Artificial Intelligence-Assisted Manual Inspection in Industrial Sites for Anomaly Detection in Connector Joints

Despite rapid advancements in artificial intelligence (AI) and robotics, facility inspections in industrial sites are still largely carried out by personnel. This study proposes an intelligent assistive solution designed to reduce workload and minimize human errors in manual inspection tasks. A two-stage computational framework based on deep learning models is developed to predict connection states in pipeline joints from color images captured on-site. The first stage identifies object types and their locations in an image, while the second stage classifies the connection state based on areas of interest extracted during pre-processing. Data augmentation techniques are applied to generate annotated synthetic data, which is combined with manually labeled real data to create a dataset for model training and validation. The framework demonstrates computational advantages over a single deep learning model for object recognition. Finally, a prototype system is implemented on a mobile device and tested in real-world environments to verify the feasibility and practical value of the proposed solution in supporting facility inspections through AI-human collaboration.