Non-intrusive Learning of Physics-Informed Spatio-temporal Surrogate for Accelerating Design

2026-04-15 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Physics-Informed Modeling · Depth: Expert, quick

Summary

A new Physics-Informed Spatio-temporal Surrogate Modeling (PISTM) framework has been developed to address the computational expense of multi-physics simulations for nonlinear spatio-temporal dynamical systems. High-fidelity simulations, while capturing fine scales, create a bottleneck in engineering design due to their high computational cost. While purely data-driven spatio-temporal surrogate models offer speed improvements, they often lack generalizability to inputs outside their training distribution. The PISTM framework integrates physics constraints with Koopman autoencoders to learn underlying spatio-temporal dynamics non-intrusively. It also incorporates a spatio-temporal surrogate model that predicts Koopman operator behavior over a specified time window for unknown operating conditions. The framework's efficacy was evaluated using a two-dimensional incompressible fluid flow around a cylinder.

Key takeaway

For engineering design teams struggling with computationally expensive multi-physics simulations, the PISTM framework offers a path to significantly reduce simulation time while maintaining accuracy and improving generalizability. You should consider integrating physics-informed machine learning techniques like PISTM to overcome the limitations of purely data-driven surrogate models, especially when dealing with complex nonlinear spatio-temporal systems and diverse operating conditions.

Key insights

PISTM combines physics constraints and Koopman autoencoders for generalizable spatio-temporal surrogate modeling.

Principles

• Physics constraints improve model generalizability.
• Koopman operators can model complex dynamics.

Method

The PISTM framework uses Koopman autoencoders to learn spatio-temporal dynamics, then employs a spatio-temporal surrogate model to predict Koopman operator behavior under new conditions, all constrained by underlying physics.

In practice

• Apply PISTM to fluid flow problems.
• Use Koopman autoencoders for dynamic systems.

Topics

• Physics-Informed Surrogate Modeling
• Spatio-temporal Dynamics
• Koopman Autoencoders
• Multi-physics Simulations
• Engineering Design

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Engineering Hybrid Physics-Informed Neural Networks for Next-Generation Electricity Systems: A State-of-the-Art Review — Physics-informed machine learning integrates domain physics into models, enhancing accuracy and efficiency for data-scarce electricity systems.
• AI-powered design comes to high-speed racing — AI surrogate models, like GIST, can drastically accelerate complex physics simulations while maintaining high accuracy.
• MoSA: Motion-constrained Stress Adaptation for Mitigating Real-to-Sim Gap in Continuum Dynamics via Learning Residual Anisotropy — MoSA learns residual anisotropy and heterogeneity using physics priors to close the real-to-sim gap in continuum dynamics.
• Mixing generative AI with physics to create personal items that work in the real world — PhysiOpt combines generative AI with physics simulations to create functional, real-world 3D-printable designs.
• Learning Physical Operators using Neural Operators — A physics-informed framework improves neural operator generalization and temporal continuity for PDE solving via operator splitting.
• Learning universal approximations for partial differential equations with Physics-Informed Broad Learning System — PIBLS offers a backpropagation-free, least-squares optimization approach for PDEs, achieving faster, more accurate solutions than PINNs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.