Toward a Domain-Agnostic Computational Architecture for Engineering Design
This paper introduces the Computational Engineering Model (CEM), a five-layer architecture — codified physics, high-fidelity solvers, learned surrogates, multidisciplinary optimization, and active learning — built so that its computational shape stays fixed while the physical domain it operates on is swappable. A strict interface boundary keeps any orchestration layer, human or AI-driven, from substituting for computation rather than directing it. We report a working implementation of the design synthesis layer for fixed-wing UAV and multirotor vehicle families, and lay out the architecture's case for generalizing to further physical domains.
A layered architecture and design synthesis implementation that separates AI orchestration from physics computation by construction, evaluated on two vehicle families as a first domain.