Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
A computer model is presented for gas explosion propagation in complex densely packed
geometries. The governing equations are formulated according to the quasi-continuum principle.
The influence of obstacles is allowed for by specifying volume and area porosities and by
including distributed resistances in the calculation domain. Distributed resistance functions are
given for various types of obstacles, such es flow parallel to, or normal to, rod bundles and single
obstacles, either sharp or rounded. Modified versions of the k-s turbulence model and the
eddy-dissipation combustion model, including the porosity/distributed resistance formulae, are
presented. The method is validated by calculating the pressure build-up in offshore module
geometries and comparisons with experimental data is generally good.
Geometries found in industrial practice may contain a (Moen et al.‘, Hjertager et al. 8*9). Similar favourable
lot of geometrical detail which can influence the results were found when the simulated results were
process to be simulated. Examples of such geometries compared to the experiments of Lee et al. lo. Their tests
are heat exchangers with thousands of tubes and were performed in a small scale tube (diameter 5 cm,
several baffles, and regenerators with many internal length 3 m) using variable concentrations of hydrogen
heat-absorbing obstructions. The geometries found -air. The model was, however, not able to predict the
inside modules on offshore oil and gas producing transition to detonation found in the hydrogen-air
platforms are relevant examples and can be modelled in experiments.
two ways: using very fine geometrical resolution or by Ghan et al. performed a series of experiments in a
using suitable bulk parameters. A detailed description channel (height 0.203 m, length 1.22 m) with repeated
requires extensive computer memory and fast calcu- obstacles and variable top venting”. Bakke and Hjer-
lation speeds and is not feasible with currently compu- tager used the above model and simulated several of
ters, and may not be to implement the possible with the test cases”. The model was able to reproduce the
future computers. We therefore used the second effect of variable confinement on flame speed and the
approach, which incorporates the so-called effect of changing position of the obstacles. Moen et al.
porosity/distributed resistance (PDR) formulation of performed experiments in a large-scale top-vented
the governing equations. This method was proposed by channel (cross-section 1.8 m X 1.8 m, length 15.5 m)
Patankar and Spalding and has been applied to analysis using acetylene-air, propane-air and hydrogen sul-
of heat exchangers, regenerators and nuclear reactors. ’ phide-air mixtures13. They also present simulation
Sha et al. have extended the method to include results using the above-mentioned model and show that
advanced turbulence modelling2~3. the model is able to predict the difference between the
Hjertager presented a computation method cap- three fuels. Namely, large flame acceleration in ace-
able of calculating explosion development in volume tylene and no significant acceleration in propane and
containing simple obstructions4,5 and a summary of the hydrogen sulphide. The model was not able to predict
results from calculations of idealized geometries using the transition from deflagration to detonation in
this mode16. The characteristics of comparisons with acetylene-air that occur at the end of the channel.
experiments include the following. Bakke and Hjertager applied the modelI to the
The model is able to simulate the peak pressures empty volume propane-air tests of Solberg15. The
and terminal flame speeds obtained in the large-scale predictions showed reasonable agreement for peak
tube (diameter 2.5 m, length 10 m) for both methane- pressure versus vent area for three different volumes
air and propane-air with variable concentrations (3.6 1,35 m3 and 425 m3).
Bjorkhaug et al. used the model to analyse the
experimental data from the radial vessel geometry with
Received 8 November I991 variable top venting l6 . They found good correlation
09504230/92/03016rM0
@ 1992 Butterworth-Heinemann Ltd
fi” =
v,+
vfvs = Vf
Axhyhz
(1)
The momentum equations for the Ui component may,
Here Vf is the volume of fluid and V, is the volume of for the situation given in Figures I and 2, be written as:
solid obstructions. Similarly, area porosities in each of
the three coordinate directions may be defined.
T& (BvP”i) + $, (PjP”jui) =
Figure2 shows the control volume with an area I
porosity in the n direction. This is expressed as:
- Pv z
ap+ & (Bjaij) + PvPgi + Ri
p _ auidjsutiace
dYdz
(2) I
x
AyAz (4)
Here U, is the velocity component in the xi coordinate
direction; p is the pressure, p is the density; oij is the
turbulent flux of momentum at the control volume
surfaces; gi is the gravitational acceleration in the xi
direction and Ri is the additional frictional resistance
caused by obstacles located inside the control volume.
The resistance per unit volume in the Xi direction,
R i, may be expressed as:
Ri = - fi * A,;# Uil Ui (5)
where A,,, is the so-called wetted area of the obstruc-
tions per unit volume. The friction factor fi may
depend on parameters like velocity, porosity, typical
dimension or hydraulic diameter, pitch between obs-
tacles, obstacle shape and orientation.
I}
0.5 1.08
3; -1
[O Turbulence and combustion models
where Re, = pi U,l * D/p1 is the Reynolds number Turbulence. In Equations (5), (10) and (12) above,
based on the average rod diameter D. P is the average the stress tensor oi,j and the fluxes Jh,i and Jl,, acting on
pitch between the rod obstacles. the control volume surfaces shown in Figures 1 and 2,
must be modelled. For this we use the k-e turbulence
Resistance due to single objects. If there is only one
model which determines the distribution of the kinetic
obstacle inside the control volume, the additional
energy of turbulence, k, and its rate of dissipation, E.
resistance can be calculated as flow over a blunt body.
The diffusive fluxes and stresses are modelled accord-
This may be expressed as3 :
ing to:
1
R,=-C+JU;.Ai
(12)
Here A, is the frontal area per unit volume of the solid
and
body and CR is the drag coefficient for the body. We
can see that the resistance goes to infinity as the
porosity approaches zero, thus imposing zero velocity
at the solid obstruction. The drag coefficient may attain
Here 8, = 1 if i = j and 6, = 0 if i # j. An effective
various values dependent on the shape of the obstacle.
viscosity peff and an effective Prandtl/Schmidt number
For a sharp edged obstacle CR is 1.2, whereas for
o@ have been introduced. The effectiire viscosity is
rounded obstacles like a tube or a sphere CR is 0.5.
related to the two variables in the turbulence model, k
Energy and chemical species conservation and E, as:
The first law of thermodynamics applied to the control
p&=K+C,,p$ (14)
volumes given in Figures I and 2 reads:
C, is a constant taken to be 0.09 (Launder and
Spalding”). The conservation equations that deter-
mine the distribution of k and E read:
- 6 (PjJh,j) + A z+ AS* + 0 (9)
$ (Pv~k)+ & (BjPUjk)=
Here h is the enthalpy; Jh,j is the enthalpy diffusive flux
at the boundaries of the control volume: Sh is the
frictional losses in the fluid inside the control volume;
’ $(eS) + - G &pe (15)
GS = Bvoij
au,
ax_ (17)
I
the two time scales are in a certain ratio modelled using approximately 100 obstructions for the
(t,,,/t,)* = D,. The rate of combustion is thus calcu- compressor module and 135 for the separator module.
lated as: Computer-aided design (CAD) software, named Auto-
CAD, has been used to represent the geometry and to
Rf,=O when tch > D, prepare input files for the EXSIM code. Figures 3 and 4
te show the geometry as represented in the calculations
zch
Rfu = - /3,: pmlim when - G Di, (29
=,
Solution procedure
Figure 3
It is noted that all conservation equations have the
same general form. Solution of these equations is
performed by finite-volume methods. Details of the
computation method are given by Hjertager4. Only a
brief description of the solution method is given here.
The calculation domain is divided into a finite
number of main grid points where the pressure p,
density p, enthalpy h, mass fraction of fuel mti,
mixture fraction f, the two turbulence quantities, k and
E, and the volume porosity &, are stored. The three
velocity components U, V, W and the three area
Figure 4 View of the separator module
porosities &, BY, p, are, on the other hand, stored at
grid points located midway between the main points.
The conservation equations are integrated over control
volumes surrounding the relevant grid points in space,
and over a time interval At. The integration is
performed using upwind differencing and implicit lOOO-
formulation.
The result of this is a set of non-linear algebraic
equations, which are solved by application of the well
known t&diagonal matrix algorithm used along the
three coordinate directions. Special care has been
7
taken to solve the pressure/velocity/density coupling of
the three momentum equations and the mass balance. :
- loo-
The ‘SIMPLE’ method developed by Patankar and
s
Spalding for three-dimensional parabolic flows’* has t
been extended by Hjertager to compressible flows4 and z
a
is used to handle this coupling. The method introduces Y
I
a new variable, the so-called pressure correction which 0.
makes the necessary corrections to the velocity com-
ponents, pressure and density to make them obey the lo-
mass balance constraint at the new time level. The
pressure correction is determined by solving a set of
algebraic equations derived from the linearized
momentum equations and the mass balance equation.
Znfruence of scaling
Figure 7 show comparison between predicted and
measured peak pressures for methane-air clouds as
function of scale. The figure also shows the predicted
extrapolation to full scale. The predicted peak pressure
increases with scale. In 15 scale the peak pressures
were between 300 and 900 mbar, whereas the pressures
for 1: 1 scale is between 2 and 3 bars.
0 1 2 3 4 5
Figure 8 shows the predicted peak pressures in the
Vent parameter (A/V*) full-scale separator module for both methane-air and
Figure6 Peak pressure as function of vent parameter for propane-air explosions. Figure 8 is therefore the scaled
centrally ignited explosions in the separator module. Compari- up version of Figure 6. In Figure 8 the predicted peak
sons between experiment (open symbols) and simulations overpressure for propane-air explosions is increased
(filled symbols). 0, 0, Propane; 0, n , methane; A, A, methane
1:33 scale; *, *, methane, empty module; 0, +, propane, from about 1.5 bar for the 15 scale to between 7 and 8
empty module bars for the 1:l scale. The figure also shows the
that the pressure distribution is much more non-uni- Figure9 Pressure versus time at three different positions along
the module I:5 scale separator module. Propane-air cloud. -,
form in the 1: 1 scale compared to the 1:5 scale. pl;---,p2;--,p3
Contour plots
Figures I1 and I2 and show contour plots in the
xz-plane of the 15 scale compressor module with a
propane-air cloud ignited centrally on the lower deck
at four instants in time, namely when 95, 75, 50 and
25% of the fuel is left inside the module, respectively.
This is similar to one of the experimental cases (test 87)
reported by Hjertager et al. “. The propagation of the
b 5.0
2 t
Time (ms.1
Figurelg Pressure versus time at three different positions
along the I:1 scale separator module. Propane-air cloud. -, pl;
---,p2;-,p3
I I I I I I I I I r I I I I I I I I
(III4 - o.auKJPNm9 mx- 0.276 K-11(0 MlN - 0.02 lvs lmx- 17.09 M
b
I
t
I 3 I I I 1 I I !
ii
Figure 11 Contour plots in the xz-plane for the 1:5 scale compressor module. (a) 95% of the fuel left in the module, (b) 75% of the fuel
left in the module
n,* - o.ooo m mK- a.581 mm “,H - 0.000 KWlQP3 IRX - 6l.Sza Kwlaa
I I . .. . _: _
2;:
.,. :(
r . . . ..‘I-. I..._
.’
._._. I’...._..
_ ._..
~_._...
:_.
~
I. 1 , I I 1 . :. :.
:_,_,..::z.
‘. . ,..’ _:
7_ _ .....
1” . . . . . . T-“‘l‘ . . . .
rc,” - o.moKGPalKo tax- 0.2.nl Komhs n1n - 0.02 ws ml- 82.34 w!i
Figum 12 Contour plots in the xz-plane for the 1:5 scale compressor module. (a) 50% of the fuel left in the module, (b) 25% of the fuel
letI in the module
congested geometries is presented. Computations were 7. Moen, I. O., Lee, J. H. S., Hjertager, B. H., Fubre, K. and
reported which compare the computer model with Eckhoff, R. K. Cornbust. Flame 1982,47,31-52
8. Hjertager, B. H., Fuhre, K., Parker, S. J. and Bakke, J. R.
several sets of experimental data relevant for offshore Progr. American Institute of Aeronautics and Astronautics, New
situations. The agreement between predictions and York 1984,94,504-522
measurements is in general good. However, more work 9. Hjertager, B. H., Fuhre, K. and Bjorkhaug. M. Combusr. Sci.
Technol. 1988,62,239-256
is needed to develop and verify the porosity/distributed 10. Lee, J. H. S., Knystautas, R. and Freiman, A. Cornbust. Flame
resistance model for explosion propagation in high 1984,56,227-239
density obstacle fields; to improve the turbulent com- 11. Chan, C., Moen, I. 0. and Lee, J. H. S. Combust. Flame 1983,
49,27-39
bustion model and to develop a model for deflagration 12. Bakke. J. R. and Hjertager, B. H. The effect of explosion
to detonation transition. venting in obstructed channels, in ‘Modeling and Simulation in
More experimental data are needed to enable Engineering’, Elsevier Science Publication, Amsterdam, 1986,
pp. 237-241
verification of the model in high-density geometries 13. Moen, I. 0.. Sulmistras, A., Hjertager, B. H. and Bakke, J. R.
using homogeneous as well as non-homogenous fuel Turbulent flame propagation and transition to detonation in
-air clouds. large fuel-air clouds, 21st Symposium (Int.) on Combustion,
The Combustion Institution, Pittsburgh, 1986, pp. 1617-1627
14. Bakke, J. R. and Hjertager, B. H. ht. J. Num. Methods Eng.
1987,24,129-140
Acknowledgement 15. Solberg, D. M. Gas explosion research related to safety of ships
and offshore platforms, in ‘Fuel-Air Explosions’, University of
The work on gas exposions at SiT/T.el-Tek is financially Waterloo Press, Ontario, 1982, pp. 787-819
supported by Shell Research Ltd. 16. Bjorkhaug, M., Bakke, J. R. and Hjertager, B. H. Calculation
of gas explosion in radial geometry using PLACS-ICE-ZD, Chr.
Michelsen Institute, CM1 Report No. 865403-4, 1986, [also in
References Bjorkhaug, M., PhD Thesis, City Univerisity, London, 19861
17. Hjertager, B. H., Fuhre, K. and Bjorkhaug, M. J. Loss Prev.
1. Patankar, S. V. and Spalding, D. B. A calculation procedure for Process Ind. 1988,1,197-205
the transient and steady-state behavior of shell-and-tube heat 18. Patankar, S. V. and Spalding, D. B. Forsch. Ingenieurwes.
exchangers, in ‘Heat Exchangers: Design and Theory Source- 1978.44.47
book (Ed. N. H. Afgan and E. V. Schltlnder) McGraw-Hill, 19. Launder, B. E. and Spalding, D. B. Computer Methods Appl.
New York, 1974, pp. 155-176 Mech. Eng. 1974,3,269-289
2. Sha, W. T., Yang, C. I., Kao, T. T. and Cho, S. M. 3 Heat 20. Magnussen, B. F. and Hjertager, B. H. On the mathematical
Transfer 1982,104,417-425 modeling of turbulent combustion with special emphasis on soot
3. Sha, W. T. and Launder, B. E. A model for turbulent formation and combustion, 16th Symp. (Int) on Combustion,
momentum and heat transport in large rod bundles, Argonne Combustion Institute, Pittsburgh, 1976, pp. 719-729
National Labs., 77-73, 1979 21. Bakke, J. R. and Hjertager, B. H. Quasi-laminar/turbulent
4. Hjertager, B. H. Cornbust. Sci. Technol., 1982,41,159-170 combusion modelling, real cloud generation and boundary
5. Hjertager, B. H. Numerical simulation of flame and pressure conditions in the FLACS-ICE code, CM1 No. 865402-2, Chr.
development in gas explosions, SM study No. 16, University of Michelsen Institute, 1986, [also in Bakke, J. R., DSc Thesis,
Waterlcm Press, Ontario, Canada, 1982, pp 407-426 University of Bergen, Bergen, 19861
6. Hjertager, B. H. Modeling, identification and control, 1989, 10, 22. Patankar, S. V. and Spalding, D. B. ht. J. Heat Mass Transfer
227-247 1972,15,1787-1806