Through an integrated macroscale/mesoscale computational model, we investigate the developing shape and grain morphology during the melting and solidification of a weld. In addition to macroscale surface tension driven fluid flow and heat transfer, we predict the solidification progression using a mesoscale model accounting for realistic solidification kinetics, rather than quasi-equilibrium thermodynamics. The tight coupling between the macroscale and the mesoscale distinguishes our results from previously published studies. The inclusion of Marangoni driven fluid flow and heat transfer, both during heating and cooling, was found to be crucial for accurately predicting both weld pool shape and grain morphology. However, if only the shape of the weld pool is of interest, a thermodynamic quasi-equilibrium solidification model, neglecting solidification kinetics, was found to suffice when including fluid flow and heat transfer. We demonstrate that the addition of a sufficient concentration of approximately 1 pin diameter TiN grain refining particles effectively triggers a favorable transition from columnar dendritic to equiaxed grains, as it allows for the latter to heterogeneously nucleate in the undercooled melt ahead of the columnar dendritic front. This transition from columnar to equiaxed growth is achievable for widely differing weld conditions, and its precise nature is relatively insensitive to the concentration of particles and to inaccurately known model parameters. (C) 2015 Elsevier Ltd. All rights reserved.@en

Various physical interfacial phenomena occur during the process of welding and influence the final properties of welded structures. As the features of such interfaces depend on physics that resolve at different spatial scales, a multiscale and multiphysics numerical modeling approach is necessary. In a collaborative research project Modeling of Interface Evolution in Advanced Welding, a novel strategy of model linking is employed in a multiscale, multiphysics computational framework for fusion welding. We only directly link numerical models that are on neighboring spatial scales instead of trying to link all submodels directly together through all available spatial scales. This strategy ensures that the numerical models assist one another via smooth data transfer, avoiding the huge difficulty raised by forcing models to attempt communication over many spatial scales. Experimental activities contribute to the modeling work by providing valuable input parameters and validation data. Representative examples of the results of modeling, linking and characterization are presented.@en