Initial Proof-of-Concept for an Accelerated Galvanostatic Intermittent Titration Technique via Embedded Machine Learning

This work conceptually proves and demonstrates in silico a new method and statistical analysis for the Galvanostatic Intermittent Titration Technique (GITT) that may decrease the experiment time by an order of magnitude. The method utilizes embedded Gaussian processes and focuses on the relaxation portion of the pulse-relaxation series. These innovations enable much higher current draws at each pulse, while shortening relaxation periods. The in silico proof-of-concept was achieved using simulated conditions similar to a published GITT experiment run on a lithium-lanthanum titanate half cell, while a validation of the relaxation-based method was performed on a porous Li ζ Ni 0.92 Co 0.03 Mn 0.05 O 2 cathode.