A Damage Detection Approach for Axially Loaded Beam-like Structures Based on Gaussian Mixture Model.

Area of Science:

• Structural Health Monitoring
• Unsupervised Machine Learning
• Vibration Analysis

Background:

• Axially loaded beam-like structures pose challenges for vibration-based damage detection due to environmental and operational variations.
• Previous work utilized multivariate damage features and Mahalanobis squared distance (MSD) for unsupervised outlier detection.

Purpose of the Study:

• To develop a novel unsupervised learning approach for enhanced damage detection in axially loaded structures.
• To improve sensitivity to early-stage damage and reduce uncertainty in vibration-based structural health monitoring.

Main Methods:

• Implementation of a Gaussian mixture model (GMM) for data clustering.
• Comparison of the GMM approach with the benchmark Mahalanobis squared distance (MSD) method.
• Testing under uncontrolled environmental conditions and real corrosion-induced damage.

Main Results:

• The Gaussian mixture model approach demonstrated increased sensitivity to structural damage.
• The GMM method significantly reduced uncertainty, enabling earlier damage detection.
• The novel approach proved effective even with environmental variations and real-world damage.

Conclusions:

• Unsupervised learning data clustering, specifically using Gaussian mixture models, offers a more sensitive and reliable method for vibration-based damage detection in challenging structural scenarios.
• This approach enhances early-stage damage identification and reduces diagnostic uncertainty, outperforming traditional MSD methods.