Structural Damage Detection Based on the Correlation of Variational Autoencoder Neural Networks Using Limited Sensors

Area of Science:

• Engineering
• Structural Health Monitoring
• Artificial Intelligence

Background:

• Assessing structural safety is critical, but identifying damage without prior structural data presents a significant engineering challenge.
• Limited sensor availability often restricts traditional structural health monitoring techniques.

Purpose of the Study:

• To develop a correlation-based damage identification method using Variational Autoencoder neural networks.
• To enable structural state determination without baseline data, utilizing responses from a limited number of sensors.

Main Methods:

• Constructed a Variational Autoencoder network model for bridge damage detection.
• Optimized model parameters including loss functions and learning rates.
• Trained the model using response data from limited sensors to identify structural states.

Main Results:

• Successfully identified damage location under various damage scenarios.
• Demonstrated strong robustness in detecting multiple structural damages.
• Enhanced the accuracy of bridge structure damage identification.

Conclusions:

• The proposed method effectively identifies structural damage without baseline data, offering a cost-effective solution.
• The approach is robust and accurate, improving practical applications in structural health monitoring.