Grain growth competition during melt pool solidification -- Comparing phase-field and cellular automaton models

TL;DR

This study compares phase-field and cellular automaton models for simulating grain growth during melt pool solidification in additive manufacturing, highlighting their differences and potential for improved microstructure prediction.

Contribution

It provides a systematic quantitative comparison of PF and CA models for microstructure development in a melt pool, revealing their respective strengths and limitations.

Findings

PF captures microscopic features like interface stability and transient growth.

CA predictions improve with grid refinement but still differ from PF.

Averaged grain distributions show better agreement between models.

Abstract

A broad range of computational models have been proposed to predict microstructure development during solidification processing but they have seldom been compared to each other on a quantitative and systematic basis. In this paper, we compare phase-field (PF) and cellular automaton (CA) simulations of polycrystalline growth in a two-dimensional melt pool under conditions relevant to additive manufacturing (powder-bed fusion). We compare the resulting grain structures using local (point-by-point) measurements, as well as averaged grain orientation distributions over several simulations. We explore the effect of the CA spatial discretization level and that of the melt pool aspect ratio upon the selected grain texture. Our simulations show that detailed microscopic features related to transient growth conditions and solid-liquid interface stability (e.g. the initial planar growth stage prior to its cellular/dendritic destabilization, or the early elimination of unfavorably oriented grains due to neighbor grain sidebranching) can only be captured by PF simulations. The resulting disagreement between PF and CA predictions can only be addressed partially by a refinement of the CA grid. However, overall grain distributions averaged over the entire melt pools of several simulations seem to lead to a notably better agreement between PF and CA, with some variability with the melt pool shape and CA grid. While further research remains required, in particular to identify the appropriate selection of CA spatial discretization and its link to characteristic microstructural length scales, this research provides a useful step forward in this direction by comparing both methods quantitatively at process-relevant length and time scales.