Modeling for free dendrite growth based on physically-informed machine learning method

In order to overcome limitations of traditional theoretical and numerical models, based on machine learning (ML) method, three models were developed for dendrite growth in undercooled alloy melts to predict solid-liquid interfacial velocity, including purely data-driven model-1, model-2 based on Galenko-Danilov (GD) model and model-3 coupled with an extended GD model newly proposed by introducing thermo-kinetic correlation. A thoroughly comparative analysis was carried out by applying the three ML models to existing data of five alloys. Results verify that model-3 has the relatively better fitting ability to experimental data in case of interpolation, due to the varied effective kinetic coefficient introduced and the thermo-kinetic correlation considered. Both cases of complete and partial extrapolations were discussed. It is concluded that on the whole the two physics-based ML models, especially model-3, are superior to the purely data-driven ML model for extrapolation ability. Thus, ML model-3 is finally proposed in modeling dendrite growth.