Deep DIC: Deep learning-based digital image correlation for end-to-end displacement and strain measurement

Digital image correlation (DIC) has become an industry standard to retrieve accurate displacement and strain measurement in tensile testing other material characterization. Though traditional DIC offers a high precision estimation of deformation for general cases, the prediction becomes unstable at large or when speckle patterns start tear. In addition, requires long computation time often produces low spatial resolution output affected by filtering pattern quality. To address these challenges, we propose new deep learning-based approach--Deep DIC, which two convolutional neural networks, DisplacementNet StrainNet, are designed work together end-to-end displacements strains. predicts field adaptively tracks region interest. StrainNet directly from input without relying on prediction, significantly improves accuracy. A dataset generation method is developed synthesize realistic comprehensive dataset, including with synthetic fields. trained datasets only, Deep gives highly consistent comparable predictions those obtained commercial software real experiments, while it outperforms very robust even localized varied qualities. capable real-time calculation down milliseconds.