2OBe/Ball_Screw_Drive_Classification

Ball Screw Drive Defect Classification

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. Images ending with "_" 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 some kind of data augmentation. 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. Some images are covered in grease from light brown to darker and black tones due to the fact that the images are taken from a Ball Screw Drive under technical operation. For those images it is sometimes hard (even for the human expert) to decide about the unerlying true class - Especially for edge cases. To record the images, a camera was mounted onto the BSD-nut s.t. the camera points vertically on the spindle. The spindel is turned under the camera. This dataset does not share classes with any of the models pre-trained on imagenet or cifar or the like. Probably you have to either only extract features or fine-tune resp. train a model.

License: cc-by-nc-4.0

Citation

BibTeX:

@misc{Schlagenhauf2021_1000133819, author = {Schlagenhauf, Tobias}, year = {2021}, title = {Ball Screw Drive Surface Defect Dataset for Classification}, doi = {10.5445/IR/1000133819}, keywords = {Condition Monitoring; Surface Inspection; Dataset; Machine Learning; Classification; Mechanical Engineering}}