A Rolling Bearing Fault Diagnosis Method Based on Improved Model Migration Strategy

Abstract

A rolling bearing fault diagnosis method with an improved model migration strategy belongs to the technical field of rolling bearing fault diagnosis. It is proposed to solve the problem that the data distribution of the same state in the source domain and the target domain are greatly different. Use wavelet transform to obtain the time-frequency spectrum of vibration signals of different types of bearings and construct an image dataset; select a certain type of data as the source domain, and other types of data as the target domain; use the source domain data to train the ResNet‑34 deep convolutional network to obtain the source Domain data classification model; use implicit gradient meta-learning to adaptively decide the transfer knowledge level and knowledge content to improve the model transfer strategy, avoiding the phenomenon that the gradient is not easy to converge in the heterogeneous architecture; introduce the transferred knowledge into the target domain ResNet‑152 convolution In the process of neural network data training, model migration is realized through parameter transfer; when training source domain and target domain networks, stochastic gradient descent algorithm is used to optimize network parameters, and fault diagnosis models of different types of rolling bearings are established.

Description

technical field [0001] The invention relates to a rolling bearing fault diagnosis method with improved model migration strategy, and belongs to the technical field of rolling bearing fault diagnosis. Background technique [0002] As a key component of rotating machinery, rolling bearings are widely used in industrial production, and effective fault diagnosis can prevent major accidents from occurring. [1] . There are various types of rolling bearings, which leads to the lack of or inability to obtain the training data of a certain type with labels in actual work. [2] . It is of great practical significance to identify the vibration signals of other models with unknown state information according to the vibration signal of the rolling bearing with known state information. [3] . [0003] The fault diagnosis of different positions and different damage degrees of rolling bearings is essentially the identification of the running state of rolling bearings [4] . The tradition...

AI Technical Summary

Problems solved by technology

[0009] The present invention aims at the problems that it is difficult to obtain the vibration data of a certain type of rolling bearing in practice, the data distribution of the same state in the source domain and the target domain is greatly different, and when the vibration data of different types of rolling bearings are selected for different domains, different fault locations and different The accuracy of damage degree identification is not high, and then a rolling bearing fault diagnosis method with improved model migration strategy is proposed

Images