Research On Fault Diagnosis Of Gearbox Based On Deep Learning

Mechanical equipment plays an irreplaceable role in industrial society.When mechanical equipment breaking down,it will cause major accidents and cause irreparable damage to personnel and the economy.Therefore,it is very important to perform fault diagnosis and condition monitoring on mechanical equipment.As an important part of mechanical equipment,the gearbox is also of great significance for the fault diagnosis of the gearbox.In gearbox fault diagnosis and condition monitoring,the factor that determines the accuracy of fault diagnosis is feature extraction.The traditional fault diagnosis method mainly extracts features from the original signal according to various signal processing methods.The extracted features are then categorized into a shallow learning network to complete fault identification and status monitoring.However,the traditional fault diagnosis method relies too much on the experience of experts and has too many human factors.Therefore,in the process of feature extraction,the personnel who require feature extraction operations need to have a large amount of solid signal analysis knowledge.At the same time,gearbox fault diagnosis and condition monitoring of mechanical equipment are facing big data problems.The traditional method of fault diagnosis seems to be inadequate.In this paper,we will focus on the above problems,aiming at intelligent and accurate fault diagnosis of gearbox in big data environment.The stacking automatic encoder and convolutional neural network in deep learning are the research methods,and the following work is done:(1)Studied the characteristics of shallow learning and deep learning structure,and the connection and difference between deep learning and shallow learning.(2)Studied the working principle of stacking automatic encoder network and design a stacking automatic encoder network,and combine it with the overall local mean decomposition method and particle swarm optimization algorithm to study the gearbox fault diagnosis.This study protocol achieved an accuracy of 94.2% in the experiment.(3)Constructied a stacking automatic encoder network with different depths,and obtaining four hidden layers as the optimal automatic encoder depth,and combining the input layer and the output layer to construct a six-layer deep stacking automatic encoder network.The network and principal component analysis methods are combined to carry out research on gearbox faultdiagnosis,and the automatic feature extraction and state recognition of data slimming and deep learning networks are completed.This study protocol achieved an accuracy of 96.4% in the experiment.(4)Studied the composition of the convolutional neural network and the functions and working principles of each layer.Combining the characteristics of convolutional neural networks with the characteristics of two-dimensional image data,a network model is constructed to automatically reconstruct one-dimensional signals into two-dimensional signals and input convolutional neural networks to complete fault diagnosis automatically.The network model reconstructs 1936 data points contained in each data sample into 4444? twodimensional matrix data,and inputs the reconstructed two-dimensional data into an input layer,two convolution layers,and two pooling layers.,a fully connected layer and a classifier in a seven-layer convolutional neural network to complete the end-to-end gearbox fault diagnosis method.The research program obtained 99.7% accuracy in the experiment,and truly realized the intelligent fault diagnosis method of the gearbox from the human factor in the big data environment.