Extreme Function Theory for Novelty Detection: A Case Study for Vehicles—Guideway System

Abstract

This paper focuses on detecting the irregularity of track geometry using the extreme value theory. The vehicle–railway system is simplified into a system of 2-degree-of-freedom motions, and the random dynamic load input is the displacement of vehicle caused by the original defects or deformation of rails. Blackman-Turkey method is adopted to cope with the surrogate model and to generate the dynamic response of vehicle. The responses of vehicle are used for observations of normal and abnormal models to conduct novelty classification by the extreme value theory. The difference between normal and abnormal models is converted to the distinction of the tail distributions, which converges to the generalized extreme value distribution regardless of the parent distribution. The parameters of the generalized extreme value distribution are estimated with the method of maximum likelihood. The threshold for irregularity classification is estimated according to the pre-desired confidence level. A case study of running trains with the dynamic response to vertical profile irregularity verifies the method well.