As the world transitions to a renewable energy economy, the need for efficient, sustainable energy storage solutions is becoming increasingly important. Advancements in battery technology that prioritize durability, sustainability, safety, and cost-effectiveness, advancements in battery technology can help to overcome the challenges that lie ahead.
Sodium-based batteries (SBBs), which offer a promising alternative to traditional Li-based counterparts, show potential advantages in terms of cost, availability of raw materials, and reduced environmental impact.1–3 By leveraging the natural abundance of sodium, SBBs could provide a more scalable and sustainable solution for energy storage, helping to mitigate the resource constraints and ecological concerns associated with lithium-ion batteries. Despite their potential, sodium-based batteries face significant technical hurdles. Ongoing research and development are needed to overcome these challenges. If successful, SBBs could become a game-changing technology for large-scale energy storage, enabling faster adoption of renewable energy sources.4,5
In this Study, we establish a high-throughput, data-driven workflow to model non-aqueous sodium-based electrolyte formulations using batch-wise active learning. Unlike closed-loop studies, which aim to identify a single optimal composition via Bayesian optimization, the central objective was to construct a predictive, physically meaningful surrogate model first. The model covers the full space of feasible electrolyte formulation, including sub-optimal regions, and thereby is able to quantify the influence of each individual components on the ionic conductivity across different temperatures.
To enable efficient modeling, a novel sodium-based electrolyte dataset was created using a method compatible with high-throughput experimentation (HTE), resulting in a model with a reduced number of experimental trials and active learning iterations. Electrolyte compositions based on NaPF6 in mixtures of ethylene carbonate (EC), propylene carbonate (PC), and ethyl methyl carbonate (EMC) are investigated between –20 °C and 60 °C as a well-defined proof-of-concept system. The selected formulation space can be reliably described using only 72 different electrolyte compositions acquired in two active learning iterations, corresponding to a reduction in required experiments by a factor of 4–7 compared to random sampling. Obtained experimental results were bundled with required metadata in an in-house developed .json file format for each individual experiment. Overall, this work establishes a transferable, physics-informed framework for data-efficient electrolyte exploration.
References:
1. N. Tapia-Ruiz et al., J. Phys. Energy, 3, 031503 (2021).
2. N. Aslfattahi et al., Journal of Energy Storage, 72, 108781 (2023).
3. Y. Li et al., Chemical Society Reviews, 51, 4484–4536 (2022).
4. E. Goikolea et al., Advanced Energy Materials, 10, 2002055 (2020).
5. C. Delmas, Advanced Energy Materials, 8, 1703137 (2018).