A Machine Learning Approach to Bridge-Damage Detection Using Responses Measured on a Passing Vehicle.

Area of Science:

• Structural Health Monitoring
• Machine Learning Applications
• Vehicle-Bridge Interaction

Background:

• Traditional bridge inspection methods are often labor-intensive and costly.
• Developing non-destructive, continuous monitoring techniques is crucial for infrastructure safety.
• Vehicle-bridge interaction presents a unique opportunity for remote structural assessment.

Purpose of the Study:

• To propose a novel two-stage machine learning approach for bridge damage detection.
• To utilize vehicle responses measured during transit for identifying structural damage.
• To develop a robust damage indicator resilient to environmental factors and noise.

Main Methods:

• Stage 1: Training an artificial neural network (ANN) on healthy bridge data to predict vehicle responses based on speed.
• Stage 2: Employing a Gaussian process to define a damage indicator based on the distribution of prediction errors.
• Utilizing root-mean-square error to quantify discrepancies between predicted and measured vehicle responses.

Main Results:

• The proposed method successfully detected damage in numerical simulations of vehicle-bridge interaction.
• The approach demonstrated robustness against road roughness profiles and measurement noise.
• Low damage levels were detectable, indicating high sensitivity of the method.

Conclusions:

• The two-stage machine learning approach offers a promising non-destructive method for bridge damage detection.
• Analyzing vehicle responses provides a viable strategy for continuous structural health monitoring.
• The developed damage indicator effectively signals changes in bridge condition.