基于一维卷积神经网络的滚动轴承故障程度诊断  

Fault Degree Diagnosis of Rolling Bearings Based on One-Dimensional Convolutional Neural Network

作　　者：薛妍 沈宁 窦东阳[1] XUE Yan;SHEN Ning;DOU Dongyang(School of Chemical Engineering and Technology, China University of Mining and Technology, Xuzhou 221116, China;Washing Center of Ningxia Coal Industry, Yinchuan 750409, China)

机构地区：[1]中国矿业大学化工学院,江苏徐州221116 [2]宁夏煤业洗选中心,银川750409

出　　处：《轴承》2021年第4期48-54,共7页Bearing

基　　金：中国矿业大学未来杰出人才助力计划(2020WLJCRCZL015);江苏省研究生科研与实践创新计划(KYCX20_2047);宁夏煤业科技创新项目洗煤厂动设备状态监测与故障诊断关键技术研究(宁煤技术类〔2021〕2号)。

摘　　要：针对滚动轴承性能退化状态的识别问题,提出了基于一维卷积神经网络的故障诊断方法。以轴承原始振动信号为输入,利用一维卷积神经网络自适应学习特征和分类的能力,实现由数据到识别结果的“端到端”诊断,避免了人为因素的干扰。通过凯斯西储大学不同故障尺寸的滚动轴承故障数据(模拟不同故障程度)加以验证,所建立python-Keras深度学习模型的诊断正确率达到98.2%。用辛辛那提大学滚动轴承全寿命周期数据对退化全过程进行诊断,根据轴承原始信号时域指标变化将全周期分为正常、轻微退化、中度退化、严重退化和失效5种程度,通过一维卷积神经网络对轴承原始数据进行有监督学习,所建立python-Keras深度学习模型的故障诊断平均准确率为93%。In order to identify the performance degradation state of rolling bearings,a fault diagnosis method based on one-dimensional convolutional neural network is proposed.Taking the original vibration signal of the bearings as input,the“end-to-end”diagnosis from data to recognition result is realized by using adaptive learning feature and classification ability of one-dimensional convolutional neural network,avoiding the interference of human factors.The method is verified by fault data(simulating different fault degrees)of rolling bearings with different fault sizes provided by Case Western Reserve University.The python-Keras deep learning model is established,and the diagnostic accuracy reaches 98.2%.The entire degradation process is diagnosed by using full life cycle data of rolling bearings provided by University of Cincinnati.According to time domain index changes of original signal of the bearings,the full cycle is divided into five levels:normal,slight degradation,moderate degradation,severe degradation and failure.The one-dimensional convolutional neural network performs supervised learning on original data of the bearings,and the average accuracy of fault diagnosis by established python-Keras deep learning model is 93%.

关 键 词：滚动轴承 故障诊断 一维卷积神经网络 状态监测 寿命周期 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP277[自动化与计算机技术—检测技术与自动化装置]