Publication

Abstract : Lithium-ion (Li-ion) batteries form the backbone of systems that power most modern electronic devices, most notably electric vehicles. However, they are vulnerable to issues such as thermal runaway and capacity degradation. Thus, early anomaly detection is of paramount importance to ensure safe and efficient operation. The detection of abnormal behaviors in Lithium-ion batteries, while also focusing on enhancing the model's accuracy and reliability is vital. Therefore, this paper aims to enhance and evaluate advanced anomaly detection systems for Li-ion batteries by integrating deep learning models with conventional approaches. By making use of discharge data obtained from NASA's dataset, key features such as voltage, current, temperature, and capacity were analyzed. Deep learning models such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks, combined with Isolation Forest and Local Outlier Factor (LOF) algorithms, are used to improve the anomaly detection process. Of all the four aforementioned models, the LSTM - Isolation Forest ensemble model achieved the highest accuracy of 91.26 %, which validates the effectiveness of this integrated approach. Future research could potentially aim at enhancing system robustness through real-time data processing and further refinement of models across diverse datasets.