Diagnostic-Quality Guided Wave Signals Synthesized Using Generative Adversarial Neural Networks

Area of Science:

• Engineering
• Materials Science
• Artificial Intelligence

Background:

• Guided waves are crucial for structural health monitoring (SHM) due to complex signal characteristics.
• Machine learning algorithms for SHM require extensive training data, which is often difficult and expensive to acquire, especially for damaged states.
• Data scarcity is a significant challenge in guided wave analysis and SHM.

Purpose of the Study:

• To develop a generative adversarial network (GAN) architecture for generating synthetic guided wave signals.
• To address the data scarcity problem in training machine learning models for SHM.
• To evaluate the effectiveness of GAN-generated data for damage detection tasks.

Main Methods:

• A novel GAN architecture was designed and implemented for guided wave signal generation.
• The GAN was trained and tested using pitch-catch experiment data from the OpenGuidedWaves database.
• The generated synthetic data was used to train classifiers for a damage detection scenario.

Main Results:

• The GAN successfully generated realistic random guided wave signals.
• The network demonstrated the ability to accurately reconstruct unseen guided wave signals.
• Classifiers trained solely on synthetic data achieved reliable performance in detecting damage when evaluated on real signals.
• Signal compression of 98.44% was achieved while preserving critical damage information.

Conclusions:

• The proposed GAN architecture effectively generates synthetic guided wave data, mitigating data scarcity issues in SHM.
• Synthetic data generated by the GAN can be utilized to train effective damage detection models.
• The GAN offers a promising solution for improving the efficiency and applicability of machine learning in structural health monitoring.