At SMART2023 on 4th July 2023, Lampros Leontaris, Research Assistant at Information Technologies Institute, Centre for Research and Technology Hellas (CERTH-ITI), presented the paper Inspection of surface defects in metal processing industry using UNet-based architectures.
Authors: Lampros Leontaris, Nikolaos Dimitriou, Apostolos Nikolousis, Dimitrios Tzovaras, Elpiniki Papageorgiou
Brief description: Surface inspection is a critical procedure of quality control during metal processing. Quality control typically involves human inspection at the last stages of the manufacturing process when its too late to repair defects. Due to the surface specularity, manual inspection is also error-prone and time consuming. In this study, the authors propose a vision-based system to automate quality control. The system comprises two machine vision sensors, an illumination system and a mounting frame. The UNet architecture was used to localise defects in images. The proposed method was validated in a sink manufacturing use case using high resolution image data that had been collected during production and that contained various types of surface defects.
Findings: The findings demonstrate that the system has potential to be used in production processes to provide accurate detections and thus accelerate inspection. EfficientNet-B7 as the network backbone provided the best performance.
Future work: Future work will involve enhancing the dataset with more images from production and classifying defective regions with separate defect categories. Different segmentation approaches and deep learning methods to control the illumination effect will also be examined.
Interested in finding out more? The paper will be published as an open access publication in the SMART 2023 conference proceedings.