Real-time prediction of flammable volume ratio in the coming hydrogen tunnels: A hybrid CFD-machine learning framework for leakage safety management

Hydrogen introduction into utility tunnels is a reasonable trend, yet safety risks from potential leaks cannot be ignored. The flammable cloud volume from pipeline leaks is crucial for evaluating the explosion degree, which is vital for structural protection and emergency response. Therefore, a dynamic risk assessment system is necessary to predict the flammable volume ratio, which factors in key parameters like leakage diameter, pipeline pressure, hydrogen temperature, and leakage angle. The impact of these parameters on the flammable volume ratio was investigated using computational fluid dynamics (CFD) simulations. Based on the simulation results, machine learning predictive models were constructed. Finally, a graphical user interface (GUI) was developed to predict the flammable volume ratio directly by inputting these parameters. The results indicate that the flammable and detonatable gas ratios increase with pipeline pressure and leakage diameter, decrease with leakage angle, and are mildly affected by hydrogen temperature variations. Dispersion velocity correlates positively with pipeline pressure, leakage diameter, and hydrogen temperature, however negatively with leakage angle. Four machine learning models were employed, and the Multilayer Perceptron (MLP) demonstrated the best predictive performance. The system improves accuracy and speed of prediction, which provides powerful decision support for tunnel safety management.