A Study on Component States Determination Using Deep Learning in Assembly Work and Work Management System Focusing on the Parent-Child Relationship of Components

In modern manufacturing industries, where high-mix, low-volume production systems are widely adopted, manual assembly by workers remains essential. However, manual assembly is prone to operational errors, such as incorrect assembly, missed component installation, and component detachment, which lead to increased defect rates and reduced productivity due to rework. This study proposes a method to detect assembly errors using image classification to categorize components into three states: “Normal,” “Abnormal,” and “Unclassifiable.” Based on the confidence level of these classifications, the system determines the component presence in real time as either “Component Present” or “Component Absent.” Furthermore, in assembly situation prone to instability, the proposed method incorporates control strategies that focus on the parent-child relationship of components, ensuring stable work management. The method proposed in this study is designed for workbench works where components are securely fixed to jigs, ensuring that their positions remain unchanged within the field of view of a fixed camera. The performance evaluation of this method, conducted on assembly work videos, achieved 100% Accuracy and Specificity, demonstrating the system’s ability to reliably detect missed component installation and component detachment. This approach shows promises for improving work management and quality assurance in manual assembly processes, potentially reducing errors and enhancing assembly accuracy.