Hydride Content\nControl of Perovskite Oxyhydride BaTiO<sub>3–<i>x</i></sub>H<sub><i>x</i></sub> Supported\nby Image-Based Machine Learning

Hydride content in perovskite oxyhydrides represents\na crucial\nparameter that governs properties of the materials. However, the synthesis\nof the oxyhidrides with precise control over the hydride content <i>x</i> remains a challenging task because of the time-consuming\nand/or destructive analytical methods typically used for the determination\nof <i>x</i>. Here, we report an image-based machine learning\n(ML) system for prediction of the hydride contents <i>x</i> in a perovskite oxyhydride BaTiO<sub>3–<i>x</i></sub>H<sub><i>x</i></sub> using a picture of powder material\nas a quick evaluation method of hydride content <i>x</i>. The ML system, which employs the ExtraTrees algorithm, enabled\nthe prediction of <i>x</i> with a mean absolute error of\n0.013 and a low running cost. Although a similar ML system constructed\nby convolutional neural networks (CNN) that we used in the previous\nstudy demonstrated comparable accuracy in the prediction, the system\nwith ExtraTrees was more computationally efficient, thus was applied\nas a nondestructive and quick analytical method for hydride content <i>x</i>. Using the ML system, detailed profiles of thermal hydrogen\nrelease of BaTiO<sub>3–<i>x</i></sub>H<sub><i>x</i></sub> were quantitatively analyzed to demonstrate the\nfeasibility of fine-tuning the hydride content within 0 ≤ <i>x</i> ≤ 0.4 by adjusting the processing temperature.\nIn the case of powder samples comprising mixtures of the oxyhydride\nwith different hydride contents, the ML-based prediction provided\nan almost averaged value for <i>x</i>. The present results\ndemonstrate an application of image-based ML for the fine-tuning of\noxyhidride materials.