Damage Classification Using Supervised Self-Organizing Maps in Structural Health Monitoring

Area of Science:

• Engineering
• Materials Science
• Computer Science

Background:

• Advancements in computing power enable complex machine learning (ML) algorithms for diverse applications.
• Structural Health Monitoring (SHM) increasingly utilizes ML for damage detection and classification in structures like aircraft and buildings.
• Current SHM systems require further development in robustness, reliability, and cost-effectiveness.

Purpose of the Study:

• To configure a data acquisition system for an active piezoelectric (PZT) sensor network.
• To develop a damage classification methodology using signal processing and ML algorithms.
• To experimentally validate the developed SHM system on aluminum and composite plates.

Main Methods:

• Utilized an active piezoelectric (PZT) sensor network for signal acquisition.
• Developed a damage classification methodology incorporating signal processing techniques (normalization, PCA).
• Applied machine learning algorithms, specifically counterpropagation artificial neural network (CPANN), supervised Kohonen (SKN), and X-Y fused Kohonen (XYF).

Main Results:

• Experimental validation was performed on aluminum plates with added masses and a CFRP plate with delamination and cracks.
• The SKN and XYF networks demonstrated significant utility in damage classification tasks.
• Achieved overall accuracies of 73.75% for SKN and 72.5% for XYF via cross-validation, outperforming other tested neural networks.

Conclusions:

• The developed damage classification methodology shows promise for robust and reliable SHM.
• The SKN and XYF variants of self-organizing maps are effective for damage classification in structural components.
• This research contributes to the advancement of low-cost and efficient automated structural health monitoring systems.