Introducing the AI Physics Operating System

The complete Rescale AI Physics Operating System provides a guided path for simulation teams to transform their own data into production-ready surrogate models, improving simulation speed and productivity. This integrated environment manages data, models, training, versioning, and inference, making surrogate modeling practical for non-AI experts.

Overview

The AI Physics OS makes it practical for non-AI experts to train their first surrogate model, for a range of solvers using production ready model architectures. 

Highlights

What is it? 

The Rescale AI Physics Operating System (OS) provides a guided, end-to-end way to turn simulation data into high-accuracy surrogate models with fast and flexible inference. The AI Physics OS has tools for preparing simulation data for model training, and an interactive application for reviewing and editing AI Datasets, training and validating models, and running inference. You can do deeper dataset and model analysis with the integrated Jupyter Notebook interface, and a co-pilot provides assistance based on natural language prompts. The inference framework includes a local desktop runner and client to download models from Rescale and run inference on any workstation or laptop. Additionally, the inference client has standard APIs, so that you can build out custom interfaces for running inference. 

Why did we build it?

Surrogate modeling has typically required users to develop their own scripts and model training pipelines to transform simulation data, integrate with open source model architectures, and manage GPU resources. The AI Physics OS leverages the solver-agnostic Rescale simulation platform to integrate surrogate model building into existing simulation processes with managed model architecture integration and GPU infrastructure. 

How is it used? 

At a high level, the AI Physics OS supports this flow:

  1. Prepare simulation data: Add Automations to your simulation jobs to extract global inputs and outputs, and convert simulation result files to model training compatible formats. 
  2. Create AI Datasets: Create an AI Dataset from a collection of simulation jobs representing the design space for the surrogate model to cover. 
  3. Train and validate models: 
    • Train a model based on a given AI Dataset by selecting from best-in-class model architectures (DoMINO, TwoStream Transformer, MeshGraphNet, and GeoTransolver), with default or advanced parameters on optimized GPU hardware. 
    • Review built‑in evaluation dashboards to verify model quality and understand applicability.
  4. Run inference:
    • Vary global inputs and geometry to visualize surrogate model predictions in AI Physics Training and AI Physics Inference
    • Build out custom inference interfaces using a standard API, allowing you to run custom inference on local environments
    • Integrate inference into CAD native environments with Rescale AI plug-ins for Blender and Autodesk Alias

Getting Started

Contact your Rescale representative to enable it for your account.

Follow the guidance and step-by-step tutorials in the AI Physics Featured Guide.

×