A preference-performance hybrid method for surrogate model updating in engineering design optimisation

This paper addresses the critical issues of effectiveness, efficiency and reliability in simulation-based design optimisation under surrogate model uncertainty. Specifically, it presents a novel method to build surrogate models iteratively with sufficient fidelity for accurately capturing global optimal design solutions at a minimal computational cost. The proposed preference-performance hybrid method is the integration of the multiple preference point method, which updates surrogate models at current local optimal points, and the performance-based maximum variance point method, which updates surrogate models at the points that are associated with maximum prediction variances. The salient features of this method include its strategic utilisation of both local exploitation and global exploration tactics and its unique ability to overcome the Smoothing Out the Best Point (SOBP) problem. Through illustrative comparison studies on 30 different optimisation scenarios from 15 different test functions, the paper demonstrates the tangible advantages of the proposed hybrid method in improving effectiveness, efficiency and reliability in surrogate model-based design for finding the global optima in a wide range of design scenarios, including those characterised by multiple global solutions, irrespective of whether the optima exist at the corners, on the boundaries or in the interiors of the design space.