The hydrotransport system of Mikhailovsky GOK PJSC used for iron ore processing tailings was studied with the aim of establishing its performance in slurry pipelines with and without polyurethane coatings of their internal surfaces under mass concentration variations of 10 to 70 %, as well as for preparing related recommendations for the enterprise. Theoretical and laboratory studies were completed with the development of a methodology for recalculating the experimental laboratory data for the specific head loss in industrial pipelines. It has been established that the use of internal polyurethane coatings significantly reduces the specific head losses in hydraulic transportation of thickened tailings slurry for iron ore processing tailings. The respective feasibility calculations confirm the economic efficiency of using steel pipes with internal polyurethane coating in hydrotransport slurry pipelines. The calculated specific head losses for the pipelines with a polyurethane coating in the hydraulic transportation of processing tailings slurry with the mass concentration of the solid phase of 6.95 % are almost 1.8 times lower than in steel pipelines without such coating. It is proposed to use the results obtained in the reconstruction of the tailings slurry hydrotransport system at Mikhailovsky GOK PJSC with a conversion to hydraulic transportation of bulk concentration tailings thickened to 35–40 % in slurry pipelines with an internal polyurethane coating, which will ensure energy savings in the hydrotransport process.

In this article actual topics concerned with mechanical properties of bulk materials, usage of computer vision and artificial neural networks in this research are discussed. The main principles of the system for analysis of bulk materials mechanical characteristics are described. Bulk material outflow behaviour with predefined parameters (particles shapes and radius, coefficients of friction, etc.) was modelled. The outflow was modelled from the calibrated conical funnel. Obtained dependencies between mechanical characteristics and pile geometrical properties are represented as diagrams and graphs.

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.