#2 2026

CONTENTS №2

Discussing a Topic…

Equipment monitoring, maintenance, and repair strategies

Lapkis A.A., Novoselov G.N., Kashchenko P.V., Povarov P.V.  Raw data acquisition for implementing risk-based strategy of A-plant equipment maintenance

Practical aspects of a risk-based strategy for maintenance and repair of pipeline valves at A-plants are considered. The paper argues that the profound reliability data analysis is a key to the comprehensive risk-based strategy. The authors propose a methodology for assessing operational risk based on the analysis of failure rates extracted from electronic fault logs, taking into account the equipment mean time between failures. The study identifies key issues related to data heterogeneity and limitations and proposes solutions by aggregating valves types and failure modes. Using data from operating power units as an example, the presence of both random and wear-related components in defects is demonstrated. It has been ascertained that for Soviet-built power units, there is a significant increase in the failure rate after 25-30 years of operation, while for new power units this effect is insignificant. The obtained results are valuable for the development of a “live” probabilistic safety analysis, which allows for the actual equipment state, as well as for optimizing maintenance and repair strategies to improve the safety and profitability of A-plant operation.

Keywords: reliability, valves, A-plant, failure rate, mean time between failures, ageing, maintenance and repair, probabilistic analysis.

Pechenin V.A.  Cutting tool condition monitoring based on digital datasheets

The paper proposes an approach to cutting tool condition monitoring during machine working based on its digital datasheet. The digital datasheet is a statistical reference of the tool’s design condition developed using vibration signals, wavelet decomposition, principal component method, and Mahalanobis distance. It enables real-time detection of wear and abnormality, thus preventing defective parts and equipment downtime. Experiments on a milling machine have confirmed the method’s effectiveness for critical wear identification. The approach is adapted for automated manufacturing and complies with GOST R ISO 7870-1-2011.

Keywords: cutting tools, digital datasheets, vibration displacement, wear, wavelet decomposition, principal component method, statistical tests.

Papulovskata N.V., Chebotarev K.V.  Machine vision in urban agriculture: classification of microgreen seed images

The paper describes a system for automated sowing of microgreens on vertical city farms using machine vision technologies. The main goal is to create an effective neural network for recognizing and classifying seeds of various agricultural crops. During the study, a comparative analysis of five neural network architectures was undertaken, including an in-house convolutional network, Inception_v3, Xception, ResNet-101, and EfficientNet_b3. Key results showed that the self-trained network had demonstrated the highest classification accuracy (88%), thus outperforming pre-trained models. The study found that background removal procedures would worsen the recognition quality rather than improving it. The optimal solution turned out to be a strategy of processing the images by slicing them into 224×224 pixel patches. The practical significance of the work is confirmed by the successful deployment of the system on a Raspberry Pi microcomputer with acceptable values of performance indicators.

Keywords: machine vision, vertical city farm, dataset, seed recognition, automatized sowing, neural network.

Kobrunov I.V.  Evaluation of the efficiency of network interaction in systems based on the CAN protocol

The paper discusses a method for conducting a study of upper-level CAN protocols for estimating network interaction efficiency. The generated traffic on the CAN bus is taken into account, that makes it possible to evaluate its effectiveness by collecting statistics on certain characteristics. To this end, a special criterion is introduced that considers sequences of individual message instances as atomic elements with subsequent generalization of the set of resulting values into a final assessment. An application example of studying the system bus protocol of an industrial logistics robot is included.

Keywords: performance evaluation criterion, network interaction, CAN protocol, distributed control system.

Discussing a Topic…

Automation and digitalization in power engineering tasks

Nenashev S.A., Pisklenov T.A., Nenashev V.A.  A method for monitoring dangerous woodland areas along power lines based on point clouds with the help of machine vision systems

A method for automated monitoring of dangerous woodland areas in protection zones of power lines is presented. The method is based on the processing of a cloud of points obtained by means of airborne laser scanning with machine vision techniques. The objective is to detect the trees which may damage the integrity of power grid infrastructure upon falling. The method is exemplified in a software suite whose testing included seasonal (summer/autumn) analysis. which had confirmed the method’s efficiency in the differentiation of permanent threats from the temporary ones caused by the seasonal dynamics of hardwood crowns. The visualization of results in the geoinformation environment has demonstrated the possibility of detecting critical segments and developing dangerous zone maps.

Keywords: machine vision, machine learning, point cloud, neural networks, woodland monitoring, power lines, airborne laser scanning.

Matrenin P.V.  Methodology for creating trusted intelligent forecasting systems for power industry applications

The paper discusses the design principles of trustworthy artificial intelligence systems for predicting electric power generation and consumption. It introduces the concept and describes the properties of a of trustworthy intelligent forecasting system based on the existing Russian regulatory and technical guidelines. The application of an ontological approach to taking into account expert knowledge about the subject area when constructing predictive machine learning models is substantiated. An extension of the CIA Triad by including the property of relevance is proposed, and its use is justified for trusted intelligent forecasting systems in the electric power industry. The paper shows how the author’s models and methods ensure the fulfillment of trustworthiness requirements and expand the boundaries of applicability of artificial intelligence technologies in industrial applications with a high cost of error.

Keywords: 6/10 KV power line, single-phase ground faults, noncontact sensors, short-time and fast Fourier transform, spectral signal processing, machine learning, classification.

Bramm A.M., Halyasmaa A.I.  A contactless control system for changing operating modes of 6/10 kV power transmission lines using electric and magnetic field signals

The paper presents a contactless system for monitoring the changes in the operating modes of a 6/10 kV power transmission line based on the signals from noncontact electric and magnetic field sensors. The work aims at the automatic detection of the mode change moment and changes classification without connecting devices to live parts. Three capacitive and six inductive contactless sensors were used to record the electric and magnetic fields of the line. The detection of the mode change moment is performed on the basis of a short-term Fourier transform (window of 256 samples, step 128) and calculation of the change in the energy of the fundamental harmonic of the sum of signals from a group of contactless sensors. To describe an event, a vector of spectral features Z is formed. The classification is implemented using an interpretable decision tree with a depth of up to three branches with aggregation of decisions by channels using the soft-voting method and a threshold value of 0.55 for the model confidence for decision making. The model was trained on three classes (normal mode, single-phase ground fault before and after the sensor installation point) and tested on 129 files. The average classification accuracy was 70.5%. The majority of errors were caused by the model’s low confidence. Experimental verification was performed through laboratory tests under single-phase ground fault conditions, as well as through field tests on an active 10 kV line under voltage reduction during design operation. The obtained results confirm the applicability of the approach for monitoring events in 6/10 kV networks and forming a database for further analysis.

Keywords: 6/10 KV power line, single-phase ground faults, noncontact sensors, short-time and fast Fourier transform, spectral signal processing, machine learning, classification.

Shvyrov V.V., Kapustin D.A., Sentyay R.N.  Methods for using large reasoning language models for security analysis of software code

Large language models (LLMs) with reasoning capabilities have been actively developed in recent years. The most prominent example is the DeepSeek-R1 model family. However, there are practically no studies of reasoning models in the field of security analysis and classification of software code vulnerabilities. Detection and classification of various vulnerability types is a problem which arises in practice. The paper presents an assessment of the capabilities of a number of LLMs with reasoning support for detecting and classifying the most dangerous types of vulnerabilities. In addition, it discusses various methods of prompt engineering for increasing the accuracy of LLM responses, in the context of the task of analyzing the security of software code. Examples of prompts for detecting common types of vulnerabilities are included. The risks of using LMM during software development are outlined.

Keywords: software code security analysis, large language models, reasoning support, vulnerability detection, prompt engineering.

Pankov A.N., Kopylova E.V., Yakushevich A.N.  The impact of training data quality on metrological provision of intelligent instrumentation based on machine learning

The development of artificial intelligence technologies accelerates the implementation of machine learning (ML) algorithms for direct and indirect measurements and condition monitoring of objects and processes. This changes the requirements for metrological provision: along with traditional sources of error, it is necessary to take into account the influence of training data quality on the accuracy and uncertainty of the result. The paper overviews the guidelines and scientific publications on metrological provision of measuring instruments based on AI and uncertainty assessment in ML tasks. It examines approaches to data quality management in the GOST R 71484 series, the requirements of GOST R 71562, and the requirements for datasets for the development and verification of models for indirect measurements (GOST R 71688). It is shown how data quality indicators (completeness, accuracy, representativeness, etc.) can be compared with the components of error in intelligent measuring instruments. Uncertainty estimation methods are analyzed, including NPL MS 34 report for regression, and the study of the impact of measurement uncertainty on ML results in smart sensors, as well as probabilistic models of virtual measurements with separation of aletoric and epistemic uncertainty. A conclusion is made about the need to integrate data quality requirements and develop methods for decomposing uncertainty by sources: data, model, and information sources.

Keywords: aletoric and epistemic uncertainty, virtual measurements, intelligent instrumentation, artificial intelligence, data quality, indirect measurements, machine learning, metrological provision, datasets, epistemic uncertainty.

Droshnev V.A., Kirillov I.V., Farafontov V.A., Palgov A.V.  Application of embedded cryptographic information protection tools for data transmission over wireless networks in automated systems using LoRa technology

The paper discusses the problem of monitoring wells in the late stage of development in the absence of automation systems and power supply lines. A solution is proposed based on the creation of energy-independent telemetry systems using LoRa long-range radio communication technology. The main focus is on the problem of ensuring information security of transmitted data under conditions of strict energy consumption restrictions. The impossibility of using traditional overlay security tools is demonstrated. The proposed solution is based on the use of built-in cryptographic information protection tools. Specific implementations are considered, namely, the ViPNet SIES Core Nano cryptochip and the ViPNet SIES Core cryptomodule, which provide data encryption before transmitting it over the radio channel. The results of the practical implementation of the solution at the sites of Gazprom Dobycha Orenburg LLC are presented, which confirm the system's compliance with information security requirements along with its high energy efficiency.

Keywords: long-range radio communication technology, cryptochip, telemetry, energy independence, information security, embedded systems, cryptographic protection, well.

Balabanov A.V., Timoshev P.A., Kasimov A.M., Sovlukov A.S., Vinogradova E.P.  A test bench for studying a microturbine of a hybrid pneumatic-electric motor with a jet-membrane control system

The paper examines a microturbine, which can be used, for example, as a power drive in a radiation-resistant unmanned vehicle with a jet-membrane control system. It presents the results of developing hard- and software for a test bench for studying the speed and energy characteristics of a microturbine.

Keywords: microturbine, speed characteristics, energy characteristics, test bench.