Uncertainty characterization of a CNN method for Lithium-Ion Batteries state of charge estimation using EIS data

Estimating the state of charge of batteries is a critical task for every battery-powered device. In this work, we propose a machine learning approach based on electrochemical impedance spectroscopy and 2D convolutional neural networks. A case study based on Samsung ICR18650-26J lithium ion batteries is also presented and discussed in detail. An application-specific data augmentation technique is developed and applied. The proposed system achieves a classification accuracy of 93% on a test dataset of new measurements from the same battery and 88% accuracy on a different battery without prior calibration. The uncertainty of the state of charge classification provided by the proposed method is evaluated using Monte Carlo Simulations and Monte Carlo dropout methods.