Deep learning-based approach to mitigating reference pattern loss in Digital Image Correlation

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract In this study, a deep learning-based Digital Image Correlation (Deep DIC) method is applied to mitigate some limitations of conventional DIC techniques. Traditional DIC suffers from a significant reduction in displacement measurement accuracy when the reference pattern within the region of interest is partially or completely lost due to deformation. The proposed method addresses this issue by enabling robust displacement estimation even when the reference pattern becomes partially degraded or moves outside the observable region. While conventional DIC methods struggle to maintain measurement accuracy under high magnification during tensile tests, the proposed Deep DIC approach enables precise displacement measurements throughout the entire testing period. This provides displacement information that can support the mechanical characterization of thin-film materials. To validate the effectiveness of the proposed method, tensile tests were conducted on thin-film specimens using a high-magnification Scanning Electron Microscope (SEM) setup. When the Deep DIC algorithm was applied to surface images obtained during the test, the relative error in displacement measurement was found to be 1.28%, demonstrating the reliability and practical applicability of the proposed approach.