Abstract: In order to implement the fault diagnosis of Chongqing light-rail’s screws，a new method of screw fault diagnosis based on Support Vector Data Description（SVDD） is proposed，which can found one-class classifier with normal screw samples only and doesn’t need feature extracting.The method solves the difficult problem of lacking fault screw samples.Compared with three familiar one-class methods，it is found that the SVDD has stronger classification ability and higher efficiency，which can distinguish the normal and abnormal screws ideally，and provides a new method for light-rail’s screws fault diagnosis.

摘要： 为了实现对重庆市轻轨轨道梁锚固螺杆的故障检测，提出了一种基于支持向量数据描述的锚固螺杆故障诊断方法，该方法只需要正常螺杆样本，且不需要对原始数据进行特征提取，就可以建立单值分类器，解决了缺少故障螺杆样本的难题。通过与常见的三种单值分类方法比较，表明SVDD分类器具有很好的分类效果和计算效率，能较好地区分正常螺杆和非正常螺杆，为轻轨锚固螺杆故障检测提供了新的诊断方法。