基于互无量纲特征与迁移学习的变工况滚动轴承故障诊断

Variable Condition Rolling Bearing Fault Diagnosis Based on Mutual Dimensionless Features and Transfer Learning

滚动轴承的可靠性是工业生产连续性与安全性的重要保证之一,但由于其运行环境的高温、高压、高时长特性,振动信号表现出一定的非平稳性与非线性,并且由于故障样本的缺失,基于数据驱动的诊断方法很难进行应用。文中提出一种基于互无量纲特征与迁移学习的变工况滚动轴承故障诊断方法,提取轴承振动互无量纲特征用于解决轴承信息非线性问题;针对故障样本缺失问题,提出一种改进的卷积神经网络迁移模型,将大数据模型迁移到小样本模型中来。通过美国西储大学轴承实验平台的验证所提出的方法平均诊断准确率达到95.9%,为滚动轴承的故障诊断提供一定的理论参考。

The reliability of rolling bearings is one of the important guarantees for the continuity and safety of industrial production.However,due to the high temperature,high pressure,and long duration characteristics of their operating environment,vibration signals often exhibit certain non-stationary and nonlinear characteristics.Moreover,due to the lack of fault samples,data-driven diagnostic methods are difficult to apply.This study proposes a fault diagnosis method for variable condition rolling bearings based on cross dimensionless features and transfer learning.Firstly,the cross dimensionless features of bearing vibration are extracted to solve the nonlinear problem of bearing information.To address the problem of missing fault samples,an improved convolutional neural network transfer model is proposed to transfer the big data model to the small sample model.The average diagnostic accuracy of the proposed method reached 95.9%through the validation of the bearing experimental platform at Western Reserve University in the United States,providing a certain theoretical basis for fault diagnosis of rolling bearings.