Semantic Multi-Modal Framework for Early Stage Anomaly Detection in Lithium-Ion Batteries.

In the era of intelligent energy systems, lithium-ion batteries play a pivotal role in electric mobility and large scale energy storage. Ensuring their safe and reliable operation requires advanced anomaly detection methods capable of interpreting complex and heterogeneous data streams. Conventional monitoring approaches, however, struggle to integrate multimodal information and contextual knowledge in real time, limiting their effectiveness for early detection of safety critical events. To address this challenge, we present a novel ontology-based framework for anomaly detection in lithium ion batteries. The framework combines heterogeneous data sources, including time series sensor measurements and thermal imaging, within a modular ontology that encodes battery processes, safety thresholds, and causal dependencies. Stream reasoning is performed through C-SPARQL queries, enabling continuous analysis of evolving data and the identification of anomalies when semantic constraints are violated or precursors of hazardous conditions are observed. The proposed approach is validated on case studies of overheating and thermal imbalance, demonstrating improved detection and providing interpretable explanations beyond conventional threshold monitoring. These results highlight the potential of combining symbolic knowledge representation with real time data analytics to support transparent and scalable battery management in safety critical applications.