(Keynote) Inferring Degradation Modes from Battery Cycling Data

Battery research and development efforts necessarily involve large-scale experimental studies where cells are cycled under different conditions. Physics based models (PBM) of the type pioneered by John Newman, Ralph White, and others can play a key role in interpreting the resulting cycling data, and of course can be used to improve performance by modifications of cell design to optimize for a specific application. Long-term studies that characterize degradation in cell performance are an important type of cycling experiment that can be used to estimate useful battery life in a real-world application. The establishment of correlations between duty cycle and life is valuable and can inform battery management systems with a goal to extend life. During the course of such experiments, large data sets are created and we explore how a PBM might be used to infer the modes of cathode degradation. From degradation modes, mechanisms of degradations might be hypothesized, and this may lead to insights to further improve design and operation. Here, two PBM are reformulated to determine how cell parameters such as the amount of active cathode material appears to change with cycle number. Parameter estimation is applied to cycling data, and parameter changes are tracked for cycling data at different C-rates and different cutoff voltages. Parameter estimates are challenged with complementary experimental data for NMC and for a second cathode material that has been characterized extensively. The comparisons suggest that the methodology is effective and can provide physical insights into degradation mechanisms. To facilitate the “workflow,” inverse surrogate machine-learning techniques, in which simulation data are used for training neural networks to rapidly estimate parameters from experimental data, were utilized. To allow for both imperfect PBM and for non-ideal experimental data, data augmentation methods were introduced to improve the accuracy of parameter estimation algorithms. Examples are shown, with an emphasis on the physics over the mathematics and with an eye towards next steps.