This repository contains the implementation of a cutting-edge approach to Structural Health Monitoring (SHM) of the Z24 Bridge dataset, leveraging the powerful capabilities of WaveNet and MiniRocket algorithms. Our project aims to provide a comprehensive toolset for the analysis, prediction, and understanding of the structural integrity of the Z24 Bridge, a well-known case study in the field of civil engineering and Structural Health Monitoring.
The Z24 Bridge dataset is a crucial resource in the SHM community, offering extensive sensor readings from the now-demolished Z24 Bridge in Switzerland. This dataset includes data on various parameters like temperature, strain, and displacement, providing a rich source for analyzing structural behaviors and anomalies. For this work, the Progressive Damage Test (PDT) section is used, in particular the ambient vibration test (avt).
The dataset is not provided by this github and must be requested from the owner at the link: https://bwk.kuleuven.be/bwm/z24
- Clone this git repository in an appropriate folder
git clone https://github.com/Elios-Lab/WaveNet_MiniRocket_Z24_Bridge_Structural_Health_Monitoring.git
-
Request the dataset from the owner at the link: https://bwk.kuleuven.be/bwm/z24
-
Setup conda environment
conda env create -f Z24_Bridge_env.yml
conda activate BridgeZ24_1
- Install the required packages
pip install tensorflow
pip install sktime
We suggest to install TensorFlow with GPU support if you have a compatible GPU.
- Add the Dataset
Find in the downloaded dataset the files for Progressive Damage Test (PDT) and inside the folder Ambient Vibration Test (avt) and place each element in the correct folder: e.g. in DatasetPDT/01/avt/ copy files from 01setup01.mat to 01setup09.mat.
Do this for all classes from 1 to 17.
For class 03 remove the file 03setup01.mat as declared in the journal paper.
Inside this repository you can find files relative to WaveNet and to the Minirocket deep neural networks.
The script are ready to be launched after the dataset has been correctly inserted in the folder.
The WaveNet is implemented in the file WaveNet.py with the construction of the model.
The training can be started by running the WavenetRun.py script and the model can be evaluated through the WaveNetEvaluate.py script after specifying the model path in the script.
The Wavenet files datasetManagement.py and utils.py regard the preparation of the dataset and utility functions.
For the MiniRocket there is one single file MiniRocket.py which also implements a Ridge Classifier after MiniRocket's feature extraction.
In the folder History it will be saved the summary of the training at the end of a training session.
To select 5 or 15 classes (as declared in the journal paper) choose in the code one of these two lines:
#5 classes
classes = ['01', '03', '04', '05', '06']
#15 classes
classes = ['01', '03', '04', '05', '06','07','09','10','11','12','13','14','15','16','17']