Data Mining for Early Cycle Life Prediction in Lithium-Ion Battery Production

The efficient prediction of product quality is a major challenge in lithium-ion battery production, as conventional measures such as aging are time-consuming and costly. This study presents a comprehensive data mining approach to predict the quality of lithium-ion batteries using linear and non-linear support vector machines. A methodology for extracting and selecting features from data sources within production is presented, and several feature selection algorithms - as well as models - are compared with regard to their predictive power. A minimum test error of 8.8 % for the early cycle life prediction was achieved, along with a classification accuracy of 96.6 %, when dividing the lithium-ion batteries into two quality grades with high and low cycle life.