Introduction

Fire modeling is something which is often found to be mysterious by attorneys. Yet,

understanding what it is, what it can do, and what it cannot do can be vital to successful
development of some types of fire cases. The purpose of this note is to present the basic
ideas so that they are understandable by the non-scientist. Thus, the information should be
of value also to fire investigators, claims adjusters, and other individuals involved with fire
losses. Most of them are not aware of either the strengths or the limitations of the fire
modeling. Thus, in this note the objective is to explain the process in simple terms, so that
a clear picture will emerge how fire modeling can and cannot be used. 

What is a model?
Before we can discuss fire models, we must explain what a scientist means by 'model.' The
meaning of this crucial term is essential to understand. A model of anything is, simply, a
systematic representation of that thing. Thus, for example, we can have

 thought models (or conceptual models)

 scale models, and
 mathematical models.

Fire modelling is used primarily to predict the spread of smoke and heat from fires.

The above three examples are probably the main 'representations' which are used by
scientists. A thought model is simply a proposed schema explaining how something works.
Scale models are often used in structural engineering, fluid dynamics, and have occasionally
been used in fire science. Model trains are familiar to all. A scale model in scientific work is
simply a reduced-size object on which certain measurements will be made. The category
which we want to discuss in this Note is the last type, the mathematical model. In general,
a mathematical model will be a series of equations which describe a certain process. If the
equations are simple enough, they can be solved on the hand calculator. More commonly,
the equations are not so simple. Consequently, a computer is required for their solution.
Thus, in the fire field, we would speak of "computer fire models." Nowadays, when one
speaks of a "fire model," it is usually understood that one is referring to a "computer fire
model." This is unnecessarily restrictive, however, and other types of models (such as scale
models) remain legitimate scientific forms of model.

A "computer fire model" is normally realized as a computer program. This again, is most
common, but not necessarily always true. A computer fire model, for example, could be
realized as only a flowchart. From the above, one can understand why fire modeling is often
taken to mean "use of computer programs for predicting fire," although this would be too
restrictive a definition.

What do fire models do?

By now, fire modeling has been in use for more than two decades. This author's computer
program COMPF was released in 1975 [1] and was the first computer program for predicting
room fires to be developed in the U.S. Research in several other countries, however, goes
back further. During the subsequent two decades, tremendous progress was made in the
field. Today, many persons who have only a limited knowledge of fire science have already
had a slight exposure to fire modeling. From this, they are apt to conclude that fire
modeling is something which allows scientists/engineers to 'wave a magic wand' and to
calculate the history of a fire just by working at their computer. On rare occasions, this can
be true. But normally, the situation is not so straightforward.
The function of COMPF was to predict the fire history within a single room. The history was
represented only after the time of 'flashover' within the room. Flashover is the point in a fire
(it does not occur in all fires) when the room "fills with flames." The hazard greatly
increases from that point on. Nowadays, various other types of computer fire models are
also available to the scientist. What kind of fire characteristics, then, can a fire model
predict? The list is limited only by the ingenuity of scientists, but we can cite
characteristics which are already routinely being computed:

 gas and surface temperatures

 flow rates of gas through openings
 heat fluxes impinging on surfaces
 smoke obscuration
 production of certain toxic gas species
 strength reduction and structural failure of building elements
 activation times for sprinklers and detectors

It can be noted that this list is weighted towards fluid mechanics and related themes. This is
not surprising, since a majority of the researchers creating fire models have been fluid
mechanics specialists. Models also exist for certain human behavior aspects (e.g., exiting
through corridors and stairs) although these have so far been very little used for practical
problem solving; thus their validity is generally unknown.

It should be noted that certain characteristics are usually not being computed. These

include:

 the ignitability of objects from small flames

 the spread of fire over surfaces
 the actual 'size' of the fire, that is, its heat release rate

A list of other fire characteristics that we cannot yet routinely predict has recently been
publicized [2]. The three characteristics above are three exceedingly important aspects of
fire, indeed heat release rate (HRR) has been referred to as the single most important
variable in describing fire hazard [3]. Likewise, there will not be a fire without ignition and,
in most cases, flame spread is also an essential trait of fire. The way that today's fire
models normally solve a problem is by being given the HRR as input. The flame spread
aspects are usually not made explicit. The most important role of flame spread is to
progressively involve greater areas in burning, that is, to cause a growth of HRR. Thus, if
we have a HRR versus time curve, the flame spread issue has already been solved. The
initial ignition is, simply, assumed to have taken place, so no computation is made there
either.

To make a computation using one of our state-of-the-art models, such as HAZARD [4], then
requires that the modeler supply a HRR curve as input. In some cases, the HRR curve may
already have been published in the literature for a 'similar' burning object. Compendia of
data are available which present some useful, non-proprietary data [5]. However, the
variety of items which can burn is essentially infinite, while the amount of publicly available
data is quite tiny.

The situation is even more complicated when one realizes that more than one item can
burn. Methods have been suggested for estimating second-item involvement [6]. However,
under most conditions, such procedures entail a great deal of uncertainty. This can be due
to: (a) irregular geometry of the item in question; (b) not well enough studied ignition
response of the item; (c) inadequately detailed knowledge of local heat fluxes, etc. When
one contemplates the uncertainties then associated with estimating the ignition for the
third, fourth, etc. item, it becomes clear that the ignition sequence of a roomful of diverse
items cannot be predicted with a reasonable degree of confidence.

Fire Testing
The solution to the above difficulty is actually straightforward: when data are not available,
run a fire test. Model development is a difficult, specialist task. Thus, one cannot expect to
say "improve the models," since progress could hardly be made on a time schedule to suit
fire litigation needs, even if the resources were available. What is possible to do on
relatively short notice is to organize fire tests.

Fire tests have their limitations, too. The largest fire that can be conducted indoors in a
laboratory, under controlled and instrumented conditions is about 20 megawatts. Physically,
this corresponds to one room or a couple of smallish rooms joined together. Fire models are
much less restricted in that respect. They are available for computing multi-story, multi-
room arrangements, and the rooms do not have to be small enough to fit under a
laboratory's exhaust hood.

Thus, the practical solution is to combine fire modeling with fire testing.

Normally, the objects, walls, etc. associated with ignition and early fire growth are directly
reconstructed in the laboratory by procuring exemplars and creating what is normally
termed a sectional full-scale mockup. Full-scale denotes that real appliances are used, real
wall thicknesses are employed, etc. Sectional denotes that only a slice out of the building is
constructed in the laboratory and not the whole fire environment.

The presumed or alleged ignition sequence is then started in the laboratory test and
measurements are taken of HRR, smoke production, temperatures, heat fluxes, and other
fire variables. Fire modeling is then used to take the laboratory data of the initial fire stages
as an input and to compute the subsequent stages of fire development. Thus, fire modeling
can be viewed as a direct extension of fire testing, or vice versa.

The confidence in the results produced by the fire model is normally greater for the
intermediate stages of the fire than for the late stages. During the late stages of fire, a
number of additional events can happen. These include burn-through of partitions, collapse
of beams, collapse of occupant goods (e.g., rack storage) and similar. Also, it may be
expected that firefighting will make some difference on the outcome of the fire, and this
may not be reasonable to try to predict mathematically. Models do exist which can allow the
prediction of the collapse of structural members, but these require input data which may
often be unavailable.

Tests vs. demonstrations

It is important to distinguish between a field demonstration and a large-scale laboratory
test. Both involve setting up of an environment intended to recreate the scene of the fire
origin. Both can be used to produce videos for jury viewing. However, a field demonstration
does not collect HRR nor other fire data which could usefully serve as input to a fire model.
Thus, demonstrations can only be used for video purposes.

The advantage of a demonstration is that it can be conducted in every town and city. A
laboratory test, by contrast, requires use of a fire testing laboratory, and there are only a
handful of such facilities in the country.

The costs, however, are not necessarily much lower for a demonstration. The bulk of the
cost is normally associated with procuring exemplars, constructing the mockup, setting up
video and other documentation, and witnessing of the test. Since a fire test laboratory
already has the HRR and other instrumentation necessary, the marginal cost is small for
setting up the instrumentation and collecting the necessary data. The actual laboratory test
procedures [7] are, by now, quite well worked out, and time does not need to be allocated
to research in this area.

Conclusions
Fire modeling can normally be considered as the prediction of fire characteristics by the use
of a mathematical method which is expressed as a computer program.

The needs of fire litigation from fire modeling are specialized. Usually, there is a great deal
of specificity about the sequence of fire ignition and the materials involved in the process.
This commonly precludes the use of handbook data as input to fire models. Instead, it will
usually be necessary to conduct a sectional full-scale mockup to obtain appropriate data
describing the initial part of the fire. This information then serves as input to a fire model,
using which the later fire development can be approximately predicted.
Introduction of Fire modelling

Of critical importance when using models for Fire Safety Engineering is that the user knows the
limitations and the appropriateness of the specific model being used. The misuse/incorrect use could
have great consequences; it could be that a life safety system (for example a smoke control system) is
designed using a tool that is not appropriate for that purpose. This could lead to an incorrect design, with
a possible result being a reduced level of safety for the occupants – particularly if an under-engineered
solution results in a system unable to deal with the actual situation, such as more smoke production than
a ventilation system was designed to manage because the design tool used was not appropriate for that
situation. Engineers are often responsible for the life safety analysis, and we are equipped with a number
of tools to aid the design process.  One challenge is to understand how to use such tools, but of equal
and paramount importance is understanding their limitations, and thus when their use is not appropriate."

Algebraic Models
Algebraic models are generally the first step employed during a fire dynamics analysis. Equations have
been developed to estimate the behaviour of fire effects such as flame height, heat release rate, plume
and ceiling jet velocities, gas layer temperatures and depth, radiation, and other phenomena.
Algebraic models employ equations that have been developed either empirically using observations from
fire test data or mathematically through direct calculation of physical properties and behaviours. These
equations often yield steady state results, providing information at a moment in time; however, many can
also be iterated to provide transient data by solving the equation(s) over a number of time steps.

The benefits of algebraic models are that they can provide an immediate answer to simple fire dynamics
questions and basic fire phenomena and that the results of these equations are directly based on
empirical data from testing or fundamental physics. The drawbacks of algebraic models are that they are
often tailored for specific simplified scenarios and that they generally provide information about a single
property of interest, rather than a more detailed representation of a fire’s behaviour as a whole. The
correlation’s predictive capability will most closely match the actual conditions when the scenario most
closely matches the test conditions from which the correlation came. The greater the differences between
the scenario and the original test conditions the lesser the confidence in the predictive answer of the
correlation. Even in cases where algebraic models are not chosen as the final method of analysis, it is
often useful to use these simple calculations to provide a check of the output obtained from more
complicated forms of modelling. Although the solutions to these models may not exactly match the output
from a zone model or computational fluid dynamics model, they are often of the same order of magnitude
as the results provided by the more complex models. If the results of an algebraic model and another
type of model are significantly different, further analysis should be conducted to determine why this is and
whether or not it is representative of an error in part of the analysis that should be corrected.  The
drawbacks of algebraic models are that they are often tailored for specific simplified scenarios and that
they generally provide information about a single property of interest, rather than a more detailed
representation of a fire’s behavior as a whole.  The correlation’s predictive capability will most closely
match the actual conditions when the scenario most closely matches the test conditions from which the
correlation came.  The greater the differences between the scenario and the original test conditions the
lesser the confidence in the predictive answer of the correlation.

Zone Models
When an algebraic equation does not provide enough detail or accuracy to determine the fire phenomena
of interest, the next step is often to use a zone model or lumped parameter model.
Zone models and lumped parameter models operate under the assumption that a given space can be
broken into a small number of volumes, or “zones”, over which a number of algebraic equations are
solved to determine the relevant flow properties.
For the purposes of computer fire modelling, these models generally consider two or three zones,
including the lower layer, the upper layer and sometimes a zone used to represent the plume or ceiling jet
in the compartment with the fire. Temperatures, velocities and other properties are assumed to be
uniform within these zones; the transfer of mass, energy, momentum and species are tracked from one
zone to another via equations that have been tailored for the “zone assumption”.

The two-layer approach to fire zone models is based on the phenomena that hot gases will collect at the
ceiling of a compartment while a cooler layer will be present below the hot gasses. Although experimental
results indicate that there are in reality variations in conditions across these zones, they have been found
to be smaller than the differences between the layers, thereby allowing a realistic simulation of many
common fire conditions using zone models. As such, zone models are often employed in order to track
properties associated with the transport of smoke and hot gases from the compartment of fire origin to
other portions of buildings via connected volumes. The outputs of zone models depend upon a user-
defined characterization of the design fire.
With advances in the capabilities of personal computers, it has become possible to run zone models of
multi-compartment fire scenarios in a matter of seconds or minutes. This allows fire modellers to obtain
more information than might be available through the use of algebraic models while still being able to test
a wide range of variables that may affect the behaviour of a fire. The drawbacks of the zone model are
that there are times when the assumption of uniform zones is not accurate, and that it is difficult (or at
times impossible) to model spaces that have complex geometries that are not accurately simulated using
equivalent rectangular volumes. Similarly, some zone model assumptions, such as the movement of hot
 gases to the ceiling, may not be true under a scenario with a large ceiling height where stratification
occurs.

Some zone models have additional sub-models to simulate specific phenomena not inherently consistent
with the zone assumption. For example, some zone models simulate detector or sprinkler actuation
through the use of ceiling jet calculations. They may calculate the velocity and temperature within a fire
room, but assume a velocity, outside the fire room, as the result of a door plume. In the case of sprinkler
actuation, the two-zone assumption breaks down when the sprinkler actuates. While this may be
adequately covered through suppression correlations when a single sprinkler operates, such an approach
does not currently work for multiple sprinklers simultaneously operating.

Field Models
Field models, or computational fluid dynamics (CFD) models, provide a method for modelling the fluid
flow through a volume using numerical solutions of the Navier-Stokes equations. The solutions of partial
differential equations for conservation of mass, momentum and energy are approximated as finite
differences over a number of control volumes. In most CFD software packages, these calculations can be
performed over a period of time to provide a transient solution. Rather than using two or three control
volumes as with a zone model, a CFD model is typically comprised of hundreds of thousands (or millions)
of control volumes. This achieves a finer level of resolution than models that utilize the zone assumption,
but at the cost of requiring greater computational resources and extended simulation times of days,
weeks or even months.

In addition to addressing fluid phenomena such as turbulence and buoyancy, the use of CFD models to
simulate a fire-driven flow should also account for physical phenomena such as the spread of hot gasses,
thermal radiation, and tracking of particles of smoke and water.
Benefits of CFD models include their basis in the fundamental physics of fluid flow, a flexibility that allows the
simulation of a large range of boundary conditions, and the ability to model much more detailed geometries
than can be achieved using algebraic or zone models. Drawbacks of CFD models include increased
complexity, elevated need for training of the model user, greater computational requirements, and a longer
timeline between initiating a project and completing the necessary simulations.
The most advanced and sophisticated fire modelling technique is the use of Computational Fluid Dynamics
(CFD) models to predict fire growth and compartment temperatures. CFD models have been shown to be
successful in the modelling of smoke movement and have recently been applied to the modelling of fires. They
are capable of modelling pre-flashover and localised fires in complex geometries with smoke movement in
multi-compartments.

Typical CFD models analyse systems involving fluid flow, heat transfer and associated phenomena by
solving the fundamental equations of the fluid flow. These equations represent the mathematical
statements of the conservation laws of physics:

 the mass of a fluid is conserved;

 the rate of change of momentum equals the sum of the forces on a fluid particle – the Newton’s
second law;
 the rate of change of energy is equal to the sum of the rate of heat increase and the rate of
work done on a fluid particle – the first law of thermodynamics.
Basically, in a CFD model, the partial differential equations of the thermodynamic and aerodynamic
variables (Navier-Stokes equations) are solved in a very large number of points in the compartments.
Most CFD models for enclosure fires are appropriate for low-speed, thermally-driven flow with an
emphasis on smoke and heat transport from fires.
Applications of scale modeling are common among many
engineering disciplines such as the design and analysis of
aircrafts, ships, vehicles, and bridges. The use of scale modeling
for fire applications has been explored by fire researchers for
many years, resulting in the successful development of
correlations for a wide range of fire phenomena. 1  As with other
disciplines, the main advantages to scale fire modeling include
the ease, cost, and feasibility of construction.

Recently, interest in scale fire modeling has increased in the fire

investigation community. 2 , 3  A series of full scale experiments
were completed in 2008 at the Bureau of Alcohol, Tobacco,
Firearms and Explosives (ATF) Fire Research Laboratory in
Beltsville, MD (Figure 1). These experiments were designed to
examine the fire dynamics of ventilation-limited compartment
fires in a scenario often encountered by fire investigators: fire
growth in a furnished bedroom. During these experiments,
temperature, heat flux, gas species, and video data were
collected, which provided a unique opportunity to explore a new
scaling technique for a realistic growing fire. Based on the 2008
full scale experiments, quarter (¼) scale experiments were
designed and completed in 2011 (Figure 2), pushing the envelope
of scaling by exploring fire growth in a furnished room. 4
Figure 1. 2008 Full Scale Tests Experiments

Figure 2. 2011 Quarter Scale Experiments

Scaling Theory
In order to properly use scale modeling to gain understanding of
the fire phenomena in full scale structures of interest, it is
necessary to understand the underlying scaling theory to identify
and include the important variables. Due to the complex and
multivariable nature of fire, it is not possible to get perfect or
complete scaling. Therefore, the fire dynamics must be
appropriately represented through partial scaling. This can
quickly become a difficult endeavor with complex geometries and
fuel packages.

Partial scaling is achieved by assessing the governing

conservation equations (e.g. mass, momentum, energy, and
species) and selecting the appropriate dimensionless groups to
keep consistent between the full scale and small scale
experiments. These dimensionless groups are determined by
deciding which aspects of the fire behavior are priorities for the
particular scenario. In other words, the scaling of the heat
transfer, fluid flow, and burning rate cannot all be perfectly
matched, but one can decide to try to scale the fire power and
heat losses in order to maintain the same temperature and
species concentrations at similar times for the model and full
scale. More detailed description of scaling theory and previous
variations, including applications of Froude scaling which were
used in these experiments, are provided by Quintiere 1 , 5  and
Carey. 2

While the fire growth in a furnished bedroom is a common

scenario among fire investigators, it is difficult to scale such a
fire. To explore this, a new scaling methodology based on the
principles of Froude scaling was used to complete quarter scaled
experiments at the ATF Fire Research Laboratory in 2011. Quarter
scaling implies that the ratio of a length scale   (l)  in the model to
that of the full scale is one to four ( 1 m /1 F S   = 1/4 ). Froude scaling
requires that the fire power and heat loss follow
or  1  or  (1 /1 ) , but this cannot be achieved for a growing fire
5/2 m FS 5/2

due to conflicting behavior of radiation and convection. The

criterion for scaling for these experiments was decided based on
previous work, understanding of fire behavior, as well as the ease
of construction and convenience. The requirements for the
quarter scale model included:

1. Scaling the dimensions, openings, and objects geometrically

[11]
2. Maintaining the same materials and same thicknesses of
construction and contents [ 1 0 ]

In order to meet both of the above requirements, the full scale

material thicknesses were contained within the geometric
boundaries of the quarter scale objects as much as possible.
Keeping all of the materials the same allows for the heat transfer
and burning to be the same over time, given the same heat flux.

The overall layout of the bedroom and location of instrumentation

that was included in both the full scale and quarter scale
experiments are displayed in Figure 3. For both series of
experiments, the fire was ignited in the trash can next to the
chair in the rear of the compartment.

Figure 3. General room layout with instrumentation locatio

(Note: THF = Total Heat Flux Gauge, GS = Gas species, TC = Thermocouple)

Results
A comparison of the longest duration full scale (FS3) and quarter
scale (QS2) experiments demonstrated that this scaling approach
provided a practical means of conducting scaled experiments,
similar overall fire progression, and a means of predicting specific
relationships among the variables measured.
Table 1 lists a comparison of the times that events occurred in
the full scale and quarter scale experiments. Figure 4 – Figure 6
show a summary of the data comparison between FS3 and QS2
experiments. These fires were allowed to burn with steady flames
out of the door for approximately two minutes prior to
extinguishment.

Table 1. Summary of events with approximate times

Event FS Time (s) QS Time (s) Time FS - Time QS (s)

Accelerated flame spread over chair 117 103 + 14

Oxygen conc. zero near ceiling & chair 170 150 + 20
Similar fire shapes and near equal
temperatures and concentrations 165 165 =
Flames move from origin 225 255 - 30
Flames emerge from door 255 300 - 45
Center vertical temperatures uniform at
about 750 to 800 ºC ~255 ~300 - 45
Figure 4. Temperature at "TC-B" for three heights: TC1 – near the
floor, TC4 – near the middle,
TC7 – near the ceiling
Figure 5. Heat flux at "THF-B” and "THF-C"
Figure 6. Gas Species at "GS-B (high)”

Analysis
The early fire growth was expected to be faster in the model
based on lateral flame spread theory. The flame spread velocity is
directly proportional to the induced air flow, and in Froude scaling
the air velocity follows  l 1 / 2 , therefore:

tspread,QS  - tspread, FS   (1/4) 1 / 2   =  t s p r e a d , FS

/2

This faster lateral spread was the likely reason for higher
temperatures in the quarter scale experiment during early growth
of the fire (Figure 4). Gas species followed this behavior as well
(Figure 6). Additionally, interior video showed the onset of rapid
flame spread over the chair occurred at 103 seconds in the model
and 117 seconds for the full scale.

At about 150 to 170 seconds, the model and full scale

temperatures and concentrations were nearly the same. A photo
of the fire conditions at 165 seconds in Figure 7 indicates that the
fire shapes were also nearly the same.

Figure 7. Flames similar shape between FS3 (left) and QS2

(right).

Theory indicates that Froude scaling should work when fire sizes
are the same relative size (i.e. flame height and diameter ~   l 1 )
because  Q  follows  l 5 / 2 ; therefore the temperatures and
concentrations should be the same. This was supported by the
data in these experiments.

Oxygen concentrations measured above the fire near the ceiling

were zero at about 150 seconds for the quarter scale and about
170 seconds for the full scale, which is an indication of
ventilation-limited conditions. Theory indicates that for the
ventilation-limited conditions, the concentrations of the model
and full scale should be the same because the burning rate is
controlled by the inflow of air, which follows   l 5 / 2 . Figure 6 shows
that these equalities persisted from about 150 to about 380
seconds, when the fires were terminated.

After the fire sizes were the same relative size (approximately
165 seconds), the full scale temperatures and heat fluxes were
greater because of an increase in radiant heating by the thicker
smoke layer. This led to more rapid fire growth in the full scale,
and flames were observed to move away from the origin earlier in
the full scale (225 seconds) than the quarter scale (255 seconds).
The measurements of the corner heat flux gauges dropped at
about these times (Figure 5), indicative of the flame moving away
from the corner.

Later, at about 300 seconds in the model and 254 seconds in the
full scale, the flame emerged from the door and moved away from
the opposite far wall. Roughly correlating with the flames
emerging from the door, the temperatures from floor to ceiling
within the compartment merged to a unified value in both the
model (300 seconds) and full scale (255 seconds).

References

1. Quintiere, J.G., "Scaling Applications in Fire Research,”   Fire

Safety Journal,  15, 1989 p. 3-29.
2. Carey, A.C., "Scale Modeling of Static Fires in a Complex
Geometry for Forensic Fire Applications,” MS thesis,
University of Maryland, College Park, 2010.
3. Campbell, M.A., "Current Research and Future Expectations
of RSE Burning,”  Proceeding of the 5th International
Symposium on Fire Investigation Science and Technology ,
2012.
4. Quintiere, J.G., Reeves, L.S., McCarthy, L.K. "Scale Model
Testing of Fire Growth in a Furnished Room,” Presentation at
The International Association of Arson Investigators, Inc.
(IAAI) 2012 Annual Training Conference, Dove, DE, 2012.
5. Quintiere, J.G.,  Fundamentals of Fire Phenomena , Wiley, p.
377-403, 2006.