We presented here two cases of data-driven electrolyte additive design, using two cathodes as examples. In scenario 1, LiNi0.5Mn1.5O4 (LNMO), a spinel-structured material with an average lithiation/de-lithiation potential at ca. 4.6-4.7 V, was used as an example for the extreme high working potentials and the resultant strains on electrolytes. In this study, we first selected and tested a diverse collection of 28 single and dual additives for the LNMO//Gr system using descriptors representative of chemical functionality.(1) Subsequently, we trained and employed machine learning (ML) models to suggest 6 binary compositions out of 125 based on predicted final area specific impedance (ASI), impedance rise (ΔASI), and final specific capacity(Q). Notably, this approach led to the discovery of a new dual additive which outperforms the initial dataset.In Scenario 2, the AI-guided workflow undergoes further enhancement for accelerated additive optimization employing new data featuring an earth-abundant cathode material of 0.3Li2MnO3·0.7LiMn0.5Ni0.5O2. This phase of the study introduces a more robust machine learning (ML) model for performance prediction, utilizing Figure of Merit Energy (FOME) and Figure of Merit Power (FOMP) as predictive metrics. Following three , Bayesian optimization iterations, 15 out of 78,000 binary and tertiary additive combinations was suggested for experiments, and several novel addtive compositions exhibiting unexpectedly high performance were subsequently identified. The optimal formulation surpasses the current standard by achieving a performance enhancement of 15-19%, as confirmed through experimental testing conducted in coin cells.Overall, our findings not only underscores the efficacy of ML in identifying new additives combinations, but also introduces an accelerated material discovery workflow that directly integrates data-drive methods with battery testing experiments.(1) ACS Omega 2018, 3, 7, 7868–7874
Read full abstract