Reduced Order Modelling for Multi-disciplinary Design Optimisation

Aircraft design and optimization in the industry require multiple design rounds usually a trade-off between various objectives and constraints. When multi-disciplinary design optimization (MDO) is of particular interest, the problem becomes more complex because it requires the interaction of different disciplines such as aerodynamics and structural analysis to attain the objectives. Although the field of numerical simulation for solving partial differential equations (PDEs) has significantly developed, enabling the solution to complex dynamics, especially large-scale industrial applications, is still computationally expensive. To meet this interactive requirement in aircraft design optimization, computational methodologies that are fast, reliable, and accurate for routine industry analysis are essential. This work proposes a non-intrusive, data-driven approach for constructing reduced-order models (ROMs) with machine learning (ML) techniques capable of solving PDE-constrained MDO problems. This proposed methodology intends to embed the high-dimensional nonlinear data onto low-dimensional subspace with ML algorithms based on the method of snapshots collected from high-fidelity simulations to make the process of repeatedly solving large-scale MDO problems feasible. However, the training of these reduced-order models is challenging when there are many design parameters to consider, such as in MDO problems. In order to address the challenge associated with training the ROMs, this work will present the algorithms for training the ROM with a piecewise-global reduced-order basis with a confined area of design space. In this article, initial findings pertaining to the prediction of flowfield, coefficients of pressure, as well as forces and moments, using the proposed methodology are presented. These preliminary results are a fundamental step toward the application of this methodology in solving MDO cases. This research brings data-driven multidisciplinary design optimization one step closer to being a practical tool for developing reliable and energy-efficient aircraft configurations, which require routine analyses early in the design cycle.