Real-time fastener detection for automated disassembly of e-waste

The rapid increase in electronic waste (e-waste) due to technological advancements and shorter product life cycles presents significant environmental and economic challenges. Efficient recycling and remanufacturing require automated disassembly systems, which would reduce labor costs, enhance processing speed, and mitigate human exposure to hazardous materials. A critical component of such systems is the accurate and fast detection and classification of screws and rivets, prevalent fasteners in electronic devices. In this paper, we present an automated real-time inspection system which is based on the YOLO object detection algorithm. It addresses common challenges arising in fastener detection and classification in e-waste, specifically in old and discarded PC housings and mainboards. The small size of the fasteners, partial occlusions and uneven lighting are common issues for optical inspection system. Existing solutions often fall short in comprehensively addressing these issues due to limitations in available datasets and detection methodologies. To overcome these challenges, we introduce the IFF Fastener Dataset (IFF-FD) on the one side, which is a comprehensive collection of high-resolution images depicting realistic disassembly scenarios with the most common fastener types under various positions, lighting conditions and camera exposures. On the other side, we demonstrate that the usage of multiple exposure images of the same scene can improve the accuracy and robustness of fastener detection models for real-world scenarios. Our experiments demonstrate that the proposed inspection system achieves a mean average precision >99.0% with an inference speed of about 50ms on common consumer hardware.