Research on a Multi-Fidelity Surrogate Model Based Model Updating Strategy

Abstract

In vehicle design modeling and simulation, surrogate model is commonly used to replace the high fidelity Finite Element (FE) model. A lot of simulation data from the high-fidelity FE model are utilized to construct an accurate surrogate model requires. However, computational time of FE model increases significantly with the growing complexities of vehicle engineering systems. In order to attain a surrogate model with satisfactory accuracy as well as acceptable computational time, this paper presents a model updated strategy based on multi-fidelity surrogate models. Based on a high-fidelity FE model and a low-fidelity FE model, an accurate multi-fidelity surrogate model is modeled. Firstly, the original full vehicle FE model is simplified to get a sub-model with acceptable accuracy, and it is able to capture the essential behaviors in the vehicle side impact simulations. Next, a primary response surface model (RSM) is built based on the simplified sub-model simulation data. Bayesian inference based bias term is modeled using the difference between the high-fidelity full vehicle FE model simulation data and the primary RSM running results. The bias is then incorporated to update the original RSM. This method can enhance the precision of surrogate model while saving computational time. A real-world side impact vehicle design case is utilized to demonstrate the validity of the proposed strategy.