This study presents and validates an adaptive CFD–AI digital twin framework for methane-leak prediction and plume monitoring in industrial environments by integrating high-fidelity computational fluid dynamics, artificial neural network surrogate modeling, and real-time data assimilation using the Ensemble Kalman Filter. The CFD model simulates methane-dispersion in the atmospheric boundary layer by solving the governing conservation equations under buoyancy-driven flow conditions. The CFD-generated data are then used to train a feed-forward ANN surrogate model capable of predicting plume characteristics in real time with significantly reduced computational cost. The digital twin continuously updates plume predictions by assimilating sensor measurements, allowing the system to remain synchronized with evolving leak conditions and atmospheric variability. Validation is performed using experimental methane-dispersion data and field-sensor measurements reported in the literature. The results show strong agreement between the digital twin predictions and observed methane concentrations, achieving a coefficient of determination of R2=0.9933 and an RMSE of 1.13 ppm. The study demonstrates that combining physics-based CFD modeling with machine-learning acceleration and real-time data assimilation significantly improves prediction accuracy and computational efficiency compared with a standalone CFD model. The proposed framework enables real-time methane monitoring, rapid leak detection, and emission-aware operational decision support in industrial facilities.
