A new metamodel method using Gaussian process based bias function for vehicle crashworthiness design

Abstract

Metamodel-based design optimisation is widely used to obtain a better design efficiently in a vehicle lightweight design. However, one of the major drawbacks of using metamodel in design optimisation is that the accuracy of any metamodel is inherently unpredictable, as the behaviours of vehicle performances are highly nonlinear and most metamodels do not consider the model uncertainty or discrepancy between physical experiments and computer model or metamodel. In this article, the bias function is used to statistically correct the model discrepancy, which is represented by a Gaussian process. A new metamodel method using Gaussian process based bias function is proposed to improve the accuracy of the metamodel. The proposed method is demonstrated through a mathematical example and examined by a vehicle crashworthiness design problem aiming at minimising the weight of front-end structure while satisfying the design target.