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.

This article is a review of current scientific problems in the field of automation of the electromagnetic levitation melting process control of non-ferrous metals and potential solutions using modern digital technologies. The article describes the technological process of electromagnetic levitation melting as a method of obtaining ultrapure metals and the main problems of the automation of this process taking into account domestic and international experience. Promising approaches to control the position of the melt in the inductor in real time on the basis of vision systems are considered. The main problems and factors preventing the mass introduction of levitation melting in the electromagnetic field to the industry are highlighted. The problem of passing the Curie point by the heated billet and the effect of the billet’s loss of magnetism on the vibrational circuit of the installation and the temperature of the inductor are also considered. The article also reflects key areas of research development in the field of levitation melting, including: optimization of energy costs, stabilization of the position of the melt in the inductor, predictive process control, and scaling of levitation melting units. The concept of a digital twin based on a numerical model as a component of an automatic process control system for the implementation of inductor control and prediction of process parameters of the melt is presented. The possibility of using vision for visual control of the melt position in the inductor based on video images for its further stabilization in the inductor and increasing the accuracy of numerical simulation results by specifying the real geometry of the melt in parallel with the calculation of the model itself is considered.

No abstract

The research is dedicated to the development of special devices (capsules) that can be used to control the mining ore behavior in the technological unit in order to increase processes efficiency. In the first part of the article, the choice of the discrete element method for generating various particle trajectories in the unit (drum pelletizer) was substantiated. This part describes the specific technologies that were used to recognize the pelletizing mode. In particular, conversation of paths to sensor readings is implemented using the Matlab Sensor Fusion and Tracking Toolbox. The obtained readings were processed using two neural network classifiers (DNN and LSTM). As a result, stable models for recognizing the pelletizing modes of the unit were obtained. LSTM recognition accuracy is greater than DNN. The developed approach can be used to recognize the operating modes of other technological units. In addition, data on particles trajectories can be used to improve DEM models of technological processes. Future work consists of the capsule physical implementation and testing the recognition algorithm on a real unit.

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.