Rail Roughness Identification Via On-Board Acceleration Data and Bayesian Filtering

Abstract

The global demand for transportation has resulted in extensive expansion of railway networks. However, ensuring the safety, reliability, and efficiency of these rapidly expanding railway infrastructures requires monitoring of their structural health. Focusing on tracks, traditional visual inspections and portable measuring devices are commonly used to gather geometric data for diagnosing and predicting track defects. In recent years, railway operators worldwide have employed specialized diagnostic vehicles equipped with optical and inertial sensors to collect track data and assess its condition. This approach has revolutionized rail condition assessment by introducing a mobile data acquisition platform for track inspection. Nevertheless, deploying these specialized vehicles disrupts regular rail service, limiting their frequency of operation and the continuous collection of rail data. To address this limitation, this study explores an on-board monitoring (OBM) method that focuses on collecting vibration data from traveling trains. The proposed methodology involves gathering acceleration data from axle boxes of trains running at normal speeds. What sets this approach apart is its use of realistic train models and the consideration of the dynamic interaction between the trains and tracks, which is typically oversimplified. The train model employed is simplified to reduce computational requirements. The identification process relies on sequential Bayesian inference for joint input and state estimation. By estimating the input, the relevant rail roughness profile can be identified, thereby providing information on the presence of isolated defects, such as welded joints and squats, along the track system.