Structural damage identification using single-point vibration data processing.

Area of Science:

• Engineering
• Artificial Intelligence
• Structural Health Monitoring

Background:

• Cracks in structures cause nonlinear dynamics, challenging traditional linear analysis.
• Existing damage identification methods often require numerous sensors, limiting practical use.

Purpose of the Study:

• To develop a method for identifying crack depth and location using single-point vibration measurements.
• To assess the effectiveness of a nonlinear analysis framework and convolutional neural networks for damage detection.

Main Methods:

• Simulating cracked beam dynamics using a nonlinear analysis framework.
• Generating spectrograms of tip response via short-time Fourier transform.
• Training a convolutional neural network on spectrograms for crack identification.

Main Results:

• High coefficients of determination achieved for both crack depth and location identification.
• Accuracy is dependent on sufficient training data coverage for damage conditions.
• Data augmentation significantly improves identification accuracy.

Conclusions:

• Single-point vibration measurements combined with AI can effectively identify crack depth and location.
• The proposed method offers a practical solution for structural health monitoring with reduced sensor requirements.
• Further implementation is feasible, especially with data augmentation for limited measurement scenarios.