Answers are grounded in this paper and cite their sources.
A surrogate model for non-equilibrium chemistry, leveraging artificial intelligence (AI) and deep machine learning (ML), has been developed to enhance reactive flow simulations. The AI/ML library, CHEM-ML, includes several models designed to efficiently simulate reactive flows in applications such as hypersonic flight, offering both detailed chemical kinetics and fast computational performance. A Deep Operator Network (DeepONet) was used to create CHEM-ML models for both simple and complex, stiff chemical reaction mechanisms, including a 5-species air reactor. These models were trained and validated using data from the Cantera chemical solver, and they demonstrate strong accuracy in predicting validation data that was not part of the training set. Furthermore, the CHEM-ML models were successfully integrated with high-fidelity CFD codes for reactive flow simulations. The accuracy and efficiency of CFD simulations using CHEM-ML were validated for hypersonic flow around a cylinder, with results showing good agreement between simulations coupled with CHEM-ML and those coupled with Cantera, achieving a remarkable speedup of 9 times.
No takes yet. Share an insight, caveat, or question.
Wei et al. (2025) studied this question.