Data-Driven Prediction of Battery Cycle Life Using (Dis)Charge Cell Temperature

As many researchers in academic and industry fields made great improvements in Li-ion battery (LIB), they have accumulated a lot of data on LIB and consequently made big data. Since the data are the result of battery components, (dis)charge method, and many other conditions, they include many important key factors of the batteries and have the meaning of the battery cycle life. Many researchers have focused on these properties of the battery big data and made a great prediction of battery cycle life utilizing the data itself, especially discharge capacity and internal resistance data. To measure these important properties during the (dis)charge process, researchers need some expensive machine or special technique. On the other hand, the cell temperature is cheap and easy to measure and does not need special techniques. The cell temperature is closely related to the increase of internal resistance during (dis)charge, therefore cell temperature can replace the role of internal resistance. As the charging/discharging process proceeds, which is the movement of the Li-ion in LIB, the Li-ion must overcome some small energy barrier. Lots of Li-ion can overcome this energy barrier during the early cycle, but as the cycle continues, the number of Li-ion that fail to overcome the energy barrier increases. These remaining Li-ion contribute to the resistance of battery cells and consequently make irreversible charge and discharge processes. Since the cell temperature is closely related and can be substitution for the resistance of the battery, cell temperature will be an important data to predict battery cycle life.<br/>In this research, I have used open data of fast-charged lithium iron phosphate (LFP). The charge and discharge process were performed at a constant temperature of 30°C in an environmental chamber. The cell temperature was recorded by stripping a small section of the plastic insulation and contacting the thermocouple to a bare metal casing. The battery cells were charged from 0% to 80% state-of-charge (SOC) using one of 72 different single-step and two-step charging methods. Subsequently, all cells were charged from 80% to 100% SOC using a 1C constant current-constant voltage (CC-CV) charging step, up to 3.6V, with a current cutoff at C/50. The discharge process is same in all cells by CC-CV discharge at 4C to 2.0V with a current cutoff of C/50. By using Machine Learning Interatomic Potential (MLIP), I have found the energy barrier that Li-ion must overcome in LFP. The remaining Li-ion on the cathode makes resistance and affects the cell temperature. Therefore, temperature will be related to the battery cycle life.<br/>Discharge capacity data shows a relation with cycle life by Pearson coefficient. The value of the Pearson coefficient is low during the early cycle, but as the cycle increases, the Pearson coefficient increases and converges to about 0.7, which means that discharge capacity and cycle life are related. However, temperature data shows a low value of the Pearson coefficient during the early cycle, and it decreases as the cycle increases. From these different tendencies of Pearson coefficient value, it is easy to think that the temperature data is useless to predict cycle life. However, through many trials and errors of the statistical data processing, I have made a variance of charge and discharge temperature difference between two cycles, and it showed a great relationship with cycle life. These two features I have found are used to predict cycle life by making a linear regression machine learning model and result in good prediction with high accuracy. This research shows that statistical data processed from temperature can be a promising machine-learning feature even when the temperature data itself are not closely related to battery cycle life.