Efficient battery electrolyte design for electric aircraft driven by physics-augmented AI

Rapid discovery of optimal materials from vast design spaces is essential to accelerating technological breakthroughs, such as the design of high-performance battery electrolytes for electric aircraft. Conventional trial-and-error and purely data-driven design approaches struggle with high-dimensional electrolyte composition spaces. Here, we present a physics-augmented machine learning framework that integrates a data-efficient model to predict ionic conductivity and the diffusion coefficient for enhancing the efficiency and interpretability of electrolyte design. Validated on experimental datasets across diverse electrolyte chemistries, our model reduces property-prediction errors to below 10% while requiring 45% less training data than conventional techniques. The design framework, augmented by accurate property predictions, achieves an approximately 4-fold acceleration in discovering the optimal electrolyte and enriches electrolyte design strategies through interpretable decision-making for demanding battery operations. Overall, this work offers a generalizable paradigm for efficiently integrating material property representations into AI to accelerate the development of advanced materials for next-generation clean transportation tools.