Search

The paper presents a methodology for training neural networks for vision tasks on synthesized data on the example of steel defect recognition in automated production control systems. The article describes the process of dataset procedural generation of steel slab defects with a symmetrical distribution. The results of training two neural networks Unet and Xception on a generated data grid and testing them on real data are presented. The performance of these neural networks was assessed using real data from the Severstal: Steel Defect Detection set. In both cases, the neural networks showed good results in the classification and segmentation of surface defects of steel workpieces in the image. Dice score on synthetic data reaches 0.62, and accuracy—0.81. [ABSTRACT FROM AUTHOR]

DEM parameters calibration is the most important step in preparing a DEM model. At the same time, the lack of a universal approach to DEM parameters calibration complicates this process. The paper presents the author's approach to creating a universal calibration approach based on the physical meaning of the friction coefficients and conducting symmetrical experiments at full scale and in a simulation, as well as the implementation of the approach in the form of a physical test rig. Several experiments were carried out to determine the DEM parameters of six material–boundary pairs. The resulting parameters were adjusted using a refinement experiment. The results confirmed the adequacy of the developed approach, as well as its applicability in various conditions. The limitations of both the approach itself and its specific implementation in the form of a test rig were identified. [ABSTRACT FROM AUTHOR]