AI for Physics and Engineering Simulations

Our team is leveraging AI to develop novel algorithms that accelerate computer simulations of complex physical and engineering systems.

Many natural and engineered phenomena are governed by complex ordinary and partial differential equations (ODEs/PDEs). Traditional solvers require significant domain expertise and can only handle these problems on the world’s largest supercomputers—expensive resources typically reserved for a small number of applications. Even then, a single simulation may take weeks to complete, and its results cannot be directly reused.

AI-driven methods are changing this paradigm. By learning the underlying physics from simulation data, AI can build surrogate models that deliver high-fidelity results in seconds, enabling faster innovation across science and engineering.

Our team is actively advancing new algorithmic and architectural approaches for AI-driven fluid simulation. We are developing neural surrogate models that more efficiently learn complex flow behavior from high-fidelity simulation data, with an emphasis on generalization across geometries and operating conditions. In a recent collaboration with a major race car manufacturer, these methods showed that, given only the vehicle geometry, it is possible to rapidly approximate full fluid flow and pressure fields over the car. This work highlights how continued innovation in algorithms, and not just compute, can unlock faster, more accurate design workflows and help push the boundaries of what’s possible in scientific and engineering simulation.