FEM Simulation-Based Adversarial Domain Adaptation for Fatigue Crack Detection Using Lamb Wave

Area of Science:

• Structural Health Monitoring
• Non-Destructive Testing
• Acoustic Wave Propagation

Background:

• Lamb wave-based methods are promising for structural integrity assessment.
• Conventional and data-driven damage detection methods require extensive labeled data, which is costly and time-consuming to acquire.
• A significant challenge is the distribution discrepancy between simulated and experimental data, hindering classifier performance.

Purpose of the Study:

• To propose an automated fatigue crack detection method using Lamb waves.
• To address the issue of insufficient labeled data in practical applications.
• To achieve accurate damage detection across different data domains (simulation vs. experiment).

Main Methods:

• Utilized finite element method (FEM) simulations to generate response signals.
• Employed Domain-Adversarial Neural Network (DANN) with Maximum Mean Discrepancy (MMD) for feature extraction.
• Developed a method for discriminative and domain-invariant feature learning between simulation and experimental data.

Main Results:

• The proposed method demonstrated superior detection ability compared to existing techniques.
• Successfully classified unlabeled experimental signals by bridging the simulation-experiment domain gap.
• Validated through fatigue tests on center-hole metal specimens.

Conclusions:

• The developed automated method effectively detects fatigue cracks using Lamb waves.
• Adversarial domain adaptation successfully overcomes the limitations of data scarcity and domain discrepancy.
• The technique offers a robust and effective tool for cross-domain damage detection in structural health monitoring.