An Expensive Optimization based Computational Intelligence Method for Railway Track Parameter Identification

January 1, 2016 in

    Conference Paper

    Alfredo Núñez

    Maider Oregui, Maria Molodova


    Proceedings of the 12th International Conference on Computational Structures Technology (CST2014)
    Year: 2014
    Location: Naples, Italy


    Link or Download
    Not available


    To reduce the subjectivity in the identification of the railway tracks parameters, in this paper we propose an expensive optimization procedure using a modified version of Particle Swarm Optimization (PSO) to cope with multiobjective optimization with subjective decisions. From the optimization point of view, the railway track model is a black-box from where given a set of track parameters a non-parametric response is provided. The final goal is to identify in-service railway track parameters which are unknown by fitting simulations to measurements. The simulation is computationally expensive and only 16 licences of the software LS-Dyna are used. The optimization algorithm search over the set of possible track parameters the ones that provide the best performance in terms of multiple objectives. In summary, the methods aim to (1) mathematically determine a good-fit while dealing with a global optimal search over a non-convex optimization problem, (2) speed-up the fitting process, (3) include multiple objectives in the same framework such as robustness to cope with an statistically reliable number of different measurements in one track, good fitting of the main seven characteristics of the railway track, and the overall good fitting of the non-parametric representation of the track response, and (4) include subjectivity of the expert by selecting the best solutions from an α-Pareto solution set.