Taniike Lab.
MENU
2025.01.21 by 藤原 綾 Aya FujiwaraConventional catalyst discovery has been limited to the exploration and optimization of areas surrounding known high-performance catalysts, making it challenging to grasp the overall landscape of the design space or transfer insights across different catalyst systems. In this study, we aimed to overcome these limitations by employing two original techniques on high-throughput experimentation and feature engineering. As a result, comprehensive catalyst knowledge was obtained for the five catalyst systems applied to oxidative coupling of methane. This enabled a systematic understanding of catalyst design rules and facilitated knowledge transfer across catalyst systems during machine learning.
2025.01.05 by Wentao DUThe traditional approach to designing solid catalysts has been time-consuming and labor-intensive, relying on trials and errors guided by prior knowledge and hypotheses. Recently, combining machine learning with high-throughput experimentation, researchers are exploring uncharted chemistry without preconceptions or biases. Our study exemplifies this shift, focusing on methane dry reforming—a key reaction for sustainable energy and greenhouse gas mitigation. We constructed an unbiased dataset of 256 γ-Al2O3-supported catalysts, each with random combinations of 17 elements from the periodic table. This eliminates assumptions about optimal elements, allowing us to discover surprising trends and catalysts through machine learning.
2024.04.24 by 谷池 俊明 Toshiaki TaniikeInternal donors (IDs), organic Lewis bases, have been the major drivers in the evolution of the Ziegler-Natta catalyst for propylene polymerization. However, the discovery of new ID molecules remains a highly empirical and demanding task. Artificial intelligence methods offer promise but lack a quality dataset. Here, the teams of JAIST and U-Naples collaborate to establish an integrated high-throughput workflow encompassing catalyst synthesis, propylene polymerization, and polypropylene characterization. Its application to an ID library generated a robust and consistent dataset, allowing to establish accurate machine learning models via molecular fingerprinting and feature selection. This synergy will lead to accelerated research and development of the Ziegler-Natta catalyst.
2024.01.19 by Joao Marcos DA SILVEIRAInternal donors are essential organic components of the heterogeneous Ziegler-Natta catalyst in producing high-quality polypropylene. However, their capability of shaping the catalyst structure during its preparation has been long overglanced, though it constitutes the basis of generating desired active sites. To address this issue, we applied our previously established method combining density functional theory and a genetic algorithm to the non-empirical structure determination of the catalyst. The result is an unprecedented view on the internal donor-induced support reconstruction at a molecular level, that shows how donor’s adsorption preferences and capabilities cause a drastic increase in the stereospecific Ti species and dictate the diversity in the Ti local environment, possible cause of the widely known effect of donors on the molecular weight distribution.
2024.01.12 by 谷池 俊明 Toshiaki TaniikeThere is a growing interest in leveraging machine learning to accelerate research and development of practical materials such as catalysts. This involves the use of data to train machines and variables (descriptors) to predict the functions of materials. Particularly, for accurate prediction, descriptors that efficiently and comprehensively incorporate factors influencing the functions are essential. We developed an automatic feature engineering technique that works with only a small amount of training data without requiring any prior knowledge. It generates and screens a large number of features (hypotheses) to select descriptors, essentially serving as hypothesis screening. The technique outperformed traditional techniques in the prediction accuracy, regardless of target catalysis, and was able to pinpoint various high-performing catalysts from a vast pool of catalysts.