simulation of swirling gas–particle flows using a nonlinear k–ε–kp two-phase turbulence model
TRANSCRIPT
Simulation of swirling gas–particle flows using a nonlinear
k–e–kp two-phase turbulence model
L.X. Zhou *, H.X. Gu
Department of Engineering Mechanics, Tsinghua University, Beijing 100084, China
Received 1 January 2002; received in revised form 1 August 2002; accepted 2 August 2002
Abstract
The linear k–e–kp two-phase turbulence model is rather simple, but it cannot predict the anisotropic turbulence of strongly swirling gas–
particle flows. The second-order moment, two-phase turbulence model can better predict strongly swirling flows, but it is rather complex.
Hence, the algebraic Reynolds stress expressions are derived based on two-phase Reynolds stress equations, and then the nonlinear
relationships of two-phase Reynolds stresses and two-phase velocity correlation with the strain rates are obtained. These relationships,
together with the transport equations of gas and particle turbulent kinetic energy and the two-phase correlation turbulent kinetic energy
constitute the nonlinear k–e–kp turbulence model. The proposed model is applied to simulate swirling gas–particle flows. Predictions give
the two-phase time-averaged velocities and Reynolds stresses. The prediction results are compared with phase Doppler particle anemometer
(PDPA) measurements and those predicted using the second-order moment model. The results indicate that the nonlinear k–e–kp model has
modeling capability nearly equal to that of the second-order moment model, but the former can save much computation time.
D 2002 Elsevier Science B.V. All rights reserved.
Keywords: Gas–particle flows; Turbulence model; Nonlinear model
1. Introduction
Recently, remarkable strides have been made in devel-
oping nonlinear k–e models for single-phase flows. The
nonlinear stress–strain relationships can be obtained by
different approaches: the generalized Cayley–Hamilton
formula and the invariant principle [1,2], the renormaliza-
tion group (RNG) theory [3], the two-scale DIA method [4]
and the algebraic Reynolds stress model [5]. In the nonlinear
stress–strain model, the equivalent eddy viscosity is aniso-
tropic. Therefore, the nonlinear model can simulate the
anisotropic turbulence of swirling flows and separating
flows. The nonlinear k–e model has been validated in
simple flows (homogeneous shear flows and channel flows)
and has been applied to simulate complex flows (backward-
facing step flows, swirling flows) [1,6]. Comparison with
experiments shows that it does make significant improve-
ment over the standard isotropic k–e model and is almost as
economical as the linear k–e model.
As for the two-phase turbulence model in developing
two-fluid models of turbulent gas–particle flows, several
years ago, the two-phase Reynolds stress or unified second-
order moment (USM) and k–e–kp models were proposed by
the first author of this paper based on Reynolds averaging
and a closure method similar to that used in single-phase
turbulence modeling [7,8]. These models are used to simu-
late swirling gas–particle flows [9,10]. The second-order
moment models of only particle turbulence were also devel-
oped by Refs. [11,12] based on the PDF transport equation.
It has been found in application that the k–e–kp model is
rather simple and can simulate well nonswirling and weakly
swirling gas–particle flows. However, for strongly swirling
flows, the USM model should be better, but the USM model
is rather complex and is not convenient for engineering
application. A best compromise between the applicability
and simplicity is either an implicit algebraic two-phase
Reynolds stress model, or a nonlinear k–e–kp two-phase
turbulence model, i.e. an explicit algebraic two-phase Rey-
nolds stress model. Since the algebraic Reynolds stress
models frequently cause some divergence problems due to
lack of diffusion terms in the momentum equation, partic-
ularly in 3-D flows, a nonlinear k–e–kp two-phase turbu-
0032-5910/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved.
PII: S0032 -5910 (02 )00153 -5
* Corresponding author. Tel.: +86-10-6278-2231; fax: +86-10-6278-
1824.
E-mail address: [email protected] (L.X. Zhou).
www.elsevier.com/locate/powtec
Powder Technology 128 (2002) 47–55
lence model is preferred. This is because the momentum
equations and k, e, kp equations have the same form for both
linear and nonlinear k–e–kp models, so it is easier to obtain
convergent results.
In this paper, algebraic two-phase Reynolds stress
expressions are derived based on two-phase Reynolds stress
equations, and then the nonlinear relationships of two-phase
Reynolds stresses and two-phase velocity correlation with
the strain rates are obtained. The nonlinear k–e–kp model is
applied to simulate swirling gas–particle flows. Comparison
is made between the predictions using the nonlinear k–e–kpmodel, the USM model and the PDPA measurement data to
assess the proposed model.
2. The two-phase Reynolds stress equations and
algebraic stress expressions
Previously, a second-order moment, two-phase turbu-
lence model [7,8] was proposed. The second-order moment,
two-phase turbulence model was used to simulate swirling
gas–particle flows [9,10]. In the second-order moment
model, the closed form of the two-phase Reynolds stress
equations is:
B
BtðqvivjÞ þ
B
BxkðqVkvivjÞ ¼ Dij þ Pij þ Gpij þ Pij � eij
ð1Þ
B
BtðNpvpivpjÞ þ
B
BxkðNpVpkvpivpjÞ ¼ Dp;ij þ Pp;ij þ ep;ij ð2Þ
where Dij, Pij, Pij, eij are terms having the same meanings as
those in single-phase Reynolds stress equations, thus, we
have
Dij ¼B
Bxkcsk
evkv1
Bvivj
Bxk
� �;
Pij ¼ �q vivkBVj
Bxkþ vjvk
BVi
Bxk
� �;
Pij ¼ Pij;1 þ Pij;2; Pij;1 ¼ �c1ek
q vivj �2
3dijk
� �;
Pij;2 ¼ �c2 Gij �2
3dijG
� �; eij ¼
2
3dije;
G ¼ �qvivkBVi
Bxk
and the source term in case of two-phase flows
Gp;ij ¼Xp
qp
srpðvpivj þ vpjvi � 2vivjÞ
is the gas Reynolds stress production/destruction due to the
particle drag force. The transport equation of dissipation rate
of gas turbulent kinetic energy is:
B
BtðqeÞ þ B
BxkðqVkeÞ ¼
B
Bxcek
evkv1
BeBx1
� �
þ ek½ce1ðGþ GpÞ � ce2qe� ð3Þ
where the new source term is
Gp ¼Xp
qp
srpðvpivi � viviÞ
Dp,ij, Pp,ij, ep,ij are the diffusion, production terms of particle
Reynolds stress and the production/destruction term due to
gas turbulence, respectively, and they can be given as
follows:
Dp;ij ¼B
BxkNpc
sp
kp
epvpkvp1
B
Bx1ðvpivpjÞ
� �
Pp;ij ¼ �ðVpknpvpj þ NpvpkvpjÞBVpi
Bxk� ðVpknpvpi
þ NpvpkvpiÞBVpj
Bxkþ npvpjgi þ npvpigj
ep;ij ¼1
srpNpðvpivj þ vpjvi � 2vpivpjÞ þ ðVi � VpiÞnpvpj�
þ ðVj � VpjÞnpvpi�
For a closed system, besides Eqs. (1)–(3), the transport
equations of npvpi; npvpj; npnp; vpivj; vpjvi are also used.
For example, the transport equation of vpjvi based on our
closure method is:
B
BtðvpivjÞ þ ðVk þ VpkÞ
B
BxkðvpivjÞ
¼ B
Bxkðme þ mpÞ
B
BxkðvpivjÞ
� �þ 1
qsrp
� ½qvpivpj þ qvivj � ðq þ qpÞvpivj�
� vpkvjBVpi
Bxkþ vkvpi
BVj
Bxk
� �� e
kvpividij ð4Þ
When using this full second-order moment model for a
three-dimensional flow, it is necessary to solve 26 differ-
ential equations, including six gas Reynolds stress equa-
tions, six particle Reynolds stress equations, nine two-phase
L.X. Zhou, H.X. Gu / Powder Technology 128 (2002) 47–5548
velocity correlation equations, one dissipation rate equation
for gas turbulent kinetic energy, three particle diffusion mass
flux equations and one equation for the mean square value
of particle number density fluctuation.
In order to reduce the computation time and simulta-
neously retain the anisotropic features of the turbulence
model, as that done in the single-phase turbulence model, an
algebraic two-phase stress model is obtained by simplifying
the stress transport equations. Neglecting the convection and
diffusion terns in (Eqs. (1),(2) and (4), the algebraic expres-
sions of two-phase Reynolds stresses and two-phase veloc-
ity correlation can be obtained as:
vivj ¼ ð1� kÞ 23kdij þ k
k
evivk
BVj
Bxkþ vjvk
BVi
Bxk
� �
þ k
c1qeðvpivj þ vivpj � 2vivjÞ ð5Þ
vpivpj ¼ � srp2
vpivpkBVpj
Bxkþ vpjvpk
BVpi
Bxk
� �
þ 1
2ðvivpj þ vpivjÞ ð6Þ
vpivj ¼ � qsrpq þ qp
vpivkBVj
Bxkþ vjvpk
BVpi
Bxk
� �
þ qq þ qp
vivj þqp
q þ qp
vpivpj �qsrp
q þ qp
1
sevpividij
ð7Þ
where vivj and vpivpj are gas phase and particle phase
Reynolds stresses, respectively. vpivj is the two-phase
velocity correlation. The last term on the right-hand-side
of the expression for vpivj is the dissipation term. It is
assumed that the dissipation is an isotropic one, which is
proportional to the summation of the normal components
divided by a time scale. This time scale can be either the gas
turbulence scale k/e, or the particle relaxation time srp, or theeddy interaction time accounting for the crossing trajectory
effect, se =min[k/e, srp]. Predictions indicate that the last oneis better than other time scales [13].
3. The nonlinear k–e–kp two-phase turbulence model
It has been found that the above stated implicit algebraic
stress model, in which the unknown stresses exist on both
two sides of Eqs. (5)–(7), frequently causes divergent
problems, in particular in 3-D flows. Therefore, in devel-
oping the single-phase turbulence model, the nonlinear k–emodel, or in other words, the explicit algebraic stress model
is preferred. To construct a nonlinear k–e–kp two-phase
turbulence model, we can transform Eqs. (5)–(7) into the
following explicit form, on the right-hand side of which
there are no terms containing the unknowns vpivj; vpivpj;vivj.
After transformation, we have
vivj ¼ Að1Þdij þ Að2Þ vivkBVj
Bxkþ vjvk
BVi
Bxk
� �
þ Að3Þ vpivpkBVpj
Bxkþ vpjvpk
BVpi
Bxk
� �
þ Að4Þ vpivkBVj
Bxkþ vjvpk
BVpi
Bxk
�
þ vivpkBVpj
Bxkþ vpjvk
BVi
Bxk
�
vpivpj ¼ Bð1Þdij þ Bð2Þ vivkBVj
Bxkþ vjvk
BVi
Bxk
� �
þ Bð3Þ vpivpkBVpj
Bxkþ vpjvpk
BVpi
Bxk
� �
þ Bð4Þ vpivkBVj
Bxkþ vjvpk
BVpi
Bxk
�
þ vivpkBVpj
Bxkþ vpjvk
BVi
Bxk
�
vpivj ¼ Cð1Þdij þ Cð2Þ vivkBVj
Bxkþ vjvk
BVi
Bxk
� �
þ Cð3Þ vpivpkBVpj
Bxkþ vpjvpk
BVpi
Bxk
� �
þ Cð4Þ vivpkBVpj
Bxkþ vpjvk
BVi
Bxk
� �
þ Cð5Þ vpivkBVj
Bxkþ vjvpk
BVpi
Bxk
� �
where A(1)–A(4), B(1)–B(4), C(1)–C(5) are all functions
of k, e, kp, kpg, qp, q and srp
Að1Þ ¼ ð1� kÞ 23k �
2qp
c1q2
3kpg
Bð1Þ ¼ Cð1Þ ¼ ð1� kÞ 23k � srp
ekþ
2qp
c1q
� �2
3kpg
Að2Þ ¼ Bð2Þ ¼ Cð2Þ ¼ �kk
eAð3Þ ¼ �
qpMsrpq
Bð3Þ ¼ � srp2q
ðq þ qp þ 2MqpÞ
Cð3Þ ¼ �qpsrp
q1
2þM
� �
Að4Þ ¼ �Msrp Bð4Þ ¼ �srp M þ 1
2
� �
L.X. Zhou, H.X. Gu / Powder Technology 128 (2002) 47–55 49
Cð4Þ ¼ �srp M þqp
2ðq þ qpÞ
!
Cð5Þ ¼ �srp M þqp
2ðq þ qpÞþ q
q þ qp
!
M ¼kqp
c1qesrpsrp ¼
d2pqp
18l1þ Re
2=3p
6
!�1
Rep ¼A!vp �!vAdp=m
The variables k, e, kp and krg will be determined by the
governing equations, where k is the gas turbulent kinetic
energy, e is its dissipation rate, kp is the particle turbulent
kinetic energy and kpg is defined by kpg ¼ vivpi.
We can write the expressions of vmvk ; vpmvpk ; vpmvk ;vmvpk ; ðm ¼ i; jÞ in a similar form, then these expressions
are iterated into the equations of vivj; vpivpj; and vpivj .
Finally, obtained expressions should take the following
form:
vivj ¼ Cdij þ OBVm
Bxn
� �þ O
BVm
Bxn
� �2 !
þ OBVm
Bxn
� �3 !
þ : : : : : : þ OBVm
Bxn
� �l!
where {C} is a constant
For practical application, we must cut the series to a
certain term. For single-phase flows, it is suggested that a
quadratic form with appropriate coefficients can predict
turbulent shear flows and swirling flows quite well. Here,
the obtained relationships are written to quadratic terms of
the strain rates. In fact, it is a perturbation method. The
resulting nonlinear stress–strain rate relationships written to
quadratic power terms of the strain rates are:
vivj ¼ G1dij þ G2
BVi
Bxjþ BVj
Bxi
� �þ G3
BVpi
Bxjþ BVpj
Bxi
� �
þ G4
BVi
Bxk
BVj
Bxkþ BVk
Bxj
� �þ BVj
Bxk
BVi
Bxkþ BVk
Bxi
� �� �
þ G5
BVpi
Bxk
BVpj
Bxkþ BVpk
Bxj
� ��
þ BVpj
Bxk
BVpi
Bxkþ BVpk
Bxi
� ��
þ G6
BVi
Bxk
BVpj
Bxkþ BVpk
Bxj
� �þ BVj
Bxk
BVpi
Bxkþ BVpk
Bxi
� �� �
þ G7
BVpi
Bxk
BVj
Bxkþ BVk
Bxj
� �þ BVpj
Bxk
BVi
Bxkþ BVk
Bxi
� �� �
þ G8
BVi
Bxk� BVpi
Bxk
� �BVj
Bxk� BVpj
Bxk
� �� �ð8Þ
vpivpj ¼ P1dij þ P2
BVi
Bxjþ BVj
Bxi
� �þ P3
BVpi
Bxjþ BVpj
Bxi
� �
þ P4
BVi
Bxk
BVj
Bxkþ BVk
Bxj
� �þ BVj
Bxk
BVi
Bxkþ BVk
Bxi
� �� �
þ P5
BVpi
Bxk
BVpj
Bxkþ BVpk
Bxj
� ��
þ BVpj
Bxk
BVpi
Bxkþ BVpk
Bxi
� ��
þ P6
BVi
Bxk
BVpj
Bxkþ BVpk
Bxj
� ��
þ BVj
Bxk
BVpi
Bxkþ BVpk
Bxi
� ��þ P7
BVpi
Bxk
BVj
Bxkþ BVk
Bxj
� ��
þ BVpj
Bxk
BVi
Bxkþ BVk
Bxi
� ��þ P8
BVi
Bxk� BVpi
Bxk
� ��
� BVj
Bxk� BVpj
Bxk
� ��ð9Þ
vpivj ¼ T1dij þ T2BVi
Bxjþ BVj
Bxi
� �þ T3
BVpi
Bxjþ BVpj
Bxi
� �
þ T4BVj
Bxiþ BVpi
Bxj
� �þ T5
BVi
Bxk
BVj
Bxkþ BVk
Bxj
� �
þ T6BVj
Bxk
BVi
Bxkþ BVk
Bxi
� �þ T7
BVpi
Bxk
BVpj
Bxkþ BVpk
Bxj
� �
þ T8BVpj
Bxk
BVpi
Bxkþ BVpk
Bxi
� �þ T9
BVpi
Bxk
BVj
Bxkþ BVk
Bxj
� �
þ T10BVpj
Bxk
BVi
Bxkþ BVk
Bxi
� �
þ T11BVi
Bxk
BVpj
Bxkþ BVpk
Bxj
� �
þ T12BVj
Bxk
BVpi
Bxkþ BVpk
Bxi
� �
þ T13BVi
Bxk
BVpj
Bxk� BVj
Bxk
� �þ BVpj
Bxk
BVi
Bxk� BVpi
Bxk
� �� �
þ T14BVj
Bxk
BVpi
Bxk� BVi
Bxk
� �þ BVpi
Bxk
BVj
Bxk� BVpj
Bxk
� �� �ð10Þ
whereallofG1–G8,P1–P8,T1–T14arefunctionsofk,e,kp,kpg,kpg, qp, q and srp. These coefficients are determined by the
following expressions, for example:
G1 ¼ Að1Þ; G2 ¼ Að2ÞAð1Þ þ Að4ÞCð1Þ;
G3 ¼ Að3ÞBð1Þ þ Að4ÞCð1Þ;
G4 ¼ Að1ÞðAð2ÞAð2Þ þ Að4ÞCð2ÞÞþ Cð1ÞðAð2ÞAð4Þ þ Að4ÞCð5ÞÞ;
L.X. Zhou, H.X. Gu / Powder Technology 128 (2002) 47–5550
G5 ¼ Bð1ÞðAð3ÞBð3Þ þ Að4ÞCð3ÞÞþ Cð1ÞðAð3ÞBð4Þ þ Að4ÞCð5ÞÞ;
G6 ¼ Bð1ÞðAð2ÞAð3Þ þ Að4ÞCð3ÞÞþ Cð1ÞðAð2ÞAð4Þ þ Að4ÞCð4ÞÞ;
G7 ¼ Að1ÞðAð3ÞBð2Þ þ Að4ÞCð2ÞÞ þ Cð1ÞðAð3ÞBð4Þ
þ Að4ÞCð4ÞÞ;
G8 ¼ 2Cð1ÞAð4ÞðCð4Þ � Cð5ÞÞ
where all of A(1)–A(4), B(1)–B(4), C(1)–C(5) are func-
tions of k, e, kp, kpg, qp, q and srp.
Að1Þ ¼ ð1� kÞ 23k �
2qp
c1q2
3kpg;
Bð1Þ ¼ Cð1Þ ¼ ð1� kÞ 23k � srp
ekþ
2qp
c1q
� �2
3kpg;
Að2Þ ¼ Bð2Þ ¼ Cð2Þ ¼ �kk
e; Að3Þ ¼ �
qpMsrpq
;
Bð3Þ ¼ � srp2q
ðq þ qp þ 2MqpÞ;
Cð3Þ ¼ �qpsrp
q1
2þM
� �; Að4Þ ¼ �Msrp;
Bð4Þ ¼ �srp M þ 1
2
� �; Cð4Þ ¼ �srp M þ
qp
2ðq þ qpÞ
!;
Cð5Þ ¼ �srp M þqp
2ðq þ qpÞþ q
q þ qp
!;
M ¼kqp
c1qesrpsrp ¼
d2pqp
18l1þ Re
2=3p
6
!�1
;
Rep ¼A!V p �!VAdp=m
The expressions for coefficients P1–P8, T1–T14 are
obtained in a similar manner. The variables k, e, kp and
kpg are determined by the governing equations given in the
following section, where k is the gas turbulent kinetic
energy, e is its dissipation rate, kp is the particle turbulent
kinetic energy and kpg is defined by kpg ¼ vivpi.
The transport equations of k, e, kp and kpg can be obtainedfrom Eqs. (1)–(4) by putting i= j:
B
BtðqkÞ þ B
BxkðqVkkÞ
¼ B
Bxkqcs
k
evkv1
Bk
Bx1
� �� qvivk
Bvi
Bxk� qe
þqp
srpð2kpg � 2kÞ ð11Þ
B
BtðqeÞ þ B
BxkðqVkeÞ
¼ B
Bxkqce
k
evkv1
BeBx1
� �
þ ek
ce1 �qvivkBvi
Bxkþ
qp
srpð2kpg � 2kÞ
� �� ce2qe
� �ð12Þ
B
BtðqpkpÞ þ
B
BxkðqpVpkkpÞ
¼ B
Bxkqpckp
kp
epvpkvpl
Bkp
Bx1
� �
� qpvpivpkBvpi
Bxkþ
qp
srpð2kpg � 2kpÞ ð13Þ
B
BtðkpgÞ þ ðVk þ VpkÞ
B
BxkðkpgÞ
¼ B
Bxkcsk
evkv1 þ ckp
kp
epvpkvpl
� �Bkpg
Bx1
� �
þ 1
qsrpðqpkp þ qk � ðq þ qpÞkpgÞ
� 1
2vivpk
Bvpi
Bxkþ vpivk
Bvi
Bxk
� �� 1
sekpg ð14Þ
The algebraic expressions (8)–(10) together with the
differential equations (Eqs. (11)–(14)) constitute the non-
linear k–e–kp two-phase turbulence model. It can be seen
that in comparison with the USM model, in this model the
Fig. 1. Swirl chamber.
L.X. Zhou, H.X. Gu / Powder Technology 128 (2002) 47–55 51
number of differential equations is reduced more than six
times (from 26 to 4) in simulating 3-D flows.
4. Simulation results and discussion
The nonlinear k–e–kp (NKP) model was used to simu-
late swirling gas–particle flows measured by Sommerfeld
and Qiu [14] using phase Doppler particle anemometer
(PDPA). The adopted grid nodes are 20� 43, but a fine
grid system of 40� 86 was also used, and it gives the same
results. The differential equations are integrated in the
computational cells to obtain finite difference equations
using a hybrid scheme. The FDEs are solved using the
SIMPLEC algorithm. The criterion for convergence is the
summation of residual mass sources less than 10� 4 for both
Fig. 2. Gas axial velocity (m/s, s = 0.47).
Fig. 3. Gas tangential velocity (m/s, s= 0.47).
Fig. 4. Axial velocity of 45-Am particles (m/s, s = 0.47).
L.X. Zhou, H.X. Gu / Powder Technology 128 (2002) 47–5552
gas and particle phases. A computer code called NKP-2 is
developed. Running a case in the Pentium-2 PC takes about
8 min. Using USM code it takes about 15 min. Comparison
is made between the predictions using the nonlinear k–e–kp(NKP) model, the USM model, and the PDPA measure-
ments. The geometrical configuration and the sizes of the
chamber are shown in Fig. 1. The particle size is 45 Am. The
inlet flow parameters are the same as those in experiments.
Figs. 2 and 3 give the predicted gas axial and tangential
velocities using both NKP and USM models. There is
almost no difference between these two models, and all of
them give results in good agreement with measurements.
For the axial and tangential velocities of 45-Am particles
(Figs. 4 and 5), in most regions of the flow field, the
difference between two model predictions is small and both
of them are in good agreement with experiments. Figs. 6 and
7 show the predicted gas axial and tangential fluctuation
velocities. Both NKP and USM models underpredict the
measured values, and the USM predictions are somewhat
but not much better than the NKP predictions. Figs. 8 and 9
Fig. 6. Gas axial fluctuation velocity (m/s, s = 0.47).
Fig. 5. Tangential velocity of 45-Am particles (m/s, s = 0.47).
Fig. 7. Gas tangential fluctuation velocity (m/s, s = 0.47).
L.X. Zhou, H.X. Gu / Powder Technology 128 (2002) 47–55 53
give the predicted axial and tangential fluctuation velocities
of 45-Am particles. The particle fluctuation velocities are
also underpredicted using both two models. Except for the
axial fluctuation velocity profile in the third section, the
difference between the two model predictions is smaller
than that for the gas phase. Both models predict that the gas
and particle axial fluctuation velocities are larger than the
tangential ones. This is in qualitative agreement with the
experimental results. In general, the NKP model can predict
what the USM model can predict, but the former can save
almost 50% computation time for a 2-D flow with small
geometrical sizes. Keeping in mind that in engineering
applications the accuracy of predicting the two-phase aver-
aged velocities is more important, one can consider that the
NKP model can be used instead of the USM model.
The discrepancy between predicted two-phase fluctua-
tion velocities (using either of two models) and measured
ones might be caused by the following reasons. First, in the
PDPA measurements each size group of particles has a
certain size range, so the measured gas or particle fluctua-
tion velocity includes the effect of particle size range. Next,
the particle–wall interaction may increase the particle
turbulence, which is not taken into account in either model.
Finally, the closure model of dissipation rate for the gas
turbulent kinetic energy in two-phase flows and the gas–
particle velocity correlation needs to be further improved.
5. Conclusions
(1) The NKP model has a capability nearly equal to that of
the USM model in simulating the two-phase averaged
velocities and fluctuation velocities of swirling gas–
particle flows.
(2) Both NKP andUSMmodels can give predicted two-phase
averaged velocities in good agreement with those
measured.
(3) Both of these models can simulate anisotropic two-phase
turbulence, but underpredict the two-phase fluctuation
velocities.
(4) The nonlinear k–e–kp model has no problem of conver-
gence encountered in the algebraic stress model.
(5) The NKP computer code is easier to obtain by modifying
a KP computer code.
(6) In 2-D flows in small geometries, the NKPmodel can save
about 50% of the computation time of the USM model.
However, in 3-D flows with large geometrical sizes, the
NKP model can save much more computation time.
Fig. 9. Tangential fluctuation velocity of 45-Am particles (m/s, s = 0.47).
Fig. 8. Axial fluctuation velocity of 45-Am particles (m/s, s = 0.47).
L.X. Zhou, H.X. Gu / Powder Technology 128 (2002) 47–5554
Acknowledgements
This study was supported by the Special Funds for Major
State Basic Research G-1999-0222-08, PRC.
References
[1] T.-H. Shih, J. Zhu, J.L. Lumley, Comput. Methods Appl. Mech. Eng.
125 (1995) 287–302.
[2] T.-H. Shih, J.L. Lumley, Remarks on turbulent constitutive relations,
Math. Comput. Model. 18 (1993) 9–16.
[3] R. Rubinstein, J.M. Barton, Nonlinear Reynolds stress models and the
renormalization group, Phys. Fluids, A 2 (1990) 1472–1476.
[4] A. Yoshizawa, Statistical analysis of the derivation of the Reynolds
stress from its eddy viscosity representation, Phys. Fluids 27 (1984)
1377–1387.
[5] D.B. Taulbee, An improved algebraic Reynolds stress model and
corresponding nonlinear stress model, Phys. Fluids, A 4 (1992)
2555–2561.
[6] C. Hirsch, A.E. Khodak, Modeling of complex internal flows with
Reynolds stress algebraic equation model, AIAA 95-2246 (1995).
[7] L.X. Zhou, Theory and Numerical Modeling of Turbulent Gas–par-
ticle Flows and Combustion, CRC Press, Boca Raton, Florida, 1993.
[8] L.X. Zhou, C.M. Liao, T. Chen, A unified second-order moment two-
phase turbulence model for simulating gas–particle flows, ASME-
FED 185 (1994) 307–313 (Lake Tahoe).
[9] L.X. Zhou, T. Chen, Y. Li, Comparison between different two-phase
turbulence models for simulating swirling gas–particle flows, ASME-
FED-v.245, Paper 5033, Washington, DC, 1998.
[10] L.X. Zhou, T. Chen, Simulation of swirling gas–particle flows using
USM and k–kp two-phase turbulence models, Powder Technol. 114
(1–3) (2001) 1–11.
[11] I.V. Derevich, L.I. Zaichik, The equation for the probability density of
the particle velocity and temperature in a turbulent flow simulated by
the gauss stochastic field, Prikl. Mat. Meh. 54 (5) (1990) 767.
[12] O. Simonin, Continuum modeling of dispersed turbulent two-phase
flows, VKI Lectures: Combustion in Two-Phase Flows, Jan. 29–
Feb. 2, 1996, von Karmen Institute for Fluid Dynamics, Brussels,
pp. 1–47.
[13] L.X. Zhou, Y. Xu, Simulation of swirling gas–particle flows using an
improved second-order moment two-phase turbulence model, Powder
Technol. 116 (2001) 178–189.
[14] M. Sommerfeld, H.H. Qiu, Detailed measurement in a particulate two-
phase flow by a phase Doppler particle anemometer. Int. J. Heat Fluid
Flow 12 (1991) 20–28.
L.X. Zhou, H.X. Gu / Powder Technology 128 (2002) 47–55 55