Towards Generative Design Using Optimal Transport for Shape Exploration and Solution Field Interpolation

Abstract

Generative Design (GD) combines artificial intelligence (AI), physics-based modeling, and multi-objective optimization to autonomously explore and refine engineering designs. Despite its promise in aerospace, automotive, and other high-performance applications, current GD methods face critical challenges: AI approaches require large datasets and often struggle to generalize; topology optimization is computationally intensive and difficult to extend to multiphysics problems; and model order reduction for evolving geometries remains underdeveloped. To address these challenges, we introduce a unified, structure-preserving framework for GD based on optimal transport (OT), enabling simultaneous interpolation of complex geometries and their associated physical solution fields across evolving design spaces, even with non-matching meshes and substantial shape changes. This capability leverages Gaussian splatting to provide a continuous, mesh-independent representation of the solution and Wasserstein barycenters to enable smooth, mathematically ''mass''-preserving blending of geometries, offering a major advance over surrogate models tied to static meshes. Our framework efficiently interpolates positive scalar fields across arbitrarily shaped, evolving geometries without requiring identical mesh topology or dimensionality. OT also naturally preserves localized physical features -- such as stress concentrations or sharp gradients -- by conserving the spatial distribution of quantities, interpreted as ''mass'' in a mathematical sense, rather than averaging them, avoiding artificial smoothing. Preliminary extensions to signed and vector fields are presented. Representative test cases demonstrate enhanced efficiency, adaptability, and physical fidelity, establishing a foundation for future foundation-model-powered generative design workflows.