Experimental and simulation investigation of stereo-DIC via a deep learning algorithm based on initial speckle...

Area of Science:

• Mechanical Engineering
• Computer Vision
• Materials Science

Background:

• Accurate initial speckle positioning is critical for deep-learning-based stereo-digital image correlation (stereo-DIC).
• Inaccurate positioning leads to errors in generated datasets and deformation field calculations.
• Existing methods require optimization for precise extrinsic parameter estimation.

Purpose of the Study:

• To propose an optimized extrinsic parameter estimation algorithm for accurate speckle positioning in stereo-DIC.
• To develop and validate a deep learning model for simultaneous displacement and strain field measurement.
• To enhance the overall accuracy and reliability of stereo-DIC techniques.

Main Methods:

• Simulations were used to evaluate the accuracy of various extrinsic parameter estimation algorithms.
• An optimized algorithm was applied to generate a stereo speckle image dataset.
• An improved dual-branch CNN deconvolution architecture (DAS-Net) was developed for displacement and strain output.

Main Results:

• The optimized extrinsic parameter estimation minimized discrepancies between experimental and dataset images (mean absolute percentage error < 2%).
• DAS-Net demonstrated reduced displacement errors compared to previous methods in simulations.
• Experimental validation confirmed the accurate measurement of displacement and strain fields by DAS-Net.

Conclusions:

• Optimized extrinsic parameter estimation significantly improves initial speckle positioning accuracy for stereo-DIC.
• The proposed DAS-Net algorithm effectively measures displacement and strain fields with high accuracy.
• This work enhances the reliability and applicability of deep-learning-based stereo-DIC for experimental mechanics.