Automated feature extraction to integrate field and laboratory data for aging diagnosis of automotive lithium-ion batteries
Valentin Steininger, Katharina Rumpf, Peter Hüsson, Weihan Li, Dirk Uwe Sauer
Abstract
Battery aging diagnosis using field data readouts presents distinct challenges compared with using laboratory data. These challenges stem from the complexity of the data structure and potential inconsistencies in aging values obtained from variations in battery management system software versions. Consequently, the efficacy of a data-driven approach to identify pertinent aging features from field data becomes susceptible to these factors. In this work, we investigate different feature extraction methods and propose a framework designed to mitigate issues arising from compromised data quality. For this purpose, we leverage the benefits of precise laboratory aging data alongside authentic driving data acquired from a cohort exceeding 600,000 customers to improve the aging diagnosis of vehicle batteries. Moreover, we provide functional fitting of statistical data, addressing the challenges posed by incomplete data structures. We validate our methods by comparing them with state-of-the-art feature extraction techniques, yielding a 57% enhancement in aging estimation accuracy.