Development of an optimization framework for solving engineering design problems.

Liu, Chengyang (2020) Development of an optimization framework for solving engineering design problems. Doctoral thesis, University of East Anglia.

[thumbnail of 2020LiuCPhD.pdf]
Download (2MB) | Preview


The integration of optimization methodologies with computational simulations plays a profound role in the product design. Such integration, however, faces multiple challenges arising from computation-intensive simulations, unknown function properties (i.e., black-box functions), complex constraints, and high-dimensionality of problems. To address these challenges, metamodel-based methods which apply metamodels as a cheaper alternative to costly analysis tools prove to be a practical way in design optimization and have gained continuous development. In this thesis, an intrinsically linear function (ILF) assisted and trust region based optimization method (IATRO) is proposed first for solving low-dimensional constrained black-box problems. Then, the economical sampling strategy (ESS), modified trust region strategy and self-adaptive normalization strategy (SANS) are developed to enhance the overall optimization capability. Moreover, as the radial basis function (RBF) interpolation is found to better approximate both objective and constraint functions than ILF, a RBF-assisted optimization framework is established by the combination of the balanced trust region strategy (BTRS), global intelligence selection strategy (GIS) and early termination strategy (ETS). Following that, the fast computation strategy (FCS) and successive refinement strategy (SRS) are proposed for solving large-scale constrained black-box problems and the final optimization framework is called as RATRLO (radial basis function assisted and trust region based large-scale optimization framework). By testing a set of well-known benchmark problems including 22 G-problems, 4 engineering design problems and 1 high-dimensional automotive problem, RATRLO shows remarkable advantages in achieving high-quality results with very few function evaluations and slight parameter tuning. Compared with various state-of-the-art algorithms, RATRLO can be considered one of the best global optimizers for solving constrained optimization problems. Further more, RATRLO provides a valuable insight into the development of algorithms for efficient large-scale optimization.

Item Type: Thesis (Doctoral)
Faculty \ School: Faculty of Science > School of Mathematics
Depositing User: Chris White
Date Deposited: 21 Apr 2021 11:20
Last Modified: 21 Apr 2021 11:20


Downloads per month over past year

Actions (login required)

View Item View Item