Quantification of Epistemic Uncertainty in Grain Attachment Models for Equiaxed Solidification

Recent work has investigated various schemes for the attachment of free-floating grains in models of equiaxed solidification in multicomponent alloys. However, these models are deterministic in nature, and simply investigating their differences for a limited number of results would not constitute an adequate comparison of their predictions. Instead, the models are compared in the context of the uncertainty in the most important input parameters. This approach is especially important in light of the effort required to implement a new model. If the predictions are essentially the same, then either model will suffice, or one may be selected for ease of implementation, numerical robustness, or computational efficiency. If, however, the models are significantly different, then the most accurate should be selected. In order to investigate the effects of input uncertainty on the output of grain attachment models, the PRISM Uncertainty Quantification framework was employed. The three models investigated were a constant packing fraction (CPF) scheme, an average solid velocity method (AVM), and a continuum attachment approach. Comparisons were made between the CPF and AVM models to estimate the importance of the local velocity field and between the CPF and continuum models to determine the sensitivity of the macrosegregation to new parameters unique to the continuum model.