Augmented semantic segmentation for the digitization of grinding tools based on deep learning

In order to analyze various process characteristics, grinding simulations can be used, which need accurate models of the tool and the individual grains. For this purpose, grinding tools can be digitized. To identify characteristic grains from a large number of measurements, each individual grain has to be analyzed and separated from the bond manually. Therefore, a deep learning-based methodology was developed to achieve a high segmentation accuracy of the grain boundaries efficiently. Additionally, a data augmentation approach was investigated to limit the data necessary for learning. The model transferability was quantified by analyzing different states of tool wear.