Acceleration of Chemical Kinetics Computation with the Learned Intelligent Tabulation (LIT) Method

In this work, a data-driven methodology for modeling combustion kinetics, Learned Intelligent Tabulation (LIT), is presented. LIT aims to accelerate the tabulation of mechanisms via machine learning algorithms such as Deep Neural Networks (DNNs). The high-dimensional composition space sampled from high-fidelity simulations covering wide range initial conditions train these DNNs. input data are clustered into subspaces, while each subspace trained with DNN regression model targeted particular part space. This localized approach has proven be more tractable than having global ANN model, which fails generalize across various spaces. clustering performed using an unsupervised method, Self-Organizing Map (SOM), automatically subdivides A dense network comprised fully connected layers considered hyper parameters optimized Bayesian optimization. nonlinear transformation used improve sensitivity minor species and enhance prediction ignition delay. method employed chemistry kinetics zero-dimensional H2–O2 CH4-air combustion. achieves good agreement benchmark being cheaper in terms computational cost. naturally extensible different models flamelet PDF transport models.