Fault Diagnosis Based On Transfer Learning For Roll Bearings

Rolling bearings are an important component of rotating machinery systems,and their mechanical properties greatly affect the safety and reliability of the equipment.If a fault occurs in the rolling bearing during equipment operation,the device will not only result in decreased performance,in extreme cases even lead to system shutdown or equipment failure,this can cause serious economic losses and even casualties.Therefore,the fault diagnosis of rolling bearings is of great significance.In recent years,data-driven intelligent fault diagnosis methods have been widely used in the research of fault diagnosis of rolling bearings.However,due to the complex working environment of rolling bearings,the research on accurate fault diagnosis of rolling bearings is facing huge challenges.More importantly,it is not easy to collect rolling bearing signals under some certain conditions,which makes it more difficult to diagnose the faults of rolling bearings.Based on transfer learning,this paper conducts research on solving the problems existing in the fault diagnosis of rolling bearings under different working environments or conditions.The main content and innovations of this paper are as follows:(1)In order to improve the model’s performance of fault diagnosis for rolling bearings under certain working conditions,this paper introduces multi-task learning,which is a kind of transfer learning method,to improve the learning mechanism of fault diagnosis network.This method solves the problem that the feature information of some factors cannot be learned in the single-task diagnosis model,those factors may also affect the vibration response of rolling bearings.Through this way,the model can more accurately identify the working environment of the input sample,then improve the model’s generalization ability for fault classification tasks.(2)To adapt the difference of rolling bearing vibration responses under changing working conditions,this paper proposes a new fault diagnosis network,which can adjust the distance of rolling bearing signal samples under different working conditions by using the introduced transfer learning algorithm.The added Deep Coral module can optimize the Coral Loss while training the model,which is a measure of the distance between different domains,so that the model has a better diagnostic performance for rolling bearing samples under unlabeled target working conditions.(3)In some extreme cases,the real data of rolling bearings is scarce or even absent.In order to solve this problem,this paper studies the feasibility of using simulation signals as training samples and applying transfer learning methods to build fault diagnosis models that can be used for real rolling bearings.Firstly,based on the vibration response characteristics and the frequency of faults,the rolling bearing signal model is established.Then,the similarity between simulation signals and real signals is compared and its movability is analyzed.Finally,the model proposed in(2)which is trained by simulation rolling bearing samples is used to diagnose the fault of real rolling bearing samples.