基于SSAE-SVM的滚动轴承故障诊断方法研究  

Research on Fault Diagnosis Method of Rolling Bearing Based on SSAE-SVM

作　　者：徐先峰[1] 黄坤 邹浩泉 赵龙龙 XU Xianfeng;HUANG Kun;ZOU Haoquan;ZHAO Longlong(College of Electronics and Control Engineering,Chang’an University,Xi’an 710064,China)

机构地区：[1]长安大学电子与控制工程学院,陕西西安710064

出　　处：《自动化仪表》2022年第1期9-14,共6页Process Automation Instrumentation

基　　金：陕西省重点研发计划基金资助项目(2021GY-098);陕西省自然科学基础研究计划基金资助项目(2019JQ-678);长安大学中央高校基本科研业务费专项基金资助项目(300102321501、300102321503);西安市智慧高速公路信息融合与控制重点实验室基金资助项目(ZD13CG46)。

摘　　要：针对现有滚动轴承故障诊断方法过度依赖于有监督学习算法的问题,提出一种基于堆栈稀疏自编码和支持向量机(SSAE-SVM)的滚动轴承故障诊断方法。利用堆栈稀疏自编码(SSAE)的频域深层特征学习能力,对轴承故障特征进行快速傅里叶变换和批归一化处理,再输入到SSAE网络。所构建的SSAE网络通过贪婪算法逐层训练,使用梯度下降法反向微调,基于无监督式深层学习输出深层特征向量。利用构造简单、泛化性能好、分类速度较快的支持向量机(SVM)分类器,基于深层特征向量进行故障识别,实现滚动轴承故障类型的准确分类。利用美国凯斯西储大学滚动轴承数据集进行对比验证。验证结果显示,相较于对比模型,SSAE-SVM滚动轴承故障诊断模型具有更高的准确率和更快的收敛速度。应用无监督学习建立轴承故障诊断模型将成为轴承故障诊断的重要发展方向之一。Aiming at the problem that the existing rolling bearing fault diagnosis methods rely excessively on supervised learning algorithm, a fault diagnosis method of rolling bearing based on stacked sparse autoencoder and support vector machine(SSAE-SVM) is proposed.Using powerful frequency domain deep feature learning capabilities of stacked sparse autoencoder(SSAE),the bearing fault features are processed by fast Fourier transform and batch normalization, and then input into the SSAE network.The constructed SSAE network is trained layer by layer by greedy algorithm, using the gradient descent method for reverse fine-tuning, and the deep feature vector is output based on unsupervised deep learning.The support vector machine(SVM) classifier which has the advantages of simple structure, good generalization performance and fast classification speed, is used to identify faults based on deep feature vectors to achieve accurate classification of rolling bearing fault types.The rolling bearing data set of Case Western Reserve University is used for verification.The verification results show that SSAE-SVM fault diagnosis model of rolling bearing has higher accuracy and faster convergence speed than the comparison models, which shows that the application of unsupervised learning to bearing fault diagnosis model building will become one important development direction of bearing fault diagnosis.

关 键 词：滚动轴承 故障诊断 智能诊断 特征提取 堆栈稀疏自编码 支持向量机 故障分类器 无监督学习 贪婪算法 

分 类 号：TH133.33[机械工程—机械制造及自动化]