Computers and Chemical Engineering 71 (2014) 39–51

Contents lists available at ScienceDirect

Computers and Chemical Engineering

journal homepage: www.elsevier.com/locate/compchemeng

Multi-scale product property model of polypropylene produced in a

FBR: From chemical process engineering to product engineering
Ya-Ping Zhu a,b , Zheng-Hong Luo a,∗ , Jie Xiao c
a
Department of Chemical Engineering, College of Chemistry and Chemical Engineering, Shanghai Jiao Tong University, Shanghai 200240, PR China
b
Department of Chemical and Biochemical Engineering, College of Chemistry and Chemical Engineering, Xiamen University, Xiamen 361005, PR China
c
School of Chemical and Environmental Engineering, College of Chemistry, Chemical Engineering and Materials Science, Soochow University, Suzhou
215123, PR China

a r t i c l e i n f o a b s t r a c t

Article history: A multi-scale product model has been built to characterize the polypropylene (PP) formation dynamics
Received 14 April 2014 in a catalytic FBR. For the first time, the gas–solid flow field, the morphological and molecular proper-
Received in revised form 8 July 2014 ties of particles, as well as their dynamics can be simultaneously obtained by solving the unique model
Accepted 16 July 2014
that couples a CFD model, a population balance model (PBM) and moment equations. The quantitative
Available online 24 July 2014
relationships between the operating conditions and the multi-scale particle properties have been fur-
ther established. The results demonstrate that the product model can be used to guide a multi-scale
Keywords:
generalization of the polymer product from chemical process to product engineering.
Polypropylene
Fluidized bed reactor © 2014 Elsevier Ltd. All rights reserved.
Multi-scale modeling
Population balance model
Method of moments
End-use property

1. Introduction it should be noted that the development of product engineering

is relied on the basic theories of the chemical process engineering
Chemical engineering originated from the need to design pro- discipline (Wesselingh, 2001).
cesses and equipment in the chemical industry (Fogler, 2009) and Up until now, new chemical products have been convention-
as a discipline, it was formally introduced more than one cen- ally created by resorting to the knowledge of existing products
tury ago (Davis, 1901). In general, the main objective of chemical combined with trial and error experimental techniques. When the
process engineering is to convert raw materials to useful prod- existing knowledge is limited, product design then relies mainly
ucts through process design and analysis (Harold and Ogunaike, on heuristics. In addition, the experimental cost could drastically
2000). Chemical product engineering, on the other hand, mainly increase. As a result, there has been a challenging task to develop
involves the design of the product and its manufacturing process a systematic product design method to minimize experimental
in order to meet the customers’ requirements and promote the efforts in the absence of complete data (Hill, 2004, 2009). Multi-
competitiveness of enterprises (Cussler and Wei, 2003). Nowa- scale modeling that can establish the comprehensive relationships
days, development of novel chemical products becomes extremely between processing conditions and product properties from micro-
important for modern chemical corporations (Cagan and Vogel, to macro-scales is thus a promising tool for new chemical product
2002; Ulrich and Eppinger, 2003). It is, therefore, widely recog- development (Jaworski and Zakrzewska, 2011).
nized that both process engineering and product engineering are In this work, the development of new high-performance
two essential components in the chemical engineering discipline. polypropylene (PP) is used as a case study to demonstrate the
Furthermore, product engineering has been becoming a fast devel- effectiveness our unique product development methodology. Since
oping concept and attracting lots of attentions from both industry fluidized bed reactor (FBR) is one of the most popular commer-
and academia (Charpentier, 2002; Costa et al., 2006). Nevertheless, cial reactors to produce polyolefin (Shi et al., 2010), FBRs are
selected to perform the high-performance product development.
Note that the PP product in FBRs exhibits multi-scale properties,
namely, polymer particles’ morphological property (i.e., particle
∗ Corresponding author. Tel.: +86 21 54745602; fax: +86 21 54745602. size distribution, PSD) and molecular property (i.e., weight-average
E-mail address: [email protected] (Z.-H. Luo). molecular weight (Mw ) and melt index (MFI)) (Soares, 2001).

http://dx.doi.org/10.1016/j.compchemeng.2014.07.013
0098-1354/© 2014 Elsevier Ltd. All rights reserved.
40 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51

Nomenclature
˛s the volume fraction of solid phase
[AlR3 ] Aluminum alkyl concentration, kmol m−3 q turbulent Prandtl numbers for q phase
[C] catalyst concentration, kmol m−3  sg and  gs dispersion Prandtl number between gas and
[C *] vacant active site concentration of catalyst, solid, 0.75
kmol m−3 εq turbulent dissipation rate of q phase, m2 s−3
Cp heat capacity, kJ kg−1 K−1 g the density of gas phase, kg m−3
Bag (L ; x, t) the birth rate of particles diameter (L) due to s the density of solid phase, kg m−3
aggregation m the mth dead matrix
Bbr (L ; x, t) the birth rate of particles diameter (L) due to  angle of internal friction, degree
breakage s granular temperature, m2 s−2
Dag (L ; x, t) the death rate of particles diameter (L) due to  mixed fluid viscosity, Pa s
aggregation g gas viscosity, Pa s
Dbr (L ; x, t) the death rate of particles diameter (L) due to m the mth live matrix
breakage s solid shear viscosity, Pa s
Dim diffusion coefficient, m2 s−1 s,coi solid collision viscosity, Pa s
es particle–particle restitution coefficient s,kin solid kinetic viscosity, Pa s
E active energy, kJ kmol−1 s,fr solid frictional viscosity, Pa s
g gravitational acceleration, m s−2 t,q turbulent viscosity of q phase
g0 radial ditribution function ¯¯ q shear stress, N m−2
G(L)n(L ; x, t) particle flux due to molecular growth rate
s collision dissipation of energy
[H2 ] hydrogen concentration, kmol m−3 gs energy exchange between gas and solid
h0i formation enthalpy of species i, J kg−1
hq enthalpy of q phase, J kg−1
Hsg and Hgs interphase exchange coefficient of energy, Meanwhile, these multi-scale properties are directly linked to the
W m−3 K−1 multi-scale polymerization FBR behaviors and are strongly influ-
Ī¯ identity matrix enced by the reactor operating conditions, e.g., the gas–solid flow
I2D the second invariant of the deviator stress tensor field (i.e., gas and solid fractions). It has been reported that differ-
k0 pre-exponential factor. ent length scales (multi-scale) are involved in polymerization FBRs
kd kinetic parameter of chain termination, s−1 (Grof et al., 2003; Thierry and Keurentjes, 2005). However, detailed
keff,q thermal coefficient of q phase, W m−1 K−1 modeling of multi-scale phenomena remains to be a very challeng-
kP kinetic parameter of chain propagation, ing task. Modeling multi-scale reactor needs to take into account
m3 kmol−1 s−1 not only complex gas–solid two-phase flow and particle–particle
kq turbulence kinetic energy tensor of q phase and particle–reactor interactions, but also nanoscale phenom-
ktrM kinetic parameter of chain transfer to monomer, ena such as the chemistry and kinetics of the active sites of
propylene, m3 kmol−1 s−1 the catalyst and molecular transport and collision within parti-
ktrH kinetic parameter of chain transfer to hydrogen, cles (Kiparissides et al., 2010). A multi-scale product engineering
m3 kmol−1 s−1 approach provides a potential solution for addressing the above-
ktrAl kinetic parameter of chain transfer to aluminum mentioned complexities and establishing the relationship between
alkyl, m3 kmol−1 s−1 the process operating condition and the multi-scale polymer
ks coefficient for granular energy property.
Kgs interphase exchange coefficient of momentum, Several contributions have been identified to address multi-
kg m−3 s−1 scale aspects of the olefin polymerization process. Dompazis
(L) internal coordinate, particle diameter et al. (2008) developed a comprehensive multi-scale, multi-phase,
ṁgs the mass transfer from gas phase to solid phase multi-compartment dynamic model to analyze the extent of par-
[M] propylene concentration, kmol m−3 ticle segregation in catalytic olefin polymerization FBRs. In their
M̄n number-averaged PP molecular weight work, the multi-scale description of the FBRs utilized “linking”
M̄w weight-averaged PP molecular weight models at four different length and time scales, namely, a kinetic
Mw propylene molecular weight, kg kmol−1 model, a single particle model, a PBM and a multi-phase, multi-
MFI melt index zone mixing model. Unfortunately, four models at individual scales
n(L ; x, t) the number density function have not been coupled together in their efforts. Recently, Chen et al.
p gas pressure, Pa (2011a) and Yan et al. (2012) in Luo’s group proposed a multi-scale
ps particle phase pressure, Pa polymerization FBR model, i.e., a CFD model coupled with a PBM-
[Pi ] the ith dead polymer concentration Polymeric Multilayer Model (PMLM) (CFD-PBM-PMLM), to describe
[Pi∗ ] the ith live polymer concentration the multi-scale flow, transport and reaction behaviors in fluidized
R ideal gas constant, kJ kmol−1 K−1 bed polymerization reactors. The PMLM was applied to describe the
Rd chain termination rate, kg m−3 s−1 growth of a single particle, and it was solved together with a PBM to
RP chain propagation rate, kg m−3 s−1 predict the dynamic evolution of PSD. Furthermore, the CFD-PBM-
Rtr chain transfer rate, kg m−3 s−1 PMLM integrated model was implemented to describe the gas-solid
Rε,q addition term in ε equation of q phase flow field in FBRs. However, the multi-scale PP product properties
T temperature, K were not included in Luo et al.’s efforts (Chen et al., 2011a; Yan et al.,
vg the gas phase velocity, m s−1 2012). In summary, the above listed investigations can be catego-
vs solid phase velocity, m s−1 rized as chemical process engineering efforts rather than chemical
˛g the volume fraction of gas phase product engineering efforts in propylene polymerization (Soares,
2001).
 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51 41

Table 1
The reaction steps of propylene polymerization.

kiM
1, chain initiation: C ∗ + M −→P1∗
kp
2, Chain propagation: Pi∗ + M −→Pi+1
∗

3, Chain transfer
ktrM
Transfer to monomer: Pi∗ + M −→Pi + C ∗
ktrH
Transfer to hydrogen: Pi∗ + H2 −→Pi + C ∗
ktrAl
Transfer to aluminum alkyl: Pi∗ + AlR3 −→Pi + C ∗
4, Chain termination
kd
Self abrogation: Pi∗ −→Pi + Cd
kd
Catalyst deactivation: C ∗ −→Cd

Zacca et al., 1996; Soga and Shiono, 1997; Khare et al., 2004; Luo
et al., 2007, 2009; Zheng et al., 2011):propagation rate:

RP = kP [M][C], (1)

transfer rate:

Rtr = ktrM [M][C] + ktrH [H2 ]0.5 [C] + ktrAl [AlR3 ][C], (2)

catalyst deactivation rate:

Rd = kd [C]. (3)

Arrhenius equation:
Fig. 1. The configuration and initial patch height of FBR selected in this work.
 E

k = k0 exp − . (4)
RT
In this work, it is our objective to develop a new chemical prod-
uct engineering technology that can minimize experimental costs
in the absence of complete data. A case study to design the high- 3. Multi-scale PP property model and its solving method
performance PP product with added values is performed based on
a multi-scale modeling technology. Since the multi-scale model The propylene polymerization in a FBR is a typical multi-scale
can be used to predict the relationship between reactor operating reaction system (Grof et al., 2003; Thierry and Keurentjes, 2005;
conditions and multi-scale PP properties, some new PP products Kiparissides et al., 2010), where the reactor operating conditions
with excellent product properties and their corresponding pro- (e.g., the feed material concentration and the feed catalyst particle
cessing conditions can be obtained in advance for guiding their size) can directly influence the gas–solid mixing in the FBR at the
industrial/experimental development. macro-scale. The dynamic gas–solid flow field in turn influences
the gas and polymer particle fractions in the FBR. Furthermore, the
macro-scale flow pattern determines the micro-scale polymeriza-
2. Propylene polymerization FBR and kinetics
tion kinetics and the solid particle kinetics. As a result, the FBR
operating condition can influence the multi-scale polymer product
As described above, we have conducted a multi-scale CFD study
property, and modeling this effect can construct the relationship
of the gas–solid flow field in a two-dimensional (2D) FBR (Yan et al.,
between the process operation conditions (i.e., the multi-scale
2012) and developed a CFD-PBM-PMLM model (Chen et al., 2011a)
process behaviors) and the product property (i.e., the product
to describe the multi-scale flow, transport and reaction behaviors
engineering). The gap between individual molecules and the final
in the polymerization FBR from the chemical process engineering
product can be bridged accordingly, which combines chemical pro-
viewpoint. The same FBR is adopted in the current study (Fig. 1),
cess and product engineering (Villadsen, 1997; Coppens, 2005;
which has an inner diameter of 0.33 m, a height of 0.90 m, and an
O’Connell et al., 2009; Gernaey and Gani, 2010).
initial bed height of 0.2 m.
For propylene polymerization in an FBR, a novel multi-scale PP
As the first step towards the multi-scale product model, a sim-
property model based on an Eulerian–Eulerian two-fluid model,
ple kinetic model is employed to describe the kinetics of propylene
incorporating the method of moments and the PBM was devel-
polymerization on the Ziegler–Natta catalyst (Zacca and Ray, 1993;
oped to bridge the gap between the operating conditions and
Zacca et al., 1996; Khare et al., 2004; Luo et al., 2007, 2009; Zheng
the product properties. The temporal–spatial evolution of the
et al., 2011). The polymerization scheme comprises a series of ele-
gas–solid flow field, polymer particle morphological property and
mentary reactions, namely, site activation, chain initiation, chain
molecular property were calculated simultaneously. The extended
propagation, chain transfer and site transformation and catalyst
Eulerian–Eulerian two-fluid CFD model, PBM, method of moments
deactivation reactions (Zacca and Ray, 1993; Zacca et al., 1996;
as well as the multi-scale coupling approach will be described
Khare et al., 2004; Luo et al., 2007, 2009; Zheng et al., 2011).
below.
Among them, the propagation reaction, chain transfer and cat-
alyst deactivation reaction play important roles in determining
the polymerization kinetic behaviors related to the polymer prod- 3.1. Micro-scale PP property modeling: the method of moments
uct micro-scale and particle property. By using the simple kinetic
model (see Table 1) and the pseudo-kinetic rate constant method, Based on the polymerization kinetics described in Section 2, the
the following equations can be obtained (Zacca and Ray, 1993; following assumptions are made:
42 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51

(1) all catalyst active sites have been activated simultaneously, d1
(2) all the reaction rate constants are independent on the chain + ∇ · (0 vs ) = (ktrM [M] + ktrH [H2 ]0.5 + ktrA [AlR3 ] + kd )1 ,
dt
length, i.e., the equal activity assumption. (13)
(3) Transfer resistance between gas-solid two phases and diffusion
can be ignored,
d2
(4) AlR3 and the produced active and dead polymers are considered + ∇ · (0 vs ) = (ktrM [M] + ktrH [H2 ]0.5 + ktrA [AlR3 ] + kd )2 .
dt
to be adsorbed on the catalyst surface.
(14)

Therefore, according to the reaction steps shown in Table 1,

Similar equations can be derived for monomer, active site of the
the mass balance equations can be obtained (Zacca and Ray, 1993;
catalyst, hydrogen and AlR3 :
Zacca et al., 1996; Soga and Shiono, 1997; Khare et al., 2004; Luo
et al., 2007, 2009; Zheng et al., 2011). d[M]
+ ∇ · ([M]vs )
For the active chain with length i = 1, we have: dt

d[P1∗ ] = ∇ (Dim ∇ ([M])) − kiM [M][C ∗ ] − (kp + ktrM )[M]0 , (15)

+ ∇ · ([P1∗ ]vs )
dt
= kiM [C ∗ ][M] − kp [P1∗ ][M] d[C ∗ ]
+ ∇ · (C ∗ vs ) = (ktrM [M] + ktrH [H2 ]0.5 + ktrA [AlR3 ])0
dt
− (ktrM [M] + ktrH [H2 ]0.5 + ktrA [AlR3 ] + kd )[P1∗ ], (5)
− kiM [M][C ∗ ] − kd [C ∗ ], (16)
for the active chains with length i ≥ 2, a slightly different equation
can be used:
d[H2 ]
d[Pi∗ ] + ∇ · ([H2 ]vs ) = ∇ (Dim ∇ ([H2 ])) − ktrH [H2 ]0.5 0 , (17)
+ ∇ · ([Pi∗ ]vs ) dt
dt
d[AlR3 ]
∗
= kp [M][Pi−1 ] + ∇ · ([AlR3 ]vs ) = −ktrA [AlR3 ]0 . (18)
dt
− (ktrM [M] + ktrH [H2 ]0.5 + ktrA [AlR3 ] + kp [M] + kd )[Pi∗ ]. (6) The PP density is about 910 kg/m3 and the gas propylene density
can be calculated according to the relationship between temper-
ature and pressure (Luo et al., 2010). Accordingly, the following
A similar equation can be obtained for the dead polymers: equations can be used to calculate the PP micro-scale property
d[Pi ] parameters based on the method of moments.
+ ∇ · ([Pi ]vs ) = (ktrM [M] + ktrH [H2 ]0.5 + ktrA [AlR3 ] + kd )[Pi∗ ].
dt 1 + 1 1
(7) M̄n = Mw = Mw , (19)
0 + 0 0
2 + 2 2
M̄w = Mw = Mw , (20)
The above equations are a series of balance equations based on 1 + 1 1
the chain length. In order to solve them, the method of moments
lg MFI = 18.56 − 3.36 lg Mw . (21)
is applied (Zhu, 1999a,b; Wang et al., 2006). The corresponding
moments are defined as follows: In addition, we also point out that the temporal–spatial evolu-
∞ ∞ 
∞ tion of the gas–solid flow field, especially the species concentration
m = im [Pi ], m = im [Pi∗ ] ≈ im [Pi∗ ]. (8) field, must be obtained in advance when solving the above moment
i=1 i=1 i=2
equations. It means that the CFD model must be added into the
multi-scale product property model (see Section 3.3).
Therefore, we get:
d0 3.2. Meso-scale PP morphological property modeling: the PBM
+ ∇ · (0 vs ) = kiM [C ∗ ][M] − (ktrM [M] + ktrH [H2 ]0.5
dt
During the propylene polymerization, the polymer particle
+ ktrA [AlR3 ] + kd )0 , (9)
size increases due to the polymerization reaction. Meanwhile, the
particle formation kinetics (i.e., single particle growth due to poly-
d1 merization, particle aggregation and breakage dynamics) leads to
+ ∇ · (1 vs ) the change of the PP morphological property/PSD. In turn, the drag
dt
force acting on the solid phase can be influenced due to the change
= kiM [C ∗ ][M] + kp [M]0 of the PSD. Therefore, the PBM for PSD quantification must also
− (ktrM [M] + ktrH [H2 ]0.5 + ktrA [AlR3 ] + kd )1 , (10) be added into the multi-scale product property model. In addi-
tion, there are still many arguments on the particle aggregation and
breakage dynamics, herein, only the particle growth kernel is con-
d2 sidered in the PBM due to the excellent dispersion characteristics
+ ∇ · (2 vs ) of the FBR (Gidaspow, 1994).
dt
Based on Hulburt et al.’s work (1964), a general form of the
= kiM [C ∗ ][M] + kp [M](0 + 21 )
population balance equation is given as follows:
− (ktrM [M] + ktrH [H2 ]0.5 + ktrA [AlR3 ] + kd )2 , (11) ∂n(L; x, t) →
 
+ ∇ · vs n(L; x, t)
∂t
d0 ∂
+ ∇ · (0 vs ) = (ktrM [M] + ktrH [H2 ]0.5 + ktrA [AlR3 ] + kd )0 , =− [G(L)n(L; x, t)]
dt ∂L
(12) + Bag (L; x, t) − Dag (L; x, t) + Bbr (L; x, t) − Dbr (L; x, t), (22)
 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51 43

where, n(L ; x, t) is the number density function with particle diam-

eter (L) as the internal coordinate, G(L)n(L ; x, t) is the particle flux
due to molecular growth rate, Bag (L ; x, t) and Dag (L ; x, t) are the
birth and death rates of particles diameter (L) due to aggregation,
respectively, and Bbr (L ; x, t) and Dbr (L ; x, t) are the birth and death
rate of particles diameter (L) due to breakage, respectively. In Eq.
(22), the first term on the left hand side is the transient term, the
second term is the convective term, and the terms on the right hand
side are the source terms describing particle growth, aggregation,
and breakage dynamics, respectively. When only considering the
particle growth kernel, one can obtain the following PBM:

∂n(L; x, t) ∂
+ ∇ · [vs n(L; x, t)] = − [G(L)n(L; x, t)]. (23)
∂t ∂L
Solving Eq. (23), one can obtain the PP morphological prop-
erty parameter and, it will be fed into the reactor CFD model (see
Sections 3.3 and 3.4) to obtain the macro-scale product property Fig. 2. The coupling mechanism of the multi-scale product property model.
parameter.

2

3.3. Macro-scale PP property modeling: the CFD model T ¯
¯¯ s = ˛s s (∇ · vs + ∇ · vs ) + ˛s s − s ∇ · vs · Ī. (31)
3
The PP product coexists with the gas mixture in the FBR and its The energy balance equations for gas and solid phases can be
macro-scale property can be described via the solid fraction and written as:
the species concentrations, which can be predicted via the CFD ∂
model based on the Eulerian–Eulerian approach. Accordingly, the (˛q q hq ) + ∇ · (˛q q vq hq )
∂t
two-fluid CFD model is used to model the macro-scale PP product

property. = ˛q ∇ (keff,q ∇ T ) − ∇ · q
 q + ¯¯ q : ∇ vq + Sq + Ksq (Ts − Tq ), (32)
In this work, the CFD model is an extension of the two-phase
model for gas–solid flows for describing the gas–solid flow field
∂
in the FBR. Since this model has been widely applied in gas–solid (˛s s cps Ts ) + ∇ · (˛s s cps −
→
vs Ts ) = ˛s ∇ (keff,s ∇ T ) + Kqs (Tq − Ts ),
reactors, it has been well documented in the literature. Due to the ∂t
limited space, the CFD model is not presented here (see Gunn, 1978; (33)
Lun et al., 1984; Ding and Gidspow, 1990; Syamlal et al., 1993; van where,
Wachem et al., 2001; Gao et al., 2009; Wang et al., 2010; Chen et al.,  
n 
n
2011a,b; Sun et al., 2011; Peng et al., 2012 for details). Herein, only hq = ( Yi Cpi )dTq + h0i , (34)
the main governing equations are listed.
i=1 i=1
The continuity equations for phase q (q = g for gas, s for solid
phases, respectively) can be written as 
n
q =
q hiji , (35)
∂
(˛g g ) + ∇ · (˛g g vg ) = −ṁgs , (24) i=1
∂t 
Sq = ˛q
i · Hi . (36)
∂
(˛s s ) + ∇ · (˛s s vs ) = ṁgs , (25)
∂t
3.4. Multi-scale coupling mechanism and solving procedure
and,
As described above, a multi-scale product property model
˛q + ˛s = 1. (26)
that integrates the moment equations, PBM and CFD model was
The mass transfer from the gas phase to the solid phase can be employed to describe the multi-scale PP property in polymerization
calculated as FBRs. The micro-scale, meso-scale and macro-scale PP properties
were described via the moment equations (i.e., Eqs. (5)–(21)), PBM
1
ṁgs = 2
s Gm2 . (27) (i.e., Eqs. (22) and (23)) and CFD model (i.e., Eqs. (24)–(36)), respec-
The momentum balance equations for gas and solid phases can tively. Furthermore, the moment equations were solved together
be expressed as: with the PBM to implement the coupling of micro- and meso-scale
models. Finally, an iterative multi-scale modeling approach was
∂ used to implement the combination of the CFD model and the cou-
(˛g g vg ) + ∇ · (˛g g vg · vg )
∂t pled model of the moment equations and the PBM (see Fig. 2).
→ More detailed coupling schemes developed in this work have
= −˛g ∇ p + ∇ · g + Kgs (vs − vg ) − ṁgs vg + ˛g g g, (28) been illustrated in Fig. 3. There are strong and complex interactions
between elements/components at different scales and information
is transferred among different scales. An iterative multi-scale prod-
T
¯¯ g = ˛g g (∇ · vg + ∇ · vg ), (29) uct property model, which couples models at various scales to allow
the interaction between scales iteratively or (should we use “and”
∂ rather than “or”?) simultaneously, is thus constructed in this work
(˛s s vs ) + ∇ · (˛s s vs · vs ) (Bi and Li, 2004).
∂t
The pressure, temperature, species mass fraction distributions
= −˛s ∇ p − ∇ ps + ∇ · s + Kgs (vq − vs ) + ṁgs vs + ˛s s g, (30) in the FBR at initial stage are specified in the initiation step. The
44 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51

Fig. 3. The calculation procedure of the coupling model.

CFD model is then solved to obtain the macro-scale flow field. Table 3
The kinetic parameters of polymerization reactions.
After that, the information of the PP particle position and velocity
is transferred from the macro-scale CFD model to the meso-scale Description Value
PBM model. Accordingly, the PBM model can be solved to obtain Chain initiation 7.04 × 102 m3 /kmol s
new PSD data, which are transferred back to the macro-scale CFD Chain propagation 4.97 × 107 m3 /kmol s
model. On the other hand, the information of the pressure, temper- Transfer to monomer 6.16 × 103 m3 /kmol s
ature, gas and solid densities and species mass fractions obtained Transfer to hydrogen 4.4 × 106 m3 /kmol s
Transfer to aluminum alkyl 3.85 × 102 m3 /kmol s
via solving the CFD model, are incorporated into the solution of
Chain termination 7.92 × 103 1/s
the moment model. Meanwhile, the species rate data are obtained Active energy 50.23 kJ/mol
via the moment model and are in turn transferred from the micro-
scale to the macro-scale. Based on the meso-scale and micro-scale
Table 4
product data, the next iterative solution of the CFD model can be
The moments used in multi-scale PP property model.
obtained. Finally, the PBM and the moment model can be coupled
via the transfer of the PSD data from the meso-scale to the micro- Moments of catalyst diameter (m)
M0 5.105 × 108 M2 2.042 × 101
scale and the particle growth rate/polymerization rate data from
M1 1.021 × 105 M3 4.084 × 10−3
the micro-scale to the meso-scale. Therefore, three models are cou- Moments of polymers (live polymer and dead polymer, kmol m−3 )
pled with each other by the above information exchange scheme. 0 0.01 0 0.005
The solution program is executed in a loop with the above solution 1 2.857 × 101 1 1.429 × 101
steps until the reaction time meets a given criterion. 2 8.163 × 104 2 4.082 × 104

3.5. Simulation conditions and modeling method USA) in the double precision mode. In addition, to simulate the 2D
reactor, a commercial grid-generation tool, GAMBIT 2.3.16 (Ansys
Some model parameters reported in our previous works (Luo Inc., USA) was used to generate the 2D geometries of the reactor
et al., 2007, 2009; Chen et al., 2011a,b; Yan et al., 2012) are adopted and the computational grids. The governing equations in the CFD
in this work (see Tables 2–4). Other parameter values in the multi- model were discretized into a uniform structural mesh by a finite
scale product property model are listed in Tables 2–4. Unless volume method. All terms in the CFD model were discretized using
otherwise noted, the parameters used for the forthcoming simu- a first order upwind method. Pressure and velocity are coupled by
lations are those in Tables 2–4 as well. the SIMPLE algorithm. And the equations and source terms of the
The 2D simulations based on the multi-scale model were per- PBM and the moment equations were incorporated into the CFD
formed with the commercial CFD code FLUENT 6.3.26 (Ansys Inc., model via external user defined functions (UDFs). For the CFD sim-
ulations, the sub-relaxation iteration method was used to ensure
Table 2 the simulations to converge and the simulations were executed in
The boundary and initial conditions of FBR. a 2.83 GHz Pentium 4 CPU desktop with 4 GB of RAM.
Descriptions Values
4. Results and discussion
Catalyst diameter 2.0 × 10−4 m
Mole fraction of hydrogen 2000 ppm
Feed rate of catalyst 0.025 kmol/m3 4.1. Validation of the multi-scale model
Feed rate of aluminum alkyl 8.696 × 10−3 kmol/m3
Reaction heat 85,830 J/mol All the Eulerian–Eulerian model, PBM and the method of
Polymer density 910 kg/m3
moments adopted in this work are standard and have been vali-
Weight-average molecular weight 1.2 × 105
dated for the gas–solid catalytic polymerization process in the FBR
 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51 45

Fig. 4. The solid volume fraction change with time in the FBR.

in our previous studies (Luo et al., 2007; Chen et al., 2011a; Yan 4.2. The multi-scale PP property distributions
et al., 2012). They plus the model girds were all validated, respec-
tively (Luo et al., 2007; Chen et al., 2011a; Yan et al., 2012). The The detailed dynamics of the field distribution and the PP prop-
indeterminacy of this coupled model is the coupling mechanism erty can be obtained via solving the multi-scale PP property model
among the three classic models. In addition, as an important param- with the selected parameters. Fig. 4 shows the solid volume fraction
eter for describing the gas and solid interaction in a FBR, pressure evolution in the FBR. From Fig. 4, the solid bed expands and surges
drop can be used to identify the effectiveness of model parameters. upward firstly, and falls after a peak, finally it mixes uniformly and
Herein, we validated the multi-scale model by the verification of reaches a stable state at about 9 s. The average solid volume fraction
pressure drop in FBR firstly. An empirical correlation can be used is about 0.35.
to obtain the pressure drop, which is shown as follows: Figs. 5 and 6 illustrate the dynamic distributions of propylene
and hydrogen mass fractions, respectively. From Fig. 5, the whole
FBR has the same propylene mass fraction distribution in the
P = (1 − ˛g )hmf (s − g )g. (37) initial stage. As the polymerization reactions proceed, propylene
and hydrogen are consumed and PP is generated. This changes the
distribution of propylene mass fraction until the solid distribution
The pressure drop calculated by Eq. (37) is 1097.7 Pa, while the
and flows of gas and solid phases reach a stable state (i.e., 9 s).
pressure drop obtained by the multi-scale PP property model is
The propylene mass fraction increases with the reactor height,
1056.5 Pa. The relative error between two values is less than 3.75%,
this may be because the feed of hydrogen is limited and is quickly
which indicates that the model parameters applied in this work are
consumed to a relatively low level, but propylene is continuously
effective.
polymerized to form PP, and also the catalyst concentration is the
The experiment data collected from the FBR in the Research
highest at the FBR bottom, so the propylene mass fraction reaches
Institute of Petroleum and Chemical Industry (Lanzhou) of China
a minimum at the FBR bottom, and increases slightly upward
National Petroleum Corporation (CNPC) (see Shi et al.’s work
due to the decreases of hydrogen mass fraction. In addition, one
(2010)) are used to validate the multi-scale model. Herein, we must
point out that the FBR shown in Shi et al.’s work (2010), which
is an agitated FBR and consists of a vertical 0.5-m-i.d. cylinder of
Table 5
height 1.5 m, is different from that described in Fig. 1. Table 5 gives
Comparison between the experiment data and the simulation data.
the comparisons between the experimental data and simulation
data (the multi-scale model). From Table 5, all the relative errors Experiment data Simulation data
at the reactor at the reactor
between experimental and simulation results for the multi-scale PP
outlet outlet
property at the FBR outlet are less than 10%, which demonstrates
Macro-scale Propylene mass 0.89 0.93
that the simulated data using the multi-scale model are in good
fraction (wt/wt)
agreement with the experiment data. Meso-scale Mean particle 0.311 0.285
On the basis of these results, the selected grid and multi-scale diameter (mm)
model are sufficiently effective to simulate the multi-scale PP prop- Micro-scale Mw (g/mol) 1,03,000 1,05,268
erty in the FBR. It is expected that this validated model can be MFI (g/min) 44.8 48.8

used to predict the multi-scale product property that are not easily The obtained data are at steady-state conditions and the simulated FBR and operat-
obtainable from experiments. ing conditions are from our previous works (Shi et al., 2010; Zheng et al., 2011).
46 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51

Fig. 5. The distribution change of propylene mass fraction with time in the FBR.

can also identify the back flow at 25 s in Fig. 5. Fig. 6 shows that low, so the catalysts hold the same activity in a long time and their
the distribution of hydrogen mass fraction reaches stable at 9 s concentration can be only affected by the solid distribution.
and decreases along the reactor height. It is clear that the sum of Fig. 8 gives PP Mw dynamics. In the starting stage, PP parti-
hydrogen and propylene mass fraction is 1. cles with high Mw are generated at the bed bottom. Then particles
Fig. 7 describes the distribution of catalyst concentration as a with different Mws mix evenly with gas and solid flows. The Mw
function of time. The distribution of catalyst concentration keeps of polymer particles in FBR keep at a stable growth rate, but it
consistent with solid volume fraction and reaches a stable state is relatively slow due to decentralized catalyst particles and low
at 5 s. In fact, the catalysts are covered by solid particles (please propylene and hydrogen concentrations. As hydrogen is consumed
make sure it is correct), and the catalyst deactivation rate is very and it decreases along the bed height, the polymerization rate at

Fig. 6. The distribution change of hydrogen mass fraction with time in the FBR.
 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51 47

Fig. 7. The distribution change of catalyst concentration with time in the FBR.

the surface of solid bed is high, which forms a slightly higher Mw. a drastic change of MFI, as Mw increases, the MFI of PP reduces
The dynamics of PP MFI distributions is also simulated here as gradually.
shown in Fig. 9. The MFI is calculated via Eq. (21). From Fig. 9, the As the gas phase polymerization rate is small, the growth rate of
MFI distribution is not uniform. These polymer particles with high PP particle diameter is also slow. So the change of the PP PSD with
MFIs aggregate together due to different Mw s caused by different time is difficult to be presented using contours, while the change of
local polymerization rates in the FBR. A slight change of Mw causes the average particle diameter with time is illustrated in line pictures

Fig. 8. The distribution change of polypropylene molecular weight with time in the FBR.
48 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51

Fig. 9. The distribution change of polypropylene melt index with time in the FBR.

in Fig. 10. There is an abnormal change during 0–5 s, which is due property (e.g., PSD) changes in this time scope has no additional
to the uneven flow field and the mix of particles in the initial stage. effect on reactor behavior.
Accordingly, the average particle diameter increases at a constant
rate, as reflected by its linear relationship with residence time (see
4.3. The effect of process operating conditions
Fig. 10).
As the distributions of PP Mw and MFI are relatively even in the
Generally hydrogen is used as molecular weight regulator to
FBR, and in order to show their changes clearly in macro tempo-
control the polymer Mw in the industry production, and hence the
ral level, the average values of those properties, which are of our
polymer property, such as PSD, Mw and MFI. This is because the
major concern, are given in Fig. 11. In Fig. 11, Mw increases and MFI
hydrogen concentration can influence the active site concentration
reduces as the polymerization proceeds. Both their changes are a
of catalyst by changing the chain transfer rate. In this paper we
linear relationship with time, which may due to that the reactor is
halve and double hydrogen concentration on the basis of main cases
small and the contact time between gas and catalyst is short, the

Fig. 11. The average molecular weight and melt index changes of polypropylene
Fig. 10. The average polypropylene diameter change with time in the FBR. with time in the FBR.
 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51 49

Fig. 12. The average particle diameter change with time under different feed hydro-
gen concentrations in the FBR. Fig. 14. The normalized average particle diameter changes with time under differ-
ent initial feed particle sizes in the FBR.

(e.g., 1000, 2000 and 3000 PPM), to study the influence of hydrogen
on product property.
Fig. 12 shows the change of average particle diameter with time
under different feed hydrogen concentrations. From Fig. 12, the the product diameter. Herein, we also study the effect of feed par-
average particle diameter increases linearly with time. As the feed ticle size by using different initial particle diameters (e.g., 0.15, 0.20
hydrogen concentration increases, the growth rate of average par- and 0.25 mm).
ticle diameter reduces slightly. This indicates that more active sites Fig. 14 shows the normalized average particle diameter under
have been transferred to hydrogen, but it is not obvious due to different initial feed particle sizes. From Fig. 14, the feed particles
the low transfer rate and low-level hydrogen concentration. On with smaller initial diameter have a faster diameter growth rate.
the other hand, the raw materials stayed in FBR for only about 9 s This is because that small-sized particle produces small surface
and they contact with catalyst for about 4 s, which may also con- area, as the volume of polymers gained per unit time is identical
tribute to the un-obvious change. Fig. 13 shows how the PP Mw for each particle (due to the same catalyst kernel), the diameter of
and MFI changes with time under different feed hydrogen concen- small-sized particles increases larger diameter growth rate. Fig. 14
trations. The Mw increases and MFI reduces with time, which is also illustrates that small particles grow quickly and their diame-
because the additional hydrogen reduces the polymerization rate. ters are difficult to control, while large particles grow slowly, which
From Fig. 13, both the Mw growth rate and the MFI reduction rate means it would take a long time to get big particles. Fig. 15 gives
decrease with the increase of the hydrogen concentration. It indi- the PP Mw and MFI evolution under different initial feed particle
cates that hydrogen does have a contribution to product property sizes. As the feed particle diameter increases, the growth rate of
control, but there is a saturation concentration, above which the Mw decreases, and the decrease rate of MFI slowed down, too.
hydrogen concentration has a negligible affect. As the feed particles diameter increases, the fluidized bed height
On the other hand, for propylene gas polymerization technol- decreases, which means the reaction time is shortened, at the same
ogy, both the PP Mw and MFI are important product quality index, time the increase of catalyst concentrate is insufficient to cover the
and the average diameter is another one. Our previous work has shorter reaction time, the result is that single conversion ratio is
proved that the feed particle diameter has a critical influence on reduced, and so the growth rate of Mw decreases.

Fig. 13. The average of polypropylene molecular weight and melt index changes with time under different feed hydrogen concentrations in the FBR.
50 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51

Fig. 15. The average polypropylene molecular weight and melt index changes with time under different initial feed particle sizes in the FBR.

Fig. 16. The multi-scale software interface for developing PP product in FBRs.

4.4. Application the multi-scale model to new PP product Table 6

The main performance index of the new PP product.
development
Item Data
Based on the above multi-scale PP property model, we have MFI (g/min.) 44.8
designed and developed a set of software, which is used to generate Tensile yield strength (MPa) ≥20
the relationship between PP properties and process operating con- Bending elastic modulus (MPa) ≥1200
ditions in FBRs (see Fig. 16 for its interface). Using this software, Izod notched impact strength (kJ/m2 , 23 ◦ C) ≥40

we can develop new PP product with less experimental efforts.

For instance, a new PP product named MF40 used for automobile
with superior flow performance was developed and manufactured 5. Conclusions
in a pilot-scale FBR. The new multi-scale product properties and
process operating conditions are listed in Table 5. In this work, we first developed a multi-scale product model
After the test, we got the main performance index of the new to describe the multi-scale PP product property in a catalytic
PP product (see Table 6). For instance, the new product with propylene polymerization FBR. The multi-scale product model con-
high flow performance not only has high bending elastic mod- sists of a two-phase CFD model, polymer moment equations and
ulus (≥1200 MPa) but also high Izod notched impact strength a PBM. Next, the multi-scale model was validated by empirical
(≥40 kJ/m2 at 23 ◦ C). It means that the multi-scale product property equations and industrial data. Based on the verified model, the
model can be used to guide the development of a new homo- multi-scale product properties in a lab-scale FBR were predicted
PP product at a pilot-scale FBR with less experimental efforts. and the effects of two important fluidized bed operating condi-
Furthermore, the new homo-PP product has the similar excellent- tions (i.e., feed hydrogen concentration and feed catalyst particle
performance of a co-PP product titled SP179 in the market (Hong, size) on the multi-scale polymerization characteristics were ana-
2011). lyzed to establish the relationship between operating conditions
 Y.-P. Zhu et al. / Computers and Chemical Engineering 71 (2014) 39–51 51

and multi-scale polymer properties. Therefore, a typical chemical Hill M. Product and process design for structured products. AIChE J
process engineering effort was related with the chemical product 2004;50:1656–62.
Hill M. Chemical product engineering-the third paradigm. Comput Chem Eng
engineering. Finally, a case study to design the high-performance 2009;33:947–53.
PP product with added values was performed based on the obtained Hong DY. Polypropylene: principle, process and technology. second ed. Beijing:
relationship between process operating conditions and multi-scale China Petrochemical Press; 2011.
Jaworski Z, Zakrzewska B. Towards multiscale modeling in product engineering.
polymer properties. Comput Chem Eng 2011;35:434–45.
In summary, this work preliminarily demonstrated that a multi- Khare NP, Luca B, Seavey KC, Liu YA. Steady-state and dynamic modeling of gas-phase
scale product property model could be used to guide the integration PP processes using stirred-bed reactors. Ind Eng Chem Res 2004;43:884–900.
Kiparissides C, Krallis A, Meimaroglou D, Pladis P, Baltsas A. From molecular to plant-
of process engineering efforts with product engineering efforts in
scale modeling of polymerization processes: a digital high-pressure low-density
the broad chemical engineering field. polyethylene production paradigm. Chem Eng Technol 2010;33:1754–66.
Lun CKK, Savage SB, Jeffrey DJ, Chepurniy N. Kinetic theories for granular flow-
inelastic particles in Couette-flow and slightly inelastic particles in a general
Acknowledgments
flow field. J Fluid Mech 1984;140:223–32.
Luo ZH, Zheng Y, Cao ZK, Wen SH. Mathematical modeling of the molecular
The authors thank the National Ministry of Science and Technol- weight distribution of polypropylene produced in a loop reactor. Polym Eng
Sci 2007;47:1643–9.
ogy of China (No. 2012CB21500402), the National 863 plan project
Luo ZH, Su PL, Shi DP, Zheng ZW. Steady-state and dynamic modeling of commercial
of China (No. 2013AA032302), the Research Fund for the Doctoral bulk polypropylene process of Hypol technology. Chem Eng J 2009;149:370–82.
Program of Higher Education (No. 20130073110077), the State- Luo ZH, Su PL, Wu W. Industrial loop reactor for catalytic propylene poly-
Key Laboratory of Chemical Engineering of Tsinghua University merization: dynamic modeling at emergency accidents. Ind Eng Chem Res
2010;49:11232–43.
(No. SKL-ChE-10A03) and China National Petroleum Corporation O’Connell JP, Gani R, Mathias PM, Maurer G, Olson JD, Crafts PA. Thermodynamic
for supporting this work. property modeling for chemical process and product engineering: some per-
spectives. Ind Eng Chem Res 2009;48:4619–37.
Peng B, Zhu J, Zhang C. A new approach to specify the inlet boundary conditions
References for computation fluid dynamics (CFD) modeling of hydrodynamic behavior of a
circulating fluidized bed (CFB) riser. Ind Eng Chem Res 2012;51:2152–65.
Bi HT, Li JH. Multiscale analysis and modeling of multiphase chemical reactors. Adv Syamlal M, Rogers W, O’Brien TJ. MFIX documentation: volume 1, Theory guide.
Powder Technol 2004;15:607–27. Springfield: National Technical Information Service; 1993.
Cussler EL, Wei J. Chemical product engineering. AIChE J 2003;49:1072–5. Soga K, Shiono T. Ziegler–Natta catalysts for olefin polymerizations. Prog Polym Sci
Cagan J, Vogel C. Creating breakthrough products. Upper Saddle River, NJ: Prentice 1997;22:1503–46.
Hall; 2002. Soares JBP. Mathematical modelling of the microstructure of polyolefins made by
Charpentier JC. The triplet “molecular processes-product-process” engineering: the coordination polymerization: a review. Chem Eng Sci 2001;56:4131–53.
future of chemical engineering? Chem Eng Sci 2002;57:4667–90. Shi DP, Luo ZH, Guo AY. Numerical simulation of the gas–solid flow in fluidized-bed
Coppens MO. Scaling-up and -down in a nature-inspired way. Ind Eng Chem Res polymerization reactors. Ind Eng Chem Res 2010;49:4070–9.
2005;44:5011–9. Sun JY, Zhou YF, Ren CJ, Wang JD, Yang YR. CFD simulation and experiments of
Costa R, Moggridge GD, Saraiva PM. Chemical product engineering: an emerging dynamic parameters in gas–solid fluidized bed. Chem Eng Sci 2011;66:4972–82.
paradigm within chemical engineering. AIChE J 2006;52:1976–86. Thierry M, Keurentjes J. Handbook of polymer reaction engineering. Weinheim:
Chen XZ, Luo ZH, Yan WC, Lu YH, Ng IS. Three-dimensional CFD-PBM coupled Wiley-VCH; 2005.
model of the temperature fields in fluidized bed polymerization reactors. AIChE Ulrich K, Eppinger S. Product design and development. third ed. New York: McGraw-
J 2011a;57:3351–66. Hill; 2003.
Chen XZ, Shi DP, Gao X, Luo ZH. A fundamental CFD study of the gas–solid flow field Villadsen J. Putting structure into chemical engineering. Chem Eng Sci
in fluidized bed polymerization reactors. Powder Technol 2011b;205:276–88. 1997;52:2857–64.
Davis GE. A Handbook of chemical engineering. Manchester, UK: Davis Brothers; van Wachem BGM, Schouten JC, van den Bleek CM, Krishna R, Sinclair JL.
1901. Comparative analysis of CFD models of dense gas–solid systems. AIChE J
Dompazis G, Kanellopoulos V, Toulopides V, Kiparissides C. Development of a 2001;47:1035–51.
multi-scale, multi-phase, multi-zone dynamic model for the prediction of Wesselingh JA. Structuring of products and education of product engineering. Pow-
particle segregation in catalytic olefin polymerization FBRs. Chem Eng Sci der Technol 2001;119:2–8.
2008;63:4735–53. Wang R, Luo Y, Li BG, Sun XY, Zhu SP. Design and control of copolymer composition
Ding J, Gidspow D. A bubbling fluidization model using kinetic-theory of granular distribution in living radical polymerization using semi-batch feeding policies:
flow. AIChE J 1990;36:523–38. a model simulation. Macromol Theory Simul 2006;15:356–68.
Fogler HS. Chemical engineering perspectives for the next decade. In: Presented at Wang JW, van der Hoef MA, Kuipers JAM. CFD study of the minimum bubbling veloc-
the 8th World Congress of Chemical Engineering; 2009. ity of Geldart A particles in gas-fluidized beds. Chem Eng Sci 2010;65:3772–85.
Gunn DJ. Transfer of heat or mass to particles in fixed and fluidized beds. Int J Heat Yan WC, Luo ZH, Lu YH, Chen XD. A CFD-PBM-PMLM integrated model for
Mass Transfer 1978;21:467–75. the gas–solid flow fields in fluidized bed polymerization reactors. AIChE J
Gidaspow D. Multiphase flow and fluidization: continuum and kinetic theory 2012;58:1717–32.
descriptions. Boston: Academic Press; 1994. Zacca JJ, Ray WH. Modelling of the liquid phase polymerization of olefins in loop
Grof Z, Kosek J, Marek M, Adler PM. Modeling of morphogenesis of polyolefin parti- reactors. Chem Eng Sci 1993;48:3743–65.
cles: catalyst fragmentation. AIChE J 2003;49:1002–13. Zacca JJ, Debling JA, Ray WH. Reactor residence time distribution effects on the mul-
Gao JS, Lan XY, Fan YP, Chang J, Wang G, Lu CX, et al. CFD modeling and validation tistage polymerization of olefins – I. Basic principles and illustrative examples,
of the turbulent fluidized bed of FCC particles. AIChE J 2009;55:1680–94. polypropylene. Chem Eng Sci 1996;51:4859–68.
Gernaey KV, Gani R. A model-based systems approach to pharmaceutical product- Zhu SP. Modeling of molecular weight development in atom transfer radical poly-
process design and analysis. Chem Eng Sci 2010;65:5757–69. merization. Macromol Theory Simul 1999a;8:29–37.
Hulburt H, Katz S. Some problems in particle technology: a statistical mechanical Zhu SP. Modeling stable free-radical polymerization. J Polym Sci Polym Phys
formulation. Chem Eng Sci 1964;19:555–74. 1999b;37:2692–704.
Harold MP, Ogunaike BA. Process engineering in the evolving chemical industry. Zheng ZW, Shi DP, Su PL, Luo ZH, Li XJ. Steady-state and dynamic modeling of Basell
AIChE J 2000;46:2123–7. multi-reactor olefin polymerization process. Ind Eng Chem Res 2011;50:322–31.