Damage identification based on the inner product matrix and parallel convolution neural network for frame structure.

Area of Science:

• Structural Engineering
• Machine Learning
• Signal Processing

Background:

• Vibration signal analysis is crucial for structural health monitoring and damage identification.
• Conventional machine learning methods, like convolutional neural networks (CNNs), often use single-domain inputs, limiting damage identification accuracy.
• Extracting multidimensional features is key to overcoming these limitations.

Purpose of the Study:

• To propose a novel approach, the Inner Product Matrix-Parallel Convolutional Neural Network (IPM-PCNN), for enhanced structural damage detection.
• To leverage multidimensional features from vibration signals for more accurate identification of structural damage.
• To validate the IPM-PCNN's effectiveness and superiority against existing methods.

Main Methods:

• Developed a novel IPM-PCNN framework integrating one-dimensional (1D) CNN for time series data and two-dimensional (2D) CNN for modal data.
• Utilized a five-story steel frame model for experimental validation.
• Compared the IPM-PCNN model against five other machine learning methods under identical conditions.

Main Results:

• The IPM-PCNN model achieved a high accuracy of 96.60% in identifying structural damage on the test set.
• The proposed method demonstrated superior performance compared to other evaluated machine learning techniques.
• The study provided visualizations of internal inference processes for better understanding of the model's decision-making.

Conclusions:

• The IPM-PCNN effectively extracts multidimensional temporal and modal features from vibration signals for accurate structural damage identification.
• This novel approach significantly outperforms traditional machine learning methods in structural health monitoring.
• The findings offer valuable insights into advanced techniques for ensuring structural integrity.