About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
Energy Materials for Sustainable Development
|
| Presentation Title |
Machine Learning Pathways To High Thermoelectric Figure-Of-Merit |
| Author(s) |
Joseph S. Poon, C. T. Ma, Wenjie Li, Bed Poudel |
| On-Site Speaker (Planned) |
Joseph S. Poon |
| Abstract Scope |
Thermoelectric (TE) materials convert waste heat to electricity with minimal environmental footprints. TE alloys with a high figure of merit are desirable. Compositionally complex alloy (CCA) approach offers enormous flexibility to attain exceptional TE properties. Promising alloys identified from high-throughput machine learning (ML) screening is an essential component of closed-loop studies for discovery high-performance materials with the vision of achieving a conversion efficiency ~20%. We employ active/deep learning to navigate the CCA landscape, extending exploration beyond existing domains, to achieve synthesizability and desirable TE properties. The complexity of ML models involving more than one hundred features and thousands of data points are often time intensive, requiring several days to train a given model even with the most advanced ML. We demonstrate how quantum annealing (QA) can lead to significant speedup, by three to four orders of magnitude. QA classification and regression models are being developed, the findings will be shown.
|