A Data-Driven Damage Identification Framework Based on Transmissibility Function Datasets and One-Dimensional...

Area of Science:

• Structural Engineering
• Data-Driven Methods
• Machine Learning for SHM

Background:

• Vibration-based structural damage identification is popular but limited by insufficient damage-sensitive features and instability under random excitations.
• Conventional intelligent algorithms struggle with damage feature extraction and noise suppression.

Purpose of the Study:

• To develop a novel damage identification framework integrating structural transmissibility functions (TFs) and one-dimensional convolutional neural networks (1D CNNs).
• To address limitations of existing methods in feature extraction, stability, and noise immunity.

Main Methods:

• Constructed massive datasets using structural transmissibility functions (TFs).
• Employed a deep learning strategy based on one-dimensional convolutional neural networks (1D CNNs).
• Validated the framework on an American Society of Civil Engineers (ASCE) benchmark structure under various damage scenarios and white noise excitations.

Main Results:

• TF signals demonstrated more significant damage-sensitive features and stronger stability compared to time series (TS) and fast Fourier transform (FFT) signals.
• The 1D CNN approach showed advantages in computation efficiency, generalization ability, and noise immunity over traditional artificial neural networks (ANNs) for high-dimensional datasets.
• The TF-1D CNN framework achieved high damage identification accuracy.

Conclusions:

• The developed TF-1D CNN framework offers a robust and efficient solution for data-driven structural damage identification.
• This integrated approach shows practical value for future structural health monitoring applications.