在涡轮泵等机械设备的故障诊断中,多类故障诊断是经常出现的问题。为提高多类故障诊断速度,在球结构支持向量机的基础上,对其分类规则进行改进,充分考虑分类球的大小不同,经过理论分析和实验验证得到样本点落在分类球外和分类球重叠区域的最佳分类公式。用该算法和其它几种常用算法对涡轮泵模拟故障进行分类比较,结果表明,基于改进型球结构支持向量机的故障诊断算法学习速度更快,诊断效果好。
Multi-category faults were common in turbo-pump and other machines. In order to speed up the multi-category faults diagnosis, an improved sphere support vector machine was proposed in this paper. The sizes of class spheres are fully taken into account, and better classification rules outside the spheres and in the intersections of the spheres are derivated by many theoretic analyses and experiments. This new method and several other normal methods are used in the diagnosis of simulated multi-category faults in turbo-pump. Results show that the diagnosis time is much shorter and the diagnosis correct rate is high with the improved sphere support vector machines.