Product Based Screw Detection and Classification Support System for Smart Disassembly Cell for Circularity

The Circular Economy (EC) model and the various R strategies have accelerated the recovery of electrical components from End of Life (EoL) products with the aim of minimizing the amount of waste and the use of new raw materials. Due to the higher degree of product complexity and the uncertainty and variability of the components to be disassembled, automation in this context is a challenging task. For robotic dismantling cell, computer vision solutions are mandatory to detect parts ranging from screws and small electrical components to the entire product. These vision systems provide data such as object location, classification, and tool suggestions, which are essential for implementing robots for disassembly tasks. This paper proposes a computer vision framework to form a screw detection and classification model with the least number of images for training the model. The location and the screw type are known, thus this model aims to check the presence of the screw in the specified location and then classify it to confirm the type of tool to be utilized for the removal process in the received EoL product to ease the integration of robots into disassembly. For both detection and classification, the YOLOv8 algorithm was used. As the model is product based, the dataset is also created for individual products by taking 10 pictures of each type of screw present in that particular product. A case study with an automotive electric motor and a power inverter was carried out to study the performance of the proposed methodology. The electric motor had Hex and Torx type screws and the power inverter had slotted hex type screws. The average precision of the two models was promising given the dataset is small providing the possibility of implementing a detection check system to support disassembly automation without the need for a huge dataset.