High-Dimensional Bayesian Optimisation with Large-Scale Constraints via Latent Space Gaussian Processes
Hauke Maathuis, Roeland De Breuker, Saullo G.P. Castro · 2024
Read High-Dimensional Bayesian Optimisation with Large-Scale Constraints via Latent Space Gaussian Processes by Hauke Maathuis, Roeland De Breuker, Saullo G.P. Castro (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-15679.txt.
Introduction
Hauke Maathuis, Roeland De Breuker, and Saullo Castro bring a global search method to aircraft design problems with many variables and many limits, in a 2024 paper. Bayesian optimization searches widely and needs no gradients, but it strains when the design space grows. Their test case is aeroelastic tailoring, where stiffness is tuned so wings flex in helpful ways. Safety rules from many fields must hold at once.
The team shrinks the design space first, using Gaussian process models in the smaller latent space. The method handled a known benchmark with hidden constraints. It then worked on the real tailoring problem, with its large stack of limits. The authors say the result shows global Bayesian search can stay practical for high-dimensional engineering design, not just small problems.
Catalog record
| CREATOR | Hauke Maathuis, Roeland De Breuker, Saullo G.P. Castro |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-15679
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
High-Dimensional Bayesian Optimisation with Large-Scale Constraints via Latent Space Gaussian Processes Hauke Maathuis, Roeland De Breuker, Saullo G.P. Castro Abstract Design optimisation offers the potential to develop lightweight aircraft structures with reduced environmental impact. Due to the high number of design variables and constraints, these challenges are typically addressed using gradient-based optimisation methods to maintain efficiency. However, this approach often results in a local solution, overlooking the global design space. Moreover, gradients are frequently unavailable. Bayesian Optimisation presents a promising alternative, enabling sample-efficient global optimisation through probabilistic surrogate models that do not depend on gradients. Although Bayesian Optimisation has shown its effectiveness for problems with a small number of design variables, it struggles to scale to high-dimensional problems, particularly when incorporating large-scale constraints. This challenge is especially pronounced in aeroelastic tailoring, where directional stiffness properties are integrated into the structural design to manage aeroelastic deformations and enhance both aerodynamic and structural performance. Ensuring the safe operation of the system requires simultaneously addressing constraints from various analysis disciplines, making global design space exploration even more complex. This study seeks to address this issue by employing high-dimensional Bayesian Optimisation combined with a dimensionality reduction technique to tackle the optimisation challenges in aeroelastic tailoring. The proposed approach is validated through experiments on a well-known benchmark case with black-box constraints, as well as its application to the aeroelastic tailoring problem, demonstrating the feasibility of Bayesian Optimisation for high-dimensional problems with large-scale constraints. A Preprint Keywords Bayesian Optimisation \cdot High-Dimensional Input and Output Problems \cdot Aeroelastic Tailoring \cdot Multidisciplinary Design Optimisation \cdot Preliminary Structural Aircraft Design 1 Introduction The design of modern aircraft with enhanced efficiency is crucial for enabling more sustainable aviation. Achieving this involves optimising structural designs to reduce energy consumption. Aeroelastic tailoring emerges as a key technique that has the potential to reduce the weight of aeroelastically efficient high aspect ratio wings. Pioneered by Shirk et al. (1986) , aeroelastic tailoring incorporates directional stiffness properties to effectively carry and control the aeroelastic deformations. Performing aeroelastic tailoring is a multidisciplinary design and optimisation (MDO) effort, involving aerodynamics for the outer-mould shape definition and calculation of the loads acting over the wing; structural design that usually defines the layout of the main structural components of the wingbox; structural analysis to define and evaluate the relevant failure modes that should be considered as constraints; aeroelasticity that couples the aerodynamic loads with the inertial and elastic properties of the wing in order to characterise the flutter behaviour; and optimisation, to properly explore the design space. Other disciplines are also involved, such as manufacturing, typically resulting in additional constraints for the design variables. Evaluating these complex aeroelastic models is computationally expensive, therefore necessitating efficient optimisation algorithms that require fewer analyses before finding an optimum solution. Due to the high number of design variables, describing the structural properties of the system, commonly gradient-based optimisation algorithms are used, leading to an efficient convergence towards the optimal solution. However, the computation of gradients is not always feasible, especially if the model’s source code is unavailable. In such cases, the model must be treated as a black box, relying on methods like finite differences to obtain the design sensitivities, which can lead to prohibitively high computational costs that would ultimately motivate the use of gradient-free methods. Furthermore, many engineering problems, such as noisy responses or experimental results, possess inherent complexities that can render gradient-based approaches less effective or even impractical. Additionally, the response surface for feasible designs in aeroelastic tailoring is often multi-modal. This complexity can cause gradient-based methods to become trapped in local optima, overlooking the broader global design space and hindering the discovery of superior designs. Therefore, it is essential to develop methods that efficiently explore the global design space, optimising structures to achieve lighter aircraft configurations. The optimisation problem at hand can be formulated as follows: \min_{\textbf{x}\in\mathcal{X}\subset\mathbb{R}^{D}}f(\textbf{x})\ \text{s.t.}\forall i\in\{1,...,G\},c_{i}(\textbf{x})\leq 0, (1) where \mathcal{X}\subset\mathbb{R}^{D} is a D -dimensional space of potential designs, f(\textbf{x}):\textbf{x}\in\mathcal{X}\to\mathbb{R} the objective function and G constraints arising from the multi-disciplinary analyses. Overall, aeroelastic tailoring can be seen as an optimisation problem consisting of high-dimensional inputs and outputs, where the utilised models are able to map the vector of design variables to the objective function f(\textbf{x})\in\mathbb{R} and all G constraints \textbf{c}(\textbf{x})\in\mathbb{R}^{G} . The simultaneous consideration of multiple disciplines can lead to large-scale constraints where G\gg 10^{3} , combining buckling, aeroelastic stability, maximum stress, maximum strain, and various others. In aeroelastic tailoring, the optimal stiffness distribution is achieved by means of a sizing optimisation that, in the case of laminated composite wings consists of finding the best set of lamination parameters and the optimum thickness for one or more composite regions ( Werter, 2017 ) . Lamination parameters allow a condensed and theoretical representation of the membrane, bending, and coupled stiffness terms of a laminate with continuous variables ( Dillinger et al., 2013 ) , making the sizing optimisation more convex and more adequate to established continuous optimisation techniques, where the design variables can be treated as continuous variables. Once this sizing optimisation is complete, a second discrete optimisation is performed to retrieve a manufacturable set of ply orientations. Yet, the presence of multiple design regions to maintain design freedom can still result in the number of design variables being in the order of hundreds or thousands. Given the expensive nature of evaluating an aeroelastic model to obtain the objective function values and associated constraints, a sample-efficient optimisation algorithm is crucial. Compared to other gradient-free approaches like Random Search, Genetic Algorithms, and others, Bayesian Optimisation (BO) has proven to be a powerful method for complex and computationally costly problems ( Mockus, 1989 ) and has been extensively applied across various domains, including aircraft design ( Saves et al., 2022 ) . BO addresses the challenge of expensive evaluations by using computationally inexpensive probabilistic surrogate models, such as Gaussian Processes ( \mathcal{GP} ). These models replace the black-box functions representing the objective and constraints, significantly improving optimisation efficiency ( Frazier, 2018 ) . While many authors have shown that for lower dimensional problems, BO methods perform well, high-dimensional cases pose significant challenges due to the curse of dimensionality ( Eriksson and Jankowiak, 2021 ; Priem, 2020 ) , resulting from the fact that high dimensional search spaces are difficult to explore exhaustively. However, BO offers a probabilistic approach to efficiently search the design space to find promising regions and to reduce the computational burden. While these algorithms offer a variety of advantages, including the learning-from-data aspect, uncertainty quantification, the lack of need for gradients, the ability to fuse data in a multi-fidelity context, and the capability to learn the correlation between simulation and experimental data, their scalability to high-dimensional problems with many constraints, as is often the case in engineering design, remains a significant challenge. The present study focuses on employing high-dimensional BO algorithms for aeroelastic tailoring while considering large-scale constraints arising from the multidisciplinary analyses, as formulated in Equation 1 . The novelty of this paper lies in the formulation of a high-dimensional BO method with a dimensionality reduction approach that significantly lowers the computational burden arising from the incorporation of a large number of constraints. Subsequently, the methodology is applied to the 7D speed reducer problem with