Enhancing collaborative multidisciplinary optimization through surrogate modeling

by Sharmili, Nowsheen, Lee, Yong Hoon and Allison, James T.

Abstract:

This paper presents a thorough examination of Multidisciplinary Optimization (MDO) strategies, with a particular focus on refining Collaborative Optimization (CO) techniques tailored for managing large, complex discipline groups. It delves into the complexities inherent in MDO, a field that necessitates the harmonious integration of multiple engineering disciplines to optimize overall system performance efficiently. The analysis spotlights the limitations of traditional CO methods, notably the tendency for optimization searches to converge on local subsystem minima and the overall slower convergence rates these methods generally exhibit. To address these challenges, this article introduces an innovative approach designed to improve optimization accuracy while simultaneously reducing computational costs. This method integrates initial design variable sampling and surrogate modeling at the subsystem-level optimization process. By employing these techniques, the proposed approach aims to refine the accuracy of optimization outcomes and significantly reduce the computational burdens often associated with traditional CO methods. The findings from this study highlight the benefits of the proposed method, particularly in terms of reduced computational demands, making it a practical and attractive option for industrial applications, especially in settings where disciplinary groups operate largely independently and benefit may be derived from parallel subproblem formulation and refinement. This enhanced efficiency and reduced cost make the new approach a compelling alternative for complex, interdisciplinary system optimizations.

Reference:

Nowsheen Sharmili, Yong Hoon Lee and James T. Allison, "Enhancing collaborative multidisciplinary optimization through surrogate modeling", in ASME IDETC/CIE Conference, DETC2024-148473, Washington, DC, USA, August 2024, pp. 1–4 (Extended Abstract).

Bibtex Entry:

@presentation{Sharmili2024IDETC,
    author = {Sharmili, Nowsheen and Lee, Yong Hoon and Allison, James T.},
    title = {Enhancing collaborative multidisciplinary optimization through surrogate modeling},
    booktitle = {ASME IDETC/CIE Conference},
    address = {Washington, DC, USA},
    year = {2024},
    month = aug,
    day = {25--28},
    number = {DETC2024-148473},
    pages = {1--4},
%    pdf = {},
%    doi = {},
%    gsid = "",
    note = {Extended Abstract},
    abstract = {This paper presents a thorough examination of Multidisciplinary Optimization (MDO) strategies, with a particular focus on refining Collaborative Optimization (CO) techniques tailored for managing large, complex discipline groups. It delves into the complexities inherent in MDO, a field that necessitates the harmonious integration of multiple engineering disciplines to optimize overall system performance efficiently. The analysis spotlights the limitations of traditional CO methods, notably the tendency for optimization searches to converge on local subsystem minima and the overall slower convergence rates these methods generally exhibit. To address these challenges, this article introduces an innovative approach designed to improve optimization accuracy while simultaneously reducing computational costs. This method integrates initial design variable sampling and surrogate modeling at the subsystem-level optimization process. By employing these techniques, the proposed approach aims to refine the accuracy of optimization outcomes and significantly reduce the computational burdens often associated with traditional CO methods. The findings from this study highlight the benefits of the proposed method, particularly in terms of reduced computational demands, making it a practical and attractive option for industrial applications, especially in settings where disciplinary groups operate largely independently and benefit may be derived from parallel subproblem formulation and refinement. This enhanced efficiency and reduced cost make the new approach a compelling alternative for complex, interdisciplinary system optimizations.},
}