A Frequency-Based Approach for the Detection and Classification of Structural Changes Using t-SNE †

Area of Science:

• Engineering
• Materials Science
• Data Science

Background:

• Structural health monitoring (SHM) is crucial for assessing infrastructure integrity.
• Existing methods often struggle with high-dimensional data from sensors.
• Nonlinear dimensionality reduction techniques offer potential for improved SHM.

Purpose of the Study:

• To develop and evaluate a novel SHM approach for detecting and classifying structural changes.
• To leverage t-distributed stochastic neighbor embedding (t-SNE) for enhanced data representation.
• To improve the accuracy and robustness of structural state diagnosis.

Main Methods:

• Data preprocessing using mean-centered group scaling (MCGS).
• Dimensionality reduction via principal component analysis (PCA).
• Nonlinear embedding using t-distributed stochastic neighbor embedding (t-SNE) to form clusters representing structural states.
• Classification of current structural states using distance-based and voting strategies.

Main Results:

• The combination of PCA and t-SNE effectively clusters data according to structural states.
• Experimental validation on an aluminum plate with piezoelectric transducers demonstrated high classification accuracy.
• The proposed method shows strong performance in identifying structural changes.

Conclusions:

• The developed SHM approach using PCA and t-SNE is effective for structural change detection and classification.
• The method offers a robust way to diagnose structural conditions based on sensor data.
• This technique holds promise for advanced structural health monitoring applications.