Importance of raw material features for the prediction of flux growth of Al2O3 crystals using machine learning†
Abstract
The flux method is an efficient liquid-phase crystal growth technique. Accordingly, it is expected to be one of the key technologies for the development of innovative inorganic materials in future because it enables the production of high-quality crystals. However, owing to the complexity of the mechanism of crystal growth in fluxes, it is difficult to establish guidelines for the experimental recipe to grow crystals. Thus, flux crystal growth still requires the long process of trial and error. Consequently, our goal is to develop a “process informatics” (PI)-assisted flux method, supported by machine learning. To predict flux crystal growth by linking it to the process, essentially, the experimental parameters must be converted into explanatory variables. However, the explanatory power of describing crystal growth is limited using only process conditions, such as raw materials and flux species, their preparation amounts, and heating conditions. In this study, we focused on using information on raw materials (raw material information) as explanatory variables and investigated their influence on the prediction of flux crystal growth. Aluminum oxide (Al2O3), in which raw materials have abundant lot numbers, was selected as the target material. After performing 185 growth experiments, we created regression models composed of process conditions and wide raw material information as explanatory variables and Al2O3 particle size distribution as the objective variable. The obtained models clarified the effect of the raw material information on the accuracy of the prediction of crystal growth. Our findings provide new insights into the PI-assisted flux method in terms of the importance of raw material information and effective descriptions. This can contribute to the development of highly accurate prediction models for data-driven experimental suggestion and clarification of important factors in flux crystal growth.
- This article is part of the themed collections: Machine Learning and Artificial Intelligence: A cross-journal collection and Data Driven Crystal Engineering