Unsupervised Learning Methods for Data-Driven Vibration-Based Structural Health Monitoring: A Review

Area of Science:

• Structural Engineering
• Machine Learning
• Data Science

Background:

• Structural health monitoring (SHM) research increasingly focuses on unsupervised learning methods.
• Unsupervised learning for SHM utilizes data from intact structures, offering practical advantages over supervised methods for early damage detection systems.
• Novelty detection using vibration data is a prevalent unsupervised approach in SHM.

Purpose of the Study:

• To review data-driven structural health monitoring publications from the last decade employing unsupervised learning methods.
• To focus on real-world applications and practicality of these methods.
• To identify challenges and knowledge gaps in translating research to practical SHM applications.

Main Methods:

• Review of recent literature on unsupervised learning in structural health monitoring.
• Categorization of state-of-the-art studies by machine learning method type.
• Examination of common benchmarks for validating unsupervised SHM methods.
• Analysis of challenges hindering practical implementation of SHM techniques.

Main Results:

• Novelty detection using vibration data is the most common unsupervised learning SHM approach.
• The review categorizes various machine learning methods applied to unsupervised SHM.
• Commonly used benchmarks for validating these methods are identified and examined.
• Key challenges and limitations in practical SHM application are discussed.

Conclusions:

• Unsupervised learning offers practical advantages for early damage detection in civil structures.
• Significant challenges remain in bridging the gap between research and real-world SHM applications.
• Recommendations for future research are provided to enhance the reliability and practicality of SHM methods.