Multi-physics field study of CO low-temperature catalytic reactor based on CFD-GNN approach

This study addresses excessive CO emissions and associated energy waste in flue gas by employing low-temperature catalytic oxidation of CO, combined with computational fluid dynamics (CFD) and orthogonal experiments to investigate key process parameters affecting catalytic performance and temperature rise. The investigation reveals that catalyst layer quantity substantially influences CO conversion efficiency, with CO concentration being the primary factor in flue gas temperature elevation. At 15000 ppm CO concentration, the sintering flue gas temperature increases by approximately 130 K, translating to potential savings of 43780 m 3 /h of blast furnace gas. A novel CFD-graph neural network (GNN) methodology was developed to expedite temperature and concentration field simulations within the catalytic reactor. The proposed GNN-autoencoder framework utilizes gradient-based dimensionality reduction and a weighted loss function to enhance predictive accuracy. The model demonstrates exceptional performance, achieving a minimum validation loss of 0.0106, with R 2 values of 0.9812 and 0.9831 for training and testing sets. Notably, the model predicts CO concentrations and temperatures with average relative errors of 0.8564% and 0.0336%, respectively. This approach significantly reduces computational complexity while accurately capturing intricate physical field characteristics, offering efficient technical support for CO emission mitigation and energy recovery in sintering processes. • CO reduction as well as heat recovery was achieved using a CO catalytic reactor. • Increasing the number of catalyst layers can effectively reduce CO in flue gas. • Using computational fluid dynamics and graph neural networks for field prediction. • Hybrid modeling reduces computation time and maintaining excellent accuracy.