Deep Learning-Based Subsurface Damage Localization Using Full-Field Surface Strains.

Area of Science:

• Structural Health Monitoring
• Artificial Intelligence in Engineering
• Non-Destructive Testing

Background:

• Structures are susceptible to aging and extreme events, leading to subsurface damage (SSD).
• Timely detection of SSD is crucial to prevent structural failure and ensure safety.
• Existing methods may not effectively identify internal damage non-destructively.

Purpose of the Study:

• To develop and validate a Convolutional Neural Network (CNN) for accurate SSD detection.
• To utilize surface strain measurements for pixel-level damage classification.
• To assess the network's generalizability across different materials and damage complexities.

Main Methods:

• A CNN architecture designed for pixel-level image segmentation was employed.
• Full-field strain measurements (256x256) were used as input for the CNN.
• Training data generated from numerical simulations of aluminum bars with varied damage scenarios.

Main Results:

• The CNN achieved high Intersection over Union (IoU) scores: 0.790 (validation), 0.794 (testing) on aluminum.
• Consistent performance observed on steel datasets (IoU 0.793) and for complex triple damage cases (IoU 0.764).
• Accurate predictions validated against real experimental data from Strain Sensing Smart Skin (S⁴).

Conclusions:

• The developed CNN is an effective non-destructive testing method for subsurface crack detection and localization.
• The network demonstrates strong generalizability across materials with similar stress-strain behaviors.
• The approach shows significant potential for real-world applications using advanced strain sensing technologies.