基于SVDD与距离测度的齿轮泵故障诊断方法研究

Fault diagnosis method for a gear pump based on SVDD and distance measure

提出基于支持向量域描述与距离测度的齿轮泵故障诊断方法。对齿轮泵各种工况下振动信号进行小波包分解,提取各频带能量百分比作为特征向量;利用正常工况下特征向量训练SVDD超球模型,通过定义绝对距离测度检测齿轮泵状态是否出现异常;针对每种工况的特征向量单独训练SVDD超球模型,通过定义相对距离测度准确定位齿轮泵的不同故障工况。试验结果表明,采用小波包频带能量可降低数据维数,能有效浓缩故障信息;基于绝对距离测度与相对距离测度的SVDD故障诊断方法既能检测异常状态,亦能区分各种故障工况,可实现状态监测与故障分类识别目的。

A fault diagnosis method for a gear pump based on support vector domain discription(SVDD) and distance measure was put forward.Firstly,vibration signals under various conditions of the gear pump were decomposed with wavelet packet technique to extract energy percentage of each frequency band as feature vectors.Then,a SVDD hypersphere model was trained only using feature vectors under normal condition,and an abnormal condition of gear pump was detected with the definition of the absolute distance measure.Finally,SVDD hypersphere models were trained independently by using feature vectors under various conditions,and various fault conditions of the gear pump were recognized accurately with the definition of the relative distance measure.Experimental results showed that the data dimensions are reduced and the fault information is concentrated efectively by adopting feature extraction with wavelet packet technolog;SVDD fault diagnosis methods based on absolute distance measure and relative distance measure can not only detect abnormal conditions but also distinguish various fault conditions to realize condition monitoring and fault classification.