Advanced Automation Defect Detection in Additive Manufacturing by Superior Virtual Polarization Filtering and Deep Learning

Abstract

Additive manufacturing (AM) is widely used in industries such as aerospace, medical and automotive, owing to its ability to facilitate precise and intricate fabrication processes. Within this domain, the defect detection technology emerges as a pivotal area of focus during the quality inspection phase of AM. A significant challenge lies in the enhancement of defect image quality and the corresponding detection capabilities with extreme complex condition. This paper proposed an advanced automation defect detection method with a virtual polarization filtering algorithm (IEVPF) and an improved YOLO V5-W model. The IEVPF algorithm improves image quality by enhancing the visibility of defects through the virtual manipulation of light polarization. And the improved YOLO V5-W model leverages these enhanced images for accurate defect identification under various lighting conditions, thereby boosting the overall accuracy of the detection process. Experiments with the YOLO V5 and YOLO V5-W models on original and enhanced datasets show the YOLO V5-W's superior performance, with a 40.3% reduction in loss, a 10.8% increase in precision, a 10.3% increase in recall, and a 13.7% increase in mAP. Furthermore, models trained on the enhanced datasets exhibit improvements, indicating that these techniques are superior in AM surface defect detection.

Keywords: Additive manufacturing, machine vision, Defect detection, image enhancement, deep learning.