Addressing Challenges for Autonomous Robotic Freeform Eddy Current Inspection Via Computer Vision on Complex Geometries

Eddy current inspection is a critical method for assessing the health of metallic structures, but it requires precise probe placement to avoid signal variations caused by inconsistent liftoff. Robotic systems enable scanning of complex geometries, maintaining stable liftoff and ensuring accurate data collection. This paper presents a robotic eddy current array (ECA) inspection system that operates without CAD models, using computer vision to reconstruct the part’s surface for path planning. Inaccuracies in robot calibration and the reconstructed mesh can disrupt the probe’s precise positioning, especially in ECA scanning, where probe tilting increases liftoff variability, particularly at greater distances from the tool’s center. To address these issues, we introduce a signal processing technique that reduces the impact of mesh inaccuracies and liftoff fluctuations on the acquired ECA data. The system is validated on curved steel samples with corrosion pits, approximately 50 µm deep and ranging from 1 to 10 mm² in area. The results demonstrate the system’s effectiveness in detecting defects and its potential for integration into the NDE 4.0 framework.