Data-Driven Modeling and Simulation of Turbulent Combustion

The modeling and simulation of turbulent combustion must account for the contribution of many chemical species and the effect of turbulence on their transport and reactions. Such reactions must span a wide range of time scales and often present bottlenecks in accelerating simulations. Data from experiments or simulations enables tools to accelerate simulations and develop accurate predictions of important turbulence-chemistry interactions. A number of methods designed to exploit this data are presented and discussed. They are motivated by, and rooted in, traditional paradigms in turbulent combustion that rely heavily on an adequate prediction of the composition space and its coupling with turbulent transport. The methods also rely on machine learning techniques for model order reduction, the extraction of closure from observations and models and learning predictions of reaction rates and new chemical states from low-dimensional descriptions of these states. These methods include principal component transport for combustion DNS, deep operator networks for chemistry integration and acceleration and methods to develop closure models and construct closure terms from multiscalar measurements. They provide pathways for efficient simulations of turbulent combustion and overcome the inherent limitations of predicting these complex flows.