Strategies for Solving Computationally Expensive Engineering Design Optimization Problems
Slawomir Koziel - Reykjavik University, Iceland
(Chair: Michael Lees)

Abstract : Computer simulation tools have become ubiquitous in contemporary engineering and science over the decade. High-fidelity simulations provide accuracy that is beyond capability of any theoretical models and allow reliable representation of complex components and systems, including coupled and often multi-physics phenomena therein. Modeling accuracy is critical for quality of the design process; however, it comes at a high computational cost. In many engineering areas (electrical, structural, aerospace, etc.) typical simulation times of realistic 3D models are as long as many hours, days, or even weeks. Obviously, this hinders application of simulation tools for solving designs tasks that require multiple evaluations of the system at hand, such as parametric optimization. In this talk, the strategies of reducing computational cost of simulation-driven design are discussed. The focus is on surrogate-assisted design techniques which are the most promising approaches to expedited design involving expensive computer models. Several specific algorithmic frameworks are presented, including space mapping, response correction techniques, feature-based optimization, shape-preserving response prediction, as well as multi-fidelity design. Applications for solving real-world design problems in various engineering fields including electrical and aerospace engineering are also provided.