C O M B U S T I O N A N D F L A M E 38: 159-171 (1980) 159

Flashover and Instabilities in Fire Behavior

P. H. THOMAS and M. L. BULLEN

Fire Research Station, Borehamwood, Hertfordshire. England

and

J. G. QUINTIER E and B. J. MeCAFFREY

u.s. National Bureau of Standards, Washington, D.C. 20234

One kind of flashover in compartments can result from a thermal instability caused by the energy generation rate
increasing faster with temperature than the rate of aggregated energy losses. This paper describes qualitatively how
various factors affect this instability, and demonstratesa quasi-steadyapproachthat can be used to explain the growth of
fires in enclosures. Thermal radiative feedback from the enclosure is considered to be the significant factor in
determining the fuel gasification rate and is the primary cause of the instability. For a plausible expressionfor thermal
feedback, it is found that the upper enclosure mean gas temperature at the onset of instabilitycan range from 300°C to
650°C in most cases. However, the correspondingfuel gasificationrate for a fixed surface area fuel at criticality is only
roughly 50% greater than the free or open burn value. Finally, the possibility of an '*extinction" instability for
ventilation controlled fires is also indicated.

INTRODUCTION cally, this can be interpreted as considering the fire

development to be quasi-steady. Obviously, this
In recent years there has been great interest in at- approximation is only likely to be valid before and
tempting to predict the development o f fire in a after the phenomenon of flashover has occurred,
room. In general, following the ignition of an ob- Quasi-steady changes could lead to a " j u m p " from
ject in a room, the fire will commence to spread. one state to another, and flashover might be repre-
Some have described this initial burning rate as sented by such a discontinuity in the state of the
exponential in time. Qualitatively, this growth fire. F o r a reasonably simple, yet fairly general
process is perceived as initially slow followed by a model, we shall examine the nature of steady-state
rapid acceleration in the fire growth which results solutions and show that there is generally more
in full room involvement. The transition to full than one possible solution.
room involvement is commonly referred to as Initially, fire in an enclosure develops in excess
"flashover." The mechanism (or mechanisms) air, and energy release here is regarded as being
that promotes flashover and a quantitative under- determined by the rate of heat transfer to the
standing o f this phenomenon are still being sought. burning surface. Any effects of gas-phase kinetics,
In this paper we shall examine the role o f thermal other than those that can be accommodated by
feedback from the fire and the enclosure to the changing the fuel heat of combustion, are dis-
burning fuel in bringing about flashover. regarded [2, 3]. For enclosures with limited venti-
The approach adopted will be to examine a sim- lation, the rate of energy release is primarily
plified general energy equation describing fire in governed by the flow rate o f available air. Quin-
an enclosure in its steady-state form [1, 2]. Physi- tiere [ l ] and Bullen [4] have considered this form
© Crown copyright.
Published by Elsevier North Holland, Inc.,
52 Vanderbilt Avenue, New York, NY 10017 0010-2180/80/050159+13501.75
160 P.H. THOMAS ET AL.

of energy release in their steady-state analyses of tion of time, t) and can be an explicit function of
enclosure fires. Solutions by Quintiere et al. [5] time as well. It can be shown [ 1, 4] that loss rate
to the steady-state equations describing plastic equals the sum of the energy loss by mass trans-
pool fires in enclosures have yielded multi-valued port through an opening, radiation loss through
results. For example, at certain values of the fuel the opening, and heat transfer losses to the solid
surface area there can be as many as three values boundary surfaces; that is,
of fuel gasification rate satisfying the conserva-
tion equations. Under a quasi-steady assumption L = (m~, + m r ) c ( T - To) + U A ( T - To), (2)
for a growing fire increasing in area, this multi-
valued result must be interpreted as a step increase where
in burning rate at some critical fire area. The source
rh is the mass flow rate
of this multiplicity of solutions is the positive ther-
U is an effective heat transfer coefficient
mal feedback from the enclosure that promotes an
A is an effective heat transfer area.
increase in the fuel gasification rate. We shall con-
sider this situation more generally by investigating The subscripts 0, a, and f refer to ambient condi-
the solutions to the steady equations of fire growth tions, air and fuel, respectively. Initially, we shall
in a simple form. consider L to be linear in T as a first approxima-
tion.
The energy generation term is determined by
THEORY-THREE EQUILIBRIA the air or fuel supply rate. It can be shown that

For the most part our approach will be qualita- R = (1 - ¢)mf, rAh c (3)
tive and general but will not preclude us from ul-
timately drawing some quantitative conclusions. where
For the purpose of illustrating the nature of solu-
tions, fire growth can be adequately described by Ahc is the effective heat of combustion
a single energy equation and a single representative f is the fraction of total energy release
temperature. An idealized energy balance is writ- emitted by the fire plume as radiation
ten for a fire plume and hot layer, as illustrated in which does not enter the upper layer
Fig. 1. This leads to an equation of the form* mr, r is the rate of fuel burned in the enclosure.

Under conditions of excess air (fuel-controlled

d(mr)
c = R ( T , t ) - - L ( T , t), (1) fires),
dt
m~
mr, r =rag for .-7--->~r, (4a)
where mr
m is the mass of the fluid in the layer
and for insufficient air (ventilation controlled fires)
c is the specific heat of the layer gases
we assume that
T is its temperature
t is time ma rn a
R and L represent the rate of energy gains and mr,, = - - for .--- < r, (4b)
losses, respectively. r mr

R and L can be represented as functions of the where r is the stoichiometric air to fuel mass ratio.
dependent variable T (which is, in general, a func- In order to consider the effect of thermal feed-
back, we postulate the simple condition of
* A more rigorous derivation of this equation would lead
to some additional terms not considered here. These .4r0"(r)
terms are generally small and their omission will not m r -- - - (5)
affect the general conclusions of this analysis. Ahoap
FIRE I~LASHOVER AND INSTABILITIES 161

,d UAIT-To)

y -],H:

Af
Fig. 1. Symbolic diagram of an energy balance for the hot layer in an enclosure fire.

Here, A t, is the surface area of the pyrolyzing fuel, the reader. A single generation curve is considered
Z~hvap is an effective heat of vaporization, and that is representative of a given fuel at a fixed sur-
q"(T) represents the net heat flux to the surface of face area. Three loss curves are shown that may
the fuel. The heat flux to fuel burning in the open, represent three types of wall insulation or three
outside of a compartment, would be represented compartments sizes (i.e., three effective enclosure
as q"(To); that is, the flux is expressed as a func- areas, A). For interest it is pointed out that the
tion of the mean temperature of the surroundings characteristics of these curves for R and L bear a
not that of the local flame. In general, dI"(T) close analogy to a Semenov diagram [7] used to
represents the flux that would occur under enclo- describe thermal explosion theory.
sure fire conditions with an upper layer gas tem- Figure 2 shows that there may be three possi-
perature, T. q"(T) would depend on the size of the ble intersections of the R and L curves, that is,
enclosure, the soot concentration of the layer, and three equilibria: A, B, and C. Each equilibrium
other factors. Because the heat flux from the layer point is a solution to the steady energy equation
to the fuel is in all cases partly if not wholly radia- given by Eq. 1. Equilibria of the type C are typical
tive, q"(T) is an increasing function of T. This is in of small fires and have been analyzed by Quin-
contrast to a linear function of T suggestive of tiere [1]. On the other hand, those discussed by
convective heating alone. Bullen [4] and Thomas [2] were essentially of
With this background, we consider the steady types B and A for large fully developed fires. Also
form of Eq. 1. In particular d(mT)/dt = 0, or is Babrauskas and Wickstrom [6] have made detailed
negligible (along with other transient terms that calculations to locate A and B, but did not con-
may appear in a more general treatment), but R sider C. Equilibria A and C are stable solutions,
and L may be slow-varying functions of time ex- whereas B is unstable [2, 6]. This can be shown by
plicitly. This is what we mean by quasi-steady considering small changes of temperature about
fire growth. Solutions are sought graphically in the equilibria. Perturbations about A and C pro-
Fig. 2 by plotting the functions R and L against T duce no change in state. However, a small increase
(for a fixed time) to seek their intersections. The in temperature at B results in R larger than L
simple case of rha and UA independent of T and which leads to further increase in temperature.
m r ~ rh a is considered for illustration. The restric- Ultimately, the fire state settles at point A. Simi-
tions of this simple case are not valid under all larly a small decrease in temperature at B drives
circumstances (e.g., high temperatures or ventila- the fire state to equilibrium C.
tion-controlled fires), but they are made in order State C is a fuel-controlled fire representative
to introduce the general concepts of this paper to of the developing fire. As the fire grows toward
162 P.H. THOMAS ET AL.

z
/ /
R (VENTILA_TION
CONTROLLEO)
0
Z

._J
~.-
~,-
I.bl
Z

R(T0]

TO ENCLOSURETEMPERATURE,T
Fig. 2. The three equilibria for a system including positive feedback.

flashover, insofar as growth is quasi-steady, the the constancy of m~ and U, can be relaxed and re-
curve R may rise since for given T, q"(T) rises as placed with quantitative models which express the
the wall temperature approaches T. L may fall as dependence of R and L on T more accurately•
the walls warm up and less energy is absorbed by The fire-induced flow of air into an enclosure
the walls• Hence, R and L move with time so that through a door or window can be expressed as [1]
the point C approaches B. When R and L are
tangential, C coincides with B, and an additional
increase in temperature would result in transition
to point A. As indicated in Fig. 2, this transition
would manifest itself as a jump increase in tem- x ( H - H . ) al2 - m r , (6)
perature and rate of energy release, and may be
termed an "ignition" instability• This critical con- where
dition at tangency is suggestive of the phenom-
W and H are the width and height of the opening
enon of flashover. The essential conditions for this
C, Po, g, To are constant parameters
"jump" lie in the nature of the L and R dependence
H,, is the height of the doorway or window sec-
on temperature and require relatively weak con-
tion through which air enters-this is of the
straints on the form of the model• We shall exa-
order of H/2.
mine some ways in which this instability can occur.
For small fires rh~ is usually much less than rha,
THREE TYPES OF FIRE AND CAUSES OF but rh~ can become significant for large ventilation-
INSTABILITY controlled fires• Thus, by Eq. 6, m a has a depend-
ence on opening geometry as WHa/z, the ventila-
Before we elaborate on the utility of Fig. 2 and its tion factor, and increases quickly with tempera-
application to understanding fire behavior, let us ture T, then decreases at large T, but remains
generalize• The previous restrictions, in particular, fairly independent of T between 400 and 1000 K.
FIRE FLASHOVER AND INSTABILITIES ! 63

The term U A ( T - To) in Eq. 2 represents the where hr' is an effective radiative coefficient that
rate of heat transfer to the wall and ceiling sur- increases with T and e. Thus, it follows that
faces bounding the upper layer, and also the rate
of radiative heat transfer to the cooler floor below A~
UA - + A ihr'. (70
and through openings. An examination of these 1 1
+ - -
processes will reveal the qualitative dependence of hk (hc + hr)
UA on T and t. The heat transfer rate to the upper
surfaces can be represented as Except for large enclosures that have taken a long
qu = A u ( h c + hr)( T - Ts), (7a) time to be "filled" by the fire gases, A u is roughly
equal to the upper half enclosure surface area prior
where to room flashover and is almost unchanging with
time. A i is approximately equal to the floor area.
h e is a convective heat transfer coefficient
The nature of the heat transfer coefficients indi-
h~ is a radiative heat transfer coefficient
cate that UA mainly increases with T and layer
Au is the upper surface area heated.
emissivity and decreases asymptotically with time
Ts is the upper surface temperature.
as less heat is conducted into the enclosure surfaces.
Also, the heat conducted into the solid can be Equations 2 to 7 provide sufficient information
qualitatively represented as to describe the general dependence of R and L on
T. Thus, for a fixed time; burning area, At; and
?lu = A uh~ (t)(T, -- To), (7b)
geometrical, fuel, and material properties, the R
where and L curves can be plotted. These generalizations
lead to an upward curvature of the L-curve, and a
hk is an effective conductive coefficient.
downward curvature of the ventilation-controlled
In general, h e increases slightly with mr, h~ in- fire branch of the R-curve. This is shown in Fig. 3.
creases with layer emissivity and T, and h k de- In this figure, a single loss curve, representative of
creases asymptotically with time, t. Equations 7a a fixed enclosure geometry, is shown along with R-
and 7b can be equated to solve for Ts; that is, T, = curves for three types of fire. Type 1 is a small fire
Ts(T, t). From this result, it follows that with relatively small thermal feedback. An increase
in A r with time as the fire spreads can lead to type
A . ( T - - To) . (7c) 2 in which an instability can occur. Finally, there
1 1 is type 3 which represents a large fire relative to
+ - - the available rate of air flow. This could be repre-
hk (he + hr) sentative of a spillage fire, ignition of gas issuing
from a fractured pipe, or a "small" fire in an al-
Radiation loss to the lower space, assumed to be at most closed compartment fed only by air leading
temperature To, is represented as around a door. There is no flashover of the type
given by the B-instability in a type 3 fire. In this
ih = A i o [ e T 4 + (1 -- e)Ts 4 - - To4], (7d)
case, any manifestation of flashover would be due
where to the transient flame propagation characteristics
of the fuel itself. Thus, evolution to state A for
Ai is the surface area of the layer bounded by the type 3 fire in Fig. 3 is not necessarily a quasi-
the lower space and openings steady process.
o is the Stefan-Boltzmann constant In order to consider factors that would pro-
e" is the layer emissivity. mote the occurrence of instabilities of the form B,
Again 7", can be eliminated, and Eq. 7d can be several examples are illustrated in Fig. 4.
written as In Fig. 4(a) the initial fire state is A 1, a ventila-
tion-controlled fire. If a door is opened, or burned
dll = A i h r ' ( T -- To), (7e) through, the L-curve will move up as indicated
164 P . H . T H O M A S ET A L

i,a.i

=.

C s~t

lip
To ENCLOSURETEMPERATURE,T
Fig. 3. Three types of fire.

A2,,, Lz ~ I L~ ,I.2

~, Ct .:..'-" a,c 2 '

Aw

umi'l--- ==
To TEMPERATURE To TEMPERATURE
(a) (b)

z L = L2
/
/ L~
7'-

B,c~,jy ~R,,= i ~~R1

QI::
--/c, i :
~,J / HJu~pl-- .~ ="7.'/I ~' ;

To TEMPERATURE To TEMPERATURE
(c) (d)
Fig. 4. Possible mechanism for initiating an instability leading to a "jump" increase in
the fire. (a) Increasing ventilation in a poorly ventilated fire; (b) reducing wall heat
losses; (c) increasing fire area; and (d) increasing the feedback by radiation from the
ceiling or a deepening gas layer.
FIRE FLASHOVER AND INSTABILITIES 165

from L1 to L2. At some precise opening, the L- in pool fires. In fact, Thomas and Nilsson [8]
curve becomes tangent to the R-curve at B, 6"2. and Friedman [9] report an enclosure enhance-
Any positive disturbance will cause the fire to ment of up to 60% over the open burn rate for
"jump" to state A2 on the higher ventilation-con- wood cribs, but up to six times has been reported
trolled branch of the R-curve. If the door is for plastic pool fires in enclosures [5. 9]. Hag-
opened abruptly, the transition to state A 2 may glund et al. [10] have indicated that the onset of
not be quasi-steady, but the intersection at A 2 still flashover (defined by attainment of a critical tem-
describes the final fire state. perature of 600°C) for wood crib fires occurs at
The inception of a critical condition by reduc- rates of burning somewhat below (30% at most)
ing wall heat losses is illustrated in Fig. 4(b). The those in ventilation-controlled fires. For fuel sys-
reduction in the L-curve can occur for several rea- tems more susceptible to thermal feedback the on-
sons. Following ignition of the fuel, the effective set of flashover may occur at much lower burning
conductance UA will decrease with time for a rates relative to the ventilation limit. In fact, the
fixed T, causing the L-curve to move from L 1 to data from Quintiere et al. [5] for burning hori-
L2. Thus, the criticality condition can be ap- zontal slabs of PMMA in a small scale enclosure
proached in a quasi-steady fashion. Another point suggest that the"onset of flashover" can occur for
of view is to consider an ensemble of fire experi- burning rates 70% below the ventilation-limiting
ments that burn a given fuel in enclosures of vari- value. We shall consider this class of fires and ex-
ous sizes, or a single enclosure with openings of amine its consequence in terms of our quasi-steady
various sizes. For this case, the R-curve is approxi- stability model.
mately fixed, but each enclosure size or opening Consider Eqs. 1 and 5 in which A f is held fixed.
size corresponds to a different L-curve. Hence, The condition for criticality requires that
there will be a critical enclosure size or opening
R =L (8a)
size at which the B-instability is reached.
Increasing fire area is shown in Fig. 4(c)which and
shows an increase in the R-curve as A f increases. dR dL
This could represent an increase in A f with time as - (8b)
dT dT
the fire spreads over the fuel surface, or a sequence
of experiments with different fuel areas. where R is given by the fuel-controlled case of Eq.
In the case of increasing the feedback by radia- 4(a). A solution to Eqs. 8(a) and 8(b) yields the
tion from the ceiling or a deepening gas layer, temperature Tc at the critical point. If UA + crh a
both the R and L curves increase with upward is constant, then this critical temperature, Tc, is
curvature as the emissivity of the layer increases. determined from
This could occur due to an increase in the soot
concentration, an increase in the layer depth, d
(T -- To) - ~ [q" (T)( Ahc(1 -- f ) -- c ( r -- To))]
changing the scale of the experiment or changing
the height of the opening.
= q"(T)iAhe(1 - - f ) -- c(T -- To)]. (9)
These examples illustrate several ways in which
flashover-like fire behavior can be explained in a
quasi-steady mode. Critical conditions can occur If on the other hand,
which lead to a jump increase in fire size, and criti-
cal parameters that promote this phenomenon can d
- - [UA + c(mo+ mf)]
be determined. dT

T H E R M A L F E E D B A C K A N D CRITICALITY
is positive, which is generally expected, the opera-
tion of applying Eqs. 8(a) and 8(b) becomes more
The role of thermal feedback in fires involving involved. It can be shown that dividing R and L
wood cribs is probably small compared to its role by T - T O and then equating terms and their de-
166 P.H. THOMAS ET AL.

rivatives lead to a useful expression for estimating of burning rate (or more precisely fuel generation
Te for a limited set of restrictions. A plot of the rate). A plausible form to assume for tt"(T) is
fuel-controlled branch of R / ( T - To), and of
L / ( T - To), is shown in Fig. 5. The point of q"(T) q"(To) a(T)o(T 4
= + - To4), (11)
tangency or critical temperature would occur at
a temperature higher than the temperature at where a(T) depends on the gas, ceiling, and wall
which R / ( T - To) or dl"(T)/(T - T o) reaches a temperatures, their emissivities and view factors.
minimum. This lower bound would coincide with Also the fuel surface temperature T s has been ap-
the critical temperature if (1 - f ) A h e >> c ( T - proximated as TO assuming that oto(Ts 4 - To 4)
To), and UA and m a are independent of T (which q"(T). When we apply Eq. 10 to 11 with the
are approximately valid over a general range of weak restrictions on
interest).
dc~ d2~x
-- > 0 and < 0,
dT dT ~

ENHANCEMENT OF BURNING RATE

it is possible to show (Appendix 1) that there are
We shall utilize the result that upper and lower limits to the lower bound on the
critical temperature Te and the factor of enhance-
d q"(T) ment, E = dl"(T)/dl"(To). (The lower bound applies
- - = o (lO)
a T ( T - - To) exactly for a as a constant.) These general results
are shown in Fig. 6.
establishes a lower bound on the critical tempera- With a = 1 such as for a fully emissive black hot
ture, to examine the case of radiative enhancement layer and with q"(To) = 10 kW/m z, a low value

L/(T-To)

f !
I

To Tc
ENCLOSURELAYERTEMPERATURE,T
Fig. 5. Criteria for determining a lower b o u n d on the critical temperature.
FIRE FLASHOVER AND INSTABILITIES 167

800
2.5
700

600

500 2.0

t 400
II
I..LJ
300
1.5
200

100

0 I 1 I I I I I I I I1.0
0 10 20 30 40 50 60 70 80 90 100 110 120
tf'(Xo}/a (kW/m 2 )
Fig. 6. Minimum conditions for "ignition" criticality.

representative of turbulent convection transfer the determination of Tc but would of course af-
from alcohol pool fires [11], we get T c = 625 K - fect the resulting enhancement.
a rise above ambient of about 350°C. With q"(To)= The quantitative results of the above discussion
40 kW/m 2, representative of wood crib fires [11], hold for a constant fuel areaA r, or for
the critical temperature rise ( T c - T o ) w o u l d be
about 520°C. In general, c~ may vary from 0 to 1 dA t
but values below about 0.3 would imply an in- dT
significant effect of radiation feedback. For 0.3 <
a < 1.0, Te ranges from 300°C to 650°C, but small-a slowly growing fire. The thermal feedback
more significantly, the enhancement ranges from in our model is responsible for criticality, but a
1.4 to 1.8. Thus, at the critical state, or the onset sudden enlargement of A t due to ignition could
of a "jump" suggestive of flahsover, the fuel gasi- also be described as flashover. That is, thermal
fication rate in the enclosure is likely to be only feedback raising the surface of the unburnt fuel
about 50% higher than its free burning rate value. into which the fire is spreading can lead to an ac-
It appears that the use of free burn rate data celerating fire spread as the ignition temperature is
would serve as a good approximation of the energy reached. While the sudden enlargement of A r
release rate of a fire burning in an enclosure. could result from a thermal feedback instability, it
This somewhat decouples the fuel dynamics from can also occur by an independent ignition mecha-
the enclosure effects and lessens the need for an nism.
accurate thermal feedback model.
For fuel burning in the upper space of the en-
THE "EXTINCTION" CRITICALITY
closure or for wall fires, convective heat transfer
may be important and can be added to the right The possibility of an "extinction" criticality can
side of Eq. 11 as h ( T - To). This would not affect also be seen in terms of our previous discussion.
 ! 68 P.H. THOMAS ET AL.

Increasing the airflow rate by, say, increasing can "knock it down" from state Y to C: a smaller
WH ~/2, will increase both the R and L curves in a fire.
ventilation controlled fire. In Fig. 7 the point Y Furthermore, a small increase in the ventilation
represents an extinction criticality, and the process factor WH 3/2 could lead to the same type of tran-
leading to extinction can be followed by con- sition. It is suggested that this is the boundary be-
sidering the fictitious state A'. A ' is determined by tween fuel and ventilation-controlled fires with re-
setting rh r = r'na/r, its stoichiometric value, in the spect to variations in the ventilation factor. More-
expression for L. It follows that the actual state A over, if the thermal feedback from the enclosure
is located such that T y <. T a <<. TA'. There is an has a weak effect on the burning rate, as would be
upper limit (and lower limit) to temperature be- for wood crib fires, and the slope of the loss curve
yond which T a is no longer larger than T y , and a does not exceed that of the generation curve,
solution is not possible in the ventilation-control- then there is no sharp transition. The conditions
led regime. These temperatures [2, 4] (T v and the for this case are that
lower temperature corresponding to the intersec-
tions of the R curve and fictitious L curve for m t = R = L at Y where rh a = rrn r
rha/r ) are given by the roots of the equation
and
At UA(T - To)
- - q"(r) =
Ahoa p (1 - - f ) A h c - - (1 + r)c(T-- To)'
rh a >~r rh f tiT'
(12)
which leads to
which follows from Eqs. 1-5. When temperature
Ty is reached, any negative perturbation on T
d ,,)_&o ~hoap (13)
causes the fire state to fall to a C-type intersection ( T - - T°)ctl" (Cl r At
at a lower temperature. The extent of the tempera-
ture drop depends on the feedback characteristic. under the conditions that (1 - f ) A h c >> c ( T -
For example, spraying water on an enclosure fire To) and UA, ma, and A r are independent of T.

i,m

Z
[.-
~c
Q~

Z
j-J¢/
..•IL !
with mf =ma/r

L~I
¢~
,

T Ty TA
ENCLOSURELAYERTEMPERATURE,T
Fig. 7. "Extinction" criticality in ventilation controlled fires.
FIRE FLASHOVER AND INSTABILITIES 169

DISCUSSION CONCLUSION

Let us now review possible phenomena responsi- The quasi-steady state energy balance for a com-
ble for the kind of significant increase in the partment fire shows that there are three equilibria
growth of a room fire which an observer might and two instabilities: "ignition" and "extinction."
identify as "flashover." We shall put the thermal The "ignition" criticality (B = C), leads to a "jump"
instability flashover mechanism in this context. increase in the mean equilibrium temperature; it is
These flashover processes are listed below: suggested that this is the basis of one form of flash-
over. A critical compartment gas temperature for
1. A constant area fire warms up the walls and so this form of flashover can be computed, provided
approximations can be made for describing the ther-
produces a thermal feedback which increases
mal feedback. The enhancement at this criticality is
with time to produce a " j u m p " - a thermal
found to be roughly 50% greater than the free burn
instability. (Fig. 4(b) and 4(d) are both ex-
amples of this.). value for a wide range of conditions.
2. A fire progressively increases in area in a quasi-
steady fashion to produce a similar criticality The UK contribution to this paper forms parts of
as shown in Fig. 4(c). Or the fire increases by the work of the Fire Research Station, Building Re-
sudden remote ignitions, a process that may not search Establishment of the Department of the
lead to the instability of Fig. 4(c), yet the tran- Environment. It is contributed by permission of the
sition might be identified as flashover. Director, Building Research Establishment.
3. A fire increases progressively in area to involve
all the fuel and reaches maximum size with-
out any "jump" and remains fuel-controlled;
but being a fully developed fire, it might be
described as having reached "flashover." APPENDIX
4. If the initial fire is large enough, the transition
DERIVATION OF UPPER A N D LOWER BOUNDS
to ventilation control could occur without ther-
FOR ENCLOSURE TEMPERATURE A N D
mal instability, for example, type 3, Fig. 3. This
ENHANCEMENT AT CRITICALITY
transition may also be perceived as flashover.
5. Unburnt gases accumulating in the upper part If
of a room may ignite spontaneously or because
flame from below ignites them. q " ( T ) = q " ( r o ) + ~o(T 4 - To4), (A1)

Clearly, combinations of these processes may also the lower bound for the critical condition is ob-
lead to flashover. tained from the minimum of q"(T)/(T- To). Per-
In this paper we have discussed the basis for forming this minimization leads to
"jumps" arising from instability in the energy
balance. They appear to be inherent in the general q"(To) = olo(3Te 4 - 4 T c 3 T o + TO 4 )
quasi-steady equation. The relatively low enhance-
ment at the thermal feedback criticality suggests + a ' o ( r e 4 - - To4)(rc -- TO) (A2)
that the time to criticality is most strongly de-
or
pendent on the time taken for the fire to reach a
heat release rate as might be measured outside a
roO"(ro)
compartment. For example, it may be the time for
o~o.(Tc4 - - r o 4 ) ( r c -- To)
the fire to spread to the boundaries of the pyroly-
zing surface. This spread is also affected by ther-
a' 3 +2x+x 2
mal feedback but that effect has not been discussed = -- Te + , (A3)
here. a (1 + x)(1 + x 2)
170 P.H. THOMAS ET AU

and the enhancement E is then given by Figure 6 shows the limits for T e and E as functions
of ¢t"(To)/a as given by Eqs. (A5-A8) for the above
q"(Tc)
assumptions.
q"(To)
NOMENCLATURE
=14
a' (3 + 2X + X2)(1 --X) effective heat transfer area
A
-- (Tc -- To) + (A4)
(1 + x)(1 + x 2) Ar fuel area
Au,Ai heat transfer areas (Eqs. 7(a) and (7d))
where C flow coefficient
f fraction of total energy emitted by the
To fire plume as radiation that does not
x =- and0<x< 1.
T~ enter the upper layer
g gravitational acceleration
If a' is positive, (i.e., the upper gas ceiling com-
H ventilation opening height
bination becoming effectively more emissive at high
H~ neutral plane height
temperatures), then it follows from Eq. (A3) that
hr, h e , h k heat transfer coefficients, (Eqs. 7(a)
Tcif'(To) 3 + 2x + x 2 and 6)
>1
aa(Tc 4 -- To4)(Tc -- To) (1 + xX1 + x2)" L rate of energy loss from the enclosure
r% air mass flow rate
(A5) mr fuel gasification rate
rhf, r burning rate
(The equality holds for ~ constant or a' = 0). If
R rate of energy gain or generation
a" < 0 as a approaches its upper asymptotic value
r stoichiometric ratio (air/fuel)
of unity, it can be shown that
[l u , [h rates of heat transfer from the hot layer
net thermal flux onto fuel
o < a'(Tc - T o ) < a(Tc), absolute gas temperature in hot gas
T
layer
which leads to the upper limit To absolute initial temperature
a critical temperature
{/"(To) (3 + 2x + x2)(1 - - x ) T~ upper layer ceiling and wall tempera-
< 1+ (A6)
otO(Te4 -- To 4) (1 + x)(1 + x 2) ture
t time
Under this same weak restriction of a' > 0 and a" < U effective heat transfer coefficient
0, it can be shown that the bounds on E are W opening width
Ahe effective heat of combustion of fuel
1 Z~xhvap heat of volatilization
E>~I+ E layer emissivity
(3 + 2x + x2)(1 - - x )
1+ O Stefan-Boltzmann constant
(1 + x)(1 + x 2) density of air
PO
(A7)
(1 + x)(1 + x 2)
E<I+ REFERENCES
(3 + 2x + x2)(1 - - x ) '
1. Quintiere, J. G., Growth of Fire in Building Com-
that is, partments, in Fire Standards and Safety (A. F.
Robertson, Ed.), ASTM STP 614, American Society
4 for Testing and Materials, 1976, pp. 131-167.
E< (A8) 2. Thomas, P. H., Some Problem Aspects of Fully
3 --x --x 2 --X 3
Developed Room Fires, in Fire Standards & Safety
F I R E F L A S H O V E R AND INSTABILITIES 171

(A. F. Robertson, Ed.), ASTM STP 614, American 8. Thomas, P. H., and Nilsson, L., Fully Developed
Society for Testing and Materials, 1976, pp. 112- Compartment Fires: New Correlations of Burning
130. Rates. Fire Research Station Fire Research Note No.
3. Takeda, H., and Nakaya, I., Small-Scale Model Fire 979, Borehamwood, 1973.
in Enclosure Using Liquid Fuel-Effect of Ventila- 9. Friedman, R., Behavior of Fires in Compartments,
tion Factor, Paper presented at the Second Joint Presented at International Symposium, Fire Safety
meeting of the U.S.-Japan Natural Resource Panel of Combustible Materials, Edinburgh, Scotland, Oct.
on Fire Research, UJNR, Tokyo, Oct. 19-22, 1976. 15-17, 1976, pp. 101-113.
4. Bullen, M. L, A Combined Overall and Surface En- 10. Hagglund, B., Janssen, R., and Onnermark, B., Fire
ergy Balance for Fully Developed Ventilation-Con- Development in Residential Rooms After Ignition
trolled Liquid Fuel Fires in Compartments, Fire from Nuclear Explosions, FOA Rept. C 20016-D6(A3)
Research Station Fire Research Note No. 1051, Forsvanets Forkskingsanstalt, Stockholm, Sweden,
1976. Nov. 1974.
5. Quintiere, J. G., McCaffrey, B. J., and Den Braven, 11. Thomas, P. H., Baldwin, R., and Heselden, A. J. M.,
K., Experimental and Theoretical Analysis of Quasi- Buoyant Diffusion Flames, Some Measurements of
Steady Small-Scale Enclosure Fires, Seventeenth Air Entrainment, Heat Transfer and Flame Merging,
Symposium {International) on Combustion, Leeds, Tenth Symposium (International) on Combustion,
U.K., August 1978. The Combustion Institute, Pittsburgh, 1965.
6. Babrauskas, V , and Wickstrom, U. G., Combust.
Flame 34:195-201 (1979).
7. Frank-Kamenetskii, D. A., Diffusion and Heat Trans-
fer in Chemical Kinetics, Plenum Press, New York, Received 25 June 1979; revised and accepted 1 October
1969, p. 353. 1979.