Design optimisation of vehicle roof structures: benefits of using multiple surrogates

Abstract

Rollover accidents have become more and more of an important issue for vehicle safety nowadays. On the basis of the final rule of Federal Motor Vehicle Safety Standard No. 216 (FMVSS 216), design optimisation of vehicle upper structures is performed to improve roof strength. Surrogate-based design optimisation is utilised to facilitate design space exploration and optimum search rapidly, in which the response of roof crush resistance force is approximated by surrogate models. Surrogates like kriging, radial basis function, support vector regression, etc. are used. Weighted average surrogates, including prediction -sum-of-squares-based (PRESS-based) weighted surrogate (PWS) and optimal weighted surrogate (OWS), constructed by above-mentioned individual surrogates, are also taken into consideration. The benefits of using the weighted average surrogate for function approximation and optimisation are investigated. The results demonstrate that OWS outperforms others in the context of prediction accuracy, and it also achieves the best improvement on crush resistance force by 41.7%, accompanied by 5.3% weight reduction compared with the original one. It seems an effective way to form an ensemble of surrogates to avoid a misleading optimum. The results also indicate that different surrogates would prefer different regions of design space. This indicates that simultaneous use of multiple surrogates could eliminate the surrogate dependency optimum, showing potential for surrogate-based design optimisation of non-linear problems.