About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Advanced Materials for Energy Conversion and Storage 2026
|
| Presentation Title |
Data and physics informed neural network for predicting lithium-ion battery electrode materials |
| Author(s) |
Yunhe Mo |
| On-Site Speaker (Planned) |
Yunhe Mo |
| Abstract Scope |
The urgent demand for high-conductivity electrode materials in electric vehicles motivated our development of the Data-Physics-Informed Neural Network (DPINN). This deep learning model integrates physical constraints into neural network loss functions to predict material features from target conductivity metrics. Initial Pearson correlation analysis (|R|<0.7) revealed no linearly predictive input features, necessitating physics-based constraints. Incorporating elemental, structural, and stability rules significantly enhanced performance: Rē reached 0.99 while RMSE decreased by 112.32 S/m. The optimized model successfully identified 14 promising orthorhombic-structured electrode candidates through inverse design. DPINN provides a solution to data scarcity challenges in material development, with applicability extending to other materials discovery domains. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Energy Conversion and Storage, Computational Materials Science & Engineering |