叶永伟，任设东，陆俊杰，杨　超

(浙江工业大学 特种装备制造与先进加工技术教育部重点实验室，浙江 杭州 310014)

基于SVM的汽车涂装线设备故障诊断

叶永伟，任设东，陆俊杰，杨超

(浙江工业大学 特种装备制造与先进加工技术教育部重点实验室，浙江 杭州 310014)

摘要：针对汽车涂装线设备故障无法及时发现和排除的困难，提出基于支持向量机的汽车涂装线设备故障诊断方法.该方法采用统计学习理论中结构风险最小化原则，克服了传统渐进理论机器学习算法的不足，适用于有限样本下模式识别问题，使预测结果更准确.依据烘房加热系统监测参数和故障类型构建SVM分类器，并采用交叉验证网格搜索法寻优各分类器的核参数及惩罚参数，建立SVM故障诊断模型；将PCA降维后的样本参数进行充分仿真训练，仿真结果表明该方法能够有效地对设备故障进行分类.

关键词：故障诊断；支持向量机；参数寻优；加热系统

Fault diagnosis for automobile coating equipment based on SVM

YE Yongwei, REN Shedong, LU Junjie, YANG Chao

(Key Laboratory of Special Purpose Equipment and Advanced Manufacturing Technology, Ministry of Education,

Zhejiang University of Technology, Hangzhou 310014, China)

Abstract:Aiming at the difficulty in discovering and eliminating the system faults of automobile coating equipment promptly, a new method of fault diagnosis based on SVM was proposed. The method was created by the structural risk minimization principle in the statistical learning theory, which overcame the weakness of the traditional learning method in asymptotic theory. And it was suitably used in pattern recognition with finite samples and made the output more precise. The classifiers were also structured according to the equipment monitoring parameters and fault types of the heating system. The optimal kernel parameters and punished parameters were searched by the cross validation grid search method in order to establish the SVM model for fault diagnosis. The samples of dimension reducing by the PCA were then taken into training in the model. It was finally reveals by the simulation experiment that it can successfully and precisely accomplish the fault diagnosis with the SVM method.

Key words:fault diagnosis; support vector machines; parameter optimization; heating system

汽车涂装设备运行参数空间重叠，维数大，具有耦合性，需对涂装线各设备运行状态进行监控并对潜在故障及时诊断以确保设备稳定运行.在支持向量机(Support vector machines，SVM)[1]基础上，设计一种基于SVM的设备故障诊断方法.SVM方法采用统计学习理论(Statistical learning theory，SLT)中结构风险最小化算法，尤其适合设备故障诊断中样本数据有限的情况[2]，它克服了传统学习方法易产生过学习状态或只能得到局部极值的缺陷，使测试结果更接近实际值[3].基于该方法，对汽车涂装线设备故……