Physics-informed neural networks coupled with flamelet/progress variable model for solving combustion physics considering detailed reaction mechanism.

This article presents a novel approach combining physics-informed neural networks (PINNs) with the flamelet/progress variable (FPV) model to efficiently and accurately solve combustion physics problems involving detailed chemical reaction mechanisms. The PINN/FPV method uses a flamelet library to tabulate thermophysical properties based on control variables, enabling PINNs to solve the governing equations of flow and combustion without directly handling the full complexity of detailed chemistry. Validated on a two-dimensional laminar methane/air diffusion flame, the model closely reproduces computational fluid dynamics (CFD) results for velocity, mixture fraction, progress variable, temperature, and species mass fractions, achieving mean relative errors below 10% and coefficient of determination (R²) values above 0.94, even without observation data points. The approach also demonstrates good generalization to unseen operating conditions, particularly for major combustion products and temperature, while intermediate species predictions show somewhat larger deviations. This study suggests that the PINN/FPV framework offers an effective tool for parameterized combustion simulations with detailed chemistry, with future work aimed at extending it to turbulent flows and multi-stream scenarios.