Tiny Machine Learning Implementation for Guided Wave-Based Damage Localization.

Area of Science:

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

Background:

• Ultrasonic guided waves (UGWs) are effective for detecting damage in structures.
• Current damage detection methods often require complex computational resources.
• Lightweight Artificial Intelligence (AI) offers potential for efficient on-site damage assessment.

Purpose of the Study:

• To develop and validate an AI-based technique for damage detection and localization using UGWs.
• To investigate the use of machine learning (ML) for analyzing UGW signal alterations caused by damage.
• To implement a computationally efficient AI model for real-time structural health monitoring.

Main Methods:

• Applied a physical signal processing approach to raw UGW data to reduce model parameters.
• Trained machine learning models on the effects of damage on UGW signals.
• Developed and validated an AI-based damage detection technique on an experimental benchmark dataset.
• Implemented a tiny ML model on a low-cost development board.

Main Results:

• Achieved high damage localization accuracies exceeding 90% by extracting simple signal features.
• Significantly reduced the size of AI models required for damage prediction.
• Successfully generated accurate heat maps indicating likely damage locations.
• Demonstrated a balance between reduced computational resources and high model precision.

Conclusions:

• Lightweight AI models effectively detect and localize structural damage using UGWs.
• Signal feature extraction is crucial for creating efficient and accurate damage detection systems.
• The developed technique is suitable for implementation on low-cost hardware for edge/cloud visualization.