The generalisation challenge: Assessment of the efficacy of acoustic signals for state estimation of lithium-ion batteries via machine learning

Acoustic measurements of batteries are known to be correlated to their state-of-charge and can be practicably leveraged for state estimation using parametric machine learning models. Such models can be easily designed to have millions of tuneable parameters, which endows them with tremendous but often misinterpreted fitting ability. The real performance metric, commonly omitted in the battery literature, is a model’s generalisation performance with respect to a population, which requires successful predictions to be made on data from one or more ‘held out’ cells. This study demonstrates that regression models based on neural networks can perform highly accurate state estimation on multiple cells using solely acoustic signals. However, this is shown to be conditional on all cells being represented in the training dataset, and generalisation to the wider population is more challenging than other studies claim. The explainability of this model behaviour is examined using complementary classification and unsupervised learning tests, which suggest that success on multi-cell data in the absence of wider generalisation is due to the propensity of models to learn cell-specific patterns implicitly. This represents an elusive type of ‘overfitting’, and highlights the need for caution when assessing the generalisation of machine learning models in battery research. • Caution required in lithium-ion battery state-of-charge estimation from acoustics. • Machine learning models tend to learn cell-specific patterns implicitly. • Explainable insights extracted by complementing regression with other ML tasks. • Generalisation to the wider population more challenging than previously claimed.