Deep Learning-Based Visual Defect Inspection System for Pouch Battery Packs.

Every year, billions of pouch battery packs are produced worldwide. The production of pouch battery packs has been highly automated. However, visual inspection is the last step of production that still requires a large number of workers. We present the hardware and software design of an automated visual inspection system for pouch battery packs. We have achieved a 4% false alarm rate, 0.7% missing alarm rate, and 3.5 s cycle time on this challenging task through well-designed optical hardware and the latest deep learning techniques.