Ball screw drive surface defect dataset for classification

doi:doi:10.35097/1511

Ball screw drive surface defect dataset for classification

Abstract: The dataset contains of 21835 150x150 Pixel RGB images of the surface of Ball Screw Drives. 11075 of these images are images without surface defects whereas the rest shows images with surface defects in form of so called pittings. So the dataset is evenly split over the classes. Pittings result from surface disruption and can ultimately lead to the breakdown of the component. To keep the availability of machines high it is important to find surface defects in time. The here presented dataset gives researchers and practitioners the possibility to train and test models for the classification of surface defects on machine tool elements. TechnicalRemarks: Images including *_ in the name are rotated by 90°. This originates from the process of image acquisition and does not harm the quality of the dataset but can be seen as a data augmentation technique. This can be reversed.

Images with "N" in the file name are images without defect whereas images with "P" in the file name are images showing Pittings.

BibTex:

@dataset{Schlagenhauf_Tobias_2023,
    abstract = {Abstract: The dataset contains of 21835 150x150 Pixel RGB images of the surface of Ball Screw Drives. 11075 of these images are images without surface defects whereas the rest shows images with surface defects in form of so called pittings. So the dataset is evenly split over the classes. Pittings result from surface disruption and can ultimately lead to the breakdown of the component. To keep the availability of machines high it is important to find surface defects in time. The here presented dataset gives researchers and practitioners the possibility to train and test models for the classification of surface defects on machine tool elements.
TechnicalRemarks: Images including *_ in the name are rotated by 90°. This originates from the process of image acquisition and does not harm the quality of the dataset but can be seen as a data augmentation technique. This can be reversed.

Images with "N" in the file name are images without defect whereas images with "P" in the file name are images showing Pittings.},
    author = {Schlagenhauf, Tobias},
    doi = {10.35097/1511},
    institution = {RADAR},
    keyword = {'Classification', 'Condition Monitoring', 'Dataset', 'Machine Learning', 'Mechanical Engineering', 'Surface Inspection'},
    title = {Ball screw drive surface defect dataset for classification},
    url = {https://service.tib.eu/ldmservice/vdataset/rdr-doi-10-35097-1511},
    year = {2023}
}