Accurately predicting dendrite growth during alloy solidification is crucial for enhancing the quality of metallic products. Recently, data assimilation has emerged as a promising tool for integrating two cutting-edge methods for studying dendritic growth, viz. in situ X-ray observation experiments and phase-field (PF) simulations, to elucidate important parameter(s) of the simulation and produce a “digital twin” of a growing dendrite. This study was conducted to evaluate the performance of a data assimilation system, employing an ensemble Kalman filter, through systematic twin experiments focused on columnar dendrite growth in a thin film of a binary alloy. The results show that the data assimilation system effectively estimates multiple parameters and dendrite morphology simultaneously. Furthermore, two approaches—domain decomposition method and parameter estimation method from a part of the domain—to reduce computational costs are investigated. Both methods can effectively reduce the computational cost of data assimilation. Additionally, data assimilation using voxel observation data of varying resolutions, mimicking actual X-ray images, is performed. The study findings discourage the direct use of voxel data owing to incompatible PF simulations. PF relaxation computation is explored as a solution for generating observation data with smooth PF profiles. The results show that while relaxation computation enhances the estimation accuracy for high-resolution voxel data, its efficacy diminishes for low-resolution data. These detailed investigations of data assimilation provide valuable insights for utilizing actual in situ X-ray observation data in dendrite growth studies.
Read full abstract