Nowadays pelletizing drums are widely used in the steel industry. These units are characterized by high endurance and low cost of maintenance. However, use and control of these units in the process of coarsening have a number of issues. For most of the cases pelletizing drums are “black box” and control accuracy can not be estimated exactly. It is explained by low existing theoretical basis of this production process. Particularly it is tied up with the variability of the bulk materials (charges) parameters supplied to the unit. Overcome of this issues can be reached with development of intelligent control systems for drum pelletizing machines. Main requirement for such systems is possibility to level or consider the effect of charges properties variability in control. However, it is necessary to study the behavior of bulk materials inside the units. Visual assessment of pelletization does not allow to evaluate the ongoing physical processes. Development of mathematical and numerical models can help studying the process and take a lot of parameters into account including charges properties and even interaction with water. But the adequacy of the resulting models also has to be clarified using physical devices to record or capture bulk materials behavior inside the units. This research proposes a DEM simulation test of the concept for bulk material behavior control through the recognition of the mixture movement fragments using special capsules. This part is dedicated to the simulation model set up and extracting the particles trajectories for further processing.

This article is devoted to the prediction of the residual life based on an estimate of the technical state of the induction motor. The proposed system allows to increase the accuracy and completeness of diagnostics by using an artificial neural network (ANN), and also identify and predict faulty states of an electrical equipment in dynamics. The results of the proposed system for estimation the technical condition are probability technical state diagrams and a quantitative evaluation of the residual life, taking into account electrical, vibrational, indirect parameters and detected defects. Based on the evaluation of the technical condition and the prediction of the residual life, a decision is made to change the control of the operating and maintenance modes of the electric motors.

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.