Sunteți pe pagina 1din 8

Twentieth Symposium(International)on Combustion/TheCombustion Institute, 1984/pp. 1727-1734

A MATHEMATICAL MIXTURE

FOR

COMBUSTION

MODEL

LEAN

HYDROGEN-AIR-STEAM

IN

CLOSED

VESSELS

ERDEM A. URAL X.~D ROBERT G. ZALOSH

Factory Mutual Research Corporation 1151 Boston-Providence Turnpike Norwood, Massachusetts 02062

The lean hydrogen-air-mixture combustion model described in this paper accounts for non- adiabatic, axisymmetric, buoyant flame propagation in a sphere or a cylinder. Turbulent burning velocity effects and heat losses associated with fan-stirred mixtures, equipment obstructions, and/or a water spray in the vessel are explicitly included in the model. Flame quenching and post-combustion cooldown are also modeled. Calculated results are compared with data from a variety of experiments conducted in vessels ranging in volume from 5 to 2,000 m 3. Comparisons with pressure data recorded during the accidental hydrogen deflagration in the Three Mite Island Unit 2 reactor building are also presented. Results indicate that peak pressures and burn times can be simulated quite well in many cases without using any ar- bitrarily adjusted parameter values. In other cases, such as the intermediate-scale cylindrical vessel tests, the nominal flame surface area had to be increased by a factor of 2-3 to achieve good agreement.

Introduction

The hydrogen burn during the Three Mile Island Unit 2 accident I has stimulated considerable inter- est in lean hydrogen-air-steam mixture deflagration phenomena. As described in Ref. 2, pressure de- velopment in these deflagrations is often influenced by residual unburned hydrogen and by heat losses during combustion. Several extensive test programs (many described in Ref. 3) have been conducted in recent years in order to develop a large data base for peak pressures, hydrogen burn fractions, and burn completion times. Efforts are underway to use this data base to predict pressure and thermal loads and hydrogen conversion fractions during hypoth- esized accidents in nuclear reactor containment buildings. A phenomenological model of lean mixture com- bustion and cooldown has been developed to assist in the data analysis and extrapolation. The model differs from most other deflagration models in the combustion literature in that it features buoyant flame propagation, turbulence-augmented burning velocities, heat losses leading to non-adiabatic pres- sure increases, flame quenching and post-combus- tion cooldown. In order to include all these com- plicated phenomena in the model, many compromises were made between fundamental ver- sus empirical formulations. The validity and effec- tiveness of these compromises can be ascertained by the comparison of calculated and experimental results presented after a brief description of the

current model. The detailed discussion of the model can be found in Ref. 4.

Model Description

As in most phenomenological deflagration models, the following assumptions are invoked. Burnt and unburnt gases, each with uniform temperature and composition, are assumed to beseparated by a con- tinuous flame surface. Pressure is instantaneously uniform throughout the compartment since flame speeds in these lean mixtures are small compared to sound speeds. The energy equations for the un- burnt and burnt gases, respectively, are:*

d

dVu

9

dt (M,,uu) +

.~lh,, + P T

+

Qco =

0,

.~aHc

+ .~h,,

=

d _ dVi,

~

(-~lbub) + V--ff + QLb.

(1)

t2)

Pressure at any time can be calculated using the ideal gas mixture relationship:

_

Mb

+

*Variable definitions are given in the nomencla- ture at the end of the text.

1727

1728

EXPLOSIONS/DETONATIONS

Equations (1) and (2) are coupled through the mass burning rate term:

/~1

dM~

dt

dM.

dt

puSTAF.

(4)

In calculating burnt gas composition and energy, all the hydrogen that passes through the flame surface is assumed to be consumed prior to flame extinc- tion. This assumption has been motivated by Mi- tanfs data5 which showed that 94% of the initial hydrogen is consumed behind the flame front in near-limit downward propagating premixed hydro- gen flames. Experimental confirmation for near-limit upward propagating mixtures is not available be- cause of irregular, non-contiguous flame surfaces.

FlameShapesand SurfaceAreas

Flame surface area and volumes needed in the preceding equations depend on assumed flame ge- ometry. Based on experimental observations, ~'6 the problem was separated into two phases: 1) a buoy- ant rise phase, and 2) a downward propagation phase, as shown schematically in Fig. 1. During the buoyant rise, flame surface area and burned gas volume are modeled as suggested by Crescitelli et al, 6 as an oblate spheroid. The aspect ratio of the spheroid is given in terms of the buoy- ant rise velocity, v, by

=

f pbv

1.5 -

0.5

PuSuJ

)

~'.

min {1,

/

a)

uleaard

propagation

t

t

t

t 4

t$

t 6

(5)

The buoyant rise velocity in Eq. (5) is determined by the fireball momentum equation:

1

ddt(Mav)= (Pu - pb)gV~ - ~ pu~rr2CDv2.

(6)

In the inertial term in the momentum equation, the induced motion of the unburnt gas in the vicinity of the flame kernel is also taken into account using the virtual mass which is represented as

(7)

Ma =

(l~b"~"Kpu)Vb

where

Milne-Thomson7 for creeping flow around an oblate

spheroid.

The drag coefficient Co is calculated us-

mass coefficient given by

K

is

the

added

ing the correlation:

Co

=

2.7 + 24/Re

(8)

as suggested by Cresitelli et al. 6 The location of the flame center can be determined by integrating Eq. (6) twice. The first phase of flame propagation ends when the flame reaches the top of the vessel. During the second phase, burnt gas is assumed to fill the top of the vessel and the downward prop- agating flame surface, is assumed to be planar (Fig. 1). The idealized flame geometries in both phases of flame propagation are adjusted as follows. When there are obstructions in the vessel, as in the case of the equipment and structures in a con-

tainment building, the flame folds around these ob-

jects, effectively increasing the flame surface area. Therefore, a flame surface area correction factor: CA = AF/AG is introduced. Here Ac is the geometric surface area calculated for the flame shapes shown in Figure 1 and AF is the actual flame surface area. Unfortunately AF is strongly affected by geometry of the vessel as well as the objects inside, and there is no universal correlation available. As a first ap- proximation (referred to as the nominal value) Ca is taken as the ratio of total solid surface area (in- cluding obstructions) to the vessel boundary surface area. Values of CA greater than unity also account for other flame deformation phenomena. These in-

clude: flame extensions associated with continuous ignition (using glow plug igniters), cellular flame surfaces associated with preferential diffusion,2 flame

deformation due to vessel penetrations and instru-

mentation, and flame stretch due to fireball inter-

action with the top wall. Although these effects oc- cur at different times, only one overall constant value

of CA is used in the model.

b) dc.mward props|stion

FIG. 1. Two phases of flame propagation.

HeatLosses

The heat loss rate from the burned gas, (~Lb, is

calculated from various convective and radiative contributions as described here. The unburned gas

LEAN HYDROGEN-AIR-STEAM MIXTURE COMBUSTION

1729

heat loss rate, 0Lu, is actually a heat gain rate cal- culated from the absorption of radiation emitted by the burned gas as described below. Convective heat transfer to the vessel walls and to the objects inside the vessel is taken into ac- count using Newton's cooling law. In these calcu- lations vessel walls are modeled as fiat plates while the equipment and structure inside the vessel are represented as uniformly distributed horizontal cyl- inders. The heat transfer surface area is allowed to increase in time due to increase in the burnt gas volume. During upward propagation, burnt gases sweep over solid surfaces with a velocity equal to the instantaneous buoyant rise velocity so that forced convection correlations s have been used to calcu- late the heat transfer coefficients. During down- ward propagation and the cooldown phases, resid- ual gas motion is relatively small so that natural convection correlations s are more appropriate. The total rate of heat loss to spray or fog droplets is calculated using the equation:

Qspray = ndVbhsSd(T -- Ts).

(9)

Spray droplets are assumed to be monodisperse, uniformly distributed and spherical so that the heat transfer coefficient is calculated using the equation:

hsDs

-- = Nus

k

= (2 + 0.459 Prl/3Res ~

~?n(l + B)

B

(lo)

The droplet number density and velocity have

been calculated

correlations for sprays, 9-n and for fog. 12 The details

can be found in Ref. 4. To simplify the calculation of the radiative heat exchange among burnt and unburnt gases and all the solid surfaces in the enclosure, it is assumed that reflection and emission from solid surfaces are

negligible. Part of the burnt gas radiation is ab- sorbed by the unburnt gases and the rest is ab- sorbed by the solid surfaces. The total radiative en- ergy leaving the burnt gas is determined by employing the emissivity of the burnt gas, calcu-

lated using

fraction of this energy that is absorbed by the un- burnt gas is determined using the average trans- mittance calculated as in Ref. 14. During the cooldown period following combus- tion, steam condenses over the solid surfaces, and possibly also in the gas phase. To account for these phenomena two simple condensation models, i.e., one for gas-phase, and one for surface condensation

a subroutine developed by Modak. 13 The

data

using separate

experimental

have been incorporated into the model as described in Ref. 4.

Burning Velocity

The adiabatic burning velocity equation used in the model is:

STAD ~- SLAD + U~AN "l" UtSpr~y~ + ukrr.

(11)

Values for SL^D as a function of hydrogen and steam concentration and unburned gas temperature were calculated from the Liu and MacFarlane cor- relations, is As for U~'AN, three alternative options 4 have been included in the model. Results pre- sented in this paper (for the Nevada dewar tests)

were based on the equation: t6

turbulent

energy dissipation

PFAN

p0VT

3 UFAN3

2

~e

(12)

where ee is the length scale of the energy dissi- pating eddies and is taken from Ref. 17.

The

value of USpray is calculated as one tenth of

the characteristic local spray droplet velocity at the

flame front as computed from the same empirical velocity distributions used in the spray or fog heat transfer algorithm. The rms flame-induced turbu-

u~'IT, is based on the un-

lent velocity fluctuation,

burned gas velocity, vu, at the flame surface. The isentropic compression relationship used for vu is:

 

V,

dP

vu =

--

(13)

~/uAFP dt

where ~, is the unburned gas ratio of specific heats.

The value of Vu calculated from the preceding equation is multiplied by arms turbulent intensity (typically taken as 0.10) to obtain U~IT~ Since there are heat losses during flame propa- gation, the burning velocity is actually less than the

(11). As sug-

gested by Crescitelli et al6 and others) s'19 the de-

adiabatic value represented

by Eq.

crease is represented by the one-step Arrhenius re- action rate relationship:

 

E

TAD-- T]

ST(T) = STAbexp

2R

TADT

J'

(14)

A value of 16 kcal/mote is taken for the overall ac- tivation energy of the hydrogen-air combustion re-

action as was recommended cote. is

by Fenn and Cal-

Flame Extinction

A flame extinction algorithm is incorporated into the model as a logical extension of the flame tem- perature effect on the burning velocity. Since the energy generation rate due to combustion is pro-

1730

EXPLOSIONS/DETONATIONS

portional to the burning velocity, the latter has a critical value below which heat generation cannot support the heat loss; at this point the flame is quenched. The extinction algorithm used in the model is based on the Frank-Kamenetskii formu- lation) 9 Accordingly, the critical burning velocity for quenching is

S}

STAD

e_l/2

(15)

m

9

,

.

,

.

,

9

,

.

,

9

,

.

,

9

,

9

,

9

/\

Dry e2/^~,

 
 

/

~

15z st,.o

 

/

/ \ ~

 

,

[

/

//V\•

 

Hydrogen-

8I

]

~

"~

 

:e:Ps:r:teU re 98 aBa TM

 

[

II

.\

~.

30z ste,.

 
 

2

4

8

8

1|

12

14

le

|8

i

 

lll/s

kFIERIDIIIIO~ (me)'

 

i

""

Combustion ends either when this criterion is

sat-

a

isfied (incomplete combustion) or when all the

hy-

a

drogen is consumed. When quenching occurs,

the

burnt and unburnt gases are assumed to mix

in-

stantly (conserving species mass and energy) and the subsequent cooldown period is calculated for this mixture. The preceding set of algebraic and ordinary dif- ferential equations is solved via numerical integra- tion based on the trapezoidal method. A time step of 10 milliseconds was found to provide equivalent .results to the more sophisticated predictor-correc- tor integration algorithm. Required CPU times are typically of the Order of 10 s on an IBM 4341 com- puter.

Comparisons with Data

Model calculations have been compared with premixed deflagration data obtained in spherical and cylindrical vessels of various dimensions as sum- marized in Table I. Vessel volumes span a range of four orders of magnitude. Hydrogen concentrations range from 5 to 12 vol. % and steam concentrations from 0 to 30%. These mixture composition ranges are applicable to nuclear reactor containment de- graded core accident scenarios. All calculations, un- less otherwise noted, were performed without us- ing any adjustable parameters, i.e., with CA set equal to unity. Calculated and measured pressure-time curves for three tests conducted in the 2.3 m diameter

FIG. 2. Effect of steam addition with bottom ig- nition (AECL experiments).

AECL sphere are shown in Fig. 2. Peak pressures agree to within 21% of the measured values, while the calculated pressure rise times are 1-2 s (30- 50% smaller) less than the measured values. Dif- ferences may be due to extrapolations in the burn- ing velocity correlation used in the calculations. There is good agreement in the decay portion of the pressure-time curves thus providing some ex- perimental validation of the model heat loss for- mulation. Peak overpressure data as a function of (dry) hy- drogen concentration are given in Fig. 3 for the four dry, bottom ignition tests in the AECL sphere. The upper solid curve in this figure corresponds to adi- abatic pressure, whereas the lower curve is the model predictionl The agreement between experi- mental and theoretical results appears to be quite good, particularly near the lower flammable limit where peak pressures are well below the adiabatic, complete combustion value. Similar results have also been obtained with data for central ignition in the Bureau of Mines sphere. Figure 4 shows the comparisons with Sandia ex- periments performed in a vertical cylinder with the

TABLE I Summary of the vessel dimensions used in combustion tests

 

Vessel

Diameter

Height

Volume

Organization

Reference

shape

(m)

(m)

(m3)

AECL

20

Sphere

2.29

2.29

6.3

Bureau of Mines

21

Sphere

3.66

3.66

25.7

EPRI/NRC/EG&G

22

Sphere

15.85

15.85

2,085.

Sandia

23

Cylinder

1.22

4.27

5.0

EPRI/Acurex

24

Cylinder

2.1

5.2

18.0

TMI

1

Cylinder

35.6

57.6

57,300.

LEAN HYDROGEN-AIR-STEAM MIXTURE COMBUSTION

1731

5OO

400

#

|

o

_8200

~X)~\r

~

FMRC-EPRIModel

E

8

 

5

6

7

8

9

10

111

12

HydrogenCor+centretlon(Volume %)

FIG. 3. Observed (circles) and predicted coneen- tration effect on hydrogen burn for AECL experi- ments (P0 = 98 kPa, To = 373 ~ no steam, bot- tom ignition).

40o

Sandia Data(Po = 90 kPa, To = 293"k)

Spark

9 GlowPlug(PartialPressure)

:~ GlowPlu(;(GasChromatogrpahy)

igniter located on the axis 0.91 m above the vessel bottom. Two sets of calculations are given in this figure. The curve marked CA = 1 corresponds to peak pressures calculated without any adjustable parameters, while the second curve (CA = 2) is cal- culated using an increased flame surface areal These two curves appear to reasonably bracket the data (for burn fraction as well as peak pressure) within the experimental accuracy of the hydrogen concen- trations. AECL and Sandia have performed fan-on/off data comparisons. The fans in each facility produced an average turnover of 5-6 vessel volumes per min- ute. The fans-on experiments produced markedly higher pressures and burn completeness than the

equivalent fans-off tests in lean mixtures, while having less effect in richer mixtures. Figure 5 dis- plays the measured and calculated overpressures for 7% hydrogen mixtures with and without fan-in- duced turbulence. The agreement is as good with fan turbulence as it is in the static mixture. EPRI/Acurex tests involved a 2.1 m diameter, 5.2 m high, cylindrical vessel equipped with water sprays. For the same initial composition the peak combustion overpressure was higher when the spray

was on. Comparisons shown

in Fig. 6 demonstrate

that the peak pressures calculated for CA = 3 were very close to experimental values for both spray-on and spray-off cases. The calculated cooldown with the sprays is significantly faster than the measured cooldown. This may be due to an underestimated characteristic spray drop size or overestimated droplet number density which are influenced by agglomeration and runoff effects in these test con- figurations. The largest-scale hydrogen-air-steam combustion tests were performed recently in a. 16 m diameter spherical dewar at the DOE Nevada test site.

 

15%

Steam

 

Im

Fan On

 

Hydrogen = 7%

 
 

Central Ignition

 

Temperature = 373"E Pressure - 98 IcPa

 

~.~\~

/

e

FMRC-EPRi

 

:

1 M~

--}

=

1

ta

'f

\\

I

\./-k

~

ignlt~o~

+

8

 

0

I

1

i

I

I

I

I

2

1

6

I

,I

12

,4

,8

,8

211

 

4

5

6

7

8

9

10

11

12

TIMEAFTERIIITION •=)

 

HydrogenConcentration(Volume %)

FIG 4. Model comparison with Saudia data (23).

FIG. 5. Model comparison with AECL data with and without fan-induced turbulence.

1732

EXPLOSIONS/DETONATIONS

-i

"-~::

FIG. 6. a. Acurex experiment 1.2 (no sprays) (Ref. 24). b. Acurex experiment 3.4 (sprays discharging 1.1 gpm through 9 nozzles) (Ref. 24).

Comparisons with these large-scale data indicate that calculations for CA equal to 1.5 and 2 seem to nicely bracket the data. The average deviation in the peak pressures calculated with CA equal to 2 was about 10% higher than the measured values. A sample comparison is given in Fig. 71 Calculations have been made for the Three Mile Island burn by representing the containment build- ing as a single vertical cylinder with the same vol- ume and height/diameter ratio as the actual con- tainment. The containment temperature, pressure, hydrogen, and steam concentrations as well as fan circulation are all inferred from plant data de- scribed in Ref. 1. The primary unknown in the cal- culations was the flame surface area correction fac- tor needed to simulate the complex containment internal geometry. A nominal value of 4.1 for CA was estimated from the equipment surface area in- ventory in the TMI containment. A value of 3.5 for CA provided the most reasonable fit to data as shown in Fig. 8.

Summary

and Conclusions

Comparisons

between

the calculated

mea-

data have pro-

and

sured pressure and burn fraction

 

SEL'~I(~

AFTER ll~lTl6g

 

-Z;.2

-12.2

E.I

17.4

~P

I

,i?.ii

12.|

T/.I

I~,1

lIP/*II

 

--

C~ILA~

(C75-4)

 
 

SpraYs Came On

 
 

~

~

~-~',~l

'

",:~

"

1~"

"l~s

"

"1~

"

"1~

I

 

5ECOl~

AFTER 1349

HOtl~

ON ~3/29/79

 

FIG. 8. Calculated pressure

spike (XHe :

 

7.5%,

Xa~o = 5.5%, CA = 3.5) compared with TMI data.

vided some important insights into the validity of the model formulation and the appropriate values of the input parameters used in the model. In par- ticular, the compilation of results shown in Fig. 9 indicate that the model can calculate peak pres- sures that (with only a few exceptions) agree quite well with the measured values over the entire range of lean hydrogen-steam concentrations and vessel configurations. However, in certain cases, this agreement does require specific values of the em- pirical surface area correction factor, CA, which de- pend on vessel configuration. For spherical vessels with diameters of 2-4 m, this agreement can be obtained with CA equal to 1.0, i.e., without any adjustments. For much larger spheres, and/or for cylindrical vessels values of CA in the range 2-3 are needed to obtain this agreement, as indicated in Fig. 9. The larger values of CA are due to the ob-

 

*

9

&S

ILl

,

9

,

&!

1. I.

I~

111

m

,

--

.

,

.

,

.

,

.

,

9

~=~.~.~

. t~lE (CA.Z.In, mF~llm

-

,

i

/

1

~'

' ~

~

x

NEVADA

8

~*

,,.s oom

Hydrogen

31

a

TI~

~II

m

(SEC{~[~)

~

m

9

,

M

9

'

Im

FIG. 7. Code comparisons with large-scale Ne- vada Test P-4.

gl

~

Ikl.~u

~

Ntm*

~D

9

~

9

§

§

o

alB

+m

i/,

IW

m

"1

d

.~+dg

E!

AI

9

I

,

I

,

I

,

t

9

i

9

FIG. 9. Comparison of calculated peak overpressures.

 

and

measured

1732

EXPLOSIONS/DETONATIONS

-i

"-~::

FIG. 6. a. Acurex experiment 1.2 (no sprays) (Ref. 24). b. Acurex experiment 3.4 (sprays discharging 1.1 gpm through 9 nozzles) (Ref. 24).

Comparisons with these large-scale data indicate that calculations for CA equal to 1.5 and 2 seem to nicely bracket the data. The average deviation in the peak pressures calculated with CA equal to 2 was about 10% higher than the measured values. A sample comparison is given in Fig. 71 Calculations have been made for the Three Mile Island burn by representing the containment build- ing as a single vertical cylinder with the same vol- ume and height/diameter ratio as the actual con- tainment. The containment temperature, pressure, hydrogen, and steam concentrations as well as fan circulation are all inferred from plant data de- scribed in Ref. 1. The primary unknown in the cal- culations was the flame surface area correction fac- tor needed to simulate the complex containment internal geometry. A nominal value of 4.1 for CA was estimated from the equipment surface area in- ventory in the TMI containment. A value of 3.5 for CA provided the most reasonable fit to data as shown in Fig. 8.

Summary

and Conclusions

Comparisons

between

the calculated

mea-

data have pro-

and

sured pressure and burn fraction

 

SEL'~I(~

AFTER ll~lTl6g

 

-Z;.2

-12.2

E.I

17.4

~P

I

,i?.ii

12.|

T/.I

I~,1

lIP/*II

 

--

C~ILA~

(C75-4)

 
 

SpraYs Came On

 
 

~

~

~-~',~l

'

",:~

"

1~"

"l~s

"

"1~

"

"1~

I

 

5ECOl~

AFTER 1349

HOtl~

ON ~3/29/79

 

FIG. 8. Calculated pressure

spike (XHe :

 

7.5%,

Xa~o = 5.5%, CA = 3.5) compared with TMI data.

vided some important insights into the validity of the model formulation and the appropriate values of the input parameters used in the model. In par- ticular, the compilation of results shown in Fig. 9 indicate that the model can calculate peak pres- sures that (with only a few exceptions) agree quite well with the measured values over the entire range of lean hydrogen-steam concentrations and vessel configurations. However, in certain cases, this agreement does require specific values of the em- pirical surface area correction factor, CA, which de- pend on vessel configuration. For spherical vessels with diameters of 2-4 m, this agreement can be obtained with CA equal to 1.0, i.e., without any adjustments. For much larger spheres, and/or for cylindrical vessels values of CA in the range 2-3 are needed to obtain this agreement, as indicated in Fig. 9. The larger values of CA are due to the ob-

 

*

9

&S

ILl

,

9

,

&!

1. I.

I~

111

m

,

--

.

,

.

,

.

,

.

,

9

~=~.~.~

. t~lE (CA.Z.In, mF~llm

-

,

i

/

1

~'

' ~

~

x

NEVADA

8

~*

,,.s oom

Hydrogen

31

a

TI~

~II

m

(SEC{~[~)

~

m

9

,

M

9

'

Im

FIG. 7. Code comparisons with large-scale Ne- vada Test P-4.

gl

~

Ikl.~u

~

Ntm*

~D

9

~

9

§

§

o

alB

+m

i/,

IW

m

"1

d

.~+dg

E!

AI

9

I

,

I

,

I

,

t

9

i

9

FIG. 9. Comparison of calculated peak overpressures.

 

and

measured

1734

EXPLOSIONS/DETONATIONS

and Spray Curtains, ASME Paper 76-WA/FE- 40, 1976.

11. ALPERT, R. L., AND MATHEWS, M. K:, Calcu-

12.

lation of Large Scale Flow Fields Induced by

Droplet Sprays, FMRC Report J.I. OEJ4.BU, RC79-BT-14, December 1979.

CAMP, A.

Program

Semi-Annual Report, April-September 1981, (M.

Berman, Ed.), NUREG/CR-2481, SAND82- 0006, February 1982, pp. 292-312.

L.,

LWR

Safety Research

13. MODAK, A. T., Fire Research, V. 1,

pp.

339-

361, 1979.

14. GROSSHANDLER,W. L., Int. J. of Heat and Mass Transfer, 23, 1447-1459, 1980.

15. Liu, D. D. S., AND MACFARLANE,R., Com- bustion and Flame, 49, pp. 59-72, 1983.

16. BATCHELOR, G. K., The Theory of Homoge-

neous Turbulence, Cambridge University Press, London, 1953.

17. GUNKEL, A. A., AND WEBER, M. E., AICHE Journal, 21, 931-949, 1975.

18, FENN, J. B., AND CALCOTE,H. F., Fourth Sym-

p. 231,

Institute,

posium (International) on Combustion,

The Combustion

1953.

19. FaANK-KAMENETSKI1,D. A., Diffusion and Heat Transfer in Chemical Kinetics, p. 365, Plenum Press, 1969.

20. TAMM, H., KUMAR, R. K., AND HARRISON, W.

C., Proceedings of the Second International Workshop on the Impact of Hydrogen on Water

Reactor Safety, pp. 6,33-649, Albuquerque, New Mexico, October 1982, also EPRI Report NP- 2955, 1983.

21. FURNO, A. L., COOK, E. B., KUCHTA, J.

M.,

AND BURGESS, D. S., Thirteenth Symposium (International) on Combustion, pp. 593-599,

The Combustion Institute 1971.

22. THOMPSON,L. B., Private Communication (1983). The test results to be published as an EPRI- NRC Report in 1985.

23. BENEDICK,W. B., CUMMINGS,J. C.,

AND PRAS-

SINOS, P. G., Proceedings of the Second Inter- national Workshop on the Impact of Hydrogen on Water Reactor Safety, pp. 665-679, Albu-

24.

querque,

New Mexico, October 1982.

et al, Hydrogen Combustion and

Control Studies in Intermediate Scale, EPRI NP-

TOROK, R.,

2953,

Project 1932-7,

Final Report,

1983.