Deep learning framework for state of health estimation of NMC and LFP Li-ion batteries for vehicular applications

Abstract

This paper proposes a data-driven estimator of capacity and energy variations of lithium-ion batteries for health monitoring purposes. The proposal uses voltage and temperature measurements as inputs to multi-layer perceptron auto-encoders to extract features and reduce their dimensionality. Then, a long-short-term memory neural network estimates capacity and energy from these features. This machine learning-based estimation framework is trained and validated with experimental ageing data for two different batteries' chemistries: lithium‑nickel‑manganese‑cobalt-oxide/graphite (NMC) and lithium‑iron-phosphate/graphite (LFP). The cells are cycled at 35 °C with a current profile adapted from the Worldwide Harmonized Light Vehicles Test Cycle. The training process is optimized to achieve high estimation accuracy and robustness to cell dispersion. The validation results show that the mean absolute percentage error is lower than 1.6 % for capacity estimation and 3.6 % for energy estimation for both battery chemistries.