Machine Learning Guided Design of Doped Lithium Lanthanum Zirconium Oxide (LLZO) for Improved Performance

Publication Type

Journal Article

Name of Author

• Bernice Ngwi Abraham, Department of Materials Science and Engineering, African University of Science and Technology, Abuja, NigeriaFollow
• Elvis Twumasi, Department of Electrical and Electronic Engineering, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana
• Desmond Edem Primus Klenam, Global Center of Advanced Materials and Manufacturing (GCAMM), College of Engineering, SUNY Polytechnic Institute, Utica, New York, USA
• Tabiri Kwayie Asumadu Asumadu, Global Center of Advanced Materials and Manufacturing (GCAMM), College of Engineering, SUNY Polytechnic Institute, Utica, New York, USA
• Abdulhakeem Bello, Department of Materials Science and Engineering, African University of Science and Technology, Abuja, Nigeria
• Vitalis Chioh Anye, Department of Materials Science and Engineering, African University of Science and Technology, Abuja, Nigeria
• Nancy A Burnham, Worcester Polytechnic Institute, Worcester,Massachusetts, USA
• Rodica Neamtu, Worcester Polytechnic Institute, Worcester,Massachusetts, USA

Publication Date (Issue Year)

2026

Journal Name

Energy Storage

Abstract

This paper presents a machine-learning approach to the development of all-solid-state batteries (ASSBs) with doped Lithium Lanthanum Zirconium Oxide (LLZO) electrolytes. LLZO electrolytes are promising solid electrolytes that exhibit high ionic conductivities of approximately ~10−3–10−4 S cm−1. However, optimizing the ionic conductivity of LLZO is a multiparametric challenge involving numerous compositional and processing variables. This work employs a data-driven methodology by applying machine learning to a dataset mined from the literature. This approach was used to model the intricate relationships between the compositional and synthesis variables and ionic conductivity. Tree-based and ensemble models, like Decision Tree, Random Forest, Light Gradient Boosting Machine, and CatBoost, are effective for interpreting complex relationships among variables. The models were trained with systematic hyperparameter tuning and rigorous cross-validation to predict ionic conductivity, achieving accuracies from 0.85 to 0.94. Feature importance and partial dependence analyses were used to quantify and visualize variable effects. Across the models, lithium composition and sintering conditions emerged as the most influential factors. Partial dependence plots were used to identify optimal synthesis and compositional ranges. By identifying promising search spaces, this approach can accelerate the design and discovery of high-performance LLZO electrolytes for solid-state batteries.

Keywords

dopants feature importance, lithium lanthanum, zirconium oxide, machine learning materials design partial dependence plots synthes

Rsif Scholar Name

Bernice Ngwi Abraham

Rsif Scholar Nationality

Nigeria

Cohort

Cohort 4

Thematic Area

Minerals, Mining and Materials Engineering

Africa Host University (AHU)

African University of Science and Technology (AUST), Nigeria

Recommended Citation

Abraham, B. N., Twumasi, E., Klenam, D. E., Asumadu, T. K., Bello, A., Anye, V. C., Burnham, N., & Neamtu, R. (2026). Machine Learning Guided Design of Doped Lithium Lanthanum Zirconium Oxide (LLZO) for Improved Performance. Energy Storage, 8 (4) https://doi.org/10.1002/est2.70418