A random forest mechanism to identify the initial conditions of external short circuits in 18650 lithium-ion batteries

Lithium-ion battery (LIB) is the mainstream energy storage technology (ESS) technology in this market, mainly because it has several advantages such as long lifetime, high density and capacity, and low self-discharging. Despite that, LIB is still sensitive to failures, and if it is not well managed, several types of abuse can be observed and cause performance and security issues. Therefore, it is essential to understand the main abuses, their causes, consequences, and how they happen to prevent them. Thus, this paper presents a contribution of two steps: firstly, it demonstrates the study of five applications of external short-circuit (ESC) experiments in 18650 LIB. Then, a random forest mechanism was applied to classify the conditions that determine the intensity of the consequences of the ESC. In the first part, the following experiments have been performed: (I) varying initial voltage (from 3.5 to 4.2 V), (II) changing the time between ESC with a relaxing time (2, 10, 20, 30, and 60 s), (III) varying capacity of the cell (20 mAh, 400 mAh, 940 mAh, 1202 mAh, and 1750 mA), (IV) varying external resistance (from 50 to 250 m\(\Omega \) with 50 m\(\Omega \) step), and (V) varying the ambient temperature (30 \(^{\circ }\)C, 40 \(^{\circ }\)C, 50 \(^{\circ }\)C, 60 \(^{\circ }\)C, and 70 \(^{\circ }\)C). The results indicate that the ESC current curve comprises four stages. The temperature increases significantly during the high current flow in the cell. In addition, the external resistance, the time of the ESC, the ambient temperature, the cell’s capacity, and the state of charge (SOC) play a vital role in the ESC’s intensity and the ESC current’s magnitude. The cell current is shown to be the main parameter used for ESC prevention mechanisms because it represents a similar behavior for almost every cause of ESC. Despite that, this work presents different magnitudes of the current curve depending on the causes and criticality of the ESC. Therefore, the information and expertise collected from the experiments can be used for machine learning prevention mechanisms to monitor battery abuses and failures in the first stage without the demand for new sensors and hardware, which is the second contribution of this work. It consists of applying a random forest mechanism to identify the causes/conditions of the ESC based on the main signals collected from the batteries. The results indicated that the proposed model can estimate the initial conditions of the ESC up to 0.99 of R2.