Photometric-Stereo-Based Defect Detection System for Metal Parts

Area of Science:

• Computer Vision
• Manufacturing Technology
• Surface Metrology

Background:

• Automated inspection using computer vision is prevalent in manufacturing but struggles with glossy metal surfaces due to overexposure and shadows.
• Adjusting lighting and viewing angles for diverse part geometries is tedious and inefficient for defect detection.

Purpose of the Study:

• To develop a robust defect detection system for metal parts that mitigates issues caused by surface gloss and shadows.
• To introduce a novel framework combining photometric stereo and defect detection for improved accuracy.

Main Methods:

• Designed a photometric-stereo-based defect detection system (PSBDDS) integrating photometric stereo with defect detection algorithms.
• Developed a framework that uses multiple directional light images to generate a normal map via photometric stereo.
• Utilized the normal map as input for a detection model to locate and classify defects.

Main Results:

• The PSBDDS effectively eliminates interference from highlights and shadows, improving defect detection accuracy.
• A new photometric stereo defect detection (PSDD) dataset was created to bridge the gap between existing methods.
• Experimental results validated the effectiveness of the proposed system and dataset.

Conclusions:

• The developed photometric-stereo-based defect detection system offers a viable solution for inspecting challenging metal surfaces.
• The created PSDD dataset facilitates further research and development in this specialized area of automated inspection.