A machine learning (ML) based surrogate model for Hypersonic Flows

There is a need to enhance and accelerate the execution of CFD (computational fluid dynamics) models using artificial intelligence (AI) to assist with handling of problems with a high-dimensionality field and high computational cost. Specifically, hypersonic and combustion flow problems are one of the most challenging CFD problems which involve complex reaction mechanisms and demand large computational time/resources. One of the primary challenges associated with simulating flows with chemical reactions is that the chemical reactions occur on timescales that are orders-of-magnitude smaller than the flow physics, which in turn increases the computational cost significantly. Many models spend over 90% of their runtime simulating chemical kinetics to facilitate coupling with high resolution CFD models and to support large chemical mechanisms. Stiff chemical kinetics require smaller timesteps than any other process (e.g., diffusion, heat conduction, and flow), which increases the computation time for reactive flows, including hypersonic flows.

To address these challenging problems, the K&C team has developed efficient AI and machine learning (ML) based surrogate models (CHEM-ML) for non-equilibrium chemistry in hypersonic flows which is critical in designing hypersonic vehicles for space exploration. The CHEM-ML model can be coupled with reactive Navier-Stokes equations or high fidelity CFD models. In addition, CHEM-ML will be able to support both simple and complex chemical mechanisms.

To demonstrate that the CHEM-ML model can be used for non-equilibrium chemistry in hypersonic flows and can be integrated in a CFD solver to simulate hypersonic flows with combustion in a more efficient way, two demonstration problems are considered: First, a 5-species Air Reactor is used to demonstrate that CHEM-ML model can be trained/validated/deployed successfully using training data generated by Cantera, a chemical kinetics solver, and then the pre-trained CHEM-ML model is integrated with a CFD solver to solve hypersonic flows around a cylinder. Secondly, a 10-species methane oxygen burnt products which usually occurs in are considered for training/validating a CHEM-ML model, which is also demonstrated to be used for the simulation of rocket nozzle flows.

In summary, accurate CHEM-ML models for multiple chemical reaction mechanisms were developed and integrated into a CFD solver to simulate hypersonic flows with non-equilibrium chemistry and rocket nozzle flows.

PRESENTER

Meisner, Noah
meisner@kcse.com
503-572-8225

Karagozian & Case, Inc.

CO-AUTHOR

Wei, Liang
wei@kcse.com

Abraham, Joseph
Abraham@kcse.com

CATEGORY

Computational Fluid Dynamics

SYSTEMS USED

Mustang

SECRET

No