Reducing the Capacity Loss of Lithium-Ion Batteries with Machine Learning in Real-Time—A Study Case

Abstract

Lithium-ion batteries (LIBs) are the state-of-the-art technology for energy storage systems. LIBs can store energy for longer, with higher density and power capacity than other technologies. Despite that, they are sensitive to abuses and failures. If the battery management system (BMS) operates incorrectly or some anomalies appear, performance and security issues can be observed in LIBs. BMSs are also hard-programmed, have complex circuits, and have low computational resources, which limit the use of prognoses and diagnoses systems operating in real-time and embedded in the vehicle. Therefore, some technologies, such as edge and cloud computing, data-driven approaches, and machine learning (ML) models, can be applied to help the BMS manage the LIBs. Therefore, this work presents an edge–cloud computing system composed of two ML approaches (anomaly detection and failure classification) to identify the abuses in the LIBs in real-time. To validate the work, 36 NMC cells with a nominal capacity of 2200 mAh and voltage of 3.7 V were used to build the experiments segmented into three steps. Firstly, 12 experiments under failures were realized, which resulted in a high capacity loss. Then, the data were used to build both ML models. In the second step, the anomaly approach was applied to 12 cells observing the cells’ temperature anomalies. Then, the combination of IF and RF was applied to another 12 cells. The IF could reduce the capacity loss by about 45% when multiple abuses were applied to the cells. Despite that, this approach could not avoid some failures, such as overdischarging. Conversely, combining IF and RF could significantly reduce the capacity loss by 91% for the multiple abuses. The results concluded that ML could help the BMS identify failures in the first stage and reduce the capacity loss in LIBs.

Studying Abuse Testing on Lithium-Ion Battery Packaging for Energy Storage Systems

“…As observed in Figure 8a, the cell's temperature under abuse rose significantly compared with the others and achieved about 36 °C. Despite that, the temperature of cells #1 and #3 increased to 32 °C (normal behavior for cells under discharging process as observed in the study of Gotz et al [34]). Thus, the combination of voltage and temperature should be used to identify OD failures in this type of configuration.…”

“…The Thus, it is possible to state that observing the cell's voltage is necessary to identify and locate the cell under OC at different configurations. Under OC, the temperature takes a long time to increase, as observed by Gotz et al [34], and the current can not indicate OC abuse in the cells.…”

“…During this failure, their temperature rose to about 30 • C, and the temperature of the other cells remained close to the controlled ambient (reference value). The study of Gotz et al [34] shows that the temperature increases more when the SOC decreases. Thus, as the voltage of cells #1 and #2 were above 3.5 V, their temperature remained close to the reference.…”

“…The reference used is the threshold equal to 4.2 V. Furthermore, the temperatures remained slightly during the OC. This happens because, according to Gotz et al [34], a cell needs to take a long time under OC to its temperature rise. Despite that, the reference temperature is about 25 °C.…”

Random Forest-Based Grouping for Accurate SOH Estimation in Second-Life Batteries

“…Lithium-ion batteries (LIB) are the most common type of storage system adopted these days due to the several advantages offered by this technology, such as long lifetime, high density and capacity, and others [16,17]. However, it is sensitive to failures, and unique management systems are required to maintain the cells under safe conditions during their operation [16][17][18]. Therefore, SLBs need more robustness control due to the unique and particular characteristics that a pack of them will show, with cells having different capacities and resistances [11,19,20].…”