Electrode Tab Deflection Detection for Pouch Lithium-Ion Battery Using Mask R-CNN

Recently the growing sensitivity of various governments toward a cleaner environment has increased the demand for electric vehicles (EVs). The battery is one of the most vital components in an EV. With the rapid development of EVs, the applications of lithium-ion battery (LIB) have become more and more extensive. Among the LIBs, the pouch type lithium-ion battery offers a simple, flexible, light weight, and robust solution to battery design, therefore it is considered to be the most promising technology for power battery. The pouch type LIB cell manufacturing processes include lithium battery cell assembly, electrolyte filling, formation, and aging, etc. The purpose of the formation is to form a stable SEI film on the electrode surface by charging. In the formation process, if the electrode tabs deviates from the accepted range, that will cause the failed charging. Therefore, a machine vision system should be built to automatically detect the electrode tab deflection. Recent developments in the field of deep learning have inspired a new interest in using neural network for general image classification tasks. In this paper, we adopt an instance segmentation algorithm called Mask R-CNN to detect and segment electrode tabs. Using the generated bounding boxes of the Mask R-CNN together with the Canny edge detection method, we can decide whether the electrode tab has deflected. As there was no existing dataset, we built a new dataset containing 398 pouch LIBs images. Among them, 200 samples were used as the training dataset, and the other 98 images were used as the validating dataset. Experimental results show 100% accuracy in electrode tab detection for both training and validating dataset. Another 100 pouch LIBs images were captured and used as test dataset. For the test dataset, the proposed algorithm has 100% accuracy in electrode deflection detection and 95% accuracy in polarity. It verifies the effectiveness of the proposed approach in electrode tab deflection detection for pouch LIB.