Model Bias Characterization in the Design Space under Uncertainty

摘要

This paper presents an approach to validate computational models in the design space under uncertainty. The basic idea is to first characterize the model bias; then correct the original model prediction by adding the characterized model bias in the design space. Particularly, a two-step calibration procedure is proposed and the model bias at each design configuration is approximated using the Maximum Entropy Principle (MEP) method. With the characterized model bias at several design configurations, response surface of the model bias is finally constructed to approximate the model bias at any new design configurations. Two examples including a modified vehicle side impact problem and a thermal problem are used to demonstrate the feasibility of the proposed approach.