Webinars

Computational fluid dynamics (CFD) simulations are subject to a variety of uncertainties such as initial conditions, boundary conditions, and choice of model form and parameter values. These uncertainties contribute to CFD results that disagree with test data. Direct sampling of a CFD model can be used to employ various statistical techniques to address these uncertainties; however, the relatively long run time for a CFD simulation typically makes this infeasible.

The solution is to take a surrogate modeling approach using a machine learning (ML) model. The ML model’s rapid prediction of CFD results allows more inputs, scenarios, and design possibilities to be investigated and in less time.

This 60-minute webinar will explore this approach where first an ML model is trained to predict the CFD model’s results before being used in place of the CFD model to run the desired analyses. It will examine current advanced ML tools designed for simulation, digital twin, and other engineering applications.

Topics include:

• The speed and accuracy of ML (aka surrogate) models
• Unique process models for problems with large numbers of inputs or spatially distributed outputs (e.g. full field of CFD results)
• Varying geometry ML models for spatially distributed outputs to cover cases where changes to the model inputs result in changes to the number and/or location of the spatial points
• Achieving better agreement between CFD results and test or experimental data

An audience Q&A session will follow the technical presentation.