Multi-Fidelity Design Optimization: Challenges In Complex Physics-based Computational Mechanics

Abstract

Engineering systems development is currently pushing the envelope of traditional multidisciplinary design capabilities. Bringing multiple physics into the design loop earlier in the design process has shown promise in handling the strict requirements constantly being placed on various areas of computational mechanics, such as the design of next generation military aircraft. The goal in multi-fidelity design is to aid in this process to expand traditional design capabilities through the implementation of techniques developed to mitigate inadequacies and/or obstacles associated with various levels of complex physics in a single design process. Achieving a desired level of accuracy while maintaining a low computational cost may very well be the greatest obstacle combating computational design. However, other hindrances exist such as
determining the appropriate physics (i.e. acoustic, thermal, structural), level of physics (i.e. Potential Flow, Euler, Navier Stokes), and mesh refinement to utilize in any given computational model. This work focuses on leveraging higher fidelity information to correct lower fidelity models so as to take advantage of the speed associated with the latter without compromising accuracy. Corrections are implemented via a custom Hybrid Bridge Function (HBF) while the design aspect is governed through the implementation of a special Trust Region Model Management (TRMM) methodology. Multifidelity design optimization is demonstrated on a thermal plate demonstration problem consisting of four differing levels of fidelity. Results show that employment of the described methodology succeeds in obtaining a design at a lower cost while maintaining a necessary level of accuracy.