fault Diagnosis Based On Variational Mode Decomposition And Convolutional Neural Network

Gearbox is a common component in mechanical transmission.Due to its complex structure,it is easily affected by factors such as lubrication and temperature.During long-term high-speed and heavy-load operation,it is very prone to failure.However,it is difficult to find the failure in the early stage of failure.If maintenance cannot be carried out as soon as possible,it will inevitably cause damage to the entire equipment and even the staff over time.Therefore,it is of great practical significance to monitor the condition of mechanical equipment,determine the type and location of faults,and ensure the safe and stable operation of machinery.This paper takes gearbox as the main research object,and proposes a gearbox fault diagnosis method based on variational mode decomposition(VMD)and convolutional neural network(CNN)based on parameter optimization,focusing on the noise reduction and fault characteristics of gearbox vibration signals.Extraction and failure pattern recognition are studied.Firstly,the research background and significance of gearbox fault diagnosis are introduced,and the common faults and vibration characteristics of gearbox are introduced.The noise reduction method based on parameter optimization variational modal decomposition is put forward.The decomposition effect of VMD was verified successively,and compared with the decomposition of EMD and EEMD.Finally,the effective component was selected for reconstruction with kurtosis as an index,and through envelope spectrum analysis,it was proved that VMD plays an important role in noise reduction of gearbox fault signals.the superiority.Secondly,a gearbox fault classification method based on generalized composite multi-scale permutation entropy(GCMPE)and parameter optimization support vector machine(SVM)is proposed.Aiming at the deficiency that permutation entropy and multi-scale permutation entropy cannot fully extract fault feature information,on the basis of multi-scale permutation entropy,the coarse-grained process is improved to obtain GCMPE,which is used as fault information feature vector and input to the SVM of immune particle swarm optimization.Experiments show that the method can effectively identify the failure mode,but the support vector machine in the gearbox failure mode recognition has the deficiency that the fault feature information needs to be extracted before classification.Finally,a gearbox fault classification method based on parameter optimization VMD and CNN is proposed.Convolutional neural network makes up for the shortage of SVM that needs to extract feature information.Due to its powerful computing and feature extraction capabilities,it can directly process vibration signals.This paper introduces the process of convolutional neural network,back propagation,and algorithm optimization.Then,the parameters of the convolutional neural network model were set,and the experimental process of VMD-CNN was formulated.Finally,five models of VMD-CNN,SVM,CNN,EMD-CNN and EEMD-CNN were compared,the results show that the method in this paper has higher accuracy.