Deep Learning Based 3D Resolution Recovery for Breakthrough Imaging of Electrochemical Energy Devices

Many energy materials can only be understood within the context of the (often sealed) devices they make up. Batteries and fuel cells contain intricate mixtures of materials with features and arrangements that span across many orders of magnitude in size and operate under non-ambient conditions. In these devices, the microstructures of the constituent materials and how they are arranged with respect to each other is directly linked to the device performance and degradation behaviors. Complicating matters, opening the devices for inspection and analysis can irreversibly alter the very microstructures in question and can pose safety hazards to researchers if not done correctly. Furthermore, critical material and device properties such as tortuosity or ion transport pathways can only be accurately quantified in 3D. As such, observing microstructures and their evolution in 3D and in their native state is essential for developing a robust understanding of how to engineer new battery and fuel cell materials and architectures with improved performance. X-ray microscopy (XRM) provides a unique method for observing these microstructures in 3D at high resolutions without needing to open them. However, microscopy techniques like XRM make tradeoffs between FOV and resolution, which can pose challenges for strongly multiscale systems like batteries and fuel cells. For example, if images are acquired at high enough resolution to observe critical features like electrode particle sizes in lithium-ion batteries or gas diffusion layer fiber weaves, catalyst distributions, and microporous layer cracks in polymer electrolyte fuel cells, the field of view of that image is restricted to a small local region of the sample. This means that broader defect distributions or larger scale inhomogeneities will be missed. Additionally, for quantitative analysis applications, this means that statistical relevance will be sacrificed because of the limited analysis volume at the appropriate resolution. And for computational modelling efforts, device level phenomena will not be captured well due to the restricted field of view. Advances in modern AI-based resolution recovery methods combined with the non-destructive multiscale imaging capabilities of X-ray microscopy have allowed researchers to overcome this dependency between resolution and field of view to produce 3D images with both high resolution and large field of view. We present here examples of applying these methods to applications in lithium-ion battery and polymer electrolyte fuel cell research 1. 1 Wang, Y.D., Meyer, Q., Tang, K. et al. Large-scale physically accurate modelling of real proton exchange membrane fuel cell with deep learning. Nat Commun 14, 745 (2023).