Model identification in reactor-based combustion closures using sparse symbolic regression

Data-driven models and digital twins for sustainable combustion technologies

“…An example of the first approach is to define data-driven correction terms in the form of summative or multiplicative residuals (i.e., using ML for correction terms while preserving a physics-based approach) in the framework of different closure paradigms, flamelet-like to PDF-like approaches, from conditional closures to reactor-based approaches. 99 , 100 Incorporating physical knowledge into ML algorithms can be achieved by promoting the compliance of the results with the known laws of physics for the problem at hand. In neural networks, this can be accomplished by tailoring the loss function to include terms penalizing the violation of specific conservation principles or the entire residual of a governing equation.…”