About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Late News Poster Session
|
| Presentation Title |
H-32: Accelerating Solid Electrolyte Discovery with an Interpretable Multimodal Machine Learning Framework |
| Author(s) |
Qingjie Li, Joshua Young, Taku Watanabe |
| On-Site Speaker (Planned) |
Qingjie Li |
| Abstract Scope |
High-throughput screening of solid electrolytes is challenged by ionic conductivity's complex nature, which emerges from a balance of structural and dynamic factors. Single-modality machine learning models often fail to capture this physics, limiting their predictive power and impeding materials discovery. We present a multimodal deep learning framework to overcome this barrier. It integrates diverse, physically-grounded descriptors for long-range crystal structure, local coordination, and vibrational dynamics into one model. The architecture is designed to learn the complex relationships between these data streams, constructing a more comprehensive and physically-accurate materials representation. Our results suggest this data-fusion strategy enhances predictive accuracy over single-modality methods and yields interpretable insights into features driving ion mobility. This work offers a powerful, data-driven tool to accelerate solid electrolyte design by elucidating complex structure-property relationships. |
| Proceedings Inclusion? |
Undecided |
| Keywords |
Energy Conversion and Storage, Computational Materials Science & Engineering, Machine Learning |