Research Repository

Development of an interface using penalisation method for improving computational performance of bushfire simulation tools

Singha Roy, Sesa (2019) Development of an interface using penalisation method for improving computational performance of bushfire simulation tools. Research Master thesis, Victoria University.

[img]
Preview
Text
SINGHA ROY Sesa-thesis_nosignature.pdf - Submitted Version

Download (10MB) | Preview

Abstract

Wind is the dominant environmental factor affecting wildland fire intensity and spread. Previously, fire analysts and managers have relied on local measurements and site-specific forecasts to determine how winds influence fire. The advancements in computer hardware, increased availability of electronic topographical and experimental data, and advances in numerical methods for computing winds, have led to the development of new tools capable of simulating wind flow. Several numerical models have been developed for fire prediction and forecasting. Modelling wind in physics-based models like Fire Dynamics Simulator (FDS) has been shown to produce promising results, but at an inordinate cost. Because of the high computational expense, physics-based models are not suitable for operational use. Little research has been conducted to improve the computational speed of these models. The current study intends to decrease the computational cost of physics-based fire simulations and improve physics-based models by including more complicated driving winds. Physics-based wildfire simulations are driven by inlet boundary conditions which model the atmospheric boundary layer. Various inlet conditions, such as the 1/7-powerlaw or the log law models with artificial turbulence (e.g. the synthetic eddy method, [SEM]) can be used as an inlet to generate a statistically steady wind field for a fire simulation. The power-law inlet is the default inlet condition used in FDS where the wind develops turbulence as it sweeps through the domain, and is often used with wall-of-wind type methods. The log-law inlet generates a log wind profile similar to Atmospheric Boundary Layer (ABL). Development of techniques for imposing inlet conditions and initial conditions for flow simulations have been topics of interest for the past few decades. Current inlet and initial conditions requires time in a scale order of 100s of CPU hours, for generating an appropriate condition to start a fire simulation, hence resulting in increased computational expense. A novel nesting method has been implemented, which involves two regions : penalisation and blending, named as the PenaBlending method. The initial conditions of the fire simulations in FDS are set to the initial condition prescribed by an external model or simulation. This is achieved by a one-way coupling method. External wind data, for which u,v,w can vary in space and time, can be obtained. The precursor data can be generated either from any reduced wind model such as Windninja, which gives terrain modified wind data, or by using analytical methods such as generating logarithmic windfield using Matlab. These external data can be introduced into the FDS domain through a penalization region at the inlet/outlet. A blending region has also been implemented near the specified inlet/outlet which allows a smooth mixing of a precursor wind field to that in the simulation domain. This new inlet condition allows complicated terrain modified temporally and spatially varying wind fields, obtained from precursor simulations or any other models, to be implemented relatively easily in the FDS domain. To test the implementation of this method, a flat terrain is considered in the current study. However, this method could also be used for complicated terrain structures, as a part of future studies. The PenaBlending method provides appropriate flow conditions with reduced computational effort (up to ~ 80%), to start a fire simulation, and, hence, reduces the computational expense of physics-based models. The results obtained using the PenaBlending method have been compared with that obtained using the existing inlet conditions of FDS, like the SEM method, wall-of-wind method and mean-forcing methods, using the 1/7 power-law or log-law inlets. To test these three methods, a set of fire simulations have been conducted and tested against the PenaBlending method. It was found that the results of the PenaBlending methods agree well with that of existing methods, with small variations for both the wind and fire cases. FDS 6.6.0 (the version used in this study) requires a very fine grid to obtain grid convergence. This is not feasible in the case of a large-scale simulation because of very high computation cost. FDS 6.2.0, with a reaction-rate-limiter combustion model, needs less fine grids to obtain grid convergence. Therefore, this combustion model is re-introduced into FDS 6.6.0, providing an option of choosing between two different combustion models, as a part of this study. For all the simulations, the reaction-rate-limiter combustion model has been used. The simulations are carried out in a neutral-atmospheric stability condition. However, the PenaBlending method can apply any general driving wind, and the effect of atmospheric stability, could be included, as part of future studies. The PenaBlending method could be extended in conjunction with Monin-Obukhov Similarity Theory (introduced in FDS 6.6.0) to model fire in various atmospheric stability conditions.

Item Type: Thesis (Research Master thesis)
Additional Information:

Master of Engineering

Uncontrolled Keywords: wind; fire; physics-based modelling; wildfire; fire dynamics simulator; PenaBlending; computational fluid dynamics; Monin-Obukhov similarity theory
Subjects: FOR Classification > 0915 Interdisciplinary Engineering
Faculty/School/Research Centre/Department > Institute for Sustainable Industries and Liveable Cities
Depositing User: VUIR
Date Deposited: 24 Jan 2020 03:40
Last Modified: 24 Jan 2020 03:40
URI: http://vuir.vu.edu.au/id/eprint/40023
ePrint Statistics: View download statistics for this item

Repository staff only

View Item View Item

Search Google Scholar