Convolution Neural Network Development for Identifying Damage in Vibrating Pylons with Mass Attachments

Area of Science:

• Structural Engineering
• Artificial Intelligence
• Vibration Analysis

Background:

• Pylon damage detection often relies on sensitive indicators, but their effectiveness is limited by low-amplitude responses to environmental loads.
• Existing methods struggle with early damage detection due to low sensitivity of damage indicators.

Purpose of the Study:

• To develop and evaluate a convolution neural network (CNN) for detecting pylon damage based on vibratory response.
• To interpret experimental data using a mathematical model for enhanced damage detection accuracy.

Main Methods:

• A mathematical model was created to interpret experimental data from a fixed-base pylon with transverse motion.
• Damage was simulated in the model using springs representing beam cracking.
• Numerically generated acceleration records were used to train a CNN for damage identification.

Main Results:

• The trained CNN was employed to identify damage from experimental acceleration records.
• Challenges encountered by the CNN in damage presence/absence identification were analyzed.
• Strategies for improving the CNN's detection performance were proposed.

Conclusions:

• CNNs show potential for pylon damage detection using vibration analysis.
• Further research is needed to overcome limitations and enhance CNN accuracy in structural health monitoring.
• The study highlights the importance of data quality and model refinement for AI-driven damage detection.