Fatigue Monitoring of Composites (F-MOC) dataset
Description
Dataset for reproducing the results of the third dataset, namely the Fatigue Monitoring of Composites (F-MOC) dataset, of the paper "A robust generalized deep monotonic feature extraction model for label-free prediction of degenerative phenomena". The setup comprises a 100 kN MTS fatigue controller connected to a bench machine, an impact cannon, an acoustic emission system, and two cameras used for DIC measurements. The cameras are capturing images of the specimen periodically every 50 seconds. To capture the acoustic emission signals generated during fatigue, an AMSY-6 Vallen Systeme GmbH, an 8-channel AE system with a sampling rate of 2 MHz, was utilized. A single broadband single-crystal piezoelectric transducer was affixed to the side of specimens using a clamping device situated between the lower grip of the fatigue machine and a safety aluminum cylinder. Ultrasound gel was applied for optimal acoustical coupling between the sensor and specimen surfaces. To ensure proper connectivity, a standard pencil lead break procedure verified the specimen-sensor connection before both the fatigue test and setting the threshold at 50 dB. The material at hand is a unidirectional Prepreg tape Hexply® F6376CHTS(12K)-5-35. The laminate is manufactured using a hand lay-up of [0/45/90/ -45] and is cured in an autoclave at a temperature of 180 Celsius and pressure of 9 bar for 120 minutes as recommended by the manufacturer. The laminate is consequently cut to obtain specimens of 400 mm x 45 mm with an average thickness of 2.28 mm. The steps to reproduce the dataset are shown below. Additional details about the experimental setup can be found at the relative link at the end of this page.
Files
Steps to reproduce
Extracting the "dataset.zip" file gives 2 folders, namely the "ACOUSTIC" and the "DIC" folder corresponding to the acoustic emission signals and the image data. There are 7 specimens, namely spec1, spec2, ..., spec7. In the "ACOUSTIC" folder, each specimen is a .pridb file that contains the relevant low-level features of the acoustic emission starting from the structure's perfect condition until the end of life. The "DIC" folder contains all of the images of each specimen in a sequential manner starting again from the structure's perfect condition until the end of life. Each subsequent image is taken every 50 seconds. Therefore, the name of each image is spec*_****, where the first unseen digit corresponds to the specimen number, and the last 4 digits to the corresponding timestep that the image was captured (in seconds). To use these data, simply put the 2 extracted folders inside the working directory of the code provided in GitHub (see relevant link below). The file "models.zip" contains the pretrained models of this dataset. If the users do not want to train from scratch the model utilizing the F-MOC dataset, they have to follow the instructions in the corresponding GitHub repository to properly install these models tinthe working directory.