应用于故障诊断的SVM理论研究

分析了利用支持向量机进行模式分类的原理,指出最优分类面上的样本相对于两类误判而言是等概率的而非等风险的。针对机械故障诊断中两类错误分类具有不同损失的特点,说明直接将支持向量机应用于该领域存在不足。在此基础上,提出了诊断可信度函数,并在特征空间中,对最优分类面进行重新设计,使其符合等损失的要求,文中算例说明了该方法在实际应用中的可行性。

A Study on SVM for Fault Diagnosis

The paper starts with an overview of the pattern classification principle using support vector machines(SVMs), and points out that from the point of view of two types of errors in classification, the samples on the optimal separating hyperplane are equal in probability rather than in risk. According to the characteristics that the losses to the diagnosed equipment from the two types of errors are different, we describe the drawbacks of support vector machine when it is directly used in this field. Then a diagnosis reliability function is introduced for improving the generalization performance of the fault diagnosis. From this, an optimal separating hyperplane in feature space is reconstructed to meet the need of minimization of the losses. At the end an actual example to illuminate the feasibility of the approach is given.