Work

Numerical study on hydrogen-air explosion dynamic characteristics with deep reinforcement learning

Year: 2026

Type: article

Abstract: A novel deep reinforcement learning-based chemical kinetics key parameters optimization system is developed for studying hydrogen-air explosion dynamic process. We combine the optimized mechanism with eddy-dissipation concept (EDC) model to investigate the evolution of hydrogen-air premixed flame in semi-confined space with various obstacle arrangements. The results demonstrate that the optimized hydrogen-oxygen kinetic mechanism (HKM) has excellent laminar flame speed prediction performance. The model validation results show that the error between the simulation and the experimental results is less than 10%. Based on the optimized HKM, the explosion dynamic characteristics and flame unstable behaviors are analyzed. Specifically, the peak overpressure in the case of three dispersed obstacles is 1.3 times and 2.3 times higher than that of two dispersed obstacles and a concentrated arrangement, respectively. Based on flow field structure and vorticity analysis, vortex flames and jet flames are the primary reason of sudden increase in overpressure and flame velocity, respectively. Furthermore, the growth rate of the coupled amplitude of Rayleigh-Taylor Instability (RTI) and Kelvin-Helmholtz instability (KHI) is solely determined by the acceleration difference and tangential velocity difference of the flame. Analytical expression for the growth rate of the coupled amplitude is solved.

Source: Case Studies in Thermal Engineering

DOI: 10.1016/j.csite.2026.108476