A Framework for Parallel Sampling of Design Space With Application to Vehicle Crashworthiness Optimization Available to Purchase

This paper presents a framework for simulation-based design optimization of computationally-expensive problems, where economizing the generation of sample designs is highly desirable. Various meta-modeling schemes are used in practice in order to approximate the input-output relationships in the designed system and suggest candidate locations in the design space where high quality designs are likely to be found. One such popular approach is known as Efficient Global Optimization (EGO), where an initial set of design samples is used to construct a Kriging model, which approximates the system output and provides a prediction of the uncertainty in the approximations. Variations of EGO suggest new sample designs according to various infill criteria that seek to maximize the chance of finding high quality designs. The new samples are then used to update the Kriging model and the process is iterated. This paper attempts to address one of the limitations of EGO, which is the generation of the infill samples often becoming a difficult optimization problem in its own right for a larger number of design variables. This is done by adapting a previously developed approach for locating the optimum of a Kriging model to a modified EGO infill sampling criterion. The new implementation also allows the generation of multiple new samples at a time in order to take advantage of parallel computing. After testing on analytical functions, the algorithm is applied to vehicle crashworthiness design of a full vehicle model of a Geo Metro subject to frontal crash conditions.