Savya SACHI: Coupling the solidification model with CALPHAD data for the prediction of macrosegregation and solidification structures

Abstract:
The present work aims at refining existing models in SOLID® by developing capabilities for improved prediction of the solidification process. Multiphase solidification models incorporate transport equations, which are closed by interphase transfer terms that are governed by microscopic constitutive relationships. These analytical relationships rely on the accurate representation of the microstructural phenomena such as the grain morphology and solute profile in the phases, along with the assumptions of diffusion controlled solidification with thermodynamic equilibria at the solid-liquid interfaces. This work focuses on two aspects: i) coupling solidification model with the thermodynamics of multicomponent alloys and ii) incorporating a new liquid diffusion length model for improved prediction of solute profile in the liquid phase. A methodology is proposed for incorporating phase diagram data into multiphase volume average solidification models. Previous instances of coupling the model with thermodynamic software packages include direct coupling with the so_tware and a tabulation and interpolation technique. Direct coupling is time-consuming, whereas the tabulation approach becomes infeasible with an increasing number of components in the system. We present a novel approach of using Artificial Neural Networks - Multi-layer perceptron (ANN-MLP) on tabulated thermodynamic data to obtain regression relationships, which can be easily coupled with the solidification model. This approach is computationally much more efficient than the above mentioned methods. The coupling procedure is described and validated with Thermo-Calc® Scheil solidification. Further simulations were performed on the Hebditch & Hunt benchmark case as well as an industrial ingot. Results obtained from the model, while improving the segregation prediction, also highlight the critical phase diagram parameters which help us propose modified values of these parameters for simulations which assume them to be constant. Secondly, the liquid diffusion length relationship proposed by Martorano et al. was extended to account for liquid convection. Simulation of the industrial ingot with the new diffusion length relationship shows a significant impact on the grain size and grain morphology.

Keywords:
Solidification, numerical modeling, Thermo-Calc®, artificial neural networks, liquid diffusion length

Composition of the jury:
> Reporters:
- Menghuai WU, Chair of Simulation and Modelling of Metallurgical Processes, Montanuniversitat Leoben,
- Benoit GOYEAU, Professeur, Université de Paris - Saclay,
> Examiners:
- Olga BUDENKOVA, Chargé de recherche CNRS, SIMAP,
- Marie BEDEL, Maître de conférences, IJL, ENSAM,
- Sabine DENIS, Professeur, Université de Lorraine,
> Direction of thesis:
- Hervé COMBEAU, Directeur de thèse, Professeur, IJL, Université de Lorraine
- Miha ZALOZNIK, Co-directeur de thèse, Chargé de recherche CNRS, IJL
> Encadrant:
- Charles-André GANDIN, Directeur de recherche CNRS, CEMEF