With the growth of electric vehicles and stationary energy storage, accurately assessing battery suitability for second-life applications is essential. This study presents a machine learning approach
that utilizes clustering techniques to identify aging mechanisms and determine previous usage conditions, offering a robust framework for second-life evaluation. Traditional assessments of state of health (SOH), which rely on indicators like capacity decrease or resistance increase, often fall short in predicting remaining useful life (RUL) because they cannot account for past usage conditions and the variety of aging mechanisms that occur within the battery. This novel approach leverages standardized check-up data to identify features that reveal previous usage patterns, such as temperature exposure, mean State of Charge (SOC), and Depth of Discharge (DOD) to assess a battery’s second-life potential. This work focus on Exploratory Data Analysis (EDA) for aging prediction. EDA plays a pivotal role in identifying key patterns and extracting meaningful features from large datasets, which are essential for accurate modelling of battery aging. By analysing an extensive aging dataset and preprocessing variables derived from Incremental Capacity Analysis (ICA), and other specific transformations, we demonstrate that previous usage conditions can be predicted by using a machine learning clustering algorithm: Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP). At BatterieIngenieure, we utilize our extensive battery cycling capabilities and databases to conduct in-depth EDA, providing insights into aging behaviour under various operating conditions. Through
advanced clustering methods, we aim to classify aging states and uncover hidden relationships within the data. This approach enables us to develop more sophisticated aging prediction models that adapt
to real-world conditions, ensuring more reliable battery management.