Bayesian System Identification of Rail–Sleeper–Ballast System in Time and Modal Domains: Comparative Study

Abstract

From the literature, time domain and modal domain data are commonly used in system identification and damage detection of various systems. This paper focuses on the comparison between time and modal domain system identification of a rail–sleeper–ballast system, which is modeled with the beam-on-springs theory. Linear elasticity is assumed in modal domain analyses, while the ballast layer is considered elastoplastic, in line with the behavior of ballast under large amplitude vibration in time domain analyses. A simple nonlinear model—the modified Ludwik model—was utilized to capture the strain-hardening behavior of ballast in the tensionless ballast springs. An enhanced Markov chain Monte Carlo (MCMC)-based Bayesian algorithm is utilized to handle the uncertainties associated with the identified system parameters from a probabilistic sense. This algorithm caters for cases that are unidentifiable and where the posterior probability density functions (PDF) are possibly nonGaussian. System identification was carried out using measured data obtained from impact hammer tests under laboratory conditions. Analysis results prove the applicability of the Bayesian algorithm in accurately identifying the severity and location of ballast damage in ballasted tracks. The results also showcase the limitations and merits of system identification of a highly damped system in the time and modal domains. It is concluded that the time domain is more favored than the modal domain for system identification of the considered rail–sleeper–ballast system owing to the effects of ballast nonlinearity under large amplitude vibration.