Multi-Sensor and Decision-Level Fusion-Based Structural Damage Detection Using a One-Dimensional Convolutional Neural...

Area of Science:

• Structural Health Monitoring
• Artificial Intelligence in Engineering
• Vibration Analysis

Background:

• Structural damage alters dynamic responses, detectable via vibration signals.
• One-dimensional convolutional neural networks (1-D CNNs) can extract damage information from vibration data.
• Practical limitations like non-synchronous and incomplete sensor data hinder CNN accuracy.

Purpose of the Study:

• To enhance structural damage detection accuracy using a novel 1-D CNN and decision-level fusion approach.
• To address challenges posed by incomplete and non-synchronous vibration data in practical engineering.
• To validate the proposed method's effectiveness on both numerical and experimental models.

Main Methods:

• Acquired acceleration signals from multiple points on structures.
• Trained individual 1-D CNN models for each acquisition point.
• Implemented a decision-level fusion strategy to integrate predictions from multiple CNNs.
• Compared performance against a control experiment using data-level fusion.

Main Results:

• Decision-level fusion of multiple 1-D CNN predictions significantly improved damage detection accuracy.
• Accuracy gains of 10% for numerical models and 16-30% for experimental models were observed compared to the control.
• The proposed method effectively overcomes limitations of incomplete and non-synchronous vibration data.

Conclusions:

• Training individual 1-D CNNs on separate sensor data and fusing their decisions is a robust strategy for accurate structural damage detection.
• This approach offers a substantial improvement over traditional data-level fusion methods.
• The findings demonstrate the potential of AI-driven fusion techniques for advanced structural health monitoring.