CFD–AI surrogates for species transport and emissions in an axisymmetric coal combustor

Accurately predicting wall temperature and emissions in coal combustor is essential for design and control, yet full-fidelity simulations are too slow for rapid studies. We propose a hybrid approach that couples computational fluid dynamics (CFD) of a 2-D, radiating, axisymmetric coal flame with a multilayer perceptron (MLP) surrogate. The CFD model resolves turbulence, devolatilization, and volatile/char oxidation and generates a dataset across coal/air flow rates, inlet velocity profiles, and initial particle temperature. The MLP learns the mapping from operating conditions to wall-temperature and emission fields and is validated on unseen cases. The surrogate attains ∼ 8% error while cutting evaluation from hours to milliseconds. Physically, higher initial coal temperature shifts reaction zones upstream, enhances burnout, and modifies CO and SO 2 levels. This CFD-to-ML pipeline enables rapid what-if analysis and data-driven optimization, offering a practical path toward real-time monitoring and digital-twin control of coal combustion systems.