Damage Identification Using Measured and Simulated Guided Wave Damage Interaction Coefficients Predicted Ad Hoc by...

Area of Science:

• Aerospace Engineering
• Materials Science
• Structural Health Monitoring

Background:

• Thin-walled structures are vital in aerospace, necessitating robust integrity assessment.
• Guided wave-based structural health monitoring (SHM) is crucial for detecting damage in these structures.
• Existing methods often struggle with the complexity of damage characterization in real-world scenarios.

Purpose of the Study:

• To develop a novel damage identification procedure for guided wave-based SHM using deep neural networks (DNNs).
• To utilize wave damage interaction coefficients (WDICs) as sensitive damage features for DNN training.
• To demonstrate the effectiveness of an experimental data-driven DNN approach for accurate damage characterization.

Main Methods:

• Trained deep neural networks (DNNs) with experimental data to learn relationships between damage characteristics and WDIC patterns.
• Extracted WDICs from scanning laser Doppler vibrometer measurements of surface-bonded artificial damages.
• Employed anglewise principal component analysis for efficient feature dimensionality reduction.

Main Results:

• DNNs accurately replicated WDIC patterns even with noisy experimental data, showing strong generalization capabilities.
• The methodology achieved high recognition accuracy and excellent performance under challenging conditions with limited sensors.
• Analysis revealed limitations of simulation-based approaches compared to the proposed experimental data-driven DNN method.

Conclusions:

• The experimental data-based DNN methodology offers a promising, accurate, and robust solution for guided wave-based SHM.
• This approach can precisely predict damage characteristics, even for cases not included in the training set.
• The study highlights the practical applicability of DNNs trained on experimental data for real-world aerospace structural integrity.