Predicting Cycle Life of NMC Cells by Discharge Capacity Voltage Curves

Abstract

The biggest issue with rechargeable batteries is arguably their limited lifetime. They suffer from capacity degradation and power fade, and their performance decreases as they age. Estimating the remaining useful life is therefore an important task. However, the complex internal aging mechanisms are difficult to model. Recently, machine learning has become a promising approach for predicting remaining useful life. This thesis evaluates whether a new elastic net machine learning model trained on data from LFP cells can be used to predict cycle life of NMC cells. The model uses capacity and voltage data during discharge phases to derive a feature highly correlated to cycle life. Four commercial NMC cells were cycled in Chalmers Electric Power Battery Lab to collect cycling data. The model was able to make useful cycle life predictions for these cells, which suggests that the approach is applicable to other lithium-ion cells.