NIST Special Publication 1018-5

Fire Dynamics Simulator (Version 5)

Technical Reference Guide
Volume 2: Verification

Randall McDermott
Kevin McGrattan
Simo Hostikka
Jason Floyd
 NIST Special Publication 1018-5

Fire Dynamics Simulator (Version 5)

Technical Reference Guide
Volume 2: Verification

Randall McDermott
Kevin McGrattan
Fire Research Division
Building and Fire Research Laboratory

Simo Hostikka
VTT Technical Research Centre of Finland
Espoo, Finland

Jason Floyd
Hughes Associates, Inc.
Baltimore, Maryland

December 11, 2008

FDS Version 5.2
SV NRepository Revision : 2836

E N T OF C O M
TM M
AR

ER
D EP

CE
ICA
UN
IT

ER

D
E

ST AM
ATES OF

U.S. Department of Commerce

Carlos M. Gutierrez, Secretary

National Institute of Standards and Technology

Patrick Gallagher, Acting Director
Preface

This is Volume 2 of the FDS Technical Reference Guide. Volume 1 describes the mathematical model and
numerical method. Volume 3 documents past and present experimental validation work. Instructions for
using FDS are contained in a separate User’s Guide [1].
The three volumes of the FDS Technical Reference Guide are based in part on the “Standard Guide
for Evaluating the Predictive Capability of Deterministic Fire Models,” ASTM E 1355 [2]. ASTM E 1355
defines model evaluation as “the process of quantifying the accuracy of chosen results from a model when
applied for a specific use.” The model evaluation process consists of two main components: verification
and validation. Verification is a process to check the correctness of the solution of the governing equations.
Verification does not imply that the governing equations are appropriate; only that the equations are being
solved correctly. Validation is a process to determine the appropriateness of the governing equations as a
mathematical model of the physical phenomena of interest. Typically, validation involves comparing model
results with experimental measurement. Differences that cannot be explained in terms of numerical errors
in the model or uncertainty in the measurements are attributed to the assumptions and simplifications of the
physical model.
Evaluation is critical to establishing both the acceptable uses and limitations of a model. Throughout
its development, FDS has undergone various forms of evaluation, both at NIST and beyond. This volume
provides a survey of verification work conducted to date to evaluate FDS.

i
ii
About the Authors

Randall McDermott joined the research staff of the Building and Fire Research Lab in 2008. He received
a B.S. degree from the University of Tulsa in Chemical Engineering in 1994 and a doctorate at the
University of Utah in 2005. His research interests include domain decomposition, Lagrangian particle
methods, and adaptive mesh refinement.

Kevin McGrattan is a mathematician in the Building and Fire Research Laboratory (BFRL) of NIST.
He received a bachelors of science degree from the School of Engineering and Applied Science of
Columbia University in 1987 and a doctorate at the Courant Institute of New York University in 1991.
He joined the NIST staff in 1992 and has since worked on the development of fire models, most
notably the Fire Dynamics Simulator.

Simo Hostikka is a Senior Research Scientist at VTT Technical Research Centre of Finland. He received
a master of science (technology) degree in 1997 and a doctorate in 2008 from the Department of
Engineering Physics and Mathematics of the Helsinki University of Technology. He is the principal
developer of the radiation and solid phase sub-models within FDS.

Jason Floyd is a Senior Engineer at Hughes Associates, Inc., in Baltimore, Maryland. He received a bach-
elors of science and Ph.D. in the Nuclear Engineering Program of the University of Maryland. After
graduating, he won a National Research Council Post-Doctoral Fellowship at the Building and Fire
Research Laboratory of NIST, where he developed the combustion algorithm within FDS. He is cur-
rently funded by NIST under grant 60NANB5D1205 from the Fire Research Grants Program (15 USC
278f). He is the principal developer of the multi-parameter mixture fraction combustion model and
control logic within FDS.

iii
iv
Acknowledgments

Partial support for the preparation of the FDS Technical Reference Guides has been provided by the Office
of Nuclear Regulatory Research of the US Nuclear Regulatory Commission (US NRC). Special thanks to
Mark Salley and Jason Dreisbach for their efforts.
Thanks to Chris Lautenburger and Carlos Fernandez-Pello for their assistance with the “two-reaction”
test case.

v
vi
Contents

Preface i

About the Authors iii

Acknowledgments v

1 What is Verification? 1

2 Survey of Past Verification Work 3

2.1 Analytical Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Numerical Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3.1 Grid Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3.2 Sensitivity of Large Eddy Simulation Parameters . . . . . . . . . . . . . . . . . . . 7
2.3.3 Sensitivity of Radiation Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3.4 Sensitivity of Thermophysical Properties of Solid Fuels . . . . . . . . . . . . . . . . 8
2.4 Code Checking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 The Basic Flow Solver 11

3.1 Single Mesh, Nonreacting Periodic Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.1.1 2D Analytical Solution to Navier-Stokes . . . . . . . . . . . . . . . . . . . . . . . . 11
3.1.2 Decaying Isotropic Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4 Thermal Radiation 17
4.1 Radiation inside a box (radiation_in_a_box) . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2 Radiation from a plane layer (radiation_plane_layer) . . . . . . . . . . . . . . . . . . . . 19
4.3 Wall Internal Radiation (wall_internal_radiation) . . . . . . . . . . . . . . . . . . . . . . 20

5 Heat Conduction 21
5.1 Simple Heat Conduction Through a Solid Slab (heat_conduction) . . . . . . . . . . . . . . 22
5.2 Temperature-Dependent Thermal Properties (heat_conduction_kc) . . . . . . . . . . . . . 23

6 Pyrolysis 25
6.1 A Simple Two-Step Pyrolysis Example (two_step_solid_reaction) . . . . . . . . . . . . . . 26
6.2 Development of surface emissivity (emissivity) . . . . . . . . . . . . . . . . . . . . . . . . 27
6.3 Enthalpy of solid materials (enthalpy) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

Bibliography 29

vii
viii
Chapter 1

What is Verification?

The terms verification and validation are often used interchangeably to mean the process of checking the
accuracy of a numerical model. For many, this entails comparing model predictions with experimental
measurements. However, there is now a fairly broad-based consensus that comparing model and experiment
is largely what is considered validation. So what is verification? ASTM E 1355 [2], “Standard Guide for
Evaluating the Predictive Capability of Deterministic Fire Models,” defines verification as

The process of determining that the implementation of a calculation method accurately repre-
sents the developer’s conceptual description of the calculation method and the solution to the
calculation method.

and it defines validation as

The process of determining the degree to which a calculation method is an accurate representa-
tion of the real world from the perspective of the intended uses of the calculation method.

Simply put, verification is a check of the math; validation is a check of the physics. If the model predictions
closely match the results of experiments, using whatever metric is appropriate, it is assumed by most that
the model suitably describes, via its mathematical equations, what is happening. It is also assumed that the
solution of these equations must be correct. So why do we need to perform model verification? Why not
just skip to validation and be done with it? The reason is that rarely do model and measurement agree so
well in all applications that anyone would just accept its results unquestionably. Because there is inevitably
differences between model and experiment, we need to know if these differences are due to limitations or
errors in the numerical solution, or the physical sub-models, or both.
Whereas model validation consists mainly of comparing predictions with measurements, as documented
for FDS in Volume 3 of the Technical Reference Guide, model verification consists of a much broader range
of activities, from checking the computer program itself to comparing calculations to analytical (exact)
solutions to considering the sensitivity of the dozens of numerical parameters. The next chapter discusses
these various activities, and the rest of the Guide is devoted mainly to comparisons of various sub-model
calculations with analytical solutions.

1
2
Chapter 2

Survey of Past Verification Work

This chapter documents work of the past few decades at NIST, VTT and elsewhere to verify the algorithms
within FDS.

2.1 Analytical Tests

Most complex combustion processes, including fire, are turbulent and time-dependent. There are no closed-
form mathematical solutions for the fully-turbulent, time-dependent Navier-Stokes equations. CFD provides
an approximate solution for the non-linear partial differential equations by replacing them with discretized
algebraic equations that can be solved using a powerful computer. While there is no general analytical
solution for fully-turbulent flows, certain sub-models address phenomenon that do have analytical solutions,
for example, one-dimensional heat conduction through a solid. These analytical solutions can be used to
test sub-models within a complex code such as FDS. The developers of FDS routinely use such practices to
verify the correctness of the coding of the model [3, 4]. Such verification efforts are relatively simple and
routine and the results may not always be published nor included in the documentation. Examples of routine
analytical testing include:

• The radiation solver has been verified with scenarios where simple objects, like cubes or flat plates,
are positioned in simple, sealed compartments. All convective motion is turned off, the object is given
a fixed surface temperature and emissivity of one (making it a black body radiator). The heat flux
to the cold surrounding walls is recorded and compared to analytical solutions. These studies help
determine the appropriate number of solid angles to be set as the default.

• Solid objects are heated with a fixed heat flux, and the interior and surface temperatures as a function
of time are compared to analytical solutions of the one-dimensional heat transfer equation. These
studies help determine the number of nodes to use in the solid phase heat transfer model. Similar
studies are performed to check the pyrolysis models for thermoplastic and charring solids.

• Early in its development, the hydrodynamic solver that evolved to form the core of FDS was checked
against analytical solutions of simplified fluid flow phenomena. These studies were conducted at the
National Bureau of Standards (NBS)1 by Rehm, Baum and co-workers [5, 6, 7, 8]. The emphasis
of this early work was to test the stability and consistency of the basic hydrodynamic solver, espe-
cially the velocity-pressure coupling that is vitally important in low Mach number applications. Many
numerical algorithms developed up to that point in time were intended for use in high-speed flow
applications, like aerospace. Many of the techniques adopted by FDS were originally developed for
1 The National Institute of Standards and Technology (NIST) was formerly known as the National Bureau of Standards.

3
 meteorological models, and as such needed to be tested to assess whether they would be appropriate
to describe relatively low-speed flow within enclosures.

• A fundamental decision made by Rehm and Baum early in the FDS development was to use a direct
(rather than iterative) solver for the pressure. In the low Mach number formulation of the Navier-
Stokes equations, an elliptic partial differential equation for the pressure emerges, often referred to as
the Poisson equation. Most CFD methods use iterative techniques to solve the governing conservation
equations to avoid the necessity of directly solving the Poisson equation. The reason for this is that
the equation is time-consuming to solve numerically on anything but a rectilinear grid. Because FDS
is designed specifically for rectilinear grids, it can exploit fast, direct solvers of the Poisson equation,
obtaining the pressure field with one pass through the solver to machine accuracy. FDS employs
double-precision (8 byte) arithmetic, meaning that the relative difference between the computed and
the exact solution of the discretized Poisson equation is on the order of 10−12 . The fidelity of the
numerical solution of the entire system of equations is tied to the pressure/velocity coupling because
often simulations can involve hundreds of thousands of time steps, with each time step consisting of
two solutions of the Poisson equation to preserve second-order accuracy. Without the use of the direct
Poisson solver, build-up of numerical error over the course of a simulation could produce spurious
results. Indeed, an attempt to use single-precision (4 byte) arithmetic to conserve machine memory
led to spurious results simply because the error per time step built up to an intolerable level.

2.2 Numerical Tests

Numerical techniques used to solve the governing equations within a model can be a source of error in
the predicted results. The hydrodynamic model within FDS is second-order accurate in space and time.
This means that the error terms associated with the approximation of the spatial partial derivatives by finite
differences is of the order of the square of the grid cell size, and likewise the error in the approximation of
the temporal derivatives is of the order of the square of the time step. As the numerical grid is refined, the
“discretization error” decreases, and a more faithful rendering of the flow field emerges. The issue of grid
sensitivity is extremely important to the proper use of the model and will be taken up in the next chapter.
A common technique of testing flow solvers is to systematically refine the numerical grid until the
computed solution does not change, at which point the calculation is referred to as a Direct Numerical
Solution (DNS) of the governing equations. For most practical fire scenarios, DNS is not possible on
conventional computers. However, FDS does have the option of running in DNS mode, where the Navier-
Stokes equations are solved without the use of sub-grid scale turbulence models of any kind. Because
the basic numerical method is the same for LES and DNS, DNS calculations are a very effective way to
test the basic solver, especially in cases where the solution is steady-state. Throughout its development,
FDS has been used in DNS mode for special applications. For example, FDS (or its core algorithms)
have been used at a grid resolution of roughly 1 mm to look at flames spreading over paper in a microgravity
environment [9, 10, 11, 12, 13, 14], as well as "g-jitter" effects aboard spacecraft [15]. Simulations have been
compared to experiments performed aboard the US Space Shuttle. The flames are laminar and relatively
simple in structure, and the comparisons are a qualitative assessment of the model solution. Similar studies
have been performed comparing DNS simulations of a simple burner flame to laboratory experiments [16].
Another study compared FDS simulations of a counterflow diffusion flames to experimental measurements
and the results of a one-dimensional multi-step kinetics model [17].
Early work with the hydrodynamic solver compared two-dimensional simulations of gravity currents
with salt-water experiments [18]. In these tests, the numerical grid was systematically refined until almost
perfect agreement with experiment was obtained. Such convergence would not be possible if there were a
fundamental flaw in the hydrodynamic solver.

4
2.3 Sensitivity Analysis
A sensitivity analysis considers the extent to which uncertainty in model inputs influences model output.
Model parameters can be the physical properties of solids and gases, boundary conditions, initial conditions,
etc. The parameters can also be purely numerical, like the size of the numerical grid. FDS typically requires
the user to provide several dozen different types of input parameters that describe the geometry, materials,
combustion phenomena, etc. By design, the user is not expected to provide numerical parameters besides the
grid size, although the optional numerical parameters are described in both the Technical Reference Guide
and the User’s Guide.
FDS does not limit the range of most of the input parameters because applications often push beyond
the range for which the model has been validated. FDS is still used for research at NIST and elsewhere,
and the developers do not presume to know in all cases what the acceptable range of any parameter is. Plus,
FDS solves the fundamental conservation equations and is much less susceptible to errors resulting from
input parameters that stray beyond the limits of simpler empirical models. However, the user is warned that
he/she is responsible for the prescription of all parameters. The FDS manuals can only provide guidance.
The grid size is the most important numerical parameter in the model, as it dictates the spatial and tem-
poral accuracy of the discretized partial differential equations. The heat release rate is the most important
physical parameter, as it is the source term in the energy equation. Property data, like the thermal conduc-
tivity, density, heat of vaporization, heat capacity, etc., ought to be assessed in terms of their influence on
the heat release rate. Validation studies have shown that FDS predicts well the transport of heat and smoke
when the HRR is prescribed. In such cases, minor changes in the properties of bounding surfaces do not
have a significant impact on the results. However, when the HRR is not prescribed, but rather predicted
by the model using the thermophysical properties of the fuels, the model output is sensitive to even minor
changes in these properties.
The sensitivity analyses described in this chapter are all performed in basically the same way. For a given
scenario, best estimates of all the relevant physical and numerical parameters are made, and a “baseline”
simulation is performed. Then, one by one, parameters are varied by a given percentage, and the changes
in predicted results are recorded. This is the simplest form of sensitivity analysis. More sophisticated
techniques that involve the simultaneous variation of several parameters are impractical with a CFD model
because the computation time is too long and the number of parameters too large to perform the necessary
number of calculations to generate decent statistics.

2.3.1 Grid Sensitivity

The most important decision made by a model user is the size of the numerical grid. In general, the finer the
numerical grid, the better the numerical solution of the equations. FDS is second-order accurate in space and
time, meaning that halving the grid cell size will decrease the discretization error in the governing equations
by a factor of 4. Because of the non-linearity of the equations, the decrease in discretization error does not
necessarily translate into a comparable decrease in the error of a given FDS output quantity. To find out
what effect a finer grid has on the solution, model users usually perform some form of grid sensitivity study
in which the numerical grid is systematically refined until the output quantities do not change appreciably
with each refinement. Of course, with each halving of the grid cell size, the time required for the simulation
increases by a factor of 24 = 16 (a factor of two for each spatial coordinate, plus time). In the end, a
compromise is struck between model accuracy and computer capacity.
Some grid sensitivity studies have been documented and published. Since FDS was first publicly re-
leased in 2000, significant changes in the combustion and radiation routines have been incorporated into the
model. However, the basic transport algorithm is the same, as is the critical importance of grid sensitivity. In
compiling sensitivity studies, only those that examined the sensitivity of routines no longer used have been

5
excluded.
As part of a project to evaluate the use of FDS version 1 for large scale mechanically ventilated enclo-
sures, Friday [19] performed a sensitivity analysis to find the approximate calculation time based on varying
grid sizes. A propylene fire with a nominal heat release rate was modeled in FDS. There was no mechanical
ventilation and the fire was assumed to grow as a function of the time from ignition squared. The compart-
ment was a 3 m by 3 m by 6.1 m space. Temperatures were sampled 12 cm below the ceiling. Four grid
sizes were chosen for the analysis: 30 cm, 15 cm, 10 cm, 7.5 cm. Temperature estimates were not found to
change dramatically with different grid dimensions.
Using FDS version 1, Bounagui et al. [20] studied the effect of grid size on simulation results to de-
termine the nominal grid size for future work. A propane burner 0.1 m by 0.1 m was modeled with a heat
release rate of 1500 kW. A similar analysis was performed using Alpert’s ceiling jet correlation [21] that
also showed better predictions with smaller grid sizes. In a related study, Bounagui et al. [22] used FDS
to evaluate the emergency ventilation strategies in the Louis-Hippolyte-La Fontaine Tunnel in Montreal,
Canada.
Xin [23] used FDS to model a methane fueled square burner (1 m by 1 m) in the open. Engineering
correlations for plume centerline temperature and velocity profiles were compared with model predictions to
assess the influence of the numerical grid and the size of the computational domain. The results showed that
FDS is sensitive to grid size effects, especially in the region near the fuel surface, and domain size effects
when the domain width is less than twice the plume width. FDS uses a constant pressure assumption at open
boundaries. This assumption will affect the plume behavior if the boundary of the computational domain is
too close to the plume.
Ierardi and Barnett [24] used FDS version 3 to model a 0.3 m square methane diffusion burner with heat
release rate values in the range of 14.4 kW to 57.5 kW. The physical domain used was 0.6 m by 0.6 m with
uniform grid spacings of 15, 10, 7.5, 5, 3, 1.5 cm for all three coordinate directions. For both fire sizes, a
grid spacing of 1.5 cm was found to provide the best agreement when compared to McCaffrey’s centerline
plume temperature and velocity correlations [25]. Two similar scenarios that form the basis for Alpert’s
ceiling jet correlation were also modeled with FDS. The first scenario was a 1 m by 1 m, 670 kW ethanol
fire under a 7 m high unconfined ceiling. The planar dimensions of the computational domain were 14 m by
14 m. Four uniform grid spacings of 50, 33.3, 25, and 20 cm were used in the modeling. The best agreement
for maximum ceiling jet temperature was with the 33.3 cm grid spacing. The best agreement for maximum
ceiling jet velocity was for the 50 cm grid spacing. The second scenario was a 0.6 m by 0.6 m 1000 kW
ethanol fire under a 7.2 m high unconfined ceiling. The planar dimensions of the computational domain
were 14.4 m by 14.4 m. Three uniform grid spacings of 60, 30, and 20 cm were used in the modeling. The
results show that the 60 cm grid spacing exhibits the best agreement with the correlations for both maximum
ceiling jet temperature and velocity on a qualitative basis.
Petterson [26] also completed work assessing the optimal grid size for FDS version 2. The FDS model
predictions of varying grid sizes were compared to two separate fire experiments: The University of Canter-
bury McLeans Island Tests and the US Navy Hangar Tests in Hawaii. The first set of tests utilized a room
with approximate dimensions of 2.4 m by 3.6 m by 2.4 m and fire sizes of 55 kW and 110 kW. The Navy
Hangar tests were performed in a hangar measuring 98 m by 74 m by 15 m in height and had fires in the
range of 5.5 MW to 6.6 MW. The results of this study indicate that FDS simulations with grids of 0.15 m
had temperature predictions as accurate as models with grids as small as 0.10 m. Each of these grid sizes
produced results within 15 % of the University of Canterbury temperature measurements. The 0.30 m grid
produced less accurate results. For the comparison of the Navy Hangar tests, grid sizes ranging from 0.60 m
to 1.80 m yielded results of comparable accuracy.
Musser et al. [27] investigated the use of FDS for course grid modeling of non-fire and fire scenarios.
Determining the appropriate grid size was found to be especially important with respect to heat transfer at
heated surfaces. The convective heat transfer from the heated surfaces was most accurate when the near

6
surface grid cells were smaller than the depth of the thermal boundary layer. However, a finer grid size
produced better results at the expense of computational time. Accurate contaminant dispersal modeling re-
quired a significantly finer grid. The results of her study indicate that non-fire simulations can be completed
more quickly than fire simulations because the time step is not limited by the large flow speeds in a fire
plume.

2.3.2 Sensitivity of Large Eddy Simulation Parameters

FDS uses the Smagorinsky form of the Large Eddy Simulation (LES) technique. This means that instead of
using the actual fluid viscosity, the model uses a viscosity of the form

µLES = ρ (Cs ∆)2 |S| (2.1)

where Cs is an empirical constant, ∆ is a length on the order of the size of a grid cell, and the deformation
term |S| is related to the Dissipation Function (see FDS Technical Reference Guide [28] for details). Related
to the “turbulent viscosity” are comparable expressions for the thermal conductivity and material diffusivity:
µLES c p µLES
kLES = ; (ρD)LES = (2.2)
Prt Sct
where Prt and Sct are the turbulent Prandtl and Schmidt numbers, respectively. Thus, Cs , Prt and Sct are
a set of empirical constants. Most FDS users simply use the default values of (0.2,0.5,0.5), but some have
explored their effect on the solution of the equations.
In an effort to validate FDS with some simple room temperature data, Zhang et al. [29] tried different
combinations of the Smagorinsky parameters, and suggested the current default values. Of the three pa-
rameters, the Smagorinsky constant Cs is the most sensitive. Smagorinsky [30] originally proposed a value
of 0.23, but researchers over the past three decades have used values ranging from 0.1 to 0.23. There are
also refinements of the original Smagorinsky model [31, 32, 33] that do not require the user to prescribe the
constants, but rather generate them automatically as part of the numerical scheme.

2.3.3 Sensitivity of Radiation Parameters

Radiative heat transfer is included in FDS via the solution of the radiation transport equation for a non-
scattering gray gas, and in some limited cases using a wide band model. The equation is solved using a
technique similar to finite volume methods for convective transport, thus the name given to it is the Finite
Volume Method (FVM). There are several limitations of the model. First, the absorption coefficient for
the smoke-laden gas is a complex function of its composition and temperature. Because of the simplified
combustion model, the chemical composition of the smokey gases, especially the soot content, can effect
both the absorption and emission of thermal radiation. Second, the radiation transport is discretized via
approximately 100 solid angles. For targets far away from a localized source of radiation, like a growing
fire, the discretization can lead to a non-uniform distribution of the radiant energy. This can be seen in the
visualization of surface temperatures, where “hot spots” show the effect of the finite number of solid angles.
The problem can be lessened by the inclusion of more solid angles, but at a price of longer computing
times. In most cases, the radiative flux to far-field targets is not as important as those in the near-field, where
coverage by the default number of angles is much better.
Hostikka et al. examined the sensitivity of the radiation solver to changes in the assumed soot pro-
duction, number of spectral bands, number of control angles, and flame temperature. Some of the more
interesting findings were:
• Changing the soot yield from 1 % to 2 % increased the radiative flux from a simulated methane burner
about 15 %

7
 • Lowering the soot yield to zero decreased the radiative flux about 20 %.

• Increasing the number of control angles by a factor of 3 was necessary to ensure the accuracy of the
model at the discrete measurement locations.

• Changing the number of spectral bands from 6 to 10 did not have a strong effect on the results.

• Errors of 100 % in heat flux were caused by errors of 20 % in absolute temperature.

The sensitivity to flame temperature and soot composition are consistent with combustion theory, which
states that the source term of the radiative transport equation is a function of the absorption coefficient mul-
tiplied by the absolute temperature raised to the fourth power. The number of control angles and spectral
bands are user-controlled numerical parameters whose sensitivities ought to be checked for each new sce-
nario. The default values in FDS are appropriate for most large scale fire scenarios, but may need to be
refined for more detailed simulations such as a low-sooting methane burner.

2.3.4 Sensitivity of Thermophysical Properties of Solid Fuels

An extensive amount of verification and validation work with FDS version 4 has been performed by Hi-
etaniemi, Hostikka, and Vaari at VTT, Finland [34]. The case studies are comprised of fire experiments
ranging in scale from the cone calorimeter (ISO 5660-1) to full-scale fire tests such as the room corner test
(ISO 9705). Comparisons are also made between FDS results and data obtained in the SBI (Single Burning
Item) Euro-classification test apparatus (EN 13823) as well as data obtained in two ad hoc experimental
configurations: one is similar to the room corner test but has only partial linings and the other is a space to
study fires in building cavities.
All of the case studies involve real materials whose properties must be prescribed so as to conform to
the assumption in FDS that solids are of uniform composition backed by a material that is either cold or
totally insulating. Sensitivity of the various physical properties and the boundary conditions were tested.
Some of the findings were:
• The measured burning rates of various materials often fell between two FDS predictions in which cold
or insulated backings were assumed for the solid surfaces. FDS lacks a multi-layer solid model.

• The ignition time of upholstery is sensitive to the thermal properties of the fabric covering, but the
steady burning rate is sensitive to the properties of the underlying foam.

• Moisture content of wooden fuels is very important and difficult to measure.

• Flame spread over complicated objects, like cables laid out in trays, can be modeled if the surface
area of the simplified object is comparable to that of the real object. This suggests sensitivity not
only to physical properties, but also geometry. It is difficult to quantify the extent of the geometrical
sensitivity.
There is little quantification of the observed sensitivities in the study. Fire growth curves can be linear to
exponential in form, and small changes in fuel properties can lead to order of magnitude changes in heat
release rate for unconfined fires. The subject is discussed in the FDS Validation Guide (Volume 3 of the
Technical Reference Guide). where it is noted in many of the studies that predicting fire growth is difficult.
Recently, Lautenberger, Rein and Fernandez-Pello [35] developed a method to automate the process of
estimating material properties to input into FDS. The methodology involves simulating a bench-scale test
with the model and iterating via a "genetic" algorithm to obtain an optimal set of material properties for
that particular item. Such techniques are necessary because most bench-scale apparatus do not provide a
complete set of thermal properties.

8
2.4 Code Checking
An examination of the structure of the computer program can be used to detect potential errors in the nu-
merical solution of the governing equations. The coding can be verified by a third party either manually or
automatically with profiling programs to detect irregularities and inconsistencies [2].
At NIST and elsewhere, FDS has been compiled and run on computers manufactured by IBM, Hewlett-
Packard, Sun Microsystems, Digital Equipment Corporation, Apple, Silicon Graphics, Dell, Compaq, and
various other personal computer vendors. The operating systems on these platforms include Unix, Linux,
Microsoft Windows, and Mac OSX. Compilers used include Lahey Fortran, Digital Visual Fortran, Intel
Fortran, IBM XL Fortran, HPUX Fortran, Forte Fortran for SunOS, the Portland Group Fortran, and several
others. Each combination of hardware, operating system and compiler involves a slightly different set of
compiler and run-time options and a rigorous evaluation of the source code to test its compliance with
the Fortran 90 ISO/ANSI standard [36]. Through this process, out-dated and potentially harmful code is
updated or eliminated, and often the code is streamlined to improve its optimization on the various machines.
However, simply because the FDS source code can be compiled and run on a wide variety of platforms does
not guarantee that the numerics are correct. It is only the starting point in the process because it at least rules
out the possibility that erratic or spurious results are due to the platform on which the code is running.
Beyond hardware issues, there are several useful techniques for checking the FDS source code that have
been developed over the years. One of the best ways is to exploit symmetry. FDS is filled with thousands
of lines of code in which the partial derivatives in the conservation equations are approximated as finite
differences. It is very easy in this process to make a mistake. Consider, for example, the finite difference
approximation of the thermal diffusion term in the i jkth cell of the three-dimensional grid:
 
1 Ti+1, jk − Ti jk Ti jk − Ti−1, jk
(∇ · k∇T )i jk ≈ k 1 − ki− 1 , jk +
δx i+ 2 , jk δx 2 δx
 
1 Ti, j+1,k − Ti jk Ti jk − Ti, j−1,k
k 1 − ki, j− 1 ,k +
δy i, j+ 2 ,k δy 2 δy
 
1 Ti j,k+1 − Ti jk Ti jk − Ti j,k−1
ki j,k+ 1 − ki j,k− 1
δz 2 δz 2 δz

which is written as follows in the Fortran source code:

DTDX = (TMP(I+1,J,K)-TMP(I,J,K))*RDXN(I)
KDTDX(I,J,K) = .5*(KP(I+1,J,K)+KP(I,J,K))*DTDX
DTDY = (TMP(I,J+1,K)-TMP(I,J,K))*RDYN(J)
KDTDY(I,J,K) = .5*(KP(I,J+1,K)+KP(I,J,K))*DTDY
DTDZ = (TMP(I,J,K+1)-TMP(I,J,K))*RDZN(K)
KDTDZ(I,J,K) = .5*(KP(I,J,K+1)+KP(I,J,K))*DTDZ

DELKDELT = (KDTDX(I,J,K)-KDTDX(I-1,J,K))*RDX(I) +
. (KDTDY(I,J,K)-KDTDY(I,J-1,K))*RDY(J) +
. (KDTDZ(I,J,K)-KDTDZ(I,J,K-1))*RDZ(K)

This is one of the simpler constructs because the pattern that emerges within the lines of code make it fairly
easy to check. However, a mis-typing of an I or a J, a plus or a minus sign, or any of a hundred different
mistakes can cause the code to fail, or worse produce the wrong answer. A simple way to eliminate many of
these mistakes is to run simple scenarios that have perfectly symmetric initial and boundary conditions. For
example, put a hot cube in the exact center of a larger cold compartment, turn off gravity, and watch the heat
diffuse from the hot cube into the cold gas. Any simple error in the coding of the energy equation will show

9
up almost immediately. Then, turn on gravity, and in the absence of any coding error, a perfectly symmetric
plume will rise from the hot cube. This checks both the coding of the energy and the momentum equations.
Similar checks can be made for all of the three dimensional finite difference routines. So extensive are these
types of checks that the release version of FDS has a routine that generates a tiny amount of random noise
in the initial flow field so as to eliminate any false symmetries that might arise in the numerical solution.
The process of adding new routines to FDS is as follows: typically the routine is written by one person
(not necessarily a NIST staffer) who takes the latest version of the source code, adds the new routine, and
writes a theoretical and numerical description for the FDS Technical Reference Guide, plus a description
of the input parameters for the FDS User’s Guide. The new version of FDS is then tested at NIST with
a number of benchmark scenarios that exercise the range of the new parameters. Provisional acceptance
of the new routine is based on several factors: (1) it produces more accurate results when compared to
experimental measurement, (2) the theoretical description is sound, and (3) any empirical parameters are
obtainable from the open literature or standard bench-scale apparatus. If the new routine is accepted, it is
added to a test version of the software and evaluated by external users and/or NIST grantees whose research
is related to the subject. Assuming that there are no intractable issues that arise during the testing period,
the new routine eventually becomes part of the release version of FDS.
Even with all the code checking performed at NIST, it is still possible for errors to go unnoticed. One
remedy is the fact that the source code for FDS is publicly released. Although it consists of on the order
of 30,000 lines of Fortran statements, various researchers outside of NIST have been able to work with
it, add enhancements needed for very specific applications or for research purposes, and report back to the
developers bugs that have been detected. The source code is organized into 27 separate files, each containing
subroutines related to a particular feature of the model, like the mass, momentum, and energy conservation
equations, sprinkler activation and sprays, the pressure solver, etc. The lengthiest routines are devoted to
input, output and initialization. Most of those working with the source code do not concern themselves with
these lengthy routines but rather focus on the finite-difference algorithm contained in a few of the more
important files. Most serious errors are found in these files, for they contain the core of the algorithm. The
external researchers provide feedback on the organization of the code and its internal documentation, that is,
comments within the source code itself. Plus, they must compile the code on their own computers, adding
to its portability.

10
Chapter 3

The Basic Flow Solver

3.1 Single Mesh, Nonreacting Periodic Test Cases

In this section we present test cases aimed at exercising the advective, pressure and viscous terms and the
time integration for nonreacting flows with periodic boundary conditions in all directions (this eliminates
the possibility of gravitational stratification). In Section 3.1.1 we present an analytical solution that is useful
for confirming the convergence rates of the truncation errors in the discretization schemes. In Section 3.1.2
we present a canonical flow for LES which tests that the subgrid stress model has been coded properly.

3.1.1 2D Analytical Solution to Navier-Stokes

The velocity is given by u = [u, v]T . Let A represent an arbitrary amplitude. The kinematic viscosity and
pressure are denoted ν and p, respectively, and the solution the 2D incompressible Navier-Stokes equations,

∂u
+ u · ∇u = −∇p + ν∇2 u , (3.1)
∂t
is given by [37]

u(x, y,t) = 1 − A cos(x − t) sin(y − t) e−2νt , (3.2)

v(x, y,t) = 1 + A sin(x − t) cos(y − t) e −2νt
, (3.3)

A2
p(x, y,t) = − [cos(2(x − t)) + cos(2(y − t))] e−4νt . (3.4)
4
Note that this solution satisfies continuity for all time,

∇·u = 0, (3.5)

is period on an interval 2π in each direction, and is time periodic on 2π if ν = 0, else the solution decays
exponentially. Below we present two series of tests which demonstrate the second-order accuracy of the
FDS numerical scheme and thus provide a strong form of code verification for the advective and viscous
terms which are exercised.
The physical domain of the problem is a square of side L = 2π. The grid spacing is uniform ∆x = ∆y =
L/N in each direction with N = {8, 16, 32, 64} for each test series. The staggered grid locations are denoted
xi = i∆x and y j = j∆y, and the cell centers are marked by an overbar, x̄i = xi − ∆x/2 and ȳ j = y j − ∆y/2.
First, we present qualitative results for the case in which ν = 0. Thus, only the advective discretization
and the time integration are being tested. Figure 3.1 shows the initial and final (t = 2π) numerical solution

11
 Figure 3.1: Initial and final states of the u-component of velocity.

for the case N = 64. As mentioned, with ν = 0 the solution is periodic in time and this figure demonstrates
that, as should be the case, the FDS numerical solution is unaltered after one flow-through time.
Next, in Figure 3.2, we show time histories of the u-component of velocity at the center of the domain
for the case in which ν = 0.1. It is clearly seen that the FDS solution (open circles) is converging to the
analytical solution (solid line).
Figure 3.3 is the key quantitative result of this verification test. In this figure we plot the rms error, εrms ,
in the u-component of velocity against the grid spacing. The error is defined by
s
1 M h k i2
εrms ≡ ∑ ij U − u(x , ȳ
i j k ,t ) , (3.6)
M k=1

where M is the number of time steps and k is the time step index. The spatial indices are (i = N/2, j = N/2)
and Uikj represents the FDS value for the u-component of velocity at the staggered storage location for cell
(i, j) at time step k; u(xi , ȳ j ,tk ) is the analytical solution for the u-component at the corresponding location
in space and time. The figure confirms that the advective terms, the viscous terms, and the time integration
in the FDS code are convergent and second-order accurate.

3.1.2 Decaying Isotropic Turbulence

In some cases the difference between verification and validation is not so clear. Once a model is well-
established and validated it may actually be used as a form of verification. Granted, such a test is not
as strong a verification as the convergence study shown in Section 3.1.1. Nevertheless, these tests are often
quite useful in discovering problems within the code. The case we examine in this section, decaying isotropic
turbulence, is highly sensitive to errors in the advective and diffusive terms because the underlying physics is
inherently three-dimensional and getting the problem right depends strongly on a delicate balance between
vorticity dynamics and dissipation. An even more subtle yet extremely powerful verification test is also
presented in this section when we set both the molecular and turbulent viscosities to zero and confirm that
the integrated kinetic energy within the domain remains constant. In the absence of any form of viscosity,
experience has shown that the slightest error in the advective terms or the pressure projection will cause the
code to go unstable. This verification is therefore stronger than one might initially expect.
In this section we test the FDS model against the low Reynolds number (Re) data of Comte-Bellot
and Corrsin (CBC) [38]. Viscous effects are important in this data set for a well-resolved LES, testing the
model’s Re dependence. Following [39], we use a periodic box of side L = 9 × 2π centimeters (≈ 0.566

12
 2.4 2.4
FDS FDS
2.2 Analytical Solution 2.2 Analytical Solution

2 2

1.8 1.8
u−velocity, m/s

u−velocity, m/s
1.6 1.6

1.4 1.4

1.2 1.2

1 1

0.8 0.8

0.6 0.6

0.4 0.4
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
time, sec time, sec

2 2
FDS FDS
Analytical Solution Analytical Solution
1.8 1.8

1.6 1.6
u−velocity, m/s

u−velocity, m/s
1.4 1.4

1.2 1.2

1 1

0.8 0.8

0.6 0.6

0.4 0.4
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
time, sec time, sec

Figure 3.2: Time history of the velocity at the center of the domain for a range of grid resolutions. The domain
is a square of side L = 2π m. The N × N grid is uniform. Progressing from left to right and top to bottom we
have resolutions N = {8, 16, 32, 64} clearly showing convergence of the FDS numerical solution (open circles) to the
analytical solution (solid line). The case is run with constant properties, ρ = 1 kg/m3 and µ = 0.1 kg/m/s, and a CFL
of 0.25.

m) and ν = 1.5 × 10−5 m2 /s for the kinematic viscosity. The non-dimensional times for this data set are:
x/M = 42 (initial condition), 98, and 171, where M is the characteristic mesh spacing of the CBC wind
tunnel and x is the downstream location of the data station. Considering the mean velocity in the CBC
wind tunnel experiment, these correspond to dimensional times of t = 0.00, 0.28, and 0.66 seconds in our
simulations.
The initial condition for the FDS simulation is generated by superimposing Fourier modes with random
phases such that the spectrum matches that of the initial CBC data. An iterative procedure is employed where
the field is allowed to decay for small time increments subject to Navier-Stokes physics, each wavenumber
is then injected with energy to again match the initial filtered CBC spectrum. The specific filter used here is
discussed in [40].
To provide the reader with a qualitative sense of the flow, Figure 3.4 shows the initial and final states
of the velocity field in the 3D periodic domain. The flow is unforced and so if viscosity is present the
total energy decays with time due to viscous dissipation. Because the viscous scales are unresolved, a
subgrid stress model is required. Here the stress is closed using the gradient diffusion hypothesis and the
eddy viscosity is modeled by the constant coefficient Smagorinsky model with the coefficient taken to be
Cs = 0.2 (see the Technical Reference Guide for further details).

13
 O(dx) O(dx)
2 2
O(dx ) O(dx )
u−velocity rms error

u−velocity rms error

−1
10
−1
10

−2
10

−2
10
−1 −1
10 10
dx dx

Figure 3.3: (Left) Convergence rate for the u-component of velocity with ν = 0 showing that the advective terms
in the FDS code are second-order accurate. The stars represent the rms error in the u-component for grid spacings of
∆x = L/N where L = 2π m and N = {8, 16, 32, 64}. The solid line represents first-order accuracy and the dashed line
represents second-order accuracy. The simulation is run to a time of t = 2π sec with a CFL of 0.25. The u-component
at the center of the domain is compared with the analytical solution at the same location. (Right) Same case, except
ν = 0.1, showing that the viscous terms in the FDS code are second-order accurate.

The decay curves for two grid resolutions are shown plotted on the left in Figure 3.5. For an LES code
such as FDS which uses a physically-based subgrid model, an important verification test is to run this peri-
odic isotropic turbulence simulation in the absence of both molecular and turbulent viscosity. For so-called
“energy-conserving” explicit numerics the integrated energy will remain nearly constant in time. This is
demonstrated by the blue line in the top-left plot in Figure 3.5. The deviations from identical energy conser-
vation (to machine precision) are due solely to the time discretization (the spatial terms are conservative as
discussed in [41]) and converge to zero as the time step goes to the zero. Note that strict energy conserva-
tion requires implicit time integration [42, 43] and, as shown by the red curve on the same plot where only
molecular viscosity is present in the simulation, this cost is unwarranted given that the molecular dissipation
rate clearly overshadows the relatively insignificant amount of numerical dissipation caused by the explicit
method. The FDS result using the Smagorinsky eddy viscosity (the black solid line) matches the CBC data
(red open circles) well for the 323 case (top-left). However, the FDS results are slightly too dissipative in
the 643 case (bottom-left). This is due to a well-known limitation of the constant coefficient Smagorinsky
model: namely, that the eddy viscosity does not converge to zero at the appropriate rate as the filter width
(here equivalent to the grid spacing) is decreased.
To the right of each decay curve plot in Figure 3.5 is the corresponding spectral data comparison. The
three black solid lines are the CBC spectral data for the points in time corresponding to dimensional times
of t = 0.00, 0.28, and 0.66 seconds in our simulations. As described above, the initial FDS velocity field
(represented by the black dots) is specified to match the CBC data up to the grid Nyquist limit. From there
the spectral energy decays rapidly as discussed in [40]. For each of the spectral plots on the right, the results
of interest are the values of the red and blue dots and how well these match up with the corresponding CBC
data. For the 323 case (top-right) the results are remarkably good. Interestingly, the results for the more
highly resolved 643 case are not as good. This is because the viscous scales are rather well-resolved at the
later times in the experiment and, as mentioned, it is well-known that the constant coefficient Smagorinsky
model is too dissipative under such conditions.
Overall, the agreement between the FDS simulations and the CBC data is satisfactory and any discrep-
ancies can be explained by limitations of the model. Therefore, as a verification the results here are positive
in that nothing points to coding errors.

14
Figure 3.4: Initial and final states of the isotropic turbulence field.

15
 −3
0.05 10
increasing time
0.045

0.04
kinetic energy (m2/s2)

−4
0.035 10

E(k), m3/s2
FDS no visc
0.03
FDS mol visc
FDS Smag
0.025
filtered CBC data
−5
0.02 10

0.015
LES Nyquist limit
0.01
−6
0.005 10 1 2 3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 10 10 10
time (s) k, 1/m

−3
0.07 10
FDS
increasing time
filtered CBC data
0.06

0.05
kinetic energy (m2/s2)

−4
10
E(k), m3/s2

0.04

0.03
−5
10
0.02
LES Nyquist limit

0.01

−6
0 10 1 2 3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 10 10 10
time (s) k, 1/m

Figure 3.5: (Left) Time histories of integrated kinetic energy corresponding to the grid resolutions on the right side
of the figure. In the 323 case (top), the CBC data (open circles) are obtained by applying a filter to the CBC energy
spectra at the Nyquist limit for an N = 32 grid. Similarly, for the 643 case (bottom), the CBC data are obtained from
filtered spectra for an N = 64 grid. Notice that the integrated FDS results for the 323 case compare better with the
filtered CBC data than the 643 results. This is a well-known limitation of the constant coefficient Smagorinsky model:
namely, that the eddy viscosity does not converge to zero at the appropriate rate as the filter width (here equivalent to
the grid spacing) is decreased. (Right) Energy spectra for the 323 case (top) and the 643 case (bottom). The solid black
lines are the spectral data of Comte-Bellot and Corrsin at three different points in time corresponding to downstream
positions in the turbulent wind tunnel. The initial condition for the velocity field (spectra shown as black dots) in the
FDS simulation is prescribed such that the energy spectrum matches the initial CBC data. The FDS energy spectra
corresponding to the subsequent CBC data are shown by the red and blue dots.

16
Chapter 4

Thermal Radiation

The Radiative Transport Equation (RTE) for an absorbing/emitting and scattering medium is

σs (x, λ)
h i Z
s · ∇Iλ (x, s) = − κ(x, λ) + σs (x, λ) Iλ (x, s) + B(x, λ) + Φ(s, s0 ) Iλ (x, s0 ) ds0 (4.1)
4π 4π

where Iλ (x, s) is the radiation intensity at wavelength λ, s is the direction vector of the intensity, κ(x, λ) and
σs (x, λ) are the local absorption and scattering coefficients, respectively, and B(x, λ) is the emission source
term. The integral on the right hand side describes the in-scattering from other directions. In the case of a
non-scattering gas the RTE becomes
h i
s · ∇Iλ (x, s) = κ(x, λ) Ib (x) − Iλ (x, s) (4.2)

where Ib (x) is the source term given by the Planck function (see below).
In practical simulations the spectral (λ) dependence cannot be solved accurately. Instead, the radiation
spectrum is divided into a relatively small number of bands and a separate RTE is derived for each band.
The band specific RTE is

s · ∇In (x, s) = κn (x) [Ib,n (x) − In (x, s)] , n = 1...N (4.3)

where In is the intensity integrated over the band n, and κn is the appropriate mean absorption coefficient
inside the band. The source term can be written as a fraction of the blackbody radiation

Ib,n = Fn (λmin , λmax ) σ T 4 /π (4.4)

where σ is the Stefan-Boltzmann constant. The calculation of factors Fn is explained in Ref. [44]. When the
intensities corresponding to the bands are known, the total intensity is calculated by summing over all the
bands
N
I(x, s) = ∑ In (x, s) (4.5)
n=1

There are numerous examples in the heat transfer literature of exact solutions, for simple configurations
of hot and cold objects, of the radiation transport equation.

17
4.1 Radiation inside a box (radiation_in_a_box)
This verification case tests the computation of three-dimensional configuration factor Φ inside a cube box
with one hot wall and five cold (0 K) walls. An overview of the test geometry is shown here:

H4

H3 c a

(y,z)

H2 dA

H1

The configuration factors are calculated at the diagonal of the cold wall opposite to the hot wall. The exact
values of the configuration factor from plane element dA to parallel rectangle H are calculated using the
analytical solution [44]

(y,z) ΦHdA (y,z) ΦHdA

0.025 0.1457 0.275 0.2135
0.075 0.1603 0.325 0.2233
0.125 0.1748 0.375 0.2311
0.175 0.1888 0.425 0.2364
0.225 0.2018 0.475 0.2391

Different variations of the case include the mesh resolution (203 and 1003 cells) and the number of radiation
angles (50, 100, 300, 1000, 2000). The exact and FDS results are shown here:

Spatial resolution 20x20x20 Spatial resolution 100x100x100

0.35 0.35
Exact Exact
2000 2000
0.3 1000 0.3 1000
300 300
0.25 100 0.25 100
50 50

0.2 0.2
ΦHdA
ΦHdA

0.15 0.15

0.1 0.1

0.05 0.05

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Position z Position z

18
4.2 Radiation from a plane layer (radiation_plane_layer)
This case tests the computation of three-dimensional radiation from a homogenous, infinitely wide layer
of hot gases. The temperature of the layer is 1273.15 K and the absorption coefficient, κ, is varied. The
thickness of the layer is fixed at 1 m, and the optical depth is τ = (1 κ) m−1 . Wall temperatures are set to
0 K. The results are compared against the exact solution S(τ) presented in [45]

S(τ) = Sb [1 − 2E3 (τ)] (4.6)

where Sb = σT 4 is the black-body heat flux from the radiating plane and E3 (τ) is the exponential integral
function (order 3) of the optical depth τ.
The FDS results are computed at two mesh resolutions in the x-direction (I=20 and I=150). For I=20,
both one-band and six-band versions are included to test the correct integration of heat fluxes over multiple
bands. For I=20, 2-D versions are also computed (J=1). The limiting case, τ = ∞, using a solid wall of
temperature 1273.15 K, is computed to test the wall heat flux computation. The exact values and FDS
predictions of the wall heat fluxes are given in the table below.

τ S(τ) FDS (I=20,J=20) FDS (I=20,J=1) FDS (I=150)

(m−1 ) (kW/m2 ) 1 band 6 bands 1 band 6 bands 1 band
0.01 2.8970 2.9180 2.9069 2.8364 2.8256 2.9258
0.1 24.9403 25.5501 25.4529 25.1078 25.0122 25.7045
0.5 82.9457 83.1309 82.8144 84.3719 84.0506 84.0264
1.0 116.2891 115.4051 114.9656 117.801 117.353 116.7751
10. 148.9698 148.9616 148.3947 148.9677 148.4005 148.9695
∞ 148.9709 148.9709 148.4037 147.9426 147.3793 148.9709

19
4.3 Wall Internal Radiation (wall_internal_radiation)
In-depth absorption of thermal radiation in a solid is computed using a two-flux model. In this example,
the accuracy of the two-flux model is tested in the computation of the emissive flux from a 0.10 m thick,
homogenous layer of material at a temperature of 1273.15 K surrounded by an ambient temperature of 10 K.
The absorption coefficient is varied to cover a range [0.01, 10] of optical depth, τ.
The exact solutions for radiative flux are the analytical solutions of plane layer emission [45]

S(τ) = Sb [1 − 2E3 (τ)] (4.7)

where Sb = σT 4 is the black-body heat flux from the radiating plane and E3 (τ) is the exponential integral
function (order 3) of optical depth, τ. The exact solutions and FDS results are shown in the table below.

τ S(τ) FDS
(m−1 ) (kW/m2 ) (kW/m2 )
0.01 2.897 2.950
0.1 24.94 26.98
0.5 82.95 93.90
1.0 116.3 128.4
10. 149.0 149.0

20
Chapter 5

Heat Conduction

This chapter contains examples that test the one-dimensional heat conduction solver in FDS. A one-dimensional
heat conduction equation for the solid phase temperature Ts (x,t) is applied in the direction x pointing into
the solid (the point x = 0 represents the surface)

∂Ts ∂ ∂Ts
ρs cs = ks + q̇000
s (5.1)
∂t ∂x ∂x
In cylindrical and spherical coordinates, the heat conduction equation is written
   
∂Ts 1 ∂ ∂Ts ∂Ts 1 ∂ 2 ∂Ts
ρs cs = rks + q̇s ; ρs cs
000
= 2 r ks + q̇000
s (5.2)
∂t r ∂r ∂r ∂t r ∂r ∂r

FDS offers the user these options, with the assumption that the obstruction is not actually recti-linear, but
rather cylindrical or spherical in shape. This option is useful in describing the behavior of small, complicated
“targets” like cables or heat detection devices.

21
5.1 Simple Heat Conduction Through a Solid Slab (heat_conduction)
Analytical solutions of transient, one-dimensional heat conduction through a slab can be found in Refs. [46]
and [47]. Four cases are examined here. In each, a slab of thickness L = 0.1 m is exposed on one face to
an air temperature of Tg = 120 ◦ C. The other face is insulated (adiabatic). The convective heat transfer from
the gas to the slab is q̇00c = h (Tg − Ts ), where h is constant, and Ts is the slab face temperature. No thermal
radiation is included.
Case k ρ c h Bi
(W/m/K) (kg/m3 ) (kJ/kg/K) (W/m2 /K) hL/k
A 0.1 100 1 100 100
B 0.1 100 1 10 10
C 1.0 1000 1 10 1
D 10.0 10000 1 10 0.1
150 150
Surface Temperature (heat conduction a) Surface Temperature (heat conduction b)
125 125
Temperature (◦ C)

Temperature (◦ C)
100 100
Analytical (Front) Analytical (Front)
75 Analytical (4 cm) 75 Analytical (4 cm)
Analytical (Back) Analytical (Back)
50 FDS (Front) 50 FDS (Front)
FDS (4 cm) FDS (4 cm)
25 FDS (Back) 25 FDS (Back)

0 0
0 5 10 15 20 25 30 0 50 100 150 200 250 300
Time (min) Time (min)
150 150
Surface Temperature (heat conduction c) Surface Temperature (heat conduction d)
125 125
Temperature (◦ C)

Temperature (◦ C)

100 100
Analytical (Front) Analytical (Front)
75 Analytical (4 cm) 75 Analytical (4 cm)
Analytical (Back) Analytical (Back)
50 FDS (Front) 50 FDS (Front)
FDS (4 cm) FDS (4 cm)
25 FDS (Back) 25 FDS (Back)

0 0
0 100 200 300 400 500 600 0 200 400 600 800
Time (min) Time (min)

22
5.2 Temperature-Dependent Thermal Properties (heat_conduction_kc)
This example demonstrates the 1-D heat conduction in cartesian, cylindrical and spherical geometries with
temperature-dependent thermal properties. The cartesian solution was computed using HEATING (ver-
sion 7.3), a multi-dimensional, finite-difference, general purpose heat transfer model [48]. The cylindrical
and spherical solutions were computed using a commercial finite-element solver, ABAQUS.
The sample of homogenous material is initially at 0 ◦ C and at t > 0 exposed to a gas at 700 ◦ C. A
fixed heat transfer coefficient of 10 W/m2 /K is assumed. The density of the material is 10000 kg/m3 . The
conductivity and specific heat are functions of temperature with the following values: k(0) = 0.10 W/m/K,
k(200) = 0.20 W/m/K, c(0) = 1.0 kJ/kg/K, c(100) = 1.2 kJ/kg/K, c(200) = 1.0 kJ/kg/K. The thickness
(radius) of the sample is 0.01 m. In the cartesian case, the back surface of the material is exposed to a gas at
0 ◦ C. In the figure below, the light colored solid lines are FDS results and the dark lines are the HEATING
results. An example input with cylindrical geometry looks like:
&MATL ID='MAT_1'
EMISSIVITY = 0.0
CONDUCTIVITY_RAMP='K_RAMP'
SPECIFIC_HEAT_RAMP = 'C_RAMP'
DENSITY=10000. /

&RAMP ID = 'K_RAMP' T=0, F= 0.10 /

&RAMP ID = 'K_RAMP' T=100, F= 0.15 /
&RAMP ID = 'K_RAMP' T=200, F= 0.20 /
&RAMP ID = 'C_RAMP' T=0, F= 1.00 /
&RAMP ID = 'C_RAMP' T=100, F= 1.20 /
&RAMP ID = 'C_RAMP' T=200, F= 1.00 /

&SURF ID='SLAB'
STRETCH_FACTOR = 1.0
GEOMETRY = 'CYLINDRICAL'
MATL_ID='MAT_1'
THICKNESS=0.01 /

250 50
Surface Temperature (heat conduction kc) Back/Inner Temperature (heat conduction kc)

200 40
Temperature (◦ C)

Temperature (◦ C)

150 HEATING (cart surf) 30 HEATING (cart back)

ABAQUS (cyl surf) ABAQUS (cyl back)
100 ABAQUS (sph surf) 20 ABAQUS (sph back)
FDS (cart front) FDS (cart back)
FDS (cyl front) FDS (cyl back)
50 10
FDS (sph front) FDS (sph back)

0 0
0 2.5 5 7.5 10 12.5 15 0 2.5 5 7.5 10 12.5 15
Time (min) Time (min)

23
24
Chapter 6

Pyrolysis

Solids can undergo simultaneous reactions under the following assumptions:

• instantaneous release of volatiles from solid to the gas phase,

• local thermal equilibrium between the solid and the volatiles,
• no condensation of gaseous products, and
• no porosity effects

Each material component may undergo several competing reactions, and each of these reactions may pro-
duce some other solid component (residue), gaseous fuel, and/or water vapor according to the yield coef-
ficients νs , ν f and νw , respectively. These coefficients should usually satisfy νs + ν f + νw = 1, but smaller
yields may also be used to take into account the gaseous products that are not explicitly included in the
simulation.

25
6.1 A Simple Two-Step Pyrolysis Example (two_step_solid_reaction)
Consider the set of ordinary differential equations describing the mass fraction of three components of a
solid material undergoing thermal degradation:

dYa
= −KabYa
dt
dYb
= KabYa − KbcYb (6.1)
dt
dYc
= KbcYa
dt
where the mass fraction of component a is 1 initially. The analytical solution is:

Ya (t) = exp(−Kabt)
Kab
Yb (t) = exp(−Kabt) − exp(−Kbct) (6.2)
Kbc − Kab
Yc (t) = [Kab (1 − exp(−Kbct)) + Kbc ∗ (exp(−Kabt) − 1)] /(Kab − Kbc ) (6.3)

The analytical and numerical solution for the parameters Kab = 0.389 and Kbc = 0.262 are shown here:

1
Solid Density (two step solid reaction)

0.8
Temperature (◦ C)

0.6 Analytical (Ya)

Analytical (Yb)
0.4 Analytical (Yc)
FDS (Ya)
FDS (Yb)
0.2
FDS (Yc)

0
0 5 10 15 20 25 30
Time (s)

26
6.2 Development of surface emissivity (emissivity)
For thermally thick surfaces, the surface emissivity is computed as a mass-weighted sum of the individual
material emissivities in the first condensed phase grid cell. In the verification test, the initial material, having
emissivity of 1.0, is converted to another material, having emissivity of 0.0, at a constant rate of 0.1 s−1 . As
a result, the surface emissivity should change linearly from 1.0 to 0.0 in 10 s.

1
Surface Emissivity (emissivity)

0.8

0.6
Emissivity

0.4

Analytical (Emissivity)
0.2
FDS (EMISSIVITY)

0
0 2 4 6 8 10
Time (s)

27
6.3 Enthalpy of solid materials (enthalpy)
Consider a thin plate of conductive material that is exposed on one side to an elevated temperature heat
source and exposed on the other to an ambient temperature void. In the thermally-thin limit, the temperature
of the slab is governed by the following equation

dTs q̇00front + q̇00back

= (6.4)
dt cs ρs δ

In this example, the initial exposure to the front side of the slab is 3 kW/m2 . The original material (call it A)
undergoes a reaction to form material B. The reaction rate is constant, 0.2 s−1 , which in this case means that
material A disappears in exactly 5 s. This is achieved by setting ns and E to 0 and A to 0.2 in the reaction
rate term:
ρs,A ns
   
E
r= A exp − (6.5)
ρs0 RTs
The density and conductivity of both materials are 30 kg/m3 and 10 W/m/K, respectively. The emissivity of
front and back is 1. The specific heat of material A changes from 1.0 kJ/kg/K to 0.1 kJ/kg/K above 80 ◦ C,
while the specific heat of material B is constant at 1.0 kJ/kg/K. The slab is 1 mm thick.

200
Surface Temperature (enthalpy)

150
Temperature (◦ C)

100

50 Analytical (T slab)
FDS (T slab)

0
0 1 2 3 4
Time (s)

Note that the “analytical” solution is actually a simple numerical integration of the equations above with
a small time step to ensure accuracy. This example tests a number of features, including the reaction rate,
mass weighted specific heats, and radiation boundary conditions. Note that the convective heat transfer has
been turned off, and the correct steady-state temperature is calculated by FDS.

28
Bibliography

[1] K.B. McGrattan, B.W. Klein, S. Hostikka, and J.E. Floyd. Fire Dynamics Simulator (Version 5),
User’s Guide. NIST Special Publication 1019-5, National Institute of Standards and Technology,
Gaithersburg, Maryland, October 2007. i

[2] American Society for Testing and Materials, West Conshohocken, Pennsylvania. ASTM E 1355-04,
Standard Guide for Evaluating the Predictive Capabilities of Deterministic Fire Models, 2004. i, 1, 9

[3] W. Mell, K.B. McGrattan, and H. Baum. Numerical Simulation of Combustion in Fire Plumes.
In Twenty-Sixth Symposium (International) on Combustion, pages 1523–1530. Combustion Institute,
Pittsburgh, Pennsylvania, 1996. 3

[4] K.B. McGrattan, H.R. Baum, and R.G. Rehm. Large Eddy Simulations of Smoke Movement. Fire
Safety Journal, 30:161–178, 1998. 3

[5] H.R. Baum, R.G. Rehm, P.D. Barnett, and D.M. Corley. Finite Difference Calculations of Buoyant
Convection in an Enclosure, Part I: The Basic Algorithm. SIAM Journal of Scientific and Statistical
Computing, 4(1):117–135, March 1983. 3

[6] H.R. Baum and R.G. Rehm. Finite Difference Solutions for Internal Waves in Enclosures. SIAM
Journal of Scientific and Statistical Computing, 5(4):958–977, December 1984. 3

[7] H.R. Baum and R.G. Rehm. Calculations of Three Dimensional Buoyant Plumes in Enclosures. Com-
bustion Science and Technology, 40:55–77, 1984. 3

[8] R.G. Rehm, P.D. Barnett, H.R. Baum, and D.M. Corley. Finite Difference Calculations of Buoyant
Convection in an Enclosure: Verification of the Nonlinear Algorithm. Applied Numerical Mathematics,
1:515–529, 1985. 3

[9] K.B. McGrattan, T. Kashiwagi, H.R. Baum, and S.L. Olson. Effects of Ignition and Wind on the
Transition to Flame Spread in a Microgravity Environment. Combustion and Flame, 106:377–391,
1996. 4

[10] T. Kashiwagi, K.B. McGrattan, S.L. Olson, O. Fujita, M. Kikuchi, and K. Ito. Effects of Slow Wind
on Localized Radiative Ignition and Transition to Flame Spread in Microgravity. In Twenty-Sixth
Symposium (International) on Combustion, pages 1345–1352. Combustion Institute, Pittsburgh, Penn-
sylvania, 1996. 4

[11] W. Mell and T. Kashiwagi. Dimensional Effects on the Transition from Ignition to Flame Spread
in Microgravity. In Twenty-Seventh Symposium (International) on Combustion, pages 2635–2641.
Combustion Institute, Pittsburgh, Pennsylvania, 1998. 4

29
[12] W. Mell, S.L. Olson, and T. Kashiwagi. Flame Spread Along Free Edges of Thermally-Thin Sam-
ples in Microgravity. In Twenty-Eighth Symposium (International) on Combustion, pages 2843–2849.
Combustion Institute, Pittsburgh, Pennsylvania, 2000. 4

[13] K. Prasad, Y. Nakamura, S.L. Olson, O. Fujita, K. Nishizawa, K. Ito, and T. Kashiwagi. Effect of Wind
Velocity on Flame Spread in Microgravity. In Twenty-Ninth Symposium (International) on Combus-
tion, pages 2553–2560. Combustion Institute, Pittsburgh, Pennsylvania, 2002. 4

[14] Y. Nakamura, T. Kashiwagi, K.B. McGrattan, and H.R. Baum. Enclosure Effects on Flame Spread
over Solid Fuels in Microgravity. Combustion and Flame, 130:307–321, 2002. 4

[15] W.E. Mell, K.B. McGrattan, and H.R. Baum. g-Jitter Effects on Spherical Diffusion Flames. Micro-
gravity Science and Technology, 15(4):12–30, 2004. 4

[16] A. Mukhopadhyay and I.K. Puri. An Assessment of Stretch Effects on Flame Tip Using the Thin
Flame and Thick Formulations. Combustion and Flame, 133:499–502, 2003. 4

[17] A. Hamins, M. Bundy, I.K. Puri, K.B. McGrattan, and W.C. Park. Suppression of Low Strain Rate
Non-Premixed Flames by an Agent. In Proceedings of the 6th International Microgravity Combustion
Workshop, NASA/CP-2001-210826, pages 101–104. National Aeronautics and Space Administration,
Lewis Research Center, Cleveland, Ohio, May 2001. 4

[18] K.B. McGrattan, R.G. Rehm, and H.R. Baum. Fire-Driven Flows in Enclosures. Journal of Computa-
tional Physics, 110(2):285–291, 1994. 4

[19] P. Friday and F. W. Mowrer. Comparison of FDS Model Predictions with FM/SNL Fire Test Data.
NIST GCR 01-810, National Institute of Standards and Technology, Gaithersburg, Maryland, April
2001. 6

[20] A. Bounagui, N. Benichou, C. McCartney, and A. Kashef. Optimizing the Grid Size Used in CFD
Simulations to Evaluate Fire Safety in Houses. In 3rd NRC Symposium on Computational Fluid
Dynamics, High Performance Computing and Virtual Reality, pages 1–8, Ottawa, Ontario, Canada,
December 2003. National Research Council, Canada. 6

[21] R.L. Alpert. SFPE Handbook of Fire Protection Engineering, chapter Ceiling Jet Flows. National Fire
Protection Association, Quincy, Massachusetts, 3rd edition, 2003. 6

[22] A. Bounagui, A. Kashef, and N. Benichou. Simulation of the Dynamics of the Fire for a Section of
the L.H.-La Fontaine Tunnel. IRC-RR- 140, National Research Council Canada, Ottawa, Canada,
K1A0R, September 2003. 6

[23] Y. Xin. Assessment of Fire Dynamics Simulation for Engineering Applications: Grid and Domain
Size Effects. In Proceedings of the Fire Suppression and Detection Research Application Symposium,
Orlando, Florida. National Fire Protection Association, Quincy, Massachusetts, 2004. 6

[24] J.A. Ierardi and J.R. Barnett. A Quantititive Method for Calibrating CFD Model Calculations. In Pro-
ceedings of the CIB-CTBUH International Conference on Tall Buildings, pages 507–514. International
Council for Research and Innovation in Building and Construction (CIB), 2003. 6

[25] G. Heskestad. SFPE Handbook of Fire Protection Engineering, chapter Fire Plumes, Flame Height
and Air Entrainment. National Fire Protection Association, Quincy, Massachusetts, 3rd edition, 2002.
6

30
[26] N.M. Petterson. Assessing the feasibility of reducing the grid resolution in fds field modeling. Fire
Engineering Research Report 2002/6, University of Canterbury, Christchurch, New Zealand, March
2002. 6
[27] A. Musser, K. B. McGrattan, and J. Palmer. Evaluation of a Fast, Simplified Computational Fluid
Dynamics Model for Solving Room Airflow Problems. NISTIR 6760, National Institute of Standards
and Technology, Gaithersburg, Maryland, June 2001. 6
[28] K.B. McGrattan, S. Hostikka, J.E. Floyd, H.R. Baum, and R.G. Rehm. Fire Dynamics Simulator (Ver-
sion 5), Technical Reference Guide. NIST Special Publication 1018-5, National Institute of Standards
and Technology, Gaithersburg, Maryland, October 2007. 7
[29] W. Zhang, A. Hamer, M. Klassen, D. Carpenter, and R. Roby. Turbulence Statistics in a Fire Room
Model by Large Eddy Simulation. Fire Safety Journal, 37:721–752, 2002. 7
[30] J. Smagorinsky. General Circulation Experiments with the Primitive Equations. I. The Basic Experi-
ment. Monthly Weather Review, 91(3):99–164, March 1963. 7
[31] J.W. Deardorff. Numerical Investigation of Neutral and Unstable Planetary Boundary Layers. Journal
of Atmospheric Sciences, 29:91–115, 1972. 7
[32] M. Germano, U. Piomelli, P. Moin, and W.H. Cabot. A Dynamic Subgrid-Scale Eddy Viscosity Model.
Physics of Fluids A, 3(7):1760–1765, 1991. 7
[33] D.K. Lilly. A Proposed Modification of the Germano Subgrid-Scale Closure Method. Physics of Fluids
A, 4(3):633–635, 1992. 7
[34] J. Hietaniemi, S. Hostikka, and J. Vaari. FDS Simulation of Fire Spread – Comparison of Model
Results with Experimental Data. VTT Working Papers 4, VTT Building and Transport, Espoo, Finland,
2004. 8
[35] C. Lautenberger, G. Rein, and C. Fernandez-Pello. The application of a genetic algorithm to estimate
the material properties for fire modeling from bench-scale fire test data. Fire Safety Journal, 41:204–
214, 2006. 8
[36] J.C. Adams, W.S. Brainerd, J.T. Martin, B.T. Smith, and J.L. Wagener. Fortran 95 Handbook: Com-
plete ISO/ANSI Reference. MIT Press, Cambridge, Massachusetts, 1997. 9
[37] R. McDermott. A nontrivial analytical solution to the 2-d incompressible Navier-Stokes equations.
http://randy.mcdermott.googlepages.com/NS_exact_soln.pdf, 2003. 11
[38] G. Comte-Bellot and S. Corrsin. Simple Eularian time correlation of full- and narrow-band velocity
signals in grid-generated, ‘isotropic’ turbulence. J. Fluid Mech., 48:273–337, 1971. 12
[39] Stephen M. de Bruyn Kops. Numerical simulation of non-premixed turbulent combustion. PhD thesis,
The University of Washington, 1999. 12
[40] R. McDermott, A. Kerstein, R. Schmidt, and P. Smith. Characteristics of 1D spec-
tra in finite-volume large-eddy simulations with ‘one-dimensional turbulence’ subgrid
closure. In American Physical Society, Division of Fluid Dynamics, Chicago, IL,
http://randy.mcdermott.googlepages.com/implied_filter.pdf, Nov. 2005. 13, 14
[41] Y. Morinishi, T. S. Lund, O. V. Vasilyev, and P. Moin. Fully conservative high order finite difference
schemes for incompressible flow. J. Comp. Phys., 143:90–124, 1998. 14

31
[42] F. E. Ham, F. S. Lien, and A. B. Strong. A fully conservative second-order finite difference scheme for
incompressible flow on non-uniform grids. J. Comp. Phys., 177:117–133, 2002. 14

[43] R. McDermott. Discrete kinetic energy conservation for variable-density flows on staggered
grids. In American Physical Society, Division of Fluid Dynamics, Salt Lake City, UT,
http://randy.mcdermott.googlepages.com/aps2007_notes.pdf, Nov. 2007. 14

[44] R. Siegel and J. R. Howell. Thermal Radiation Heat Transfer. Taylor & Francis, New York, 4th
edition, 2002. 17, 18

[45] Y.B. Zel’dovich and Y.P. Raizer. Physics of shock waves and high-temperature hydrodynamic phenom-
ena. Dover Publications, New York, 2002. Translated from the Russian and then edited by W.D.Hayes
and R.F. Probstein. 19, 20

[46] D. Drysdale. An Introduction to Fire Dynamics. John Wiley and Sons, New York, 2nd edition, 2002.
22

[47] H.S. Carslaw and J.C. Jaegar. Conduction of Heat in Solids. Oxford University Press, 2nd edition,
1959. 22

[48] K.W. Childs. HEATING 7: Multidimensional, Finite-Difference Heat Conduction Analysis Code
System. Technical Report PSR-199, Oak Ridge National Laboratory, Oak Ridge, TN, 1998. 23

32