2026/07/31 by Natasha Dropka, Xiao Le Ye, Kunal Meshram +2
paper · doi:10.1002/crat.70135
crossref issued 2026/07/31 · crossref published 2026/07/31 · crossref published-online 2026/07/31 · crossref created 2026/07/31 · crossref deposited 2026/07/31 · crossref indexed 2026/07/31 · crossref published-print 2026/08/01
ABSTRACT The Czochralski (Cz) method is widely used for the growth of bulk yttrium aluminum garnet (YAG) crystals, but aging of iridium crucibles introduces thermal resistance and material loss that can adversely affect crystal quality and process efficiency. In this study, 555 CFD‐generated synthetic datasets were used to investigate the influence of crucible degradation and other process parameters on heating power, interface deflection, and the ratio of growth rate to interface temperature gradient (v/G n ). Machine learning techniques, including Lazy Predict, symbolic regression (SR), and artificial neural networks (ANN), were applied to evaluate predictive performance, while Shapley value analysis was used to interpret feature importance and nonlinear interactions. ANN provided the best predictions for heating power and interface deflection, whereas SR outperformed other methods for ln(v/G n ). Shapley analysis identified iridium loss as a primarily negative factor limiting heating power, moderately influencing interface deflection, and indirectly reducing v/G n , while crystal rotation, pulling rate, and weight/size variables were also significant contributors. The results demonstrate that data‐driven models can accurately predict Cz‐YAG process outcomes and provide physically interpretable insights into the effects of crucible aging, supporting process optimization and improved crystal quality.