摘要
针对目前支持向量机和支持向量数据描述多分类方法,无法有效处理新类别样本的难题,提出一种自适应的故障诊断方法。首先,在支持向量数据描述多分类器中,增加一种新的球边界偏移判别条件,使诊断模型具有识别新的未知样本的能力;其次,根据未知样本的空间分布情况,进行样本的聚类学习,建立新的故障类别描述域,完成诊断模型的自更新。以混凝土泵车柱塞泵为研究对象进行仿真实验,结果表明:与传统多分类方法相比,该方法有更好的识别精度,特别是测试样本包含未知类别故障时,识别精度仍大于95%,显示了更好的适应能力。
For fault identification,the online diagnostic model based on support vector machines or support vector data description cannot process the sample of new type.To solve the problem,an adaptive fault diagnosis method was presented.First,in the multi-class support vector data description,a new sphere boundary offset discrimination condition was used,then the diagnostic model can identify the new and unknown sample;Second,according to the space distribution of the new samples,the model learn and form a new domain of fault class,then the model get updated.Piston pump of concrete pump truck was studied and the results show that compared with the traditional multi-classification,the adaptive method gets better recognition accuracy,especially when the test samples contain some of unknown class.Recognition accuracy is still more than 95%,which shows this fault diagnosis method has a better ability to adapt.
出处
《电子测量与仪器学报》
CSCD
2012年第10期877-882,共6页
Journal of Electronic Measurement and Instrumentation
基金
国家高技术研究发展计划(863计划)资助重点项目(编号:2008AA042802)
国防科工局军用技术推广专项(编号:2011240)资助项目
2012年度电子信息产业发展基金资助项目
湖南大学青年教师成长计划资助项目
关键词
球边界偏移判别
空间分布聚类
故障诊断
Sphere boundary offset discrimination
Space distribution clustering
Fault diagnosis
