Flow simulation software to optimize coextrusion die design

1 June 2015
Mahesh Gupta and Kim Ryckebosch
Virtual fine-tuning of coextrusion dies using a flow simulation software reduced the lead time and the number of fine-tuning cycles in experiments by 35  significantly cutting the cost of designing a die.

Coextrusion, which involves simultaneous extrusion of several different polymers through a die, combines the functionalities and benefits of several polymers into a single multilayered product. For instance, plastic films used for wrapping food and for other packaging applications typically have three to nine different layers, with each layer having a specific function such as oxygen and moisture barrier, chemical resistance, scratch resistance, or adhesion.

Despite these inherent advantages, growth of the coextrusion market has been slowed by the complexity of coextrusion die design. Because polymer layers change their shape as they flow through a die, developing a die geometry that will produce the required layer structure in the final product is extremely challenging. Furthermore, with an improperly designed die, the shape of the extruded profile can change significantly after the polymer comes out of the die. Software that accurately predicts the layer structure in the final product, as well as the post-die distortion of the extruded profile, would thus be a valuable tool for a die designer.

The first attempts to simulate coextrusion date back to the 1980s. Owing to the limited power of the computers then available, Vlachopoulos and coworkers used a 2D approximation of the flow in extrusion dies to estimate the velocity, layer structure, and other quantities of interest.1, 2 The resultant estimates were good at times, depending on the geometry of the die, but discrepancies were often encountered between experimental data and numerical predictions. In the 1990s, researchers began to attempt 3D simulation of coextrusion using the finite element method: an approach that breaks up a die geometry into smaller and simpler discrete parts called ‘elements,’ connected together to form a ‘mesh.’3

The main difficulty in simulating multilayer flow during coextrusion is enforcing the different material properties when an element in the mesh is occupied by more than one polymer. Early attempts at 3D simulation of coextrusion used a ‘moving mesh’ technique, in which the mesh is modified after each flow simulation so that the inter-element boundaries match the interfaces between adjacent polymer layers.3,4 In this way, each mesh element is occupied by only one polymer, and the different properties of adjacent polymer layers can be easily enforced in the simulation. However, the moving mesh technique is impractical for simulating complex real-world coextrusion systems.

To tackle this, we recently developed a new fixed mesh technique (implemented in the polyXtrue5 software package) that is able to accurately simulate any coextrusion system. It uses a ‘mesh partitioning’ algorithm to model coextrusion. First, a mesh of tetrahedral elements is generated over the complete coextrusion die. Then, those elements that contain an interface between adjacent polymer layers are partitioned into two separate elements, each containing only one polymer, allowing the different material properties of the two polymers to be easily enforced.6–8 Since the finite element mesh itself remains unaltered during the entire simulation, the mesh partitioning technique can easily simulate a multilayer flow no matter how complex the geometry or layer structure of the coextrusion die.

We tested the ability of the polyXtrue software to optimize coextrusion dies used in a real-world application. Figure 1 shows the geometry of a coextrusion die used to produce a bilayer PVC window profile at Deceuninck NV of Belgium. The core polymer enters the die from the inlet on the right. The polymer for the skin layer enters from the top, and is split into two channels to form thin skin layers on both sides of the profile. The skin layer, which is the visible portion of this profile when installed on a window, is made from a virgin PVC compound. The polymer used for the core is a recycled PVC compound, obtained from old PVC window profiles collected during renovation of old buildings. The finite element mesh used for the simulation had 1,246,057 elements. On a Dell Precision M6700 laptop, the polyXtrue software required 29 minutes of computation time to complete the simulation.

Geometry of a bilayer die for coextrusion of a window profile.

Figure 2(a) shows the predicted velocity distribution in 10 different cross sections of the profile die. The velocity distribution at the die exit is shown in Figure 2(b). The initial design of this die had much larger variation in the velocity distribution at the die exit.8 The die geometry shown in Figure 1 was obtained after four virtual fine-tunings using polyXtrue. Figure 3 shows the evolution of the interface shape between the core polymer and the skin layer, starting from the line of first contact between the two layers to the die exit, and in the calibrator beyond the exit. The distortion of the internal walls of the profile and the layer structure predicted by polyXtrue are shown in Figure 4. These are in good agreement with the profile shape and layer structure of the final product extruded in experiments (see Figure 5).

Velocity distribution (a) in various cross sections across the die and (b) at the die exit.

Interface shape between the two polymer layers.

Predicted shape of the extrudate profile (red), layer structure (green), and shape of the profile at the die exit (blue).

Shape of the profile and layer structure in the extrusion experiments.

In the feed block for the skin layer of this die, stagnant flow—and consequent polymer degradation—were observed experimentally (see Figure 6). The simulation also captured this stagnation (see Figure 7), but it was unfortunately not noticed before the die was machined. The stagnation was later eliminated by putting in an insert and milling the feed channel again (see Figure 8). The pressure distribution in the die predicted by the simulation software is shown in Figure 9. The predicted pressure at the entrances of the core (36.1MPa) and skin layer (30.4MPa) are in excellent agreement with the corresponding experimental values of 35.4 and 29.9MPa, respectively.

Degraded polymer near the bottom of the feed channel for the skin layer.

Predicted velocity distribution showing the stagnant flow.

Velocity distribution with no stagnant flow in the modified feed channel.

Pressure distribution in the die predicted by the software.

As noted above, it took four virtual tunings using polyXtrue before the die shown in Figure 1 was machined. Prior to adopting the polyXtrue software, Deceuninck required eight to nine physical tunings in experiments to arrive at the final geometry of a coextrusion die. Now, with the virtual tunings using polyXtrue, only five to six physical tunings are needed. At Deceuninck, each mechanical tuning of a profile die costs about $3750 (€3000) and takes 1.5 to 2 weeks of lead time. The use of polyXtrue has thus delivered savings of about $7500–$11,250 (€6000–€9000) for each die design, and cut down the lead time for designing a die from about 3 months to about 2 months.

In summary, a computationally efficient flow simulation software is an effective design tool for virtual fine-tuning of complex profile extrusion dies. Although tuning of the extrusion die still needs to be completed experimentally, use of the simulation software reduces the number of ensuing physical tuning cycles, resulting in significant savings in the cost and lead time for designing the die. In future developments we envisage that, instead of the die designer fine-tuning the die geometry after each flow iteration, the die geometry will be optimized automatically by an algorithm in the software.


Mahesh Gupta
Michigan Tech University (MTU)

Mahesh Gupta is currently a professor of mechanical engineering at MTU, as well as president of Plastic Flow, LLC. He has been working on computer simulation and optimization of polymer extrusion and injection molding since 1988. He is a fellow of SPE, and is currently serving on the board of directors of its Extrusion Division.

Kim Ryckebosch
Deceuninck NV

Kim Ryckebosch has been working at Deceuninck NV since 2001. He has been designing dies for coextrusion of PVC window profiles since 2007.


  1. H. Mavridis, A. N. Hrymak and J. Vlachopoulos, Finite element simulation of stratified multiphase flows, AIChE J. 33, pp. 410-422, 1987.

  2. J. Perdikoulias, C. Richard, J. Vlcek and J. Vlachopoulos, A study of coextrusion flows in polymer processing, SPE ANTEC 37, pp. 2461-2464, 1991.

  3. A. Karagiannis, A. N. Hyrmak and J. Vlachopoulos, Three-dimensional studies on bicomponent extrusion, Rheol. Acta 29, pp. 71-87, 1990.

  4. T. Marchal, T. Burton, G. Franceschetti, J. De rijcke, C. Chauvin and H. M. Metawally, Numerical balancing of coextrusion dies: a validation study with a TPV-based hose, SPE ANTEC 53, pp. 2418-2423, 2007.

  5. http://www.plasticflow.com/ polyXtrue software, developed by Plastic Flow, LLC. Accessed 14 April 2015.

  6. M. Gupta, Mesh partitioning technique for three-dimensional simulation of coextrusion, SPE ANTEC 54, pp. 217-222, 2008.

  7. M. Gupta, Three-dimensional simulation of coextrusion in a complex profile die, SPE ANTEC 56, pp. 2032-2036, 2010.

  8. K. Ryckebosch and M. Gupta, Virtual fine-tuning of a profile coextrusion die using a three-dimensional flow simulation software, SPE ANTEC 61, pp. 1108-1113, 2015.

DOI:  10.2417/spepro.005962

Footer Links (2nd Row)