A Stacked Neural Network Model for Damage Localization.

Area of Science:

• Structural Health Monitoring
• Artificial Intelligence in Engineering

Background:

• Traditional vibration-based damage detection methods are often manual, leading to delays and errors.
• Automated methods are needed to improve the efficiency and accuracy of structural damage assessment.

Purpose of the Study:

• To develop an accurate and automated method for detecting structural damage locations using Artificial Neural Networks (ANNs).
• To propose a novel stacked neural network architecture for enhanced damage prediction accuracy.

Main Methods:

• Feature extraction using Relative Frequency Shifts (RFSs) from vibration modes.
• Implementation of a stacked neural network approach using Multilayer Perceptron, Recurrent Neural Network, Long Short-term Memory, and Gated Recurrent Units.
• Training individual networks on segmented beam data and comparing with a standard ANN.

Main Results:

• The proposed stacked neural network approach demonstrated high accuracy in predicting damage locations.
• A specific stacked model comprising 14 two-layer feedforward networks achieved the best performance.
• The stacked approach outperformed a standard neural network trained on the entire structure.

Conclusions:

• Stacked neural networks offer a powerful tool for accurate, automated structural damage detection.
• The segmentation strategy and specific network configurations are crucial for optimizing performance.
• This approach significantly reduces the time and potential for errors associated with traditional methods.