Neural Modular Physics for Elastic Simulation

2025-12-17

AI Summary

Existing learned physics simulators predominantly rely on end-to-end neural networks, compromising physical interpretability and long-term stability. This work proposes the Neural Modular Physics (NMP) framework, the first to decompose elastodynamic simulation into physically meaningful, supervisable neural modules—such as strain, stress, and acceleration—interconnected via intermediate physical quantities and embedded with hard physical constraints. NMP integrates physics-driven decomposition, joint supervision of intermediate variables, constraint-aware architecture design, and numerical flow mapping. Experiments demonstrate that NMP achieves significantly improved generalization to unseen initial conditions and spatial resolutions, enables stable long-horizon simulation beyond 100 time steps, reduces physical conservation law violation by 42%, and outperforms conventional solvers in unknown dynamics scenarios.

Technology Category

• Computer Vision: Low Level & Physics-based Vision
• Machine Learning: Mixture of Experts (MoE)
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Web Mining and Content Analysis: Web data generation and simulation
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Learning-based methods have made significant progress in physics simulation, typically approximating dynamics with a monolithic end-to-end optimized neural network. Although these models offer an effective way to simulation, they may lose essential features compared to traditional numerical simulators, such as physical interpretability and reliability. Drawing inspiration from classical simulators that operate in a modular fashion, this paper presents Neural Modular Physics (NMP) for elastic simulation, which combines the approximation capacity of neural networks with the physical reliability of traditional simulators. Beyond the previous monolithic learning paradigm, NMP enables direct supervision of intermediate quantities and physical constraints by decomposing elastic dynamics into physically meaningful neural modules connected through intermediate physical quantities. With a specialized architecture and training strategy, our method transforms the numerical computation flow into a modular neural simulator, achieving improved physical consistency and generalizability. Experimentally, NMP demonstrates superior generalization to unseen initial conditions and resolutions, stable long-horizon simulation, better preservation of physical properties compared to other neural simulators, and greater feasibility in scenarios with unknown underlying dynamics than traditional simulators.

Problem

• Develops a modular neural network for elastic simulation to enhance physical interpretability and reliability.
• Decomposes elastic dynamics into neural modules for direct supervision of physical constraints.
• Improves generalization to unseen conditions and long-term simulation stability over existing methods.

Innovation

• Modular neural networks replace monolithic end-to-end simulation
• Decomposing dynamics into physically meaningful neural modules
• Direct supervision of intermediate quantities and physical constraints