摘要
提出一种基于支持向量机的往复泵泵阀故障诊断方法。该方法将泵阀振动信号的小波包变换系数作为特征向量 ,输入到由多个支持向量机构造的一个多值分类器中进行故障模式分类。试验结果表明 ,该方法不仅可以对发生故障的单个泵阀进行诊断 ,而且还能对同时发生故障的多个泵阀进行诊断。与常用的人工神经网络方法比较 ,该诊断方法具有更好的有效性、鲁棒性和推广性 。
A Support Vector Machines-based approach is presented for fault diagnosis of valves in reciprocating pumps. According to the approach, the input vectors of Support Vector Machines consisted of wavelet packet transform coefficients extracted from the collected vibration signals of valves as the time-frequency characteristics. And these vectors were inputted into a multi-class classifier composed of many Support Vector Machines to identify multiple fault modes of valves. The experimental diagnosis results show that, the approach can not only correctly diagnose single faulty valve but also precisely identify multiple faulty valves. Furthermore, comparing with the traditional artificial neural networks, the approach is more efficient, robust and adaptive, which indicates the potential of the SVMs techniques in machinery fault diagnosis.
出处
《机械强度》
CAS
CSCD
北大核心
2002年第3期362-364,共3页
Journal of Mechanical Strength
关键词
故障诊断
支持向量机
小波包变换
阀
Fault diagnosis
Support vector machines
Wavelet packet transform
Valve
