Georgia Institute of Technology
Multiphysics Modeling and Multiphase Flow in Vacuum Induction Melting of Aluminum Alloys
Abstract
dc:description.abstractVacuum Induction Melting (VIM) is a critical manufacturing process for high-performance aerospace and biomedical alloys, offering superior compositional control and homogeneity through electromagnetic heating under vacuum conditions. The process involves tightly coupled electromagnetic, thermal, and fluid dynamic phenomena that interact nonlinearly throughout the melting cycle, making computational modeling both necessary and chal- lenging. Mature numerical methods exist for the constituent physics individually, including enthalpy-porosity phase change formulations, level-set interface tracking, and Reynolds- averaged turbulence models, yet their rigorous and simultaneous application to VIM re- mains limited. Existing simulations tend to replicate specific experimental configurations and address parts of the coupled problem in isolation, and few provide systematic numerical verification or dimensionless analysis, undermining their utility as quantitative predictive tools. This thesis develops verified multiphysics models for three VIM scenarios in aluminum- based systems using established finite element methods implemented in COMSOL Mul- tiphysics 6.3, organized around the two electromagnetic effects (Joule heating and the Lorentz body force) that drive the process. The first model captures melting progression of a solid aluminum charge through the mushy zone transition using the enthalpy-porosity method. Fluid flow is excluded on the basis of scaling analysis: the Prandtl number of liq- uid aluminum, Pr ∼ O(10−2), places the thermal field in a conduction-dominated regime, and radiation scaling shows surface losses to be two orders of magnitude below volumet- ric Joule heating. Systematic sensitivity studies on the solidus-liquidus interval confirm that predicted melting timescales are insensitive to this numerical parameter. A paramet- ric power sweep shows that full melt is achieved at 282.7 s, 197.8 s, and 155.5 s at 5, 7.5, and 10 kW respectively, with the predicted 10 kW onset and duration underpredict- ing laboratory observations by approximately 13.3% and 3.75%; the residual discrepancy is attributed to the solid-block idealization of the pellet charge, incomplete temperature- dependent property characterization, and power losses not captured by the nominal coil input. The second model examines immiscible Al-Bi liquid mixing using level-set interface tracking. Bond- and Weber-number scaling establishes the outcome predictively: Bo ≈ 60.85 confirms that surface tension cannot resist gravitational settling of the denser Bi-rich phase, and We ≈ 22.6 confirms that inertial stresses from Lorentz-driven recirculation are sufficient to fragment the interface but not to sustain dispersion. The simulations bear this out across the full 0–10 kW range examined: a bismuth inclusion undergoes Rayleigh- Taylor-driven fragmentation within 0.2 s and settles to the crucible base within 3 to 5 s regardless of power level, establishing that homogenization in the immiscible regime is not achievable by electromagnetic stirring alone at the approximately 4:1 Al-Bi density ratio. The third model investigates miscible liquid homogenization through a variable- property species transport formulation coupled bidirectionally with turbulent flow. A qui- escent molecular-diffusion baseline, with the turbulent viscosity suppressed to eliminate spurious eddy-diffusivity contributions from the initialized turbulence model, reduces the normalized concentration standard deviation by only 17% over 30 s, leaving the melt far from homogeneous on any process-relevant timescale. Electromagnetic stirring at 5 and 10 kW drives homogenization to within 5% residual non-uniformity at 7.4 s and 4.5 s respectively, demonstrating that Lorentz-driven convective mixing is the rate-controlling mechanism in the miscible regime. All three models undergo rigorous numerical verification through mesh convergence studies, solver tolerance studies, and physics validation against analytical limits. A direct comparison of k-ε and k-ω turbulence closures reveals a 27.1% divergence in domain-averaged velocity, with k-ε adopted for the multiphase models on the basis of robust convergence and a 3.1× reduction in solve time; the velocity deviation represents a bounded, acknowledged uncertainty in the reported mixing timescales. Taken together, the three mod-els establish that electromagnetic induction controls both melting and mixing in aluminum-based VIM but cannot overcome gravitational separation when the alloy system is thermodynamically disposed to stratify, a process-level conclusion with direct implications for the use of electromagnetic stirring in immiscible alloy processing. While experimental validation is deferred pending diagnostic capabilities, the verified models provide credible baselines for future validation studies and immediate utility for parametric process analysis.
Degree
thesis:*- Name thesis:degree_name
- Mechanical Engineering, MS
- Department thesis:degree_department
- Mechanical Engineering
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Annamaraju, Maruthi
- Advisor dc:contributor.advisor
-
- Kalidindi, Surya R.
- Committee members dc:contributor.committeemember
-
- Stebner, Aaron
- Joseph, Roshan
- ORCID person:identifier.orcid
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81629