Using Machine Learning and In Situ Microscopy to Track Microbubble Dynamics during Electrolysis

Gas-evolving electrochemical reactions suffer performance losses due to bubble formation. However, studying the behaviors of these bubbles on industrially relevant porous transport layer (PTL) electrodes under operating conditions remains a challenge. Using a Pt ultramicroelectrode as a model system, a lightweight convolutional neural network was employed to systematically track microbubble dynamics. Using this tracking as an optical baseline, we demonstrated that a simple frequency-domain analysis of the voltage signal reliably captured the same characteristic bubble residence times as those optically extracted. This frequency analysis approach was then translated to complex commercial electrode geometries, including gauzes, foams, and felts, operating at industrial current densities. Because these electrodes introduce compounding multi-bubble dynamics, we utilized a weighted average of the frequency spectra to extract a lumped characteristic time representing the overall voltage fluctuations on the electrode. In this work, by bridging the gap between idealized model electrodes and practical PTLs, we provide a generalizable operando tool for the extraction of meaningful bubble-related metrics during electrolysis from electrochemical voltage noise.