You are on page 1of 18

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

COMPARISON OF TNT-EQUIVALENCE APPROACH, TNO MULTI-ENERGY APPROACH AND A CFD APPROACH IN INVESTIGATING HEMISPHERIC HYDROGEN-AIR VAPOR CLOUD EXPLOSIONS.
A. Beccantini1, A. Malczynski1,2 and E. Studer1
1

CEA Saclay, DEN/DM2S/SFME/LTMF 1,2 Warsaw University of Technology

ABSTRACT
In this work, we investigate the pressure wave generated by a so-called vapor cloud explosion (VCE), in the particular case in which the hydrogen-air mixture is homogeneous, presents a hemispheric geometry and the ignition occurs at the center of the hemisphere. Then we compare the results obtained on this particular geometry using three different approaches, namely TNT-equivalence approach, TNO multi-energy approach and ``1D point-symmetric deflagration'' approach, which solves the Fluid Dynamics equations (Computational Fluid Dynamics or CFD approach) in the case of 1D point-symmetric flows. These three approaches have been compared on a large-scale experiment performed at Fraunhofer Institute of Chemical Technology. As we will show, the TNT-equivalence method have shown not to be a good candidate to investigate VCE. The CFD approach gives a pressure field which fits the experimental one, providing that the flame speed is correctly evaluated. Concerning the multi-energy approach, it is possible to establish a relation between its strength index and the fundamental flame speed, a part from a region close to the one containing the burnt gas.

INTRODUCTION
The introduction and commercialization of hydrogen as an energy carrier of the future makes great demands on all aspects of safety [1]. For instance, the use of automobiles which use hydrogen as its primary source of power for locomotion implies the existence of gas refueling stations having hydrogen dispensing pumps. In the case of accidental hydrogen release and explosion, one of the most important issue is the safety of the working personnel and of the general public living around. In Nuclear Industry, as presented in [2], the possibility of using the high temperature gas of Nuclear Gas Cooled Reactors to massively produce hydrogen is under evaluation (see an example of layout of a combined plant in [3]). In this case, one of the most important issue is to guarantee the integrity of the nuclear power plant under the accidental case of hydrogen explosion in the coupled chemical plant as well as the safety of the working personnel and of the general public living around. There are several method to investigate the pressure wave generated by so-called vapor cloud explosions (VCE) [4]. In this work, we restrict our attention to the case in which the hydrogen-air mixture is homogeneous, presents a hemispheric geometry and the ignition occurs at the center of the hemisphere. Then we compare the results obtained on this particular geometry using three different approaches, namely TNT-equivalence approach, TNO multi-energy approach and ``1D pointsymmetric deflagration'' approach, which solves the Fluid Dynamics equations (Computational Fluid Dynamics or CFD approach) in the case of 1D point-symmetric flows. All of them present one parameter which must be specified by the ``user'' (and, as we will see, blast parameters result to be very sensitive to). In the TNT-equivalence method, a parameter between 1 and 0 must be fixed, the so-called efficiency factor, which artificially decreases the energy released by the explosion of the TNT-equivalent mass, thus pretending to take into account the real mass which burns and the combustion regime. The TNO multi-energy method involves the so-called strength index, arising from 1 (weak deflagration) to 10 (strong detonation); there exist tables which help in the choice of such parameter. In the CFD approach here considered, the fundamental flame speed must be
159 www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

specified. In the literature, there exist correlations which specify the flame speed as function of the mixture composition and of the obstacles array (if present) [13]. This paper is divided into several part. First, we describe the three approaches. Then we analyze the problem of the hemispherical vapor cloud explosions in the case of constant speed deflagration. In particular, we present the governing equations of the problem (and the hypothesis they are based on), perform the dimensional analysis, show the problem solution. Afterward, we perform a parametric study of the different approaches on this problem, namely we investigate how the efficiency for the TNT approach, the strength index for the TNO multi energy approach and the fundamental flame speed for the CFD approach affect the maximum overpressure and the positive impulse. As we will show, if we are interested in evaluating the maximum overpressure and the positive impulse, as certain blast criteria require, in the particular case of hemispherical vapor cloud explosion there exists a relation between the fundamental speed and the strength index of the TNO multi-energy method (although there is a little difference between these two approaches in the combustion region). Conversely, no relation between the parameter involved in the TNT-equivalence method and the CFD approach can be established. This is due to the difference of mode of combustion energy released in the TNT (the energy is immediately released in a small region) and the vapor cloud (the energy is distributed on a large domain in a finite time). Finally, the CFD results are compared with some experimental results provided on a large-scale experiment performed at Fraunhofer Institute of Chemical Technology and which has been the object of a code benchmark for the participants of the Hysafe project [5]. In this experiment, a 10 meter radius polyethylene hemispheric balloon has been filled with a stoichiometric and homogeneous mixture of hydrogen-air. The mixture has been ignited at the center. Experimental results show that the combustion of the mixture occurs at almost constant speed of about 63 m/s and gives a maximum overpressure of 50 millibar. As we show, the CFD approach gives a pressure field which fits the experimental one, providing that the flame speed is taken equal to the one given by experimental results. Conclusion follows.

1. DESCRIPTION OF THE APPROACHES


In this section we briefly present the TNT-equivalence method, the TNO multi-energy method and the ``1D point-symmetrical deflagration'' approach to evaluate the overpressure caused by the combustion of a ``hemispherical vapor cloud'' on the ground.

1.1 TNT-equivalence approach


The TNT-equivalence method has been widely used to express the explosive load of vapor clouds using an equivalent weight of a TNT-charge. In order to estimate the maximum overpressure and the positive impulse at a specified distance, the chemical energy available in a vapor cloud is converted into an equivalent weight of TNT using the formula

WTNT =

m H QH
2

QTNT

Where WTNT is the equivalent mass of TNT, hydrogen in the gas cloud, QH
2

is an efficiency factor, mH

is the total mass of

the heat of combustion of hydrogen (123 MJ/kg), and QTNT is the

heat of combustion of the TNT (approximately 4.184 MJ/kg). Once computed WTNT , there exist diagrams, obtained via TNT explosions, which give the non-dimensional maximum overpressure and the positive impulse as function of the Sachs scaled distance r / (WTNT / P0 )
(1 / 3 )

160

www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

Pmax r = f (W / P )1 / 3 P0 TNT 0
(1)

P0 (WTNT / P0 )
1/ 3
1/ 3

I + c0

1/ 3

r = f (W / P )1 / 3 TNT 0

Here r (m in the SI system) is the distance from the center of symmetry, P0 (Pa) is the atmospheric

((kg/Pa) ) is the Sachs scale distance (or the energy scaled distance), c0 pressure, (WTNT / P0 ) (m/s) is the speed of sound in the undisturbed air; the maximum overpressure and the positive impulse are defined by

Pmax (r ) = max t (P(r,t ) P0 )

I+

(r ) = (max(P(r,t ) P0 ),0)dt

The efficiency factor pretends to take into account the fact that in the cloud the combustion occurs at deflagration regime, by simply postulating that the effects of a VCE in which a deflagration occurs is equal to the ones of a TNT explosion with a released energy reduced of a factor that, by varying inversely proportional to the I
+

. Note from

(1)

, the graphic maximum overpressure Pmax versus radius r is scaled of a factor

1/ 3 on the r-axis; the graphic positive impulse I + versus radius r is 1/ 3 1/ 3 scaled of a factor inversely proportional to on the r-axis and of a factor proportional to on
axis.

1.2 TNO multi-energy method


This model is based on the assumption that a blast is generated in a vapor cloud explosion only where the flammable mixture is partially confined and/or obstructed while, on the other hand, a unconfined/unobstructed mixture does not generate a blast. The total domain occupied by the vapor cloud is divided into several sub-domains. In each sub-domain, the mixture is concentrated in a hemisphere and treated as homogeneous and stoichiometric. There exists abacuses which give the nondimensional maximum overpressure and the positive impulse as function of the Sachs scaled distance and of a strength index arising from 1 (weak deflagration) to 10 (strong detonation). These abacuses have been obtained by using the semi-analytical solution of the 1D point-symmetrical flame in a homogeneous medium and numerically studying its time-evolution as the combustion stops. There exist rules for the choice of the index (see for instance [7]). There exist also rules for summing nondimensional maximum overpressures and positive impulses. Nevertheless the choice of the index as well as of the domain decomposition is quite user-dependent.

1.3. 1D point-symmetrical deflagration approach


In this approach we suppose to deal with a mixture of thermally-perfect gases. We neglect viscosity effects. We suppose that the chemical combustion is governed by the one step irreversible chemical reaction

1 H 2 + O 2 H 2 O. 2
We also suppose that the mixture occupies a hemisphere close to the (rigid) ground, the solution is 1D
161 www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

point-symmetrical, the flame is infinitely thin and the fundamental flame speed (the flame speed with respect to the unburnt gas close to the flame surface) is equal to K 0 . The problem we solve will be presented in section 2. Let us briefly describe the numerical method we use (see [8] and [9] for details). In an operator splitting strategy, we first solve the non-reactive Euler equations; then we update the energy and the densities of the species by taking into account the chemical reaction. Concerning the non-reactive part, we solve the Euler equations in conservative form via an explicit algorithm in time. We suppose that we are dealing with a mixture of ideal gases. The variation of specific heats with respect to temperature is taken into account. We compute the average of the conservative variables (namely densities, momentum and total energy (sensible plus kinetic energy) per unit volume) via a cell-centered Finite Volume approach. Far from shocks and discontinuities, second order space accuracy is obtained via the linearly-exact reconstruction on primitive variables (namely total density, speed, pressure, species mass fractions) of [10]. At each interface we compute the flux vector using the shock-shock Flux Difference Splitting [9]. Concerning the reactive part, we update the total energy and the density of the species by observing that the quantity of unburnt gas which crosses the flame at each time is given by

dm = u, f K 0 S f , dt

(2)

2 where u, f is the density of unburnt gas ahead of the flame and S f = 2r f is the surface of the

hemisphere at the flame position r. As one can read in [13], the apparent speed of a spherical flame depends on the burning velocity and the ratio between the real surface of the flame and the surface of the sphere at the flame position. If the flame were perfectly spherical, this ratio would be equal to one; nevertheless, because of instability effects, the flame surface is crisped; then this ratio is larger than one. In the CFD approach here presented, the fundamental flame speed K 0 incorporates this ratio, namely its value is equal to the (apparent) flame speed Dflam divided by the expansion ratio between densities of unburnt and burnt gases).

(ratio

2. HEMISPHERIC VCE
We have a hemisphere (radius rhem ) with a stoichiometric mixture of hydrogen-air inside and air outside, everything at constant pressure and temperature ( P0 and T0 ). We suppose that the flame is initiated in the center and the flow keeps 1D point-symmetrical and occurs at the fundamental flame velocity K 0 . We also suppose to deal with a mixture of thermally-perfect gases, that viscous effects are negligible and that the chemical combustion is governed by a one step irreversible chemical reaction.

2.1 Dimensional analysis


The governing equations (Euler Equation in 1D spherical geometry) can be written as

1 + 2 ur 2 = 0 t r r u 1 2P + 2 u 2 r 2 + r 2 P = t r r r 2 u 2 1 P u & e + + 2 ur 2 e + + = j h 0 j j t 2 2 r r Y j 1 & + 2 Y j ur 2 = j t r r

(3)

162

www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

where the pressure P and the internal energy e are linked via the Equation of States (EOS)
* Y j R T = Y j R j T P= j Mj j

e = { Y j c v, j ( )d }
0

Here r (m in the SI system) is the distance from the center of symmetry, (kg/m ) the density, u
3

(m/s) the speed, e (J/kg) the sensible internal energy, T (K) the gas temperature, M
0 j

(kg/mole) the

molar weight for the j-th gas involved in the mixture, h (J/kg) its formation enthalpy at 0 K, Y j its

& mass fraction, j (s

) its reaction rate, cv, j (T ) (J/(Kg K)) its constant volume specific heat, R j
*

(J/(Kg K)) its gas constant and R the universal gas constant (8.314 J/(mole K)). Actually, because of the hypothesis of irreversible chemical reaction and infinitely thin flame (i.e. infinitely fast reaction), the evolution of the densities Y j during the combustion phase is evaluated using (2) instead of the reaction rates. Concerning the initial conditions, we have a stoichiometric mixture of hydrogen and air inside the hemisphere and air outside; everything is at ambient conditions (pressure and temperature P0 and

T0 ). Actually, because of the hypotheses previously presented, we can treat the problem as if we were
dealing with three gases only: the air (air), the unburnt mixture (u) and the burnt mixture (b). Then, at the beginning we have an hemisphere filled with the unburnt mixture and air outside. The solution of the problem can be expressed as follows: U = r, t, K 0, rhem , P0,T0, Rair , Ru , Rb , c v ,air , c v ,u , c v ,b , hu0 hb0 ,

U being the vector of the conservative variables (densities, momentum and total (sensible plus kinetic) energy per unit volume), i.e. it depends on 13 parameters1 including the independent variables r and t. Since there are 4 independent dimensions involved (mass, length, temperature and time),
according to the Buckingam -theorem, the non-dimensional solution depends on 13-4=9 parameters. If we chose as dimensionally-independent reference quantities P0 , T0 , rhem and Rair , the nondimensional conservative variables can be expressed as

K0 rhem P0 T0 Rair Ru Rb c v ,air c v ,u c v ,b hu0 hb0 ~ ~ r t RairT0 U = U , , , , , , , , , , , , rhem rhem Rair T0 rhem P0 T0 Rair Rair Rair Rair Rair Rair Rair T0 0 0 c c K0 R R c h hb ~ U ~, ~, r t , u , b , v ,air , v ,u , v ,b , u = Rair T0 Rair Rair Rair Rair Rair Rair T0
They are solutions of the Euler equations (3) subjected to the EOS

Rj ~ ~ T Y j j =u, b, air Rair ~ T c v, j (T0 ) ~= e Yj d Rair 0 j = u, b, air ~ P=

Strictly speaking, specific heats are not parameters but function of T. Then, if these functions are approximated using 4-th order temperature polynomials, each specific heat involves 5 parameters.
163 www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

As one can see, with this particular choice of the dimensionally-independent variables, the values of the initial pressure and of the radius of the hemisphere do not affect the non-dimensional solution. The initial value of temperature is involved in the non-dimensional numbers

(h

0 u

hb0 / RairT0 . The first non-dimensional number represents the ratio between the fundamental

K 0 / Rair T0 ,

combustion per mass unit and initial sensible energy in the air (divided by a factor close to (air 1) ). If the gas is thermally perfect but non-calorically perfect (as it is in this case), the initial value of the temperature also affects the non-dimensional specific heats, which are temperature functions. It follows that, if we change the value of T0 and we keep constant the other independent variables and parameters, the non-dimensional solution is affected. Conversely, if the mixture involved calorically perfect gases, the initial value of the temperature would not affect the non-dimensional solution.

flame speed and the unperturbed sound speed in the air (divided by the square root of the air specific heat ratio air ). The second one is representative of the ratio between the energy released in the

2.1 Solution of the problem


Let us restrict our attention to a problem the conditions of which are similar to the large-scale experiment performed at Fraunhofer Institute of Chemical Technology. As already mentioned in the previous section, the values of the initial pressure and of the radius of the hemisphere do not affect the non-dimensional solution, while the initial temperature does. We have a 10 m radius hemisphere with a stoichiometric mixture of hydrogen-air inside and air outside, everything at almost atmospheric conditions ( P0 and T0 equal to 0.989 bar and 283 K). This corresponds to mH = 52 kg of hydrogen
2

which, if completely burns, releases Q = 6.2910 MJ. In this section we present the solution of the problem under the hypothesis of constant fundamental speed K 0 equal to 22.6 m/s (nevertheless, from a qualitative point of view, we have similar results if we take a different value). The solution is obtained using the numerical approach previously described. We can distinguish between 3 different phases. In the first one, the flow consists of a precursor shock, followed by an isentropic compression and by the flame surface. In this phase, neither the shock wave nor the flame has reached the surface separating the combustible mixture with the outside gas. In the second phase, the shock wave has reached the separating surface but the flame has not. In the third phase, the flame has reached the separating surface and there is no more combustion. Before the precursor shock reaches the surface of the balloon (first phase), the solution is selfsimilar (and it coincides with the propagation of a constant speed flame in an homogeneous medium [11, 12]). The evolution of the pressure with respect to variable r/t is qualitatively shown in Figure 1. Here 0 is the undisturbed state ahead of the shock wave. The surface between 0 and 1 is a shock wave, moving with speed Dsh > c 0 , c0 being the sound speed in the unperturbed medium. This surface separates the perturbed and the unperturbed flow. 1 is the state behind the shock wave. The region between 1 and 2 is an isentropic compression, in which pressure, density and speed of sound increase as we approach the flame surface. In this region, the flow velocity u increases to the value Dflam K 0 just ahead of flame surface, Dflam being the speed of the flame with respect to the
3

unperturbed state (as already mentioned Dflam = K 0 , where is the expansion ratio); moreover the flow is identical to the one generated by a spherical expanding piston the (constant) speed of which is smaller than Dflam and larger than u = Dflam K 0 . 2 is the state ahead of combustion. The dotted line represents the infinitely thin flame, which separates the unburnt and the burnt mixture. 3 is the state behind the flame (flow at rest). Note that the pressure is maximum just ahead of the flame and slightly decreases once the flow crosses the flame. In Figure 2 we show the computed pressure as function of the distance, before the shock wave reaches the balloon surface. The precursor shock speed
164 www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

is about 400 m/s, i.e. almost equal to the speed of sound in the hydrogen-air mixture, accordingly to the fact that in this case the shock strength is small (note that the larger the fundamental flame speed, the stronger the shock). The flame speed Dflam is about 160 m/s (i.e. the ratio between the flame speed and the speed of sound in the unperturbed mixture of hydrogen and air is Dflam / c0,cloud 0.4 ). In Figures 3 and 4 we show P versus r and P versus r/t after the compression wave has reached the interface (phase 2). As one can see, because of this interaction, the pressure increases even in the unburnt region. This is due to that fact that the compression wave enters into a gas with different properties, i.e. air with density larger than the one of the hydrogen-air mixture. Since the compression wave can be considered as an ensemble of shock waves of infinitesimal strength, the reason can be easily understood by considering the problem of the interaction of a right-traveling (i.e. toward increasing values for r) shock wave with a contact discontinuity (see Figure 5). Once a shock wave reaches a contact discontinuity, we have the formation of a new discontinuity with a left state equal to the left state of the shock and a right state equal to the right state of the contact discontinuity. This state differs from the right state of the shock only for the densities. The discontinuity thus formed constitutes a so-called Riemann problem, i.e. in the pressure-velocity plane, we have to look for the intersection of the rarefaction and shock curves2. As one can see in [8], given the state R, the slope of the left shock curve in the pressure velocity plane in R is

dP = ( c )R = du R

P R .

It follows that, if the specific ratio is the same (as it is in our case), we can distinguish between two cases: if the density on the right of the contact discontinuity (R' in Figure 5) is larger than the density on the right of the shock (R in Figure 5), as it is in our case, then the solution of this Riemann problem involves a left and a right-traveling shock, with a pressure larger than the one of the original shock; if the density on the right of the discontinuity is smaller than the density on the right of the original shock, then the solution involves a left-traveling rarefaction and a right-traveling shock, with a pressure smaller than the one of the original shock. In both cases the discontinuity starts moving from the left to the right. In our case, the pressure increases. Note also that the shock wave is slower than in the first phase (340 m/s in phase 2 versus 400 m/s in phase 1) and this is due to the fact that the sound speed in air is smaller than the sound speed in the stoichiometric mixture of hydrogen and air. Phase 3 (t> 0.1 s) starts once the flame surface reaches the surface between the mixture and the air, i.e. once the combustion stops and there is no more chemical energy provided to the fluid. Once again, when the flame reaches the interface between the stoichiometric mixture and the air, we have the formation of a discontinuity. On the left we have zero velocity, while on the right we have a rightmoving flow. The pressure is almost constant at the interface. This structure generates two rarefaction waves, one moving left and one moving right. The latter is faster than the precursor shock, therefore it weakens the pressure in the flow head. The former moves toward the center of symmetry and then it weakens the pressure in this region; then it is reflected and starts moving in the opposite direction. At the end, in each point, the pressure reaches the atmospheric value (see Figure 6).

2.2 Characteristic scales.


Let us now present some characteristic scales for this problem. First of all, let us present the value of some quantities (density, pressure, temperature and sound speed, in SI units) inside the hemisphere before the combustion occurs and after and Adiabatic Isobaric Complete Combustion (AIbCC).

Inside

(kg / m )
3

P (bar) 0.989

T (K) 283

c (m/s) 405

0.878

Given one state, we define left (right) shock curve the set of the states which can be connected to this state by a left (right) shock. In the same manner, we define the rarefaction curve.
165 www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

AIbCC

(kg / m )
3

P (bar) 0.989

T (K) 2510

c (m/s) 1030

0.116

Let us present some characteristic scales for the velocity (m/s).

Fundamental flame speed K 0 Flame speed in the laboratory Dflam = K 0 Sound speed in the unburnt gas c0,cloud Sound speed in the burnt gas Sound speed in unperturbed air cout
Let us present some characteristic scales for the distance (m).

8.5 3 63 4 405 1030 338 50.3 10.0 19.6 105 5 0.31 6

(Sachs) Energy based length (2Q / P0 ) Initial hemisphere radius rhem

1/ 3

Final hemisphere radius rhem' = rhem 1 / 3 Acoustic distance rac rhem' cout / Dflam

As characteristic scale for time, we have the time at which the combustion stops (s).

rhem ' / Dflam

The sound speed represents the speed at which the perturbations (as well as weak shocks) travel in a medium. The energy based scale represents the dimension of a box at which the released energy generates an overpressure of the same order as the initial pressure. The acoustic distance represents the distance lasted by a weak shock once the combustion stops. We also emphasize that in the case of K 0 = 8.5 m/s (Fraunhofer experiment), once the combustion stops, 30% of the released energy has contributed to increase the internal energy inside the combustion region; the rest of the released energy has contributed to increase the total energy outside the hemisphere. We emphasize that the larger the flame speed, the larger the overpressure in the burnt region, the larger the density in the burnt region, the smaller the final radius of the vapor containing the burn mixture, the larger the quantity of released energy which contributes to increase the energy inside the combustion region. In the particular case of spherical detonation, once the combustion stops, the dimension of the hemisphere has not changed and outside of the sphere the flow is completely undisturbed.

3. PARAMETRIC STUDY OF THE DIFFERENT APPROACHES


The average of the flame speed measured in the experiment performed at Fraunhofer Institute
3 Fundamental flame speed corresponding to the average flame speed in the Fraunhofer experiment. 4 Value of the average flame speed in the Fraunhofer experiment. 5 Distance last by the acoustic wave once the reaction stops, corresponding to the average flame speed in the Fraunhofer experiment. 6 Duration of the combustion , corresponding to the average flame speed in the Fraunhofer experiment.
166 www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

of Chemical Technology is Dflam = K 0 = 63 m/s, which gives a fundamental flame speed K 0 8.5 m/s. Here we compute the propagation of the pressure waves due to a vapor cloud explosion, by taking our initial conditions and different fundamental flame speeds ( K 0 from 5.65 to 43.2 m/s, which corresponds to Dflam from 40 to 320 m/s, i.e. Dflam / c0,cloud from 0.1 to 0.8). For the TNT equivalence method we take the efficiency factor equal to 10, 50 and 100%. Generally, 10% is the value recommended for the case of a deflagration [4]. In the TNO multi-energy model, we take the initial blast strength equal from 1 for insignificant deflagration to 7 for strong deflagration. Strict application of the multi-energy method would lead to the choice of 1 since the cloud is completely unconfined and there are no obstacles in the patch of the flame [7]. As presented in [13], there exist blast damage criteria which involve maximum overpressure and positive impulse. For this reason, we compare the different approaches on these two quantities. In Figure 7 we compare the results obtained using the TNT-equivalent approach and the CFD approach on the maximum pressure and the positive impulse as function of the distance from the center. As one can see in [6, 11], the pressure field generated by a TNT-explosion (or by a point explosion) is not the same as the one originated by a vapor cloud explosion in deflagration regime. In the former case, the energy is immediately released in a small volume, while in the latter, as we have already mentioned, the combustion energy is released in a volume the dimension of which depends on the flame speed. In the case of TNT explosion, each point sees an incoming shock followed by a rarefaction wave, while in the latter case a compression wave is followed by the rarefaction wave. The decaying of the overpressure is not the same and it is more pronounced in the case of the TNT explosion. In Figure 7 we see that, although equal to 10% also reduces the slope of the overpressure as function of the distance (with respect to the one with efficiency factor equal to 100%), there is a severe intersection between this curve and the one corresponding to a fundamental speed equal to 22.6 m/s at r=80 m. This implies that, in a deflagration with fundamental speed equal to 22.6 m/s, the overpressure is less important than in a TNT explosion involving 10% of the released combustion energy before the intersection point, but more important after this intersection point. Moreover, if we check the behavior of the positive impulse versus distance in Figure 7, we see that the one given by the TNT approach with the equivalent factor equal to 10% is much smaller than the one given by the deflagration with fundamental speed equal to 22.6 m/s. Indeed, although maximum overpressure in both cases are quite close, the decaying of a blast generated by a TNT wave is more important, which explains this behavior. Then, even if we artificially reduce the released energy by introducing an efficiency parameter, it is not possible to establish a relation with this parameter and the flame speed. In our opinion, these differences make the TNT approach a bad candidate for analyzing VCE explosions. In Figure 8 we compare the results obtained using the TNO multi-energy approach and the CFD approach on the maximum pressure as function of the distance from the center. As already mentioned, abacuses in the TNO multi-energy method have been obtained by considering constant velocity 1D point-symmetric deflagration of hemispherical vapor clouds. As one can see in Figure 8, once a strength index is given, the TNO multi-energy approach predicts a constant (in-space) maximum overpressure in the region occupied by the burnt gases at the end of the combustion. This is in contradiction with the fact that there is an interaction between the separation discontinuity and the compression wave ahead of the flame; this interaction increases the value of the maximum overpressure in the region close to the surface which separates the burnt gases and the air. Nevertheless, far from the combustion region, (r > 30 m), a correlation between the blast strength and the fundamental speed can be established. TNO strength index 2 corresponds to K 0 6 m/s; strength index 3 corresponds to K 0 9 m/s; strength index 4 corresponds to K 0 15 m/s; strength index 5 corresponds to K 0 20 m/s; strength index 6 corresponds to K 0 30 m/s; finally, strength index 7 corresponds to K 0 40 m/s. Results on positive impulses are in accord with the considerations done for maximum overpressure curves.

167

www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

4. EXPERIMENTAL RESULTS VERSUS CFD RESULTS. Finally, let us compare the computational results and the ones of the large-scale experiment performed at Fraunhofer Institute of Chemical Technology. As already mentioned, in this experiment, a 10 meter radius polyethylene hemispheric balloon is filled with a stoichiometric and homogeneous mixture of hydrogen-air. The mixture is ignited at the center, by ignition pills of 150 J. When the flame reaches the half of the radius, the balloon bursts at the seams bordering the ground and along longitudinal welds. Experiments give the position of the flame as a function of time, 0<t<0.3 s; the pressure as a function of time (0<t<0.6 s) in r=2, 5, 8, 18, 35, 80 m [5]. Experimental results on maximum overpressure perfectly fit the curve corresponding to a fundamental flame speed K 0 8.5 m/s. In Figures 9, 10 and 11 we represent the experimental results on the pressure obtained on different positions as function of time. As one can see, they are in good agreement, which implies that the 1D point-symmetrical CFD approach gives good results providing that the good fundamental flame speed is chosen. Nevertheless, numerical results presented in the previous section show that the maximum overpressure as well as the positive impulse are very sensitive to the flame speed. This is true in the region close to the combustion region as well as at 180 m (we recall that the Sachs scale distance for this problem is about 50 m). This implies that if we want reliable results for safety analysis, the correct flame speed should be specified. In the literature, there exist several correlations which specify the flame speed as function of the mixture composition as well as the obstacles array in the region in which the flame propagates. Some of these correlations are presented in [13]. As observed by M. Kustnetzov [14], in the particular case the Fraunhofer experiment the average flame speed can be correctly predicted by observing that its acceleration is due to the Darrieus-Landau instability; then, using the Gostintsev correlation for free turbulent flames [15], he estimates an average flame speed of 72 m/s (which gives a fundamental flame speed of around 10 m/s). CONCLUSION After analyzing and solving the hemispherical vapor cloud explosions due to a constant-speed deflagration, in this work we have investigated the behavior of three simple approaches as applied to such problem. The TNT-equivalent approach has shown to be a bad candidate for investigating flows arising from the deflagration in the clouds. This is due to the fact that the generation and the decaying of a blast wave generated by a TNT-explosion (or a point explosion) is different from the pressure wave associated to a VCE. Then, even if we artificially reduce the released energy by introducing the efficiency parameter, and if we are interested in both evaluating maximum overpressure and positive impulse as function of the distance from the center of symmetry, it is not possible to establish a relation between this parameter and the flame speed. Conversely, if we are interested in both evaluating maximum overpressure and positive impulse as function of the distance from the center of symmetry, it is possible to establish a relation between the TNO strength index and the fundamental flame speed, a part from a region close to the one containing the burnt gas. Finally, the comparison with an experiment of a large spherical deflagration performed at Fraunhofer Institute shows that, if the flame speed is correctly evaluated, the CDF approach gives results very close to experimental ones. Nevertheless, since the maximum overpressure as well as the positive impulse are very sensitive to the value of the flame speed, it is important to correctly evaluate this value. In the literature, there exist several correlations which linked the flame speed with the mixture as well as with the array of the obstacles [13]. We emphasize that, as already mentioned in [13], a detailed analysis of the blast effect associated to a VCE requires the realization of complex 3D computations, sometimes impossible to realize. The advantage of a 1D computational approach is that, once the flame speed is correctly estimated, it can be easily provide results which give an idea of the maximum overpressure and of the positive impulse. It can be pointed out that, in the literature, there exist approaches which allow to compute blast parameters (as maximum overpressure and positive impulse) once the constant flame
168 www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

speed is given (see [13] in which analytical formulas are given or [16] in which blast curves are given). Nevertheless, 1D CFD approach can be also used for investigating flows generated by flame acceleration. Validation of this tool for such purpose will be the object of another work.

REFERENCES
[1] [2] [3] [4] [5] Safety of Hydrogen as an Energy Carrier. In http://www.hysafe.org. 2006 Y. Inaba and T. Nishihara and M.A. Groethe and Y. Nitta. Study on explosion characteristics of natural gas and methane in semi-open space for the HTTR hydrogen production system. Nuclear Engineering and Design, vol. 232. 2004. T. Takizuka. Reactor technology development under the HTTR project. Progress in Nuclear Energy, vol. 47. 2005 Y. Mouilleau and J.F. Lechaudel. Guide des mthodes d'valuation des effects d'une explosion de gaz l'air libre. INERIS Technical Report. (Available and downloadable from the INERIS web site http://www.ineris.fr/).1999 E. Gallego, J. Garcia, E. Migoya, A. Crespo, A. Kotchourko, J. Yanez, A. Beccantini, O.R. Hansen, D. Baraldi, S. Hoiset, M.M. Voort, and V. Molkov. An intercomparison exercise on the capabilities of CFD models to predict deflagration of a large-scale H2-air mixture in open atmosphere. International Conference on Hydrogen Safety. Pisa, September 8-10. 2005 W.E. Baker. Explosions in air. Wilfred Baker Engineering. 1983 M.W. Roberts and W.K. Crowley. Evaluation of flammability hazards in non-nuclear safety analysis. 14-th EFCOG Safety Analysis Workshop. San Francisco, CA. (Available and downloadable from http://www.llnl.gov/es_and_h/sawg2004/pdf/34_roberts.pdf). 2004 A. Beccantini. Solveurs de Riemann pour des mlanges des gaz parfaits avec capacits caloriphiques dpendant de la temprature. PhD Thesis/CEA Report CEA-R-5973. 2001. A. Beccantini. Colella-Glaz splitting scheme for thermally perfect gas. In Godunov methods: theory and application (E.F. Toro, ed.), Kluwer/Plenum Academic Press, 2001. T.J. Barth, D.C. Jespersen. The design and application of upwind schemes on unstructured meshes. AIAA 89-0366. 1989 L.I. Sedov. Similarity and dimensional methods in Mechanics. Academic Press. 1959 A.L. Kuhl, M.M. Kamel and A.K. Oppenheim. Pressure waves generated by steady flames. Fourteenth (international) symposium on Combustion, The Combustion Institute. 1973 S.B. Dorofeev. Evaluation of safety distances related to unconfined hydrogen explosions. International Conference on Hydrogen Safety. Pisa, September 8-10. 2005 M. Kustnetzov, J Grune. Planned HyTunnel experiments at FZK. Proc of 3rd IEF Workshop on velocity measurements in gases and flames, April 5-6 2006, HSL, Buxton, UK, 2006 Y.A. Gostintsev, A.G. Istratov, Y.V. Shulenin. Self-similar propagation of a free turbulent flame in mixed gas mixture. Fizika Gorenika i Vzryva. Vol. 24. 1988 M.J. Tang, Q.A. Baker. A New Set of Blast Curves from Vapor Cloud Explosion. Process Safety Progress (Vol 18). 1999

[6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16]

169

www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

FIGURES

Figure 1. Solution of the problem. Steady 1D point-symmetrical flame in a homogeneous medium.

Figure 2. Solution of the problem. Phase 1. Computed pressure P versus distance r.

170

www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

Figure 3.. Solution of the problem. Phase 1. Computed pressure P versus the characteristic variable r/t. As one can see, the solution is self-similar.

Figure 4. Solution of the problem. Phase 2. Computed pressure P versus distance r.

171

www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

Figure 5. Solution of the problem. Shock contact discontinuity interaction. (P,u) diagram in the phase space on the left, pressure versus distance on the right. On the top, the situation before the right traveling shock reaches the contact discontinuity. L and R are the left and right states of the shock. R and R' are the left and right states of the contact discontinuity (R' and R differs only for the densities). Once the shock reaches the contact discontinuity (pictures on the medium and on the bottom), we have the formation of a new discontinuity (new Riemann problem to deal with). In the (P,u) diagram, R and R' coincide. In the picture on the medium, R' > R and the slope of the shock curve for R' is larger than slope the of the shock curve for R (dot line). Two shocks and one contact discontinuity connect L and R'. In the pictures on the bottom R' < R and we have the formation of a left traveling rarefaction and a right traveling shock.

Figure 6. Solution of the problem. Phase 3. Computed pressure P versus distance r.


172 www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

Figure 7. Parametric study of the different approaches. CFD approach versus TNTequivalent approach. Maximum overpressure and positive impulse versus distance.

173

www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

Figure 8. Parametric study of the different approaches. CFD approach versus TNOapproach. Maximum overpressure and positive impulse versus distance.

174

www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

Figure 9. Experimental results versus CFD results. Pressure versus time in r=5 m.

Figure 10. Experimental results versus CFD results. Pressure versus time in r=35 m.
175 www.see.ed.ac.uk/feh5

Proceedings of the 5th International Seminar on Fire and Explosion Hazards, Edinburgh, UK, 23-27 April 2007

Figure 11. Experimental results versus CFD results. Pressure versus time in r=80 m.

176

www.see.ed.ac.uk/feh5