#4 2026

CONTENTS №4

Dozortsev V.M., Korostelev A.Ya., Malashkevich A.V., Anosov A.A., Agafonov D.V.  Advanced process control: origin and evolution, market state and outlook, Russian contribution, trends, and import substitution

The paper describes the origin and evolution of advanced process control systems, primarily those based on predictive models (MPC systems) with the focus on Soviet and Russian contribution to this process automation segment. It reviews the market state and outlook, analyzes MPC development trends in the context of other advanced process control and production management solutions. Import substitution problems caused by the withdrawal of foreign MPC vendors from the Russian market are discussed.

Keywords: model predictive control, advanced process control, APC systems, real-time optimization (RTO), artificial intelligence.

Discussing a Topic…

Computer vision applications

Okhotkin G.P., Potapov A.G.  Defect detection in metal-roll with the help of YOLOv8 loss function modified using physics-informed gradient analysis

The problem of automatic detection of defects in industrial metal sheets is investigated. For solving it, the paper proposes a modified method for training the YOLOv8 neural network, which integrates the loss function based on physically-informed gradient analysis (Physics-Informed Gradient Loss, PIG-Loss). Unlike traditional approaches based only on coordinate regression, the proposed method allows for the textural and geometric features of defects by analyzing the gradients of brightness and object structure anisotropy using the Sobel operator. Experimental studies on the GC10-DET dataset showed that the implementation of the PIG-Loss component enables an increase in localization accuracy (mAP50-95) and a reduction in the number of false positives caused by industrial noise (glare). The results confirm the effectiveness of a hybrid approach combining deep learning with methods for mathematical analysis of image structure.

Keywords: deep learning, machine vision, flaw detection, physics-informed neural networks, object detection, steel industry, gradient analysis, Sobel gradient.

Krasnov A.E., Turchinsky K.A. Automated system for visual quality inspection using halftone images

An automated system for visual product quality inspection using halftone images was developed for online use with limited computing resources and a specified decision-making time. Preprocessing selects the product area (region of interest, ROI) and normalizes the brightness, after which three applied recognition branches are used. For rejecting cast parts, a fast route is used on compact texture and shape features (local binary patterns and geometric silhouette characteristics) with a linear classifier. A small convolutional neural network is used for six-class classification of surface defects. Anomaly detection without labeling is implemented using a convolutional autoencoder. The solution's applicability for online sorting and rejection of products in a continuous mode is demonstrated.

Keywords: visual inspection, halftone images, region of interest, autoencoder, anomaly detection, threshold calibration.

Sharova D.E., Pakhomov A.S., Akopyan R.O., Semenov D.S.  An algorithm for recognizing water meter readings based on computer vision methods

The problem of automatic recognition of water meter readings from a photograph is examined. An approach based on the use of standard photographic recording tools and computer vision algorithms is proposed as a possible direction for the digitalization of data collection processes. The paper describes the features of forming a training dataset based on real images of metering devices used in dwelling houses, as well as measures to ensure the stability of the algorithm under various photographing conditions. A two-stage image processing algorithm has been implemented, including the selection of the reading area and the recognition of digital values. Test results using real-life photographs are presented. They confirm the solution’s applicability in municipal information systems.

Keywords: water meters, automatic reading, computer vision, image processing, digitalization, housing and communal services.

Voronov R.M., Nenashev S.A., Nenashev V.A. Automated inspection of product defects using machine vision systems

The paper presents a method for automated inspection of geometric defects of products based on the processing of point clouds obtained using three-dimensional lidar scanning technologies and machine vision systems. The problem of detecting and recognizing deviations in the shape and surface of parts compared to a reference model is solved. The method rests upon the algorithms for point cloud processing and local analysis of defect geometry by estimating model parameters based on random samples ensuring the generation of deviation maps and the determination of defect locations. A software suite has been developed that implements the proposed approach and includes modules for data preprocessing, analysis, and visualization. Experimental studies were conducted which confirmed the method’s high accuracy and efficiency in testing products of various complex shapes. The results demonstrate the outlook for integrating the technology into industrial quality control systems for reducing defects and improving the stability of production processes.

Keywords: machine vision, machine learning, point cloud, neural networks, 3D laser scanning, quality control, defects, production automation.

Kondusova V.B., Amandusov A.I., Ovechkin M.V.  Methodological support for production tooling CAD in KOMPAS-3D environment

Modern production is developing in conditions of technological sovereignty, a high need for product individualization, and a strive to reduce product lifecycle. In this context, digital technologies, including computer-aided design (CAD), are increasing their role as a basis for technological advances in production environment. The manufacturers face the need for continuous improvement the process efficiency, where the quality and speed of production tooling development plays a paramount role. The description of methods for providing a system for automated design of production tooling becomes an important mean for complex automation, making it possible to solve complex engineering problems at a basically new level.

Keywords: CAD, production tooling, verification of calculations, integration.

Kiryushin P.N.  Functional safety and cybersecurity of emergency shutdown systems in oil refining process control

Today, a pressing challenge is finding solutions to ensure the safety of hazardous industrial facilities by creating cyber-protected, functionally safe emergency shutdown (ESD) systems. The development of target requirements based on the analysis of hazards and risks is the most important stage in the creation of ESD systems. The paper overviews international and national standards on functional safety and information security of automated control systems. It presents a framework for joint analysis of process hazards, shows the distribution of protection functions across protection layers, and discusses the joint specification of requirements taking into account the provisions of the Russian Federation regulatory legal acts on the categorization of critical information infrastructure facilities.

Keywords: process control system, emergency shutdown systems, lifecycle, functional safety, information security, critical information infrastructure, instrumented safety function, standardization.

Kochueva O.N., Shcherbakov I.A.  Machine learning algorithms for automatic determination of bit position in horizontal well drilling

In horizontal oil well drilling it is necessary to determine the position of the bit in space with high accuracy. Traditionally, geosteering tasks were performed manually by experts, although in the recent 10 years, automatic geosteering systems have been appearing. The paper offers a two-level procedure to solve a subtask of the automatic geosteering, namely, determining the stratigraphic position of the bit relative to the reservoir. At the first level, a classification problem is solved for selecting a reference well, the logging data of which will subsequently be used to solve the second problem: determining the position of the bit in the target interval. To solve these problems, the traditional support vector machine (SVM) and the relevance vector machine (RVM) were applied. The second one uses the automatic relevance detection (ARD) process during training. The merits and flaws of both methods are discussed. Both algorithms were tested in the process of solving a binary classification problem: determining whether sensor readings at a given time instant correspond to a target interval or to a segment outside the target interval. It is shown that the RVM demonstrates prediction results not inferior to the SVM ones in terms of accuracy metrics, while using much fewer vectors. This results in a less calculations of the kernel function, which in the long term provides a significant reduction in the computational complexity of prediction that is important for supporting drilling operations.

Keywords: automatic geosteering, horizontal drilling, machine learning, support vector machine, relevance vector machine.

Akhunova D.G.  Prerequisites for the creation and architecture of a control loop for positioning mobile fire crews

The paper discusses the prerequisites and development stages of a hybrid fire safety system for a metropolis, including stationary depots and mobile fire crews (MFC). A comprehensive four-stage method for controlling the positioning of MPR (forecasting, placement optimization, movement planning, routing) is presented and a modular architecture of the control loop that implements it is synthesized. With the example of St. Petersburg, a hypothetical assessment of efficiency was conducted, which showed, in particular, that the optimal number of MFC is 10–12 units, and the specific efficiency of a MFC is 1.5–4 times higher than the construction of a new depot.

Keywords: hybrid system, mobile fire crew, positioning, prerequisites, control loop architecture, forecasting, optimal positioning, routing.

Journal’s Club

Smirnov S.V.  Application of powered exoskeletons in industry

Powered exoskeletons are devices equipped with electric motors, hydraulic drives or pneumatic systems, sensors, and batteries. The paper shows that such devices have received a new impetus for development owing to the advances in science and technology, automation tools and systems. It gives examples of industrial exoskeletons aimed at enhancement, support, or restoration of human movement.

Keywords: exoskeleton, industry, active, passive, servo drive, sensor.