THE 2ND INTERNATIONAL CONFERENCE ON STRUCTURAL DAMAGE MODELLING AND ASSESSMENT (SDMA 2021)
VIBRATION-BASED DAMAGE DETECTION OF Z24 BRIDGE USING TWO-DIMENSIONAL CONVOLUTIONAL NEURAL NETWORK
Hieu Tran-Nguyen, Dung Ngoc-Bui, Lan Ngoc-Nguyen, Thanh Bui-Tien, Guido De Roeck, Magd Abdel Wahab
Contact
Researcher: [email protected] Promotor: [email protected]
Universiteit Gent
@ugent Ghent University
Vibration signal 2D CNN
F u ll y c o n n e ct ed l ay e r
Spectrogram
O u tp u t la y e r
Objectives
- Improving the accurate of damage diagnosis - Take the advantaged of CNN in image processing
Method
- Transform the output-only vibration-based signal into the images - Fed the spectrogram-based images to the convolution neural network
- 2D-CNN automatically extracts the features and detects the damage Method: Workflow of the 2-D CNN-based method for damage
Add
Conv 1x1, Linear
Dwise 3x3, Relu6
Conv 1x1, Relu6
Input Stride = 1 block
Input Stride = 2 block Conv 1x1, Relu6
Dwise 3x3, Stride=2, Relu6 Conv 1x1, Linear
Results
-Apply the method to Z24 bridge’s vibration data - Get the high accuracy in damage detection
Network architecture of CNN Performance of training and testing
Confusion matrix for CNN feature
Conclusions
- The proposed damaged detection method using 2D-CNN.
- The images coverted from vibration data help to collect richer features.
- Experimental results show that the method are suitable for damaged detection.
The images with damaged (left) and undamaged (right) labels