Battery Health Index: Combination of Physical and ML-Based SoH for Continuous Health Tracking

Small devices, such as drills, are increasingly being equipped with Internet of Things (IoT) functions that make it possible to collect usage data, adapt the way they work, and also gain insights into the further development of the devices; even in application areas with limited battery capacity and noncontinuous Internet connection. Since the health of the battery is a crucial factor in the successful long-term deployment of these IoT devices, tracking their capacity-based State of Health (SoH-C) is important to avoid outages. To preserve energy, utilization data is aggregated and sent when an event occurs (e.g., charging). To avoid the need to introduce expensive intrinsic battery tracking sensors as done in large-scale IoT devices, this article uses the existing capacity tracking sensor of the battery management system (BMS) to track the SoH by applying the peak State-of-Charge (SoC) extraction technique. However, an SoH update can only be achieved and verified when the battery is peak cycled; which does not happen every charge/discharge cycle and also depends on the charging behavior of the customer. As long as the battery is shallow cycled, existing approaches would not update the SoH. To ensure continuous SoH tracking, the novel solution, presented in this article, called “battery health index” (BHI) combines physical capacity-based measurements with data-driven machine learning predictions based on utilization data to provide an always up-to-date SoH. The proposed state-of-the-art method is evaluated on a hand-held battery platform with millions of batteries and it outperforms existing solutions. The presented model enables proactive battery exchange by predicting the remaining useful lifetime (RUL) therefore increasing customer experience.