The 8th International Conference on System Reliability and Safety
Sicily, Italy - November 20-22, 2024

Ahmed Shokry, CMAP, Ecole Polytechnique, France

Dr. Ahmed Shokry Abdelaleem is a research engineer at the Centre for Applied mathematics (CMAP), Ecole Polytechnique -L’x, Palaiseau, France. In the CMAP, he contributes to the chair of “Data Science for Process Industry” to improve control and monitoring of industrial and chemical processes. He also participates in the Chair of "AI and Predictive Maintenance" to develop machine learning-based methods to detect early signs of degradation in Lithium-ion batteries.
Dr. Shokry holds a MSc and a PhD in process engineering from Polytechnic University of Catalonia, Barcelona, Spain, and achieved his post-doctoral studies in Prognostics and Health Management of energy production systems at the Polytechnic University of Milan, Italy. He received the 2022 excellence award in computer aided process engineering form the European federation of chemical engineering (EFCE).

Speech Title: Machine Learning-based Methods for Predicting the State of Health of Lithium-ion Batteries Using Non-cyclic Operational Data

Abstract: To maximize availability, ensure safety, and minimize failures during the use of Lithium-ion (Li-ion) batteries, the accurate prediction of the state of health (SOH) of these batteries is essential. Machine learning (ML) techniques have garnered significant interest from both academia and industry to achieve this objective. This interest arises from their independence from the need to comprehend complex degradation phenomena and sophisticated aging mechanisms, as well as their capacity to leverage vast amounts of data collected from batteries at a low cost, thanks to significant advancements in sensing technologies and battery management systems (BMS).
Over the past two decades, numerous ML-based methods have been developed to predict the SOH of Li-ion batteries. However, most of these methods have been designed and adapted using standardized datasets generated in fully controlled laboratory environments, often under ideal loading conditions or tests. As a result, these ML-based approaches tend to be much less effective for monitoring the SOH of batteries in real-world operations, where operating logic and loading conditions differ significantly from those in the lab.

This work presents a novel approach for predicting the SOH of rechargeable batteries, specifically tailored to align with real-world data and actual operational requirements. It introduces a new feature extraction framework capable of mining relevant information from non-cyclic and random operating conditions, relying on incremental analysis of all events experienced by the battery throughout its lifetime. Additionally, feature selection techniques are employed to identify the most significant features for SOH assessment. Finally, machine learning models, including artificial neural networks and Gaussian process (GP) models, are trained using these selected features to predict the degradation profiles of new batteries given certain operational conditions. The method is applied to benchmark datasets, demonstrating accurate and robust performance.