A deep learning-based process monitoring system for toothbrush manufacturing defect characterization

Toothbrush manufacturing process is prone to a number of defects concerning the bristle stapling, affecting the amount of scrap parts and rework. State-of-the-art inspection techniques are characterized by low efficiency, unsustainable operator fatigue, resulting in a low detection performance with the consequence of an overall final product low quality and safety issue. To enable an automatic process monitoring this paper presents a machine vision-based inspection system endowed with a deep-learning YOLOv5s-based decision-making for toothbrush bristles defects identification and characterization. The proposed system is made of three modules, respectively the image acquisition module, the image processing module and the intelligent defect classification module. A laboratory scale experimental rig was designed in order to carry out trial aimed at validating the proposed monitoring method. The results of testing demonstrated a high classification accuracy capability and high performances in terms computation time, indicating an excellent suitability for industrial applications.