Sound Damage Detection of Bridge Expansion Joints Using a Support Vector Data Description

Area of Science:

• Structural Engineering
• Acoustics
• Signal Processing

Background:

• Modal bridge expansion joints (MBEJs) are critical infrastructure components susceptible to damage.
• Effective monitoring of MBEJ integrity is essential for ensuring bridge safety and longevity.
• Traditional inspection methods can be labor-intensive and may not detect subtle damage.

Purpose of the Study:

• To develop and validate a novel, non-destructive method for identifying damage in MBEJs using acoustic signals.
• To compare the effectiveness of different signal processing and machine learning techniques for damage detection.
• To achieve high accuracy in distinguishing between healthy and damaged MBEJ states.

Main Methods:

• Fabrication of MBEJ specimens in healthy and damaged states.
• Acquisition of acoustic impact signals using a microphone.
• Feature extraction via wavelet packet energy ratio and dimensionality reduction with Principal Component Analysis (PCA).
• Damage classification using a Support Vector Data Description (SVDD) model, optimized with Bayesian optimization.

Main Results:

• The wavelet packet energy ratio effectively differentiated between healthy and damaged MBEJ support bars.
• Bayesian optimization yielded the best performance for the SVDD model.
• The proposed acoustic-based method achieved 99% accuracy in damage detection for MBEJ support bars.

Conclusions:

• Sound signal analysis, specifically wavelet packet energy ratio, is a viable technique for MBEJ damage identification.
• The SVDD model, optimized via Bayesian methods, provides a robust and accurate classification framework.
• This acoustic-based approach offers a promising, high-accuracy solution for structural health monitoring of MBEJs.