Enhancements in the aircraft preliminary design methodologies: adoption of sustainable high-fidelity simulations

Aircraft preliminary design requires the capability of rapidly exploring several different configurations without compromising the accuracy of the different physics involved, such as aerodynamics, structural analysis and low observability.
In this work, previously presented at the 10th CEAS Aerospace Europe Conference and 28th AIDAA International Congress, we propose a multi-fidelity methodology combining analysis with different levels of fidelity, where the high-fidelity is meant to be supported by Reduced Order Models (ROM).
The purpose is to significantly accelerate the decision process, increasing the quality of the design already in the preliminary phases.
Inspired by a previous ESTECO’s experience [1], the methodology is applied by Leonardo Aeronautics in a study-case based on an example configuration of a fighter jet. The ultimate goal is to include this methodology among the tools available for preliminary aircraft design [2].
The design workflow begins with a conceptual phase that involves the rapid generation of several configurations through low-fidelity simulations. The purpose is not only to get a preliminary evaluation of the performances and requirements compliance, but also to find a region of interest in the input domain (mostly sizing parameters) where to concentrate the analysis effort. This step is crucial as integrating multiple disciplines, such as aerodynamics, structural and low observability, and formulating a multi-objective optimization problem, leads to having dozens or hundreds of decisional parameters, each with a potentially wide variation range. In those cases, an exhaustive exploration of the solution domain with high fidelity simulations is almost impossible due to the combinatorial explosion.
Inside the region, identified through low-fidelity simulations, with the construction of a Design of Experiments (DOE) Table containing different possible aircraft configurations was developed. This DOE was used to generate different geometries through a parametrized CAD, and to perform the high-fidelity simulations for the two selected physics: CFD for the aerodynamics and electromagnetic fields for the estimation of the radar cross-section (RCS).
The two analyses that are performed on the same starting geometry have similar computational costs; this enables them to run in parallel to obtain a set of high-fidelity solutions.
The resulting set can be used both to confirm the quality of the selected region and as a training set for the development of the non-intrusive ROM models, which do not intervene in the Full Order Model (FOM) simulation, treated as a black box, but instead operate purely on its inputs and outputs. The ROM used in this study, based on Proper Orthogonal Decomposition (POD) techniques and non-linear regression models [3], replace the high-fidelity simulations with highly reliable predictions computed with a reduction in computational time by several orders of magnitude.
During the model’s training phase, the solutions computed using the Full-Order Model (FOM), obtained for different parametrized geometries, are projected onto the basis of the main modes (POD), determining the expansion’s coefficients.
The generation of a solution (RCS and CFD) for a new geometry is simplified by predicting, using regression models, the relevant expansion’s coefficients of the POD.
This approach enables a detailed exploration of the solution space with limited computational costs, making it possible to converge on engineering-feasible configurations within the timeframes of the preliminary design phase.
During tests conducted on realistic aircraft configurations, the mean errors on the most relevant Quantity of Interest (QOI) were below 5%, reducing the computational time for a single simulation from hours to seconds.
The management of the whole process relies on the VOLTA platform, which orchestrates the integration of heterogeneous computational environments: the Davinci HPC cluster by Leonardo (Linux) for high fidelity simulations and Windows workstations for CAD modeling. Beyond the computational aspect, VOLTA facilitates collaboration among the various roles involved in the design process: designers, CFD analysts, RCS specialists, structural engineers, and decision makers can work synergistically within a single infrastructure, tracking every change and ensuring consistency throughout the entire workflow. The presented methodology demonstrates significant potential for improving the quality of the preliminary design simulations without increasing the time required. It represents a concrete step towards the structured adoption of Digital Twins in the aerospace industry, enabling multidisciplinary, collaborative design that is both data-driven and based on efficient multifidelity simulations.

[1] Mull, K. M. (2020). Expanded MDO for Effectiveness Based Design Technologies: The EXPEDITE Program and Successes with ESTECO Technologies.
[2] Visconti, L. (2024). Ottimizzazione multidisciplinare di un velivolo fighter nella fase di progetto preliminare - Master Thesis developed at Leonardo Aircraft.
[3] Bui-Thanh, T., Damodaran, M., and Willcox, K. (2003). Proper Orthogonal Decomposition Extensions for Parametric Applications in Compressible Aerodynamics. In 21st AIAA Applied Aerodynamics Conference. Orlando, FL. AIAA 2003-4213.