Damage-Detection Approach for Bridges with Multi-Vehicle Loads Using Convolutional Autoencoder.

Area of Science:

• Structural Engineering
• Artificial Intelligence
• Civil Infrastructure

Background:

• Deep learning enhances bridge damage detection.
• Unsupervised deep learning increases the applicability of damage detection.
• Convolutional Autoencoder (CAE) is an unsupervised deep learning network for damage detection.

Purpose of the Study:

• To verify the applicability of the CAE-based damage detection approach for bridges under multi-vehicle loads.
• To assess the performance of the CAE approach on rigid-frame and reinforced-concrete-slab bridges.
• To evaluate damage detection accuracy and false-negative rates in a typical multi-vehicle load scenario.

Main Methods:

• Modeling and simulation of rigid-frame and reinforced-concrete-slab bridges.
• Acquisition of bridge behavior data under simulated loads.
• Application and testing of a Convolutional Autoencoder (CAE)-based damage detection approach.

Main Results:

• The CAE-based damage detection approach achieved satisfactory accuracy exceeding 90% for both bridge types.
• The false-negative rate for damage detection was less than 1% across all simulations.
• The approach demonstrated high performance in detecting damage under multi-vehicle load conditions.

Conclusions:

• The CAE-based damage detection approach is effective for various bridge types under multi-vehicle loads.
• The unsupervised deep learning method shows significant potential for real-world bridge health monitoring.
• The study confirms the robustness and accuracy of the CAE approach in complex loading scenarios.