A Multimodal Feature Sensing and Fusion Neural Network for Damage Localization by Ultrasonic Guided Waves

Area of Science:

• Composite Materials
• Structural Health Monitoring
• Non-Destructive Testing

Background:

• Ultrasonic guided waves (UGWs) are crucial for damage localization in composites, ensuring structural integrity.
• Current deep learning methods often fail to integrate diverse signal features, limiting damage localization accuracy.
• Effective damage localization necessitates the fusion of multiple data modalities for comprehensive analysis.

Purpose of the Study:

• To introduce a novel Multimodal Feature Sensing and Fusion Neural Network (MSFN) for enhanced damage localization using UGWs.
• To address the limitations of single-modal deep learning approaches in feature extraction and fusion.
• To improve the accuracy, robustness, and efficiency of damage localization in composite materials.

Main Methods:

• Developed an innovative multimodal input strategy using damage, scattered wave, and energy density signals.
• Employed Convolutional Neural Networks (CNNs), Gated Recurrent Units (GRUs), and Bidirectional Gated Recurrent Units (BiGRUs) as specialized encoders.
• Utilized an attention mechanism-guided feature fusion strategy for aggregating multimodal features.

Main Results:

• Achieved a damage localization accuracy of 98.13%, even with significant noise interference.
• Demonstrated superior robustness, accuracy, speed, and generalization compared to existing methods.
• Validated the MSFN's effectiveness for near-real-time structural health monitoring applications.

Conclusions:

• The proposed MSFN effectively integrates multimodal features for accurate UGW-based damage localization in composites.
• The network architecture, featuring specialized encoders and attention-guided fusion, significantly enhances performance.
• MSFN shows strong potential for real-world structural health monitoring, improving safety and reliability.