Computational Exploration and Experimental Verification for Designing SF6 Alternatives

• Published in: IEEE transactions on dielectrics and electrical insulation 2024-08, p.1-1
• Main Authors: Shimakawa, Hajime, Umemoto, Takahiro, Kumada, Akiko, Sato, Masahiro
• Format: Article
• Language: English
• Subjects: alternative gas; Climate change; Computational modeling; Dielectric breakdown; dielectric breakdown strength; Gas insulation; Global warming; global warming potential; Insulation testing; Machine learning; molecular exploration; Predictive models; quantum chemical calculation; Quantum chemistry; SF<sub xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">6; Training data
• ISSN: 1070-9878
• DOI: 10.1109/TDEI.2024.3446953

There have been numerous experimental efforts in developing SF 6 alternatives. However, promising candidates have not been identified due to the trade-off relationship among multiple requirements necessary for eco-friendly insulating gases. This study presents a computational molecular exploration and experimental verification for identifying potential SF 6 alternatives. We propose machine learning models based on quantum mechanical insights to realize extrapolative prediction of gas properties. The evaluation results demonstrate the superior reliability of the proposed models compared to existing ones, while also achieving high-throughput screening. The molecular exploration is systematically conducted to narrow down candidates that pose low global warming potential and superior insulation strength under low-temperature and high-pressure conditions. The screening results reveal new candidates with enhanced insulation performance and favorable environmental impact compared to the existing candidates, including C 4 F 7 N, C 5 F 10 O, and CF 3 I. Furthermore, we newly measure the breakdown strength of a gas material for which no experimental data are available, and whose structure is extrapolative relative to the training data. The experimental result matches the prediction well, serving as an example that indicates the extrapolative capability of the proposed model. Our methodology and exploration results contributes to a wide range of material design as well as SF 6 alternatives.