Metal anode batteries have high specific energy and are the key component towards electrification and energy sustainability. Still, they also face challenges such as dendrite formation and side reactions, causing capacity loss and inefficiency. This project aims to develop a high-throughput optical electrochemical data acquisition platform and a multi-fidelity active learning algorithm to discover materials and protocols for high-performance metal anode batteries. By integrating operando optical characterization data as the intermediate input, the surrogate model training can be faster and more accurate and contain more physical meaning for outcome interpretation and knowledge translation. This project will construct optical coin cells and use Raman microscopy and Fourier plane imaging to gather topographical and chemical information. Multi-fidelity active learning framework optimizes not only the battery performance but the experimental procedure itself, reducing the time and resources needed for battery material discovery.
“Machine learning can help us find better battery materials much faster, but it is still time consuming to gather enough detailed cell-level test data for computers to make reliable predictions. This project aims to develop operando optical microscopy methods to quantitatively detect trace amounts of battery degradation to enhance data acquisition speed, machine learning algorithm accuracy, and physical interpretability, enabling us to develop batteries that store more energy, last longer, and are safer to use.”
Po-Chun Hsu, Assistant Professor of Molecular Engineering, UChicago Pritzker School of Molecular Engineering
