A Shapelet Transform-Based Method for Structural Damage Identification: A Case Study on a Wooden Truss Bridge.

Area of Science:

• Structural Engineering
• Vibrational Analysis
• Machine Learning for Civil Infrastructure

Background:

• Structural health monitoring (SHM) faces challenges with environmental changes and sensor variations.
• Existing methods often lack interpretability or sensitivity for early damage detection.

Purpose of the Study:

• To develop a robust and interpretable damage identification method for SHM.
• To address limitations of current techniques in handling real-world complexities.

Main Methods:

• Utilized Shapelet Transform to extract interpretable local features from vibration signals.
• Employed Random Forest classifier for damage state identification.
• Tested on a timber truss bridge with simulated damage and varying environmental conditions.

Main Results:

• Achieved 100% identification accuracy with a 10-minute Shapelet extraction time.
• Successfully detected minimal damage (0.07% of structural mass).
• Demonstrated robustness across different sensor locations and environmental variations.

Conclusions:

• The Shapelet Transform and Random Forest combination offers high accuracy, robustness, and interpretability for SHM.
• This method provides a practical approach for real-time damage identification systems.
• Outperforms traditional and deep learning methods in specific scenarios, especially with limited data.