Improving the accuracy of vehicle crashworthiness response predictions using an ensemble of metamodels

Due to the scale and computational complexity of current simulation codes for vehicle crashworthiness analysis, metamodels have become indispensable tools for exploring and understanding the design space. Traditional application of metamodelling techniques is based on constructing multiple types of metamodels based on a common data set, selecting the most accurate one and discarding the rest. However, this practice does not take full advantage of the resources devoted for constructing different metamodels. This drawback can be overcome by combining individual metamodels in the form of an ensemble. Two case studies with a high-fidelity finite element vehicle model subject to offset-frontal and side impact conditions are presented for demonstration. The prediction accuracies of the individual metamodels and the ensemble of metamodels are compared, and it is found for all the crash responses of interest that the ensemble of metamodels outperforms all individual metamodels. It is also found that as the number of metamodels included in the ensemble increases, the prediction accuracy of the ensemble of metamodels increases.