You are on page 1of 20

Visco-elastic analysis of polymer melts in complex ows

W.M.H. Verbeeten, A.C.B. Bogaerds, G.W.M. Peters, and F.P.T. Baaijens Eindhoven University of Technology, Materials Technology, PO Box 513, NL-5600 MB Eindhoven, The Netherlands. March 1, 1999

Abstract A mixed low-order nite element technique based on the DEVSS/DG method has been developed for the analysis of two and three dimensional visco-elastic ows in the presence of multiple relaxation times. In order to evaluate the predictive capabilities of some nonlinear constitutive relations, results of calculations are compared with experiments for two complex ows. The well-known differential Giesekus and Phan-Thien Tanner model and the recently introduced Feta-VD model are investigated. The latter model provides enhanced independent control of the shear and elongational properties. Experiments and calculations are performed for the steady shear ow around a symmetric conned cylinder and in a cross-slot ow for an LDPE polymer melt. In particular in elongational dominated regions, the numerical / experimental evaluation shows that the multi-mode Giesekus and the PTT models are unable to describe the stress related experimental observations. The Feta-VD model proves to perform signicantly better in these regions. However, a price is paid for this model by an overprediction of the stresses in shear dominated regions.

1 Introduction
Visco-elasticity frequently has a signicant effect on polymer melt ows in industrial applications. Most present research on calculations of steady visco-elastic ows has been performed on 2D benchmark problems like the falling sphere in a tube problem (e.g. Lunsmann et al. [1993], Baaijens [1994], Sun and Tanner [1994], Yurun and Crochet [1995], Baaijens et al. [1997]) or the planar contraction problem (e.g. Yurun and Crochet [1995], Gu nette and Fortin [1995], Azaiez et al. [1996], Baaijens e et al. [1997]). Also periodic ows such as the corrugated tube ow (e.g. Pilitsis and Beris [1991], Van Kemenade and Deville [1994], Talwar and Khomami [1995b], Szady et al. [1995]) and the ow past an array of cylinders (e.g. Talwar and Khomami [1995a], Souvaliotis and Beris [1996]) have been extensively investigated. These ows are generally characterised by steep stress gradients near curved boundaries and geometrical singularities, which require the use of highly rened meshes. Accurate and realistic ow analysis of both polymer solutions and melts compels the use of multiple relaxation times. When using mixed nite element methods this results in a very large number of degrees of freedom. Thus, solution efciency, both in terms of CPU time and memory requirement, is an important issue. Over the last decade, a lot of research has been performed on solving the governing equations in an accurate and stable manner and yet still being able to efciently handle multiple relaxation times. Two basic problems need to be resolved. First, the presence of convective terms in the constitutive equation, which relative importance growths with increasing elasticity, causes problems. Second, the choice of discretisation spaces of the independent variables (velocity, pressure, extra stresses and auxiliary variables) are not independent. Several techniques have been proposed to overcome these 1

problems. Some of the most effective mixed nite element methods presently available to handle the convective terms, employ the Streamline Upwind Petrov Galerkin (SUPG) method of Marchal and Crochet [1987], or the Discontinuous Galerkin (DG) method of Fortin and Fortin [1989], based on the ideas of Lesaint and Raviart [1974]. The lack of ellipticity of the momentum equation is frequently resolved by using the Explicitly Elliptic Momentum Equation (EEME) formulation introduced by King et al. [1988], the Elastic Viscous Stress Split (EVSS) formulation [Rajagopalan et al., 1990], or, more recently, the Discrete Elastic Viscous Stress Splitting (DEVSS) method of Gu nette and Fortin e [1995]. In a mixed velocity-pressure-stress formulation, interpolation of the different variables has to satisfy a compatibility condition. Marchal and Crochet [1987] introduced a four-by-four bi-linear subdivision for the stresses on each bi-quadratic velocity element to obtain a stable discretisation scheme. The EEME method has been shown to give accurate and stable results by introducing a second-order elliptic operator to the momentum equation. This method however, is restricted to UCMlike nonlinear constitutive equations and excludes the use of a solvent viscosity. The EVSS method is obtained by splitting the deviatoric stress into a viscous and an elastic contribution. An adaptive strategy in combination with a modied SUPG method as proposed by Sun et al. [1996] has given stable results for the falling sphere in a tube benchmark problem. A disadvantage of all of these methods is the continuous interpolation of the extra stress and the subsequent large sets of global unknowns upon discretisation. On the other hand, the Discontinuous Galerkin method employs a discontinuous interpolation of the stress variables which leads to a more easily satised compatibility condition for stress and velocity and a substantial reduction of global degrees of freedom when an implicit/explicit scheme is used. Here, a combination of the DG and DEVSS methods is applied to the visco-elastic ow simulations. A change of variable leads to an extra stabilising equation. By using an implicit/explicit handling of the advective part, the extra stresses can be eliminated at element level. This results in the DEVSS/DG method introduced to the calculation of visco-elastic ows by Baaijens et al. [1997] and has successfully been used by B raudo et al. [1998]. e In this work, an efcient numerical scheme, based on the DEVSS/DG method, is presented for the analysis of 2D and 3D multi-mode visco-elastic ows. Furthermore, a numerical/experimental evaluation is presented for two steady shear complex ows: a ow around a symmetric conned cylinder and in a cross-slot ow. The behaviour of several constitutive equations is evaluated in these ows. Two well-known constitutive relations of differential type (the Giesekus and the Phan-Thien Tanner model) are applied together with a recently proposed model by Peters et al. [1999] that provides enhanced independent control of shear and elongational properties. Although the computational procedure allows for transient calculations, only steady ows will be investigated using a time-marching scheme to reach the steady state solution.

2 Problem denition
Isothermal and incompressible uid ows, neglecting inertia, are described by the equations for conservation of momentum (1) and mass (2): = 0, u = 0, (1) (2)

where is the gradient operator, denotes the Cauchy stress tensor and u the velocity eld. The Cauchy stress tensor is dened by equation (3): = pI + , 2 (3)

with a pressure term p = 1 tr() and the extra stress tensor, which has to be dened by a consti3 tutive model.

2.1 Constitutive models


The extra stress tensor can be further divided into a Newtonian solvent and a visco-elastic part. For a realistic description of the visco-elastic contribution, a multi-mode approximation of the relaxation spectrum is often necessary for most polymeric uids. This can be expressed as the sum of the separate visco-elastic modes:
M

= 2s Du +
i=1

i ,

(4)

where s denotes the viscosity of the purely viscous or solvent mode, Du = 1 (L + Lc ) the rate of 2 deformation tensor, in which L denotes the velocity gradient tensor L = ( u)c . i is the visco-elastic contribution of the ith relaxation mode and M the total number of different modes. Within the scope of this work, a sufciently general way to describe the constitutive behaviour of a single mode is obtained by using a differential equation based on the generalised Maxwell-type equation: i + i + f (i , Du ) = 2G(i )Du , (i ) (5)

with (i ) and G(i ) the stress dependent relaxation time and modulus, respectively. The upper convected time derivative of the stress i is dened as: i = i +u t i L i i Lc , (6)

where t denotes time. The functions f (i , Du ), (i ) and G(i ) depend upon the chosen constitutive equation. Notice, that by choice of f (i , Du ) = 0, (i ) = i and G(i ) = Gi , with i and Gi the (constant) relaxation time and modulus of the ith mode, respectively, the Upper Convected Maxwell model is retrieved. In this work, two conventional non-linear constitutive models are investigated for rheological behaviour. For the Giesekus model, the functions are dened as: i i , (7) f (i , Du ) = Gi i (i ) = i , (8) G(i ) = Gi , where is a material parameter. For the exponential PTT model (PTTa), the functions are dened as: f (i , Du ) = Du i + i Du , (i ) = i e
G I i

(9)

(10) (11) (12)

G(i ) = Gi ,

with and material parameters, and I = tr(i ) the rst invariant of i . More detailed information on these constitutive equations can be found in e.g. Tanner [1985], Bird et al. [1987] and Larson [1988]. 3

Also, a new visco-elastic constitutive relation is investigated, which has recently been proposed by Peters et al. [1999] and Schoonen et al. [1998]. It is based on the upper convected Maxwell model, so f (i , Du ) = 0 is chosen. To incorporate more exibility, both the relaxation time function (i ) and the modulus function G(i ) are chosen to be non-linear function of the visco-elastic stress i . For steady simple shear ows, with shear rate , the shear stress can be written as: 12 = G = . (13)

Hence, the viscosity of the uid is represented by the product of the modulus and the relaxation time 12 = G. Now, the simple shear viscosity () = can be chosen such that it describes accurately the measured shear viscosity. A suitable choice, for instance, is the extended Ellis model: (i ) = G(i )(i ) = Gi i 1+A
|II | Gi2
a b

(14)

2 Here, II is the second invariant of i dened as II = 1 (I tr(i2 )). Note, that for innitesimal 2 2 1) the linear Maxwell model is recovered, as desired. strains (|II |/Gi Since only the viscosity function is determined, a choice remains to be made for the relaxation time function (or modulus function). Here, the variable drag (VD) model [Peters, 1994] is chosen as relaxation time function:

(i ) = i e

I G i

I G

(15)

with a material parameter. This model is called the Feta-VD model, for equation (14) xes the shear viscosity (hence the prex Fixed eta). In contrast to the PTTa model, the shear viscosity of the Feta-VD model is not sensitive to the material parameter , while the rst normal stress difference N1 = 11 22 is less sensitive to . This material parameter can now be used to t the elongational behaviour of the material. Thus, the shear and elongational properties can be controlled more independently. The three non-linear functions for the Feta-VD model now read: f (i , Du ) = 0 ,

(i ) = i e G(i ) =

(16)
I I G G i i

Gi e

I G i

q,
a

(17)
b

I G i

1+A

|II | Gi2

(18)

3 Computational method
The modelling of polymer ows gives rise to some considerable characteristic problems. Looking more closely at the governing equations (1), (2) and (5), it is obvious that the use of multiple relaxation times inevitably leads to a very large system of equations, when the extra stress variables are considered as global degrees of freedom. Another problem and a challenging eld of investigation is the loss of convergence of the numerical algorithm for increasing elasticity in the visco-elastic ow (increasing Weissenberg number We). There are several computational methods available today that are more or less capable of efciently handling the above problems. The method used here is known as the Discrete Elastic Viscous Stress Splitting / Discontinuous Galerkin method (DEVSS/DG). It is basically a combination of the 4

Discontinuous Galerkin method, developed by Lesaint and Raviart [1974], and the Discrete Elastic Viscous Stress Splitting technique of Gu nette and Fortin [1995]. The combination of the two methe ods into the DEVSS/DG method was rst applied by Baaijens et al. [1997].

3.1 DEVSS/DG method


This technique nds its basis in the classical Galerkin Finite Element Method. Consider a spatial domain that is divided into K elements (e ) such that = e with boundary . Gu nette and Fortin [1995] proposed a new mixed formulation by introducing an L2 projection e of the rate of deformation tensor to yield a discrete approximation D (equation (22)), in combina tion with a stabilisation term in the discrete momentum equation (2(Du D), equation (19)). The stabilising auxiliary viscosity in equation (19) can be varied in order to give optimal results. Fol lowing Gu nette and Fortin [1995] and Baaijens et al. [1997], = e Gi i is chosen and found to give satisfactory results. Based on the ideas of Lesaint and Raviart [1974], and rst applied for the analysis of viscoelastic ows by Fortin and Fortin [1989], a discontinuous interpolation is applied to the extra stress variables. They are now considered as local degrees of freedom and can be eliminated at element level. Upwinding is performed on the element boundaries by adding integrals on the inow boundary of each element and thereby forcing a step of the stress at the element interfaces (equation (21)). Time discretisation of the constitutive equation is attained using an implicit Euler scheme, with the exception that iext is taken explicitly (i.e. iext = iext (tn )). Hence, the term Sid : un(iext ) d has no contribution to the Jacobian which allows for local elimination of the extra stress. The mixed weak formulation now reads: Problem DEVSS/DG: Given (u, p, i , D) at t = tn , nd a solution at t = tn+1 such that for all d admissible test functions (v, q, Si , G),
M

Dv , 2s Du + 2 Du D +
i=1

v , p

=0,

(19)

q,

=0,

(20)

Sid , i +

i + f (i , Du ) 2G( i )Du (i )
K

Sid : un (i iext ) d = 0
e=1 e in

i 1, 2, . . . , M , (21)

G , D Du

=0,

(22)

where ( , ) denotes the L2 -inner product on the domain , D = 1 + ( )c with = u , v, 2 e an auxiliary viscosity, in is the inow boundary of element e , n the unit vector pointing outward normal on the boundary of the element (e ) and iext denotes the extra stress tensor of the neighbouring, upwinding element. 5

3.2 Solution strategy


In order to obtain an approximation of Problem DEVSS/DG, a 2D domain is divided into K quadrilateral and a 3D domain into K hexahedral elements. A choice remains to be made about the order of the interpolation polynomials of the different variables with respect to each other. As is known from solving Stokes ow problems, velocity and pressure interpolation cannot be chosen independently and has to satisfy the Ladyzenskaya-Babuska-Brezzi condition. Likewise, interpolation of velocity and extra stress has to satisfy a similar compatibility condition in order to obtain stable results. Baaijens et al. [1997] have shown that for 2D problems, discontinuous bi-linear interpolation for extra stress, bi-linear interpolation for discrete rate of deformation and pressure with respect to bi-quadratic velocity interpolation (gure 1, left) gives stable results. This approach is extrapolated to a third dimension and satisfying the LBB condition, spatial discretisation is performed using tri-quadratic interpolation for velocity, tri-linear interpolation for pressure and discrete rate of deformation while the extra stresses are approximated by discontinuous tri-linear polynomials (gure 1, right). Integration of equation (19) to (22) over an element is performed using a quadrature rule common in nite element analysis (3 3 or 3 3 3 Gauss rule, for 2D and 3D respectively).

u i , p, D

u i , p, D

Figure 1: Mixed nite element. Left: u bi-quadratic, p, D bi-linear, i discontinuous bi tri-linear, i discontinuous tri-linear. linear. Right: u tri-quadratic, p, D To obtain the solution of the nonlinear equations, a one step Newton-Raphson iteration process is carried out. Consider the iterative change of the nodal degrees of freedom ( , u , D , p ) as variables of the algebraic set of linearised equations. This linearised set is given by: Q Q u 0 0 f f Q Quu QuD Qup u u u (23) = , fD 0 D QDu QD D 0 0 Qpu 0 0 p fp where f ( = , u, D, p) correspond to the residuals of equations (19) to (22), while Q follow from linearisation of these equations. Due to the fact that ext has been taken explicitly in equation (21), matrix Q has a block diagonal structure which allows for calculation of Q1 on the element level. Consequently, this enables the reduction of the global DOFs by static condensation of the extra stress. Despite this approach, still a rather large number of global degrees of freedom remains per element, as it is depicted in gure 1 (43 or 137 DOF/element, in the 2D and 3D case respectively). A further reduction of the size of the Jacobian is obtained by decoupling problem 23. First, the Stokes problem is solved (u, p) after which the updated solution is used to nd a new approximation for D.

The following problems now emerge: Problem DEVSS/DGa: Given (u, p, i , D) at t = tn , nd a solution at t = tn+1 of the algebraic set:
Quu Qu Q1 Q u Qpu Qup 0

u p

fu Qu Q1 f fp

(24)

and Problem DEVSS/DGb: Given D at t = tn and u at t = tn+1 , nd D from:


QDD D = fD .

(25)

n Notice that fD is now taken with respect to the new velocity approximation, i.e. fD (u n+1 , D ) rather n than fD (u n , D ). The nodal increments of the extra stress are retrieved element by element following: = Q1 f , (26)

with f also taken with respect to the new velocity approximation (f ( n , u n+1 )). Using the above procedure, for 3D calculations still a substantial number of unknowns remain per element. Although direct solvers often prove to be more stable in comparison to iterative solvers, they become impractical for the 3D visco-elastic calculations due to excessive memory requirements which inevitably leads to the application of iterative solvers. To solve the non-symmetrical system of problem DEVSS/DGa an iterative solver is used based on the Bi-CGSTAB method of Van der Vorst [1992]. The symmetrical set of algebraic equations of problem DEVSS/DGb is solved using a Conjugate-Gradient solver. Incomplete LU preconditioning is applied to both solvers. It was found that solving the coupled problem (hence, solving for u , D , p at once) led to divergence of the solver while signicantly better results were obtained for the decoupled system. In order to enhance the computational efciency of the Bi-CGSTAB solver, static condensation of the center-node velocity variables results in lling of the zero block diagonal matrix in problem DEVSS/DGa and, in addition, achieves a further reduction of global degrees of freedom. Finally, to solve the above sets of algebraic equations, both essential and natural boundary conditions must be imposed on the boundaries of the ow channels. At the entrance and the exit of the ow channels the velocity proles are prescribed. And at the entrance, the steady state stresses are prescribed along the inow boundary.

4 Rheological characterisation
The polymer melt that is investigated in this work is a commercial grade low density polyethylene (DSM, Stamylan LD 2008 XC43), further referred to as LDPE. It has been extensively characterised by Schoonen [1998] and is also given in Peters et al. [1999]. The parameters for a four mode Maxwell model are obtained from dynamic measurements. The non-linear parameter for the Giesekus model is determined on the steady shear data. Two sets of parameters for the exponential PTT model are determined. The rst set is tted only on the steady shear data, whereas the second set is also tted on elongational measurements. For the Feta-VD model, the Ellis parameters (A, a and b) are determined on the steady shear data. The t for this model is based on the steady shear data and the steady planar elongational data, taken from measurements in a cross-slot device. The parameter ts for the different numerical models at T = 190 [ C] are given in table 1. Figure 2 shows the rheological behaviour of the four numerical models and measured data in simple shear. Predictions in extension for the different models are depicted in gure 3. All models show similar 7

Maxwell parameters i 1 2 3 4 Gi [Pa] 7.598 104 1.664 104 3.518 103 3.174 102 i [s] 2.097 103 2.767 102 2.711 101 2.474 100

Feta-VD A, a, b 2, 2, 1 , 0.12, 0.44

Giesekus 0.29

PTTa-1 , 0.15, 0.08

PTTa-2 , 0.004, 0.08

Table 1: Material parameters of different models for LDPE melt (T = 190 [ C], = 0.9377 [s]).
Steady shear viscosity for LDPE melt at T=190oC First Normal Stress Difference N1 [Pa]
Steady shear Literature (cone/plate) Literature (capillary)

10

First Normal Stress Difference for LDPE melt at T=190oC


6 Steady shear Literature (cone/plate)

10

Viscosity [Pa s]

10

10

10

10 2 10

FETA (=0.12,=0.44) Giesekus (=0.29) exp. PTT (=0.150,=0.08) exp. PTT (=0.004,=0.08) Maxwell model

10

10 Shear rate [s1]

10

10 2 10

FETA (=0.12,=0.44) Giesekus (=0.29) exp. PTT (=0.150,=0.08) exp. PTT (=0.004,=0.08) Maxwell model

10 Shear rate [s1]

10

Figure 2: Model predictions and measurements of steady shear data (left) and rst normal stress difference (right) of LDPE melt at T = 190 [ C].
Transient uniaxial elongational viscosity for LDPE melt at T=120oC 7 10
= 0.03 [s1] = 0.10 [s1] 6 = 0.30 [s1]

10

Planar elongational viscosity for LDPE melt at T=150oC

Viscosity + [Pa s] u

Viscosity p [Pa s]

10

= 1.00 [s1]

10

10

10

10 1 10

FETA (=0.12,=0.44) Giesekus (=0.29) exp. PTT (=0.150,=0.08) exp. PTT (=0.004,=0.08) Maxwell model

10

4 FETA (=0.12,=0.44) Giesekus (=0.29) exp. PTT (=0.150,=0.08) exp. PTT (=0.004,=0.08)

10

10 Time t [s]

10

10

10 2 10

Crossslot 0

10

10 Strain rate [s1]

10

10

Figure 3: Model predictions and measurements of transient (left) and planar elongational data (right) of LDPE melt at T = 120 [ C] and T = 150 [ C], respectively. results for the shear data. The agreement with measurements is rather well in shear for all these ts. However, in extension, the models show quite different behaviour. For the transient elongational behaviour, only the PTTa-2 t ( = 0.004, = 0.08) can predict the upswing reasonably well. Second best is the Feta-VD model. However, for the planar extension, the elongational thickening behaviour for this PTTa-2 model is overpredicted. Here, the Feta-VD model gives the best agreement. The sharp increase clearly visible for different modes, does not seem to be very natural.

5 Complex ows of LDPE melt


A comparison is made between numerical and experimental results for two complex ows. First, the ow around a symmetric conned cylinder is reported in detail, both for 2D and 3D calculations. Then, the visco-elastic ow through a cross-slot device is presented. Both ow geometries are known to have a simple shear region, a solely elongational region, and a combined shear / elongational region, which makes them to be complex ow geometries. The experimental data of the ows described in this work consists of two parts [Schoonen, 1998]. First, eldwise velocity measurements have been carried out using Particle Tracking Velocimetry (PTV). Second, Flow Induced Birefringence (FIB) has been used to measure the stresses over the depth of the ow. For these stress measurements, a 10mW HeNe laser (wave length 0 = 633 [nm]) was used. The FIB measurements can be compared with the calculations by means of the empirical stress optical rule. For a projection of the birefringence tensor in a plane perpendicular to the optical path, the empirical stress optical rule reads: sin 2 = 2k0 dCxy , cos 2 = k0 dCN1 , (27) (28)

with the rotation angle, the phase retardation, k0 the initial propagation number (k0 = 2/0 ), d the thickness, C the stress optical coefcient, xy the mean plane shear stress and N1 = xxyy the mean rst normal stress difference of the element. A stress optical coefcient of 1.53 109 [Pa1 ] was determined for this material. Isochromatic lines are observed for retardation levels that equal multiples of 2. Then, equations (27) and (28) reduce to a single equation for the isochromatic lines ( = k2, k = 1, 2, 3, . . . ):
2 2 N1 + 4xy =

k0 , dC

k = 1, 2, 3, . . . .

(29)

5.1 Flow around a cylinder


An investigation of the planar ow around a symmetric conned cylinder is carried out. This ow is characterised by a compression of the polymer towards the cylinder, shearing along the cylinders surface and the material is stretched in the wake of the cylinder. Figure 4 shows the 3D geometry, the mesh and characteristics used to analyse numerically this ow. For symmetry reasons, only a quarter of the whole geometry has been modelled. At the entrance and exit, a fully developed velocity prole is prescribed, taken from a ow through a rectangular duct. Using the velocity prole at the entrance, the steady state stresses are computed and given as natural boundary conditions at the inlet. On rigid walls, the no-slip condition is used. Along the center plane, symmetry conditions are prescribed. To characterise the ow, the dimensionless Weissenberg number is chosen, which denotes the amount of elasticity in the ow: We = u 2D , R (30)

where is the mean relaxation time ( = ( 2 Gi )/( i Gi )), u2D is the 2D mean velocity and R i is the radius of the cylinder. The experiments are performed at a temperature of 170 [ C] at four ow rates. The characteristics are given in table 2. For the 3D calculations, only the lowest ow rate for 9

#Elements = 2856 #Nodes = 26295 #DOF(u, p) = 74069 #DOF(D) = 22512 #DOF( ) = 548352

H R y x z D

H = 4.95 mm D = 40 mm R = 1.1875 mm : flow direction

Figure 4: FE mesh, the geometry and characteristics for a planar ow around a symmetric conned cylinder. u2D [mm/s] 0.96 1.98 5.23 7.55 || = u2D /R [s1 ] 0.81 1.67 4.40 6.36 | | = 2D /R [kPa] u 3.74 7.71 20.36 28.11 We [] 1.4 2.9 7.7 11.1

Table 2: Characteristics for ow around a cylinder (T = 170 [ C], = 1.7438 [s])

the Giesekus model is shown, as it is only to point out, that the third dimension for this ow geometry is negligible. In gure 5, the normalised velocities in x-, y- and z-direction at cross-section x/R = 1.5 are depicted, together with the dimensionless normal stress differences and the main shear stress xy . As can be seen, the inuence of the front and back wall on the proles is rather limited. Only over a small depth near those conning walls, the proles change from zero to their maximum. Even at this cross-section, where the inuence of the cylinder is rather large, the z-velocity is at most only 10% of the velocity in x-direction. Therefore, a nominally 2D geometry can be assumed. For further comparison, 2D calculations sufce. In gure 6, the mesh and characteristics are shown for the 2D calculations. Again for symmetry reasons, only half of the geometry is meshed. Similar boundary conditions as for the 3D ow are prescribed. Along the centreline of the ow around a cylinder, the velocities were measured. In gure 7, the measured velocities for the four ow rates together with the Giesekus and PTTa-1 numerical results are shown. Unfortunately, the PTTa-2 model failed to give converged solutions and therefore is not presented here. The velocity proles predicted by the models are similar and in good agreement with the experimental velocities. For We = 7.7 and We = 11.1, the PTTa-1 model predicts a higher overshoot downstream near the cylinder than the Giesekus model. The latter seems to be more in agreement with the experimental data.

10

x z

1.5

1.5

1.5

ux / u 2D

uy / u 2D

1 0.5 0

1 0.5 0

uz / u 2D
2

1 0.5 0

0.5 15 10 0 1 2

0.5 15 10 0 1

0.5 15 10 0 1 2

z/R

10

15

y/R

z/R

10

15

y/R

z/R

10

15

y/R

4 2 0

2 0

xy / | |
2

N1 / | |

N2 / | |

2 0

2 4 15 10 0 1 2

2 4 15 10 0 1

2 4 15 10 0 1 2

z/R

10

15

y/R

z/R

10

15

y/R

z/R

10

15

y/R

Figure 5: Normalised velocity and stress proles for a ow around a cylinder at cross-section x/R = 1.5, Giesekus model ( = 0.29), We = 1.4 (T = 170 [ C], = 1.7438 [s], u2D = 0.96 [mm/s], | | = 3.74 [kPa]). #Elements = 1190 #Nodes = 4963 #DOF(u, p) = 8838 #DOF(D) = 3876 #DOF( ) = 57120 Figure 6: FE mesh and characteristics for a planar ow around a symmetric conned cylinder.
12 We=11.1 10 We= 7.7 We= 2.9 We= 1.4 8

u [mm/s]

0 6

Figure 7: Measured (o) and calculated velocities (: Giesekus ( = 0.29); : PTTa 1 (, = 0.15, 0.08) ), along the centreline. (T = 170 [ C], = 1.7438 [s], u2D = 0.96, 1.98, 5.23, 7.55 [mm/s])

x/R

11

W e

= 1:4
4 6 8 10 12

x/R
-4 2 1 -2 2 3 3 2 4 43 2 1 0 2 1 0

y/R
0 0

experiment
Feta-VD1

Giesekus

PTTa-1

Figure 8: Measured (top) and calculated isochromatic fringe patterns at We = 1.4 (T = 170 [ C], = 1.7438 [s], u2D = 0.96 [mm/s], Feta-VD: , = 0.12, 0.44, Giesekus: = 0.29, PTTa-1: , = 0.15, 0.08). The numbers indicate the fringe order.

The measured and calculated isochromatic fringe patterns for all four Weissenberg numbers are depicted in gures 8 to 11. At rst sight, the agreement between the measured and calculated isochromatic fringe patterns seems rather good. However, if we have a closer look, some differences can be detected. In the simple shear region (x/R = 4), both position and number of predicted fringes are very close to the measured pattern for the Giesekus and PTTa-1 models. This holds for all four ow rates. However, the Feta-VD model overpredicts the amount of fringes, especially for the higher Weissenberg numbers. This can be explained by looking at gure 2. In the shear rate region of the FIB measurements (|| = 0.81 6.36 [s1 ]), the shear viscosity of all models are almost equal. However, for the rst normal stress difference, the Feta-VD model predicts higher values and thus will show more fringes in shear dominated regions. This overprediction of fringes is also observed at cross-section x/R = 0. In the wake of the cylinder, the stresses predicted by the Giesekus and PTTa-1 models relax rather fast. The experimentally observed relaxation is signicantly slower. In this elongation dominated region, the Feta-VD model follows the experiments much better. If the planar elongational data in gure 3 is observed, the Giesekus model predicts a slight elongational thickening behaviour, which is not sufcient to capture the upswing. The PTTa-1 model even has a elongational thinning behaviour at the measured elongational rate. Only the elongational thickening behaviour of the Feta-VD model seems to be sufcient to capture the stress upswing in the wake of the cylinder. If the parameters of the models are changed, such that the upswing is captured, still the relaxation of the stresses can not be predicted satisfactory. In the Feta-VD model, the Variable Drag part is responsible for a realistic stress relaxation. Overall, the Giesekus model shows the best agreement in the shear dominated regions. Whereas the Feta-VD model gives the best predictions for the elongation dominated region. 12

W e
-4 2 1 -2 4 5 2 3 0 6 2

= 2:9
4 6 8 10 12

x/R
1 0

y/R
0 0 23 7 6 5 4

experiment
2 1

Feta-VD1

Giesekus

PTTa-1

Figure 9: Measured (top) and calculated isochromatic fringe patterns at We = 2.9 (T = 170 [ C], = 1.7438 [s], u2D = 1.98 [mm/s], Feta-VD: , = 0.12, 0.44, Giesekus: = 0.29, PTTa-1: , = 0.15, 0.08). The numbers indicate the fringe order.

W e
-4 2 3 2 1 0 1 23 4 -2 4 5 67 8 9 1 0 3 2 2

= 7:7
4 6 8 10 12

x/R
1 0

y/R
0

experiment
8 7 6 5 4 3 2 1

Feta-VD1

Giesekus

PTTa-1

Figure 10: Measured (top) and calculated isochromatic fringe patterns at We = 7.7 (T = 170 [ C], = 1.7438 [s], u2D = 5.23 [mm/s], Feta-VD: , = 0.12, 0.44, Giesekus: = 0.29, PTTa-1: , = 0.15, 0.08). The numbers indicate the fringe order.

13

W e
-4 2 3 4 -2 8 91011 5 67 0 1 5 2 34 2

= 11:1
x/R
1 0 1110 9 8 7 6 5 4 3 4 6 8 10 12

y/R
0

2 1 0

experiment
2 1

Feta-VD1 Giesekus PTTa-1

Figure 11: Measured (top) and calculated isochromatic fringe patterns at We = 11.1 (T = 170 [ C], = 1.7438 [s], u2D = 7.55 [mm/s], Feta-VD: , = 0.12, 0.44, Giesekus: = 0.29, PTTa-1: , = 0.15, 0.08). The numbers indicate the fringe order.

5.2 Cross-slot ow
As a second complex ow, the cross-slot ow is investigated. Here, two liquid ows approach each other in opposing directions, meet in a stagnation point, and leave in perpendicular opposing directions. In the stagnation point, the material experiences an innite extensional strain. Over the centreline towards and away from the stagnation point, the material is only compressed and stretched. In the in- and outow rectangular ducts, a pure shear ow is present. In the region around the stagnation point, the ow is a mixture of shear and elongation. Figure 12 shows the geometry, mesh and characteristics used to analyse numerically this ow. For symmetry reasons, only one quarter of the whole geometry is modelled. On rigid walls, no-slip conditions are used. Along centrelines, symmetry conditions are prescribed and a fully developed velocity prole is taken as boundary conditions at the inand outow. The steady shear stresses are derived and prescribed as natural boundary condition at the inow, based on the fully developed velocity prole. For characterisation of the ow, the radius R in equation 30 is replaced by half the height of the inow channel h = 1 H: 2 We = u 2D . h (31)

The experiments are performed at a temperature of 150 [ C] at two ow rates. The characteristics are given in table 3. Over the centreline, the velocities are measured and shown in gure 13, along with the calculated velocities for the different models. Again, the PTTa-2 model gave convergence problems. Near the stagnation point, the residence time of the material approaches innity. Therefore, a relatively small 14

#Elements = 1904 #Nodes = 7875 #DOF(u, p) = 13976 #DOF(D) = 6102 #DOF( ) = 91392

H = 5 mm D = 40 mm R = 1.25 mm : inflow plane : outflow plane : flow direction

D H

y x z

Figure 12: Detail of FE mesh, the geometry and characteristics for a cross-slot ow. u2D [mm/s] 3.0 4.4 || = u2D /h [s1 ] 1.20 1.76 | | = 2D /h [kPa] u 11.6 17.0 We [] 4.1 6.0

Table 3: Characteristics for cross-slot ow. (T = 150 [ C], = 3.4338 [s])


Velocity along centerline u2D = 3.0 [mm/s]
8 8

Velocity along centerline u2D = 4.4 [mm/s]

Ux [mm/s]

Ux [mm/s] 0 Vy
Experiment FetaVD, , = 0.12,0.44 Giesekus, = 0.29 exp. PTT1, , = 0.15,0.08
3 2 1 0 1 2 3 4

Vy
4

Experiment FetaVD, , = 0.12,0.44 Giesekus, = 0.29 exp. PTT1, , = 0.15,0.08


3 2 1 0 1 2 3 4

8 4

8 4

y/H

x/H

y/H

x/H

Figure 13: Measured and calculated velocities along the centreline. (T = 150 [ C], = 3.4338 [s], u2D = 3.0, 4.4 [mm/s])

amount of polymer melt ows through this area, and only a limited number of data points could be measured around this stagnation point. The velocity prole in the upstream part is for all models in good agreement with the experimental data. The Feta-VD model predicts an undershoot, which is not observed in the other models nor in the measured data. In the downstream part, the area of strong elongation, the models overestimate the velocity. Also, an overshoot is calculated, which is largest for the PTTa-1 model. This overshoot is experimentally not seen. 15

x/h
0 1 2 3 4 5 6 7 8 9 10

experiment
0 1 2 1 5 4 3 2 1 1 2 3 3

y/h

Feta-VD1

Giesekus

PTTa-1
Figure 14: Measured (top) and calculated isochromatic fringe patterns at We = 4.1 (T = 150 [ C], = 3.4338 [s], u2D = 3.0 [mm/s], Feta-VD: , = 0.12, 0.44, Giesekus: = 0.29, PTTa-1: , = 0.15, 0.08). The numbers indicate the fringe order.

16

x/h
0 1 2 3 4 5 6 7 8 9 10

experiment
1 0 1 2 3 4 4 3 2 1 5 4 3 2

y/h

Feta-VD1

Giesekus

PTTa-1
Figure 15: Measured (top) and calculated isochromatic fringe patterns at We = 6.0 (T = 150 [ C], = 3.4338 [s], u2D = 4.4 [mm/s], Feta-VD: , = 0.12, 0.44, Giesekus: = 0.29, PTTa-1: , = 0.15, 0.08). The numbers indicate the fringe order.

17

The measured and calculated isochromatic fringe patterns for the two ow rates are depicted in gure 14 and 15. In the fully developed inow region, the Feta-VD model overpredicts the amount of fringes by almost a factor two. Similar as the overprediction in the ow around a cylinder, this is caused by a higher prediction of the rst normal stress difference by this model. Also, it can partly be ascribed to the slightly higher mean velocity in the calculations (see gure 13). In this inow region, the Giesekus and PTTa-1 model slightly overpredict the amount of fringes, partly due to the higher mean velocity. Near the stagnation area, the Feta-VD model predicts the most fringes, which seems to be accurate. The PTTa-1 model has the least fringes in this region. Over the downstream centreline, away from the stagnation point, the PTTa-1 and Giesekus results relax too fast in regard to the experimental data. The Feta-VD model does a better job, and shows a good agreement with the experiments in this case. A better view of the performance over the centreline is given in gure 16. This gure clearly shows the incapability of the Giesekus and PTTa-1 model to predict the upswing and relaxation.
x 10 2 1.8 1.6
5

x 10

Experiment FetaVD, , = 0.12,0.44 Giesekus, = 0.29 exp. PTT1, , = 0.15,0.08

2 1.8 1.6

Experiment FetaVD, , = 0.12,0.44 Giesekus, = 0.29 exp. PTT1, , = 0.15,0.08

[N/m2]

1.4 1.2 1

[N/m2]
2 xy 2 1

1.4 1.2 1

N + 4

2 xy

0.8 0.6 0.4 0.2 0 10

N + 4
5 0 5 10 15 20

2 1

0.8 0.6 0.4 0.2 0 10

10

15

20

y/H

x/H

y/H

x/H

Figure 16: Measured and calculated stresses over the centreline. (T = 150 [ C], = 3.4338 [s], u2D = 3.0, 4.4 [mm/s])

6 Conclusions and discussion


A mixed low-order nite element based on the DEVSS/DG method is developed and implemented for the calculation of 2- and 3-dimensional visco-elastic ows. Steady state uid ows through two complex ow geometries for several ow rates have been calculated for an LDPE polymer melt. Several different constitutive relations have been applied for the evaluation of these complex ows. Two of them are established and well-known differential models, the Giesekus and exponential PTT model, and a third one is the recently introduced Feta-VD model, also in a differential form. By means of Particle Tracking Velocimetry (PTV) and Flow Induced Birefringence (FIB), experimental values are obtained and compared to the calculations. For both complex ows, the velocity and stress calculations with the Giesekus and PTT models show a good agreement between numerical and experimental data in shear dominated regions. The Giesekus model gives a slightly better comparison than the PTT model. However, in the elongational dominated regions, the stress relaxation of the Giesekus and PTT models is too fast compared to the experiments. Here, the upswing of the stresses can not be captured due to the low planar elongational thickening and thinning behaviour (at these rates) of the Giesekus and PTT model, respectively. Consequently, the relaxation is not in 18

agreement with the experimental data either. In these elongational dominated regions, the Feta-VD model performs signicantly better. Its higher planar elongational thickening behaviour is sufcient to capture accurately the stress upswing observed in the experiments. The Variable Drag part of the model accounts for a satisfying stress relaxation. However, the Feta-VD model pays a price by loss of accuracy in predictions of the shear induced rst normal stress difference. The overprediction of the rst normal stress difference is the main cause of the discrepancy for the stresses in shear dominated regions. This does not negatively inuence the calculated velocity proles, which are in good agreement with the experimental values.

References
J. Azaiez, R. Gu nette, and A. At-Kadi. Numerical simulation of viscoelastic ows through a planar e contraction. J. Non-Newtonian Fluid Mech., 62:253277, 1996. F.P.T. Baaijens. Application of low-order Discontinuous Galerkin methods to the analysis of viscoelastic ows. J. Non-Newtonian Fluid Mech., 52:3757, 1994. F.P.T. Baaijens, S.H.A. Selen, H.P.W. Baaijens, G.W.M. Peters, and H.E.H. Meijer. Viscoelastic ow past a conned cylinder of a LDPE melt. J. Non-Newtonian Fluid Mech., 68:173203, 1997. C. B raudo, A. Fortin, T. Coupez, Y. Demay, B. Vergnes, and J.F. Agassant. A nite element method e for computing the ow of multi-mode viscoelastic uids: Comparison with experiments. J. NonNewtonian Fluid Mech., 75:123, 1998. R.B. Bird, R.C. Armstrong, and O. Hassager. Dynamics of Polymeric Liquids, volume 1. Fluid Mechanics. John Wiley & Sons, New York, 1987. M. Fortin and A. Fortin. A new approach for the FEM simulation of viscoelastic ows. J. NonNewtonian Fluid Mech., 32:295310, 1989. R. Gu nette and M. Fortin. A new mixed nite element method for computing viscoelastic ows. J. e Non-Newtonian Fluid Mech., 60:2752, 1995. R.C. King, M.R. Apelian, R.C. Armstrong, and R.A. Brown. Numerically stable nite element techniques for viscoelastic calculations in smooth and singular domains. J. Non-Newtonian Fluid Mech., 29:147216, 1988. R.G. Larson. Constitutive Equations for Polymer Melts and Solutions. Butterworths, Boston, 1988. P. Lesaint and P.A. Raviart. On a nite element method for solving the neutron transport equation. Academic Press, New York, 1974. W.J. Lunsmann, L. Genieser, R.C. Armstrong, and R.A. Brown. Finite element analysis of steady viscoelastic ow around a sphere in a tube: calculations with constant viscosity models. J. NonNewtonian Fluid Mech., 48:6399, 1993. J.M. Marchal and M.J. Crochet. A new mixed nite element for calculating viscoelastic ow. J. Non-Newtonian Fluid Mech., 26:77114, 1987.

19

G.W.M. Peters. Thermorheological modelling of visco-elastic materials. In P.J. Halley and M.E. Mackay, editors, Proceedings 7th National Conference on Rheology, Brisbane, pages 167170, 1994. G.W.M. Peters, J.F.M. Schoonen, F.P.T. Baaijens, and H.E.H. Meijer. On the performance of enhanced constitutive models for polymer melts in a cross-slot ow. J. Non-Newtonian Fluid Mech., 82:387 427, 1999. S. Pilitsis and A.N. Beris. Viscoelastic ow in an undulating tube. part II: Effects of high elasticity, large amplitude of undulation and inertia. J. Non-Newtonian Fluid Mech., 39:375405, 1991. D. Rajagopalan, R.C. Armstrong, and R.A. Brown. Finite element methods for calculation of steady viscoelastic ow using constitutive equations with a Newtonian viscosity. J. Non-Newtonian Fluid Mech., 36:159192, 1990. J.F.M. Schoonen. Determination of Rheological Constitutive Equations using Complex Flows. PhD thesis, Eindhoven University of Technology, 1998. J.F.M. Schoonen, F.H.M. Swartjes, G.W.M. Peters, F.P.T. Baaijens, and H.E.H. Meijer. A 3D numerical/experimental study on a stagnation ow of a polyisobuthylene solution. Accepted, J. NonNewtonian Fluid Mech., 1998. A. Souvaliotis and A.N. Beris. Spectral collocation/domain decomposition method for viscoelastic ow simulations in model porous geometries. Comp. Meth. in Applied Mechanics and Engineering, 129:928, 1996. J. Sun, N. Phan-Thien, and R.I. Tanner. An adaptive visco-elastic stress splitting scheme and its applications: AVSS/SI and AVSS/SUPG. J. Non-Newtonian Fluid Mech., 65:7591, 1996. J. Sun and R.I. Tanner. Computaion of steady ow past a sphere in a tube using a PTT integral model. J. Non-Newtonian Fluid Mech., 54:379403, 1994. M.J. Szady, T.R. Salomon, A.W. Liu, D.E. Bornside, R.C. Armstrong, and R.A. Brown. A new mixed nite element method for viscoelastic ows governed by differential constitutive equations. J. Non-Newtonian Fluid Mech., 59:215243, 1995. K.K. Talwar and B. Khomami. Flow of viscoelastic uids past periodic square arrays of cylinders: Inertial and shear thinning viscosity and elasticity effects. J. Non-Newtonian Fluid Mech., 57: 177202, 1995a. K.K. Talwar and B. Khomami. Higher order nite element techniques for viscoelastic ow problems with change of type and singularities. J. Non-Newtonian Fluid Mech., 59:4972, 1995b. R.I. Tanner. Engineering Rheology. Oxford University Press, New York, 1985. H.A. van der Vorst. Bi-CGSTAB: A Fast and Smoothly Converging Variant of Bi-CG for the Solution of Nonsymmetrical Linear Systems. SIAM Journal on Scientic and statistical Computing, 13:631 644, 1992. V. van Kemenade and M.O. Deville. Application of spectral elements to viscoelastic creeping ows. J. Non-Newtonian Fluid Mech., 51:277308, 1994. F. Yurun and M.J. Crochet. High-order nite element methods for steady viscoelastic ows. J. NonNewtonian Fluid Mech., 57:283311, 1995. 20

You might also like