AI-enabled vibration sensing system for early detection of trains at active highway-rail grade crossings

Abstract

Highway-rail grade crossings (HRGCs) play an essential role in ensuring the secure traversal of road users across railway tracks. However, despite their significance, they present safety challenges, particularly when trains go undetected, heightening the risk of potential collisions between the road user and train. This paper aims to explore the viability of employing vibration sensors for detection and characterization of an approaching train s speed at HRGCs. The methodology involves analyzing rail vibrations and developing a time series predictive machine learning (ML) model. To accomplish this, a Finite Element (FE) model of a ballasted track railway is created in SAP2000, consisting of essential track components such as rails, sleepers, rail pads, ballast, and subgrade. The acceleration time history of the rails, induced by a moving single car, is recorded, and employed as input data for training and testing a Long Short-Term Memory (LSTM) network. The LSTM model predicts crucial parameters of the approaching train, including its location and speed. Successful prediction of these parameters is anticipated to enable the determination of the precise moment when the train reaches the HRGC, affording road users sufficient reaction time to safely traverse the railway track.