基于AR-SVM的转子故障诊断

Rotation machine fault diagnosis based on AR-SVM

将时间序列建模与支持向量机相结合并应用于转子故障诊断领域。用时间序列理论进行故障建模,可以在缺乏对实际故障机理了解的情况下从机组自身的运行过程中动态获取故障的统计特征信息1。而支持向量机作为模式识别领域的新工具,其具有小样本学习能力等显著优势2。这里首先对实验台振动信号建立时间序列模型,然后用模型参数来训练一个支持向量机作为故障诊断的分类器。实验结果表明,这种方法有很好的实用性。

A faults diagnosis method based on time series modeling and Support Vector Machine is presented. Using time series modeling, the fault pattern can still be recognized in a statistic way, even though there was little knowledge about the characteristics and features of faults. As a new tool for pattern recognition, SVM has a good performance despite of insufficient training samples. After modeling of signals collected from a simulative rotation machine, the AR（AutoRegression） coefficients were extracted as condition features and were sent to SVM as training and testing samples respectively. This method was proved to be practical and efficient by experiments on a test rig.