A Preliminary Study into the Conceptual Design of Aircraft using Simulation-Based Inference

In this paper, we explore the potential of likelihood-free inference for generating conceptual engineering designs of electric vertical take-off and landing (eVTOL) aircraft. As a preliminary study, our goal is to highlight two key findings: (1) simulation-based inference (SBI) can be effectively scaled to a large and complex multi-physics simulator composed of multiple interacting components, where each is governed by different physical principles; and (2) We introduce a new hierarchical diffusion-based SBI method that successfully captures the governing physical laws of conceptual aircraft design. Using conditioning, we show physical laws are captured at both the component level and the system (full aircraft) level, underscoring promise for future research.