Investigation of Frequency-Domain Dimension Reduction for A²M-Based Bridge Damage Detection Using Accelerations of...

Area of Science:

• Structural Engineering
• Vibration Analysis
• Machine Learning Applications

Background:

• Bridge health monitoring using vehicle vibrations is gaining interest.
• Existing methods often require constant vehicle speeds or labeled damage data, limiting practical application.
• Labeling bridge damage is difficult and often impractical as bridges are typically in a healthy state.

Purpose of the Study:

• To propose a novel, damaged-label-free, machine learning-based indirect bridge health monitoring method.
• To develop a method that overcomes the limitations of existing bridge health monitoring techniques.
• To improve the accuracy and practicality of bridge damage detection.

Main Methods:

• The proposed method, assumption accuracy method (A2M), uses raw frequency responses of vehicles to train a classifier.
• K-folder cross-validation accuracy scores are used to establish a threshold for determining bridge health.
• Dimension reduction techniques, specifically Principal Component Analysis (PCA) and Mel-frequency cepstral coefficients (MFCCs), are employed to handle high-dimensional frequency response data, with MFCCs showing greater damage sensitivity.

Main Results:

• Utilizing full-band vehicle responses, including higher frequency ranges, significantly improves detection accuracy compared to low-band responses.
• Mel-frequency cepstral coefficients (MFCCs) are found to be more sensitive to damage than PCA.
• Accuracy values for healthy bridges using MFCCs are dispersed around 0.5, increasing significantly to 0.89-1.0 upon damage detection.

Conclusions:

• The assumption accuracy method (A2M) offers a viable and accurate solution for indirect bridge health monitoring without requiring labeled damage data.
• The study demonstrates the importance of considering full-band vehicle frequency responses and highlights the effectiveness of MFCCs for damage detection.
• This approach enhances the practicality of bridge health monitoring in real-world engineering applications.