EVALUATING THE NON-LINEARITY OF RAILWAY BALLAST USING BAYESIAN FRAMEWORK

Non-linearity in the mechanical properties of engineering materials with discrete particles is well-established. Hence, for proper evaluation of the stiffness and resilience of the ballast layer in railway tracks, it is essential to adequately represent this phenomenon in structural models. For structural health monitoring of railway tracks, misrepresenting or neglecting the non-linearity phenomenon will result in over- or under-estimation of the stiffness and stress to which the ballast under a sleeper section is subjected; leading to an inefficient track health evaluation. The investigations reported in this paper were carried out on an indoor rail track panel under laboratory conditions, by an impact hammer test. To ensure that the developed model is able to capture the real behavior of damaged ballast, ballast damage was artificially simulated under a section of the sleeper using smaller-size ballast. A simple model of the track was utilized, by modelling individual sleeper as a beam on a continuous non-linear spring series, which represents the ballast. For model updating purpose, a MCMC-based Bayesian algorithm was employed to identify the model parameters and quantify their uncertainties, using acceleration data collected at thirteen sensors installed along the considered sleeper. Six well-known empirical models for rigid/plastic materials are considered in modelling the behavior of the non-linear spring series; and the Bayesian model class selection
method is used to evaluate the most plausible model class for capturing the mechanics of the ballast particles. The selected model class will be useful in future research works on more accurate ballast damage methodologies.