Calendar-based RuL prediction for batteries: A data-driven approach using IoT device utilization data

Abstract

Accurately predicting the Remaining Useful Life (RuL) of a battery is essential for effective maintenance scheduling and proactive replacement to avoid costly and hazardous outages. Traditional RuL predictions focus on remaining charging cycles, which do not accurately represent real-world usage where calendar time is a more relevant metric, especially for knowing when the battery will reach End of Life (EoL). We propose an innovative data-driven RuL estimation method that predicts battery life in calendar months instead of charging cycles. Our approach leverages low-frequency utilization data from IoT devices, without the need for additional internal sensors and enabling seamless integration with existing IoT platforms. Tested on a proprietary battery dataset, our method achieves higher RuL prediction accuracy compared to current models. To illustrate the benefits of our solution, we put it in the context of the automotive industry with a prominent use case of IoT battery management systems in Electric Vehicles (EVs). We propose an application of our RuL method for battery leasing contract optimization. The model shifts the uncertainty of battery performance and longevity from EV owners to leasing companies, highlighting the necessity for efficient battery stock management as the leasing market grows. Our method addresses key challenges for leasing companies, such as fixed leasing durations and post-lease battery reallocation. Although demonstrated through EV battery leasing, our method is versatile and applicable to various battery-dependent sectors, including small-scale IoT devices, laptops, and heavy machinery.