Vision-Based Screw Head Detection for Automated Disassembly for Remanufacturing

Abstract

Remanufacturing is commonly perceived as a promising field for future challenges such as resource efficient production. For an economic operation of remanufacturing facilities, an automation of the currently manual labor is mandatory. Thus, the automation plays a vital role in order to realize high rates of re-utilization and therefore a significant reduction of waste. Screw connections allow for non-destructive dismantling and are commonly used connection elements. Especially the automation of the disassembly step is a key element as products from the field are of unknown specification upon feeding to the remanufacturing line due to alterations during their life cycles. State of the art solutions for automated disassembly lack flexibility to adapt to different products and product conditions. This contribution presents a highly flexible approach for the localization and classification of screws in electric motors. The presented system utilizes a tool equipped industrial robot with an integrated eye-in-hand vision system and an industrial computer. The system is able to locate and classify six different types of screw heads of varying sizes using machine learning approaches in order to adapt the robot’s end-effector. Because of the presented hardware concept the system depends upon a minimum of constraints concerning the presentation of objects. This paper compares different network architectures and peripheral settings and presents the most suitable solution to the use case. A dataset consisting of six classes of different screw heads was created to train neural networks to detect screws in an experimental set-up consisting of metal blocks holding different screws of diverse types and conditions. Results are validated on two different electric motors from the automotive sector being processed on an automated disassembly line.