Dynamic evolution process of spill fires in longitudinally ventilated flat road tunnels: Experimental study, physical modelling, and real-time forecasting

Abstract

Spill fires caused by leaking vehicles are a typical fire scenario in road tunnels. The behaviour of a spill fire is primarily influenced by longitudinal ventilation, followed by the complex evolution of the flame geometry and unpredictable fire risks. In this paper, spill fire experiments were conducted in a 1:8 reduced-scale road tunnel model at different wind speeds (0–2 m/s) and fuel discharge rates (0.80–2.75 ml/s). The dynamic behaviour of spill fires was studied by combining physical analysis and a machine learning method. Results showed that the flame tilt angle decreases with the discharge time, while the flame base length showed the opposite trend. By dimensionless analysis, correlations for the steady stage flame tilt angle and flame base length were proposed and validated against the experimental data. Subsequently, dynamic prediction models for the spill fire flame shape (flame tilt angle and flame base length) were established by incorporating the dimensionless correlations into a PHAST (fuel layer spreading) model. Using the machine learning method, a real-time prediction algorithm for the discharge rate was established and the prediction error was found to be less than 10 %. Finally, the dynamic evolution of the flame shape was forecasted by integrating the predicted discharge rate into the physical model. Compared to traditional approaches, the combination of physical modelling and machine learning effectively improves the interpretability of fire prediction models, showcasing the potential for enhancing intelligent firefighting systems through physical theoretical models and limited experimental data.