The maintenance of oil pumps is a complex task for any operating organization, and for an industrial enterprise in the oil and gas sector of the economy, this issue has a high degree of urgency. One of the reasons for this is a wide spread of pumping equipment in all areas of oil and gas enterprises. At the same time, an aggressive environment, uneven load, remote facilities, and harsh climatic zones (especially in the areas of the Arctic region or production platforms) are factors that make it relevant to develop special systems that help or simplify the maintenance of pumping equipment. Dynamic modeling is one of the modern technologies which allows for solving the urgent issue of assessing the technical condition of equipment. It is the basis of systems that carry out diagnostics and prognostic calculations and allow for assessing the dynamic state of objects under various conditions of their operation, among other functions. Augmented reality technology is a technology that allows for reducing the time for equipment maintenance by reducing the time for searching and processing various information required in the maintenance process. This paper presents an investigation of the effectiveness of an augmented reality and a dynamic simulation system collaboration in oil pump maintenance. Since there is insufficient research on the joint application of these two technologies, the urgent issue is to prove the effectiveness of such collaboration. For this purpose, this paper provides a description of the system structure, gives a description of the development process of the augmented reality system application and tests the application using Microsoft HoloLens 2.

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 problem of DEM models calibration is considered in the article. It is proposed to divide models input parameters into those that require iterative calibration and those that are recommended to measure directly. A new method for model calibration based on the design of the experiment for iteratively calibrated parameters is proposed. The experiment is conducted using a specially designed stand. The results are processed with technical vision algorithms. Approximating functions are obtained and the error of the implemented software and hardware complex is estimated. The prospects of the obtained results are discussed.