Normal and Anomalous Audio Data Processed with the Wavelet Scattering Transform, Collected During the Operation of FastBlade, a Site for Regenerative Fatigue Testing

Dataset Information

The uploaded dataset is the output of the wavelet scattering transform (WST) applied to audio data collected during the operation of FastBlade, a site for regenerative fatigue testing: https://www.fastblade.eng.ed.ac.uk/. 

WST is applied to the audio data to extract features that can be used to detect anomalous operation of the set-up during a fatigue test of a tidal turbine blade. The machine learning model used for the classification task is trained on normal operation data. Subsequently, feature maps from the hidden layers of the deep neural network trained are examined to identify specific features that:

are common to all training samples,

do not appear in the anomalous samples.  

The following files can be found in the dataset:

Normal_Data_Training.npy - normal operation data used for training. 

Normal_Data_Validation.npy - normal operation data used for feature selection. 

Anomalous_Data_Validation.npy - anomalous operation data used for feature selection. 

Normal_Data_Test.npy - normal operation data used for model evaluation. 

Anomalous_Data_Test.npy - anomalous operation data used for model evaluation. 

The full details of the experiment, the machine learning model trained, and the results are presented in the paper published in the "Engineering Applications of Artificial Intelligence" journal, titled: "An audio-based framework for anomaly detection in large-scale structural testing", https://doi.org/10.1016/j.engappai.2024.109889. 

Loading the Dataset

The best way to access the dataset is using Python's numpy library. Example use case: 

Training_Data = np.load('Path_to_file/Normal_Data_Training.npy')

Array Dimensions

Considering array dimensions in the following format - (X, Y, Z):

X is the number of samples,

Y is the height of each sample, 

Z is the width of each sample. 

The data has the following sizes: 

Normal_Data_Training: (1347, 221, 375)Normal_Data_Validation: (1347, 221, 375)Anomalous_Data_Validation: (80, 221, 375)Normal_Data_Test: (1347, 221, 375)Anomalous_Data_Test: (79, 221, 375)