Publications

Publications / SAND Report

Modeling injection molding of net-shape active ceramic components

Rao, Rekha R.; Brooks, Carlton F.; Cote, Raymond O.; Castaneda, Jaime N.; Mondy, L.A.; Noble, David R.; Hopkins, Matthew M.; Notz, Patrick N.; Halbleib, Laura L.; Yang, Pin Y.; Burns, George B.; Grillet, Anne M.

To reduce costs and hazardous wastes associated with the production of lead-based active ceramic components, an injection molding process is being investigated to replace the current machining process. Here, lead zirconate titanate (PZT) ceramic particles are suspended in a thermoplastic resin and are injected into a mold and allowed to cool. The part is then bisque fired and sintered to complete the densification process. To help design this new process we use a finite element model to describe the injection molding of the ceramic paste. Flow solutions are obtained using a coupled, finite-element based, Newton-Raphson numerical method based on the GOMA/ARIA suite of Sandia flow solvers. The evolution of the free surface is solved with an advanced level set algorithm. This approach incorporates novel methods for representing surface tension and wetting forces that affect the evolution of the free surface. Thermal, rheological, and wetting properties of the PZT paste are measured for use as input to the model. The viscosity of the PZT is highly dependent both on temperature and shear rate. One challenge in modeling the injection process is coming up with appropriate constitutive equations that capture relevant phenomenology without being too computationally complex. For this reason we model the material as a Carreau fluid and a WLF temperature dependence. Two-dimensional (2D) modeling is performed to explore the effects of the shear in isothermal conditions. Results indicate that very low viscosity regions exist near walls and that these results look similar in terms of meniscus shape and fill times to a simple Newtonian constitutive equation at the shear-thinned viscosity for the paste. These results allow us to pick a representative viscosity to use in fully three-dimensional (3D) simulation, which because of numerical complexities are restricted to using a Newtonian constitutive equation. Further 2D modeling at nonisothermal conditions shows that the choice of representative Newtonian viscosity is dependent on the amount of heating of the initially room temperature mold. An early 3D transient model shows that the initial design of the distributor is sub-optimal. However, these simulations take several months to run on 4 processors of an HP workstation using a preconditioner/solver combination of ILUT/GMRES with fill factors of 3 and PSPG stabilization. Therefore, several modifications to the distributor geometry and orientations of the vents and molds have been investigated using much faster 3D steady-state simulations. The pressure distribution for these steady-state calculations is examined for three different distributor designs to see if this can indicate which geometry has the superior design. The second modification, with a longer distributor, is shown to have flatter, more monotonic isobars perpendicular to the flow direction indicating a better filling process. The effects of the distributor modifications, as well as effects of the mold orientation, have also been examined with laboratory experiments in which the flow of a viscous Newtonian oil entering transparent molds is recorded visually. Here, the flow front is flatter and voids are reduced for the second geometry compared to the original geometry. A horizontal orientation, as opposed to the planned vertical orientation, results in fewer voids. Recently, the Navier-Stokes equations have been stabilized with the Dohrman-Bochev PSPP stabilization method, allowing us to calculate transient 3D simulations with computational times on the order of days instead of months. Validation simulations are performed and compared to the experiments. Many of the trends of the experiments are captured by the level set modeling, though quantitative agreement is lacking mainly due to the high value of the gas phase viscosity necessary for numerical stability, though physically unrealistic. More correct trends are predicted for the vertical model than the horizontal model, which is serendipitous as the actual mold is held in a vertical geometry. The full, transient mold filling calculations indicate that the flow front is flatter and voids may be reduced for the second geometry compared to the original geometry. The validated model is used to predict mold filling for the actual process with the material properties for the PZT paste, the original distributor geometry, and the mold in a vertical orientation. This calculation shows that voids may be trapped at the four corners of the mold opposite the distributor.