Simulation of a liquid methanol fire for the MaCFP Project

Computational fluid dynamics (CFD) is applied to simulate methanol pool fires with pool diameters of 30cm and 1m. The corresponding validation experiments were carried out at Waterloo University and the National Institute of Standards and Technology (NIST). Steady-state profiles of gas temperature, velocity, species concentrations, and radial and axial heat flux profiles were recorded during the experiments which provided the basis for the numerical analysis. During the experiments, the liquid layer height was maintained constant throughout.

The CFD tool for fire simulations, Fire Dynamics Simulator v6.8.0 was used for performing the calculations [1]. The two-step simple chemistry combustion model was used to model the oxidation of methanol. FDS allows the user to prescribe the fuel mass flux into the domain and let FDS predict the energy release based on the evaporation of the liquid. Both methods have been incorporated in the study.

Sensitivity studies have been performed and included: sensitivity to the mesh resolution for cell sizes ranging from 5 mm to 4 cm; changed temporal and spatial resolution of the radiation solver; and different turbulence models. The preliminary results from the study show that good agreement can be achieved with FDS using the default settings. However, there is still the possibility to improve the predictions from FDS, especially in the near field of the pool fire just above the surface of the liquid pool. Several modifications were done on the default models used in FDS which has led to improved predictions from the simulations.

Upon comparing two different radiation solvers (grey gas model and wide band model), it was found that the predictions from the default grey gas model were superior further away from the flames. Based on the sensitivity study on the settings of the grey gas model showed that a spatial resolution of 800 angles was sufficient for good accuracy. Deardorff model and the dynamic Smagorinsky model were used to model turbulence. Upon comparison of the two models, no significant variation was found between their predictions.

Furthermore, the puffing frequency of the flames during the steady burning period was estimated using the Fast-Fourier Transform algorithm on the signals of HRR, vertical velocity component and the flame height calculated within FDS.