The paper presents a methodology for training neural networks for vision tasks on synthe-sized 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.

The causes of failure of the tangential rotary cutter bits of the road header during stonedrift in rocks of medium strength are analyzed in the article. It was revealed that the most typical cause of failure of cutter bits is premature wear of the toolholder (body) of the cutter bit. It is well known that the most effective way to improve the wear resistance is to increase hardness. The influence of the thermomechanical treatment of the material of the cutter bit toolholder on its hardness is studied. It was established that the thermomechanical treatment of the cutter bit toolholder material results in the increase of its hardness. It was found that the increase of material hardness is proportional to the increase of material strain intensity during thermomechanical treatment. It was concluded that the use of thermomechanical treatment can lead to the increase of both the hardness and wear resistance of the cutter bit material.

The problem of DEM contact models choice is described in the article. An analysis of the input data calibration, including a choosing the contact model, is performed. The main theoretical positions of the most popular models are considered, their comparison is made on the basis of a specially designed stand. Description and justification of the designed stand are considered. The analysis of the obtained results is carried out. Recommendations on the choice of the DEM contact model are given. The results show that it is time-efficient to use non-linear contact model for future model calibration.