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.

Mining enterprises are widely introducing digital technologies and automation is one of such tools. Granularity monitoring, namely, the size determination of rock mass pieces is a common operational component of the processes that extract minerals by open-pit mining. The article proposes an approach that, in addition to the lump size distribution, makes it possible to estimate the lump form distribution as well. To investigate the effectiveness of monitoring the form of blasted rock mass lumps, the authors conducted experiments in four stages related to the rock condition. They include geological occurrence, explosive crushing, trommelling, and mill crushing. The relationship between these stages is presented and the change in the lumps fragment form is traced. The present article proposes an informational and analytical model of the processes at mining enterprises, extracting minerals by open-pit mining, as well as an algorithm for determining the lumps form and obtaining their distribution in the rock mass.

The aim of this paper is to study the numerical modeling of the crushing process in a jaw crusher with complex motion jaws using Discrete Element Method (DEM). The main goals are to ensure the conformity of numerical simulation results to the experimental data, optimize the operating modes and the crusher design and develop automated verification and calibration methodology. It is offered to use automated selection of mathematical model parameters using large- or laboratory-scale experiments instead of deriving parameters from conventional intensive material tests (e.g. drop-weight test). Experimental data were obtained from a large-scale crusher designed and manufactured by the company “Mekhanobr-Tekhnika”. Both jaws have their drive shafts located asymmetrically; one jaw is driven in the upper part, the second at the base. The crushing process involves fragmentation of single smoothed rectangular granite particles of the same weight. During the experiment, the required parameters such as particle size distribution were obtained for further computer simulations. Based on the obtained experimental data, the Rocky DEM model was built. Simulation breakage parameters were calibrated using automated parameter selection methodology based on multiple set of parameter-controlled simulations. A good correlation of the experimental data and numerical simulation data was obtained. The methodology in combination with well-known capabilities of simulation modeling will help reduce the research time in the development of new mineral processing equipment.