Physics-Guided CNN Detection of Crack-Associated Events from Embedded Fiber Bragg Grating Sensors

Area of Science:

• Materials Science
• Mechanical Engineering
• Computer Science

Background:

• Crack detection in composite structures is crucial for structural health monitoring (SHM).
• Limited sensors, such as multiplexed fiber Bragg gratings (FBGs), pose challenges for accurate crack detection.
• Existing methods often struggle with the low-density sensor data from FBGs.

Purpose of the Study:

• To develop and validate a physics-guided convolutional neural network (CNN) for crack-associated event detection.
• To utilize multiplexed FBG signals and synchronized mechanical data for enhanced SHM.
• To create a robust and reproducible framework for identifying cracks in glass-fiber-reinforced polymer (GFRP) beams.

Main Methods:

• A physics-guided CNN framework was developed to analyze multiplexed FBG interrogator signals.
• The dataset included raw FBG recordings and synchronized force-displacement metadata from GFRP beams under three-point bending.
• Mechanics-guided descriptors, including Euler-Bernoulli strain terms, were used to encode experimental responses.

Main Results:

• The CNN model achieved high performance metrics, including window-level F1 scores of 0.905 and run-level F1 scores of 0.909.
• The model demonstrated strong generalization capabilities on newly manufactured specimens, reaching a window-level F1 score of 0.952.
• The framework successfully identified crack events with high precision and recall, even with a limited number of sensors.

Conclusions:

• A compact sequence CNN, enhanced with mechanics-guided strain interpretation, can reliably detect crack events from multiplexed FBG measurements.
• The proposed framework offers a simple, reproducible, and effective approach for structural health monitoring of composite materials.
• This physics-guided approach shows significant promise for advancing SHM techniques in challenging composite structures.