Forecasting ballast performance under fast heavy haul trains using an Analytical – Machine Learning Track (AMLT) model

For ballasted railway tracks catering to fast heavy-haul trains, it is pertinent to consider the dynamic amplification of ballast permanent response with speeds to optimise track maintenance. This paper presents an analytical-machine learning track (AMLT) model to analyse the influence of heavy-haul trains operating at different speeds on permanent vertical strains (εv) and breakage of ballast (BBI). Using a physics-based analytical model, the elasto-dynamic response considering Rayleigh-wave propagation is captured. This response is then used as an input to data-driven models for εv and BBI developed using a Genetic Algorithm integrated with an Artificial Neural Network (GA-ANN), trained with past laboratory data using relevant input parameters. Results showed that both εv and BBI increase with train speeds, and their amplification is significantly greater than the dynamic stress amplification factor. By using operational thresholds for εv and BBI, a new performance-based limiting speed is proposed which can be used as an alternative to critical speed for heavy-haul trains. In contrast to critical speed, the limiting speed is much lower and is also dependent on the axle load and the age of ballast. Furthermore, the influence of stiff subgrade and higher confining stress on limiting speeds are presented with implications to practice.