Emerging Methods for 2D and 3D Physics Simulation to Support Early Design Decisions

Abstract

Abstract Current approaches to developing 2D and 3D dynamic simulations can be time-intensive to develop and run early in the design process. Recently, machine learning-driven methods have produced near-real-time predictions of design performance. As these approaches rely on backpropagation methods, designers cannot inspect the meaning of results as they pertain to a single target — for example, object class identification or efficiency estimation. In addition, machine learning does not usually generate 2D and 3D field results. This paper identifies and classifies emerging dynamic field simulation methods and previews a novel method for accurate and efficient dynamic simulation. A literature review of recent peer-reviewed journals and books, as well as an exploration of emerging simulation software, will be presented in this paper. A grounded theory approach will be used to extract and classify key concepts. Preliminary examples of interest include cellular automata, decision trees, consensus or crowd-based methods, and hybrid machine learning methods. This study is intended to provide engineers with a better understanding of the emerging capabilities of state-of-the-art simulations in industries parallel and orthogonal to mechanical engineering theory, including gaming, film, and architecture. The work examines a multi-physics challenge in designing electric propulsion systems for spacecraft as dynamic field models for this design space are computationally intensive to produce.