Abstract

Exploring the vast chemical design space to find better electrode and electrolyte materials is a roadblock to electric mobility and grid scale storage as no laboratory can keep up. While graph neural networks (GNNs) are a promising method for estimating material properties from the atomic structure, limited use of GNNs in battery research is due to three limitations: predictions from GNNs are hard to interpret, they are typically trained for a single target, and they require labelled data which is limited for new chemistries. This paper provides a review of recent research on graph-based learning, explainable artificial intelligence (XAI), data-efficient learning and multi-objective optimisation for battery materials discovery focusing on eight key studies alongside supporting examples from the last five years (2021–2026). literature. It can be seen from the criterion-based comparison made that no single of the reviewed studies covers the topic of representation of the graph, Explainability, multi-objective trade-offs, limited-data learning and experimental validation. Filled with this need, we We provide a well-defined conceptual framework for graph construction, uncertainty-aware GNN prediction, explanation, Pareto-based candidate selection and DFT/experimental verification, and define an evaluation protocol for it. The framework is suggested rather than applied and open challenges and future directions are discussed.

Keywords

How to Cite This Article

Rights