Cross-Permeability of the Semisolid Region in Directional Solidification: A Combined Phase-Field and Lattice-Boltzmann Simulation Approach
- 1.1k Downloads
Based on the results of microstructure simulations, fluid flow through the semisolid region during directional solidification of the technical Ni-base alloy 718 has been studied. Three-dimensional microstructures at different positions in the semisolid region were obtained by using a multicomponent multiphase-field model that was online coupled to a commercial thermodynamic database. For the range of five different primary dendrite distances λ 1 between 50 µm and 250 µm, the flow velocity and the permeability perpendicular to the dendrite growth direction was evaluated by using a proprietary Lattice-Boltzmann model. The commercial CFD software ANSYS FLUENT was alternatively applied for reference. Consistent values of the average flow velocity along the dendrites were obtained for both methods. From the results of the fluid flow simulations, the cross-permeability was evaluated as a function of temperature and fraction liquid for each of the five different primary dendrite distances λ 1. The obtained permeability values can be approximated by a single analytical function of the fraction liquid and λ 1 and are discussed and compared with known relations from the literature.
KeywordsFraction Liquid Mushy Zone Primary Dendrite Interface Thickness ANSYS Fluent
In simulation of solidification processes, fluid flow phenomena play a crucial role. Numerous types of defects arise during solidification as the formation of dendritic structures in the mushy zone causes specific fluid flow patterns. Although most of these defects are located at the micro- or meso-scale, they can be linked to the macro-scale by the melt permeability of the mushy zone. As stated by several authors, the specific permeability is a determining parameter in fluid flow phenomena like freckles, as well as in defects like porosity and hot cracking among others.1, 2, 3, 4
Improvements of simulation techniques and the combination of multiple tools working on different length scales through Integrated Computational Materials Engineering (ICME)5 open up new possibilities of investigations on the impact of permeability on casting defects. Phase-field models have become very popular for the simulation of the microstructure evolution during solidification processes of alloys. To describe their thermodynamic properties, idealized descriptions of the phase diagrams (ideal solution approximation6 and linear phase diagrams7) have been used for binary and pseudo-binary alloys. But this approximation is not suitable for use in multicomponent multiphase systems. Instead, using Gibbs energy descriptions assessed from experimental data via the Calphad approach,8 together with software tools for Gibbs energy minimization,9 seems to be most promising.
MICRESS® 10 has been developed by Access11 at Aachen Technical University (RWTH). It is based on a phase-field concept for multiphase systems that has been consequently extended to multicomponent systems12, 13, 14 by direct coupling to thermodynamic databases via the TQ Fortran interface to Thermo-Calc.9 Since then the software has been developed further and applied to different alloy systems and to Ni-based super alloys.15,16
Although numerous phase-field studies on directional dendritic solidification have been published, the three-dimensional (3D) simulation of representative parts of the mushy zone for technical multicomponent alloys, which are needed for a reliable evaluation of the liquid permeability, is still a challenge. The present paper reports phase field simulations carried out with MICRESS® that allowed subsequent investigations concerning the permeability during the solidification process. The results of the 3D dendrite simulations were subjected to further calculations by using the Lattice-Boltzmann method and the commercial CFD software ANSYS FLUENT. This allowed for assessment of the relation between temperature, fraction liquid and cross permeability for isothermal fluid flow, which plays a crucial role for freckle formation in technical remelting processes (e.g., of alloy 718).2 The obtained results are compared with empirical relations provided by Poirier17 and Madison.18
These equations are implemented in the software package MICRESS® 10 being used for the simulations throughout this paper. Direct coupling to thermodynamic and mobility databases is accomplished via the TQ-interface of Thermo-Calc Software.9 The thermodynamic driving force ΔG αβ and the solute partitioning are calculated separately by using the quasi-equilibrium approach,13 and they are introduced into the equation for the multiple phase-fields Eq. 1. This allows the software package to be highly flexible with respect to thermodynamic data of a variety of alloy systems and not to be restricted by the number of elements or phases being considered. Various extrapolation schemes14 have been implemented to minimize the thermodynamic data handling, especially for complex alloy systems.
Simulation started with a temperature of 1733 K at the bottom of the domain. A constant temperature gradient of G = 40 K/cm and cooling rate of V = 0.40 K/s were applied in all cases, ensuring a constant secondary arm spacing,19 the average of which has been determined as 31.2 µm. TiN particles were nucleated at the beginning according to a seed-density model by using an arbitrary density-radius distribution20,21 and were given time to grow and ripen before the primary dendrite appeared. To reach stationary growth conditions as quickly as possible, nucleation of the initial seed of fcc phase was performed exactly at the stationary tip temperature of 1618.8 K at the lower left corner of the simulation domain. Due to the high composition of alloying elements in alloy 718, stationary growth is achieved almost immediately, and the tip temperature remained at this value until the dendrite reached the top boundary.
Interface energy (J cm−2)
Static anisotropy coefficient
Interface mobility (cm4 J−1 s−1)
Kinetic anisotropy coefficient
2.0 × 10−5
1.0 × 10−4
1.0 × 10−8
1.0 × 10−4
1.0 × 10−11
In the phase-field simulations, the elements Ni, Cr, Fe, Nb, Mo, Ti, Al, C, N, and B and the phases LIQUID, FCC_A1, and FCC_A1#2 (Ti(N,C)) were taken into account. Thermodynamic data were obtained directly by coupling to the thermodynamic database TTNI7. Diffusion coefficients for the solid phases were taken from the mobility database MOBNI1;9 for the liquid phase, they were estimated to 1 × 10−5 cm2 s−1 for all elements.
Calibration of Interface Kinetics
It is well known that phase-field models suffer from numerical artifacts if spatial resolution is not sufficiently high.22 If the interface thickness is not much smaller than the diffusion length of all elements, the interface kinetics deviate from the sharp interface solution due to artificial solute trapping. As an additional problem, the interface even may get instable if the driving force varies too much over the length of the diffuse interface. For the case of solidification, there have been attempts to correct for these artifacts by introducing an anti-trapping current to the diffusion equations and by applying a suitable correction to the interface mobility (the so-called thin interface limit).23 Those approaches are often referred to as quantitative phase-field models. But even when rigorous thin interface corrections are considered, the interface thickness has to be on the order of the capillary length, and the required grid would still be too fine for 3D application of the phase-field method to multicomponent and multiphase technical alloys.
Furthermore, an averaging of the driving force ΔG along the normal vector of the interface was performed to reduce “artificial solute trapping” and to stabilize the interface profile, and a small interface thickness (η = 3Δx = 3 µm) was used that further helps reducing trapping artifacts. Artifacts originating from the small number of interface grid points were minimized by using a correction scheme for numerical discretization errors.25
Results of 3D Dendrite Simulations
Figure 1 shows the shape of the growing dendrite as an iso-surface with \( \phi = 0.5 \) at three different times (t = 3, 9, and 37 s). Small spherical TiN particles can be observed in front of the dendrite tip, which have formed at a much higher temperature. The dendrite emerges and overgrows tertiary arms that compete with the primary trunk. After hitting the top boundary, dendrite coarsening is the main phenomenon that can be observed. Due to the immense calculation time, only in the case of the smaller simulation domains (lower primary spacing), an overall fraction of solid above 0.5 could be reached.
Furthermore, these tertiary branches that grow in the direction of the temperature gradient compete with the primary dendrite. When a primary distance of λ 1 = 300 µm is reached, those tertiary arms cannot be blocked any more. This indicates that the upper limit of the stability range has been passed. Hence, this simulation set has been discarded from further evaluation.
Simulation of Permeability
In the process of solidification, the mushy zone is a domain of solid phases with emerging dendritic structures and an interconnected void phase filled with liquid. It has been proven to be consistent to model the mushy zone as a porous medium showing a spatially varying permeability.27
The calculations are carried out by using the Lattice-Boltzmann method29 or alternatively the CFD software ANSYS FLUENT. As the velocities and characteristic lengths are very small, in the order of 10−8 m/s and 10−5 m, respectively, a laminar, steady flow can be assumed in both models.
The original results from the MICRESS® simulations comprise precise fraction liquid data for all grid cells. For simulation of fluid flow, the fraction liquid information had to be reduced to zero or one in each cell. This was achieved by using a threshold value for the fraction liquid close to 0.5 below which the cell was assumed to be completely solid and otherwise completely liquid. This threshold value was chosen such that the total fraction liquid in the whole simulation domain was conserved. During this procedure, all TiN particles that were not in contact with the fcc dendrite were converted into liquid phase because they moved with the melt flow and such did not contribute to the pressure drop.
With the described procedures, according to the value of λ 1 in the phase-field simulation, cubic domains with a height of 1000 cells and base areas between 50 × 25 and 250 × 125 cells with a cell size of 1 µm3 were obtained that comprise a sharp interface between solid and liquid. These were directly used as representative domains for CFD calculation of the permeability of the mushy zone for different temperature and primary spacing.
Although there is very good agreement for the cross-permeability curves for intermediate values of λ 1 (150–200 µm) and the fraction liquid range below 0.65, there are strong deviations otherwise. Most striking is the steep decay of the cross-permeability curve in the current results occurring already at high values of f L (>0.85), whereas it was predicted by Poirier for the range 0.65 < f L < 0.75.17 The decay is attributed to the transition between free growth in the dendrite tip region and restricted growth below and (at least under these conditions) is wrongly predicted by the empirical relation of Poirier by up to a factor of 100 for high fraction liquid values. For prediction of fluid-flow phenomena in the upper part of the mushy zone (e.g., the initiation of freckles during electro-slag remelting (ESR)),2 this can have significant consequences.
The second strong difference is the much smaller dependency of the cross-permeability on λ 1 in our study. This can be easily explained by the fact that in our phase-field simulations, the primary dendrite distance λ 1 was modified “artificially” without changing the local thermal parameters (temperature gradient and cooling rate). To the contrary, the formulae by Poirier17 and Madison18 assume the change of λ 1 to be attributed to a change of the local thermal condition (i.e., they assume that the dendrites have selected their primary distance according to the local conditions). Thus, no similarity should be expected with respect to the λ 1-dependency of the cross-permeability because the main effect, which is the change of λ 2 due to an altered local solidification time, has not been included in our study. Direct comparisons of the datasets should only be made for average (stable) values of the primary distance (λ 1 = 150–200 µm).
Although for application a representation as a function of temperature (Fig. 7) may be most practical, plotting permeability in terms of the interfacial surface area (ISA) has been proposed as a more comprehensive description.34 Whether the ISA, which can be easily evaluated from simulated microstructures, provides a suitable metric for prediction of permeability will be the subject of future research.
Conclusion and Outlook
The coupling between the 3D phase-field simulation of multicomponent alloy microstructures by using MICRESS® and fluid-flow simulations carried out with ANSYS FLUENT or the Lattice-Boltzmann method presents an important contribution to the still developing and improving field of computational simulation of casting processes in an ICME framework. Besides the methodological aspects, the study also gives new knowledge and insight into the cross-permeability of the interdendritic zone and points out the opportunities offered by simulation approaches. By direct simulation of the dendrite morphology and its change downward through the mushy zone and by a systematic variation of the primary dendrite distance in the phase-field simulations, the dependency of the effective cross-permeability on the fraction liquid and the effect of the primary dendrite distance could be quantitatively evaluated for the technical Ni-base alloy 718 and compared with empirical relations from the literature.
The study not only presents a methodology to link different software tools and exchange data between them but also proposes a way to aggregate detailed results on the microscale and transform them by systematic variation and averaging into empirical laws that can be applied on the macroscale. The path is now open, and more dimensions can be included by variation of other parameters like cooling rate, temperature gradient, or alloy composition to obtain more comprehensive description of the permeability and other properties of the mushy zone.
The author B. Böttger acknowledges funding by the German Research Foundation (DFG). Parts of this work were conducted in the context of the Collaborative Research Centre SFB1120 “Precision Melt Engineering” at RWTH Aachen University.
- 2.B. Böttger, G.J. Schmitz, F.-J. Wahlers, J. Klöwer, J. Tewes, and B. Gehrmann, High Temp. High Press 42, 115 (2013).Google Scholar
- 3.B. Goyeau, T. Benihaddadene, D. Gobin, and M. Quintard, Metall. Mater. Trans. B 30, 1107 (1999).Google Scholar
- 5.U. Prahl and G. Schmitz, Integrative Computational Materials Engineering: Concepts and Applications of a Modular Simulation Platform (Weinheim: Wiley, 2012).Google Scholar
- 8.N. Saunders and A. Miodownik, CALPHAD Calculation of Phase Diagrams: A Comprehensive Guide (New York: Elsevier, 1998).Google Scholar
- 9.Themo-Calc Software, http://www.thermocalc.se.
- 10.http://www.micress.de. Accessed 29 Oct 2015.
- 11.www.access-technology.de. Accessed 29 Oct 2015.
- 19.W. Kurz and D.J. Fisher, Fundamentals of Solidification (Switzerland: Trans Tech Publications, 1989), p. 85.Google Scholar
- 30.M. Renwei, W. Shyy, and Y. Dazhi, ICASE Report No. 2002–2017.Google Scholar
- 32.S. Succi, The Lattice Boltzmann Equation for Fluid Dynamics and Beyond (Numerical Mathematics and Scientific Computation) (New York: Oxford University Press, 2001).Google Scholar
- 33.M.C. Sukop and D.T. Thorne Jr, Lattice Boltzmann Modeling—An Introduction for Geoscientists and Engineers (Berlin: Springer, 2006).Google Scholar