KARAGOZIAN & CASE, INC. — National Aeronautics and Space Administration SBIR Phase I: S17

Description

The Kamp;C team plans to propose efficient artificial intelligence (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 such as FUN3D and DPLR. In addition, CHEM-ML will be able to support both simple and complex chemical mechanisms. A deep operator network (DeepONet) will be employed to model the chemical kinetics in hypersonic flows such as gas-species reactions and gas surface reactions depending on the velocity, altitude and the materials of the hypersonic vehicle. DeepONet is based on the universal approximation of nonlinear operators which is suggestive of the potential application of neural networks in learning nonlinear operators from data. DeepONet can learn the stiff temporal evolution of chemical speciesrsquo; mass fractions over a given duration during offline training, so that during a prospective simulation inference from the learned algorithm can evolve the thermo-chemical state at a rate comparable to the hydrodynamic time scale, but without sacrificing the fidelity of the chemical systemrsquo;s transition path. Note that Kamp;C team has recent experience with DeepONet models for stiff chemical kinetics problems which were successfully used in reactive flow CFD simulations to speed up the calculation by over x1000 times. The Kamp;C team is poised to develop a model for a variety of chemical reaction mechanisms despite the short period of performance for Phase I due to the extensive expertise and existing DeepONet tools already used by Kamp;C.