Inverse design of lithium-ion battery microstructures using physics-constrained generative AI

The electrochemical performance of lithium-ion batteries is strongly governed by the complex three-dimensional architecture of porous electrodes. While recent advances in generative artificial intelligence have enabled realistic reconstruction of battery microstructures, existing approaches primarily focus on reproducing morphology or correlating manufacturing parameters with structural characteristics, providing limited control over transport-relevant properties. In this work, we develop a physics-constrained conditional generative adversarial network (PC-cGAN) for the inverse generation of LiNi 0.5 Mn 0.3 Co 0.2 O 2 , NMC532 cathode microstructures conditioned on target physical descriptors. The framework integrates a physical constraint network that enforces porosity, active-material volume fraction, and tortuosity during training, enabling direct generation of microstructures consistent with desired transport characteristics. Beyond morphology reconstruction, the framework also supports inverse microstructure design by mapping user-specified transport properties directly into three-dimensional, physically consistent electrode architectures. This functionality opens a pathway to probe structure-transport relationships that are otherwise challenging to isolate experimentally, while laying the groundwork for AI-enabled digital twins of battery electrodes. Coupling the framework with electrochemical and manufacturing simulations will enable virtual screening of architectures, accelerate optimization cycles, and advance physics-guided battery design with reduced experimental effort.