International Journal of Chemical Engineering

Volume 2017, Article ID 4934956, 12 pages

https://doi.org/10.1155/2017/4934956

## Prediction of a Small-Scale Pool Fire with FireFoam

^{1}Department of Mechanical Engineering, Universidad de los Andes, Cra 1 Este N 19A-40, Bogota 111711, Colombia^{2}Department of Mechanical Engineering, Universidad del Valle, Calle 13, No. 100-00, Cali 760032, Colombia^{3}Department of Chemical Engineering, Universidad de los Andes, Cra 1 Este N 19A-40, Bogota 111711, Colombia

Correspondence should be addressed to Omar Darío López; oc.ude.sednainu@02zepol.do

Received 23 December 2016; Revised 8 April 2017; Accepted 13 April 2017; Published 1 June 2017

Academic Editor: Mahesh T. Dhotre

Copyright © 2017 Camilo Andrés Sedano et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

#### Abstract

A computational model using Large Eddy Simulation (LES) for turbulence modelling was implemented, by means of the Eddy Dissipation Concept (EDC) combustion model using the* fireFoam* solver. A small methanol pool fire experiment was simulated in order to validate and compare the numerical results, hence trying to validate the effectiveness of the solver. A detailed convergence analysis is performed showing that a mesh of approximately two million elements is sufficient to achieve satisfactory numerical results (including chemical kinetics). A good agreement was achieved with some of the experimental and previous computational results, especially in the prediction of the flame height and the average temperature contours.

#### 1. Introduction

Correct prediction and description of a fire have become one of the main concerns in safety engineering and risk analysis. Studies on fire safety have been developed mainly with emphasis on fire detection, heating of structures, and smoke-filling rates [1]. Usually this phenomenon has been analyzed through different experimental techniques. However, due to the destructive nature of fires, experimentation can be highly expensive and due to its randomness, it can be nearly impossible to replicate. Taking this into account, in recent years there has been an increasing interest in computational and numerical modelling of fires.

Nevertheless, there are several difficulties that arise when trying to fully develop models for fire dynamics simulations. First and foremost, the fact that fires consider several different physical phenomena that take place simultaneously such as turbulent flow, turbulent mixture processes, thermodynamics, heat transfer (especially through radiation, which in turn allows pyrolysis), and chemical kinetics [2]. Another difficulty is the coupling of the several analyses considered since these include a long range of length and time scales (for example the turbulent and chemical time scales). This is where different assumptions are applied on the combustion, chemical kinetics, and fluid dynamics processes [3].

These simplifications lead to the development of different models. When it comes to conflagrations (defined as an uncontrolled fire spread [2]), there are two main types of models: zone and field models. The first one basically divides the space in which the fire is taking place into two major zones: one in which predominantly remain the products of the combustion process and another one where the reacting air (oxygen) remains, until it is consumed by the reaction. On the other hand, field models are developed using Computational Fluid Dynamics (CFD) for reacting flows, in order to solve the Navier-Stokes equations coupled with the chemical kinetics solution (i.e., mixture fraction solution) [4].

While zone models are fairly straightforward, they have a major limitation in that these are only able to consider fires in an enclosure, therefore restricting the geometries and cases which can be explained through them. In contrast, field models have a higher mathematical complexity, while being able to adjust to almost any geometric domain and constraints. Therefore, there has been an increasing interest in this field to develop reliable field models which are accurate and exact enough to predict fires, complementing the experimental analysis. Some of the most notable field models are the Flame Surface Density model developed by Trouvé and Poinsot [5], the Partially Stirred Reactor model of Chen [6], or the Laminar Flamelet Model initially proposed by Peters [7].

One of the main combustion models used within field models is the Eddy Dissipation Concept (EDC), developed by Magnussen. In it, the author suggests a way to “relate the rate of combustion to the rate of dissipation of eddies” [8] assuming that the rate of reaction is a function of the mean concentration of a reacting specie, turbulent kinetic energy and its dissipation rate [9]. The EDC started as a model capable of considering both turbulent and momentum mixing, while considering the chemical kinetics solution and particularly soot formation. Most recently the interest in this model has shifted to develop an EDC that can account for Large Eddy Simulation (LES) turbulence models instead of the Reynolds Averaged Navier-Stokes (RANS) models traditionally used [9–12].

The biggest concern with the use of LES models is the fact that these filter the turbulent properties according to the different length scales for the eddies created (Integral, Taylor, Kolmogorov scale). Therefore, it becomes complicated to discuss a certain mesh convergence in computational models which use LES, since any change in the size of the cells of the mesh would also lead to a change in the size of the filter for the turbulent model, rendering different results which not necessarily tend to convergence [10].

Regarding the simulation of pool fires, there have been several studies in recent years. One of the most studied cases is the experiment proposed by Tieszen et al. in 2002 [13]. This experiment has been simulated by different authors, using mainly the Fire Dynamics Simulator (FDS). Such is the case of the simulations by Xin et al. in 2008 [14] and Cheung and Yeoh in 2009 [15]. Their results mainly show an interest in the flow’s velocity with great details of the fluctuating velocity field [14, 15]. Another example is the study of Novozhilov and Koseki [16] who use the FIRE software to simulate the results obtained in different experimental analysis including the experiment by Weckman and Strong [17]. Thus, they report results for the evolution of temperature with time, as well as normalized flame heights and burning rates for pool fires of different fuels and sizes. Finally, in 2014 Chen et al. [18] also recreate computationally the results obtained by Weckman and Strong [17], in an attempt to test the capabilities of the LES models developed in the solver* fireFoam*. Chen developed a new model for radiative emission, obtaining different results for the flame heights, heat generation, temperature, and emissive contours. However, some of these results do not agree completely with the experimental ones, particularly flame heights and the temperature root-mean-square contours.

Hence, the main objective of this work is to predict the dynamics of a small-scale pool fire and study the influence of the computational domain’s discretization especially in the vertical (flame propagation) direction, when using LES turbulence modelling with the EDC as the combustion model. As test case, the experiment proposed by Weckman and Strong in 1996 [17] was selected and simulated. The idea is to validate the numerical results obtained, by comparing them to the experimental results as well as other computational results found in literature. Numerical results show the influence of the mesh resolution on some variable contours and the flame height.

#### 2. Methods

##### 2.1. Experimental Setup

The base case used in the present computational study is the experiment developed by Weckman and Strong which consisted of a 30.5 cm methanol pool fire [17]. Weckman and Strong used different experimental techniques (for example Doppler anemometry) in order to measure a broad range of data: velocities, centerline and contour temperatures, mean and root-mean squares (rms) of fluctuations in velocities and temperatures, and flame heights [17]. This particular case has been well documented in literature, as different authors have developed computational analysis to validate numerical results (FDS [19], FIRE [16], SOFIE [20], and fireFoam [18, 21]).

The experimental facility consists of a pan burner of 30.5 cm in diameter where the methanol was injected at a rate of 1.35 cm^{3}/s to assure a heat release rate of 24.6 kW. The burner is designed to minimize the obstruction of ambient air flow into the fire, such that it is guaranteed that the pan is at least one pool diameter above the floor. It is worth mentioning that a natural draft hood was used to exhaust the combustion products.

##### 2.2. Computational Setup

In order to carry out the simulations, the open source code OpenFOAM–2.4.0 (OF) was used. This OF version was compiled with the* fireFoam* solver developed between CFD Direct and FMGlobal. This way the Navier-Stokes equations for compressible flows can be solved using the finite volume methods available as part of the OF library for turbulent flows.* FireFoam *uses a PIMPLE algorithm to iterate the calculation of the reacting flow properties. This iteration allows combining both the SIMPLE and PISO algorithm, such that it can use up to three correction equations for the pressure and velocity before solving the other transport equations [9].

##### 2.3. Computational Domain, Mesh, and Boundary Conditions

The computational domain is a cylinder of 1.8 m in height and 1.8 m in diameter (grossly 6 pool diameters) to make sure that the boundary conditions are far enough from the pool base. This way the flame can develop freely, such that the pulsating phenomena are due only to buoyancy and not an effect of interacting boundary conditions. To generate the mesh, OF’s utility* blockMesh* was used. Hexahedral elements within the domain were used and a total of 90 cells across the diameter of the pool fire were used to run the simulations [18]. The mesh generation was restrained by the vertices of the geometry, which in turn forces a total of 270 cells across the domain’s diameter. However, there is no major study on the influence of the number of divisions in the vertical direction (flame propagation) in the numerical results. Four different meshes were generated in order to pursue this study and Table 1 shows the most important details of these meshes. All the meshes were refined towards the bottom face, so that there are more elements near the burner’s exit.