Track Geometry Estimation and Hanging Sleeper Detection Using Vehicle Dynamic Responses with Unknown System Parameters

Abstract

This study presents an efficient, but effective, approach to infer railroad track geometries and locate hanging sleepers from the car body responses. A half-car model is employed to represent the bounce and pitch motions of the car body, and an augmented state space model is derived in which the track geometry is treated as a state variable. In the Kalman filter process, both the process and observation noise covariance matrices are adaptively updated by the Robbins-Monro algorithm, and the resulting state estimates are further smoothed using the Rauch-Tung-Striebel smoother. In addition, the output-only model optimization can determine the unknown vehicle parameters by minimizing a novel cost function that accounts for engineering constraints on the railroad track geometries. Local discrepancy in the inferred track geometries using two vehicles of different weight is then used as an indicator to identify hanging sleepers. The feasibility of the proposed framework is first demonstrated upon a numerical simulation, and it is furthermore applied to the experimental measurements on passenger vehicles in Japan.