SAM 기반 분할 기법을 이용한 침지 금속의 부식 모니터링에 관한 연구

Abstract

This study employed a sealed immersion test cell to conduct long-term monitoring of the corrosion resistance of metallic specimens (POSSEN, STS304, and Steel) in NaCl solution. While conventional electrochemical evaluation methods provide high precision, their application is limited in inaccessible environments. To address this, a non-contact, non-destructive image-based analysis technique was utilized to quantitatively assess corrosion progression. Leveraging the zero-shot transferability and prompt-based segmentation capability of the Segment Anything Model (SAM), specimen regions were effectively separated from the container and background, ensuring stable image acquisition even under variations in lighting and solution transparency. The segmented specimen regions were converted into CIELab and HSV color spaces, and temporal changes in color values were analyzed to quantitatively characterize the initiation and progression of corrosion. The proposed method improves the accuracy and reliability of corrosion resistance evaluation under conditions with environmental variability, and holds potential for long-term corrosion monitoring and material performance assessment in various industrial settings.