Damage identification using deep learning and long-gauge fiber Bragg grating sensors

Area of Science:

• Structural Engineering
• Artificial Intelligence
• Materials Science

Background:

• Structural health monitoring is crucial for infrastructure safety.
• Traditional damage identification methods can be time-consuming and labor-intensive.
• Advanced sensing technologies and machine learning offer promising alternatives.

Purpose of the Study:

• To develop an innovative structural damage identification method using dynamic response data.
• To leverage deep convolutional neural networks (DCNNs) for accurate damage pattern classification.
• To investigate the effectiveness of transfer learning (TL) in reducing model training time for new structures.

Main Methods:

• Utilizing dynamic response data from long-gauge fiber Bragg grating (FBG) sensors and accelerometers.
• Training DCNN models with acceleration, wavelength, and fused data.
• Applying transfer learning (TL) by pre-training a model on one structure and fine-tuning it on another (e.g., steel beam to reinforced concrete beam).
• Evaluating model performance using training history and confusion matrix analysis.

Main Results:

• The DCNN-based method accurately classifies structural damage patterns.
• CNN models trained with fused sensor data achieve high classification accuracy and faster training speeds.
• Transfer learning effectively reduces the training time required for new structural damage identification tasks.

Conclusions:

• The proposed DCNN approach with sensor fusion is a highly effective method for structural damage identification.
• Transfer learning offers significant efficiency gains in applying damage identification models to different structures.
• This research advances the field of structural health monitoring through intelligent data analysis.