Modern Modeling and Simulation Powered by Machine Learning

Precision Modeling With Digital Twins Means Faster Production and Lower Costs

Building precise and accurate models is difficult and labor-intensive with current tools. JuliaSim’s unified modeling frameworks for integrating machine learning with traditional simulation in order to overcome the challenge of building predictive models. Pre-start the modeling process with our built-in libraries, use machine learning to automatically discover higher order physics, swiftly calibrate the digital twin to real-world data, use neural surrogates to accelerate the simulation, and generate highly accurate nonlinear controls for deploying on embeded devices.

Integrate with Existing Tools

Engineers can use their models from existing tools. JuliaSim is compatible with the industry standard interface used across virtually all modeling tools - the Functional Mock-up Interface. Generate a Functional Mock-up Unit (FMU) from your existing modeling tools (such as Simulink or Dymola), import that FMU into JuliaSim, train an accelerated surrogate model, export a new FMU and import that into your FMI-compliant tools. Teams not using FMUs can use function registration to embed calls to Julia, R, and Python models from within model blocks.

Build Models Faster

JuliaSim is built on the ModelingToolkit acausal modeling platform. [This system allows for building complex models with orders of magnitude less code and with more code reuse than causal modeling platforms](link to acausal vs causal video). Further increase this productivity by integrating one of JuliaSim’s pre-made models, which span domains from two-phase flow HVAC simulation to battery packs.

Transform Models to Digital Twins

JuliaSim has all of the tools necessary to integrate data into the predictive modeling process. Nonlinear calibration, integrated automatic differenitation, and optimal design tools allow for fast and accurate parameter tuning. Meanwhile techniques like universal differential equations allow for automating the discovery of missing higher-order physics to extend models in a data-driven way.

Ensure a Robust Deployment

JuliaSim understands that the biggest impediment to deploying nonlinear model-predictive controllers and integrating machine learning models into the field is guarenteeing safety. JuliaSim has tools to tune nonlinear controllers using robust and uncertainty-aware techniques, analyze the uncertainty and errors of neural surrogate models, and quantify the parametric uncertainty of model fits.