Overdischarge Detection and Prevention With Temperature Monitoring of Li-Ion Batteries and Linear Regression-Based Machine Learning

Abstract

This work focuses on the use of linear regression analysis-based machine learning for the prediction of the end of discharge of a prismatic Li-ion cell. The cell temperature was recorded during the cycling of Li-ion cells and the relation between the open circuit voltage (OCV) and cell temperature was used in the development of the linear regression-based machine learning algorithm. The peak temperature was selected as the indicator of battery end of discharge. A battery management system (BMS) using a pyboard microcontroller was constructed to monitor the temperature of the cell under test and was also used to control a MOSFET that acted as a switch to disconnect the cell from the circuit. The method used an initial 10 charge and discharge cycles at a rate of 1C as the training data, then another charge and discharge cycle for the testing data. During the test cycling, the discharge was continued beyond the cutoff voltage to initiate an overdischarge while the temperature of the cell was continuously monitored. When the temperature of the cell exceeded the predetermined threshold, the pyboard triggered the MOSFET to disconnect the cell and stop the overdischarge. The experiment was performed on three different cells, and the overdischarge for each was secured within 0.1 V of the cutoff voltage. The results of these experiments show that a linear regression-based analysis can be implemented to detect an overdischarge condition of a cell based on the anticipated peak temperature during discharge.