This repository hosts the scripts and AI-driven tools developed as part of our innovative study on lithium brine softening. Our research introduces an AI-enabled, active learning high-throughput method that integrates carbon dioxide (CO2(g)) capture into the conversion of lithium chloride (LiCl) to lithium carbonate (Li2CO3). This approach not only streamlines the production process but also enhances the sustainability of battery manufacturing by significantly reducing carbon emissions. The materials within include the developed machine learning models, data, and key findings that underline our advancements in efficient and eco-friendly battery material production.
Utilization of CO2(g) in the lithium softening reaction.
Active learning-driven high-throughput experimentation methodology.
Focus on elemental concentrations of carbon, nitrogen, and lithium for process optimization. Enhanced sustainability in battery-grade lithium production.
Repository Structure
- data/: Contains datasets used and generated during the study.
- docs/: Additional documentation including detailed methodology, supplementary material, and figures.
- paper/: The final version of the journal paper and any draft versions.
To engage with this project:
-
Load and Preprocess Data
- Begin with the
0_Load_Clean_Data.ipynbnotebook to load and preprocess the data.
- Begin with the
-
Explore Active Learning Methodologies
- Proceed to the
1_Active_Learning_Cycle.ipynbnotebook to explore the active learning methodologies applied in this research.
- Proceed to the
-
Reproduce Paper Figures and Analyses
- Utilize the
2_Paper_Reproduction.ipynbnotebook to replicate all paper figures, analyses, and supplementary information.
- Utilize the
-
Utilize Trained Models for Prediction
- The trained GPR models, including feature standard scalers and GPR models for each of the active learning iterations, are available in the
/modelsdirectory in.pklformat. These models can be directly loaded and used for predictions without the need for retraining. Instructions for loading and using these models are provided within the notebooks, ensuring that users can easily integrate these models into their work.
- The trained GPR models, including feature standard scalers and GPR models for each of the active learning iterations, are available in the
If you use the data, code, or findings from this study in your work, please cite our paper as follows:
If you use the data, code, or findings from this study in your work, please cite our paper as follows:
Masouleh, S. S., Sanz, C. A., Jansonius, R. P., Shi, S., Romero, M. J. G., Hein, J. E., & Hattrick-Simpers, J. (2024). Artificial Intelligence-Enabled Optimization of Battery-Grade Lithium Carbonate Production. arXiv preprint arXiv:2402.07000.
@article{Mousavi2024lithium, title={Artificial Intelligence-Enabled Optimization of Battery-Grade Lithium Carbonate Production}, author={S. Shayan Mousavi Masouleh, Corey A. Sanz, Ryan P. Jansonius, Samuel Shi, Maria J. Gendron Romero, Jason E. Hein, Jason Hattrick-Simpers}, journal={arXiv preprint arXiv:2402.07000}, year={2024}, url={https://doi.org/10.48550/arXiv.2402.07000} }
We gratefully acknowledge support from the Simons Foundation, member institutions, and all contributors.
This project is licensed under the MIT License.