Gradient-enhanced robust design optimization for engineering systems under uncertainty
Loading...
Authors
Bedonian, Garo
Issue Date
2025-05
Type
Electronic thesis
Thesis
Thesis
Language
en_US
Keywords
Mechanical engineering
Alternative Title
Abstract
Engineers are interested in numerical robust design optimization (RDO) during early phases of the design process for its ability to produce system configurations that are both optimally performant and robust to sources of uncertainty in fabrication, operation, and analysis. Prohibitive to this practice is the often intractable cost of accurately estimating statistical or probabilistic measures of the optimization objectives or constraints from repeated sampling of analysis codes, particularly for problems involving expensive high-fidelity or multi-disciplinary analyses. Given the recent proliferation of efficient gradient calculation methods within these analysis codes, we explore and contribute to gradient-based methods for accelerating robust design optimization, simultaneously leveraging gradient-enhanced uncertainty quantification (UQ) and gradient-based optimization techniques. To this end, we develop a novel gradient-based partition-of-unity surrogate model and adaptive sampling method tailored to robust design optimization. Furthermore, we propose a multi-fidelity robust optimization method that uses surrogate-based UQ and adaptive sampling-based gradient error estimates to mitigate the cost of sampling effort as the optimizer approaches convergence. For a suite of analytical test problems and an aerostructural design problem involving uncertainties related to shock-boundary layer interaction, the novel surrogate and adaptive sampling method demonstrate competitive to superior global accuracy per sample compared to standard surrogates, and the proposed multi-fidelity optimization method demonstrates greatly-reduced sampling effort to achieve design convergence overall when compared to single-fidelity benchmarks.
Description
May2025
School of Engineering
School of Engineering
Full Citation
Publisher
Rensselaer Polytechnic Institute, Troy, NY