Application of the Neural Network Model U-Net for Detecting the Wear Profile of a Part

The possibility of using deep neural network models, in particular the U-Net architecture, for automated detection of the profile of a worn part is examined. Wear of working elements, especially in agricultural machinery, sharply reduces the efficiency of the equipment and increases the costs of its maintenance. Traditional diagnostic methods have a number of limitations, including subjectivity of assessments, the need to stop equipment, and the low accuracy of quantitative assessment of wear. A method of semantic segmentation of images using the U-Net neural network is proposed, which allows for the automatic formation of a profile of a worn part. The stages of data preparation, model training, and image analysis are described, which ensures accurate obtaining of the part profile. The effectiveness of the method was confirmed experimentally using a deep-ripping chisel as an example.