#6 2025

CONTENTS №6

Discussing a Topic…

Artificial intelligence technologies in industry

Introduction

Dozhdev V.S., Shantaev E.B., Daraselia L.Sh., Khramov A.E.  Integrated approach to industrial data quality assurance for solutions based on artificial intelligence

Industrial application of artificial intelligence (AI) technologies within the Industry 4.0 concept requires effective industrial data management. However, data heterogeneity, their poor quality and the difficulties real-time processing remain the key challenges hampering wider dissemination of AI technologies. The paper systematizes the approaches to industrial data management, offers data quality criteria, and makes practical recommendations on their preparation for effective AI model training at real-world plants. It presents a data quality assessment procedure based on four-level criteria, which comprise information accessibility, accuracy, completeness, and security. The approaches developed in the study enable significant improvement of industrial data management efficiency in AI applications.

Keywords: industrial data management, information-management systems, prediction, data quality criteria, artificial intelligence.

Matrenin P.V., Khamitov R.N., Sergheev N.N.  Application of artificial intelligence in short-term prediction of electricity consumption by industrial enterprises with reference to production factors

The application of artificial intelligence in the short-term prediction of electric power consumption by industrial enterprises are examined. The paper studies the effect of production and whether factors on the forecast accuracy. The study results are presented, comparative analysis of ensemble and neural network predictive models is undertaken, the application of the ensembles of regression decision trees is substantiated. The benefits of forecast accuracy improvement at the wholesale electricity and power market are estimated.

Keywords: electricity consumption prediction, artificial intelligence, industrial enterprise, ensemble models, wholesale electricity and power market.

Mironenko Ya.V., Halyasmaa A.I.  Architecture of a decision support system based on the mining of electrical system monitoring data

The paper discusses the architecture of an intelligent decision support system for monitoring and diagnosis of electrical systems of industrial enterprises with examples from compressor units used in oil refining and gas processing. Key architectural components are described, application results are presented. The architecture includes data acquisition, diagnosis, prediction, and decision support modules. The system’s featured property is the application of data mining based on ensemble models in classification tasks and neural networks in prediction ones.

Keywords: artificial intelligence, health monitoring, technical diagnosis, prediction, electrical systems, compressor units, oil refining and gas processing industries.

Mitin G.V., Panov A.V.  Data mining system architecture in flaw detection processes at complex electronic device manufacturing

The paper describes the multilevel architecture of a data stream mining system, which uses machine learning techniques able of working under considerable information noise. The system is underlain by an information pattern search mechanism, which enables the detection of regularities at various abstraction levels regardless of the original data nature. Target systems based on the proposed architecture can be used in quality control processes of complex electronic device manufacturing. Along with automatic flaw detection, such systems may reveal the causes of flaw occurrence.

Keywords: information system, data mining, machine learning, information noise filtration.

Molchanov D.V.  Analysis of flame sensor readings based on data recovery with the help of autoencoder for anomaly detection

The paper discusses the detection of anomalies in flame sensor readings in gas preheaters. The problem is important for ensuring safety and reliability of industrial systems. Sensor failures or false positives may entail serious consequences such as emergency equipment shutdowns, slowdowns, higher maintenance costs, and personnel safety threats. To overcome the challenge, a method based on autoencoder application is offered. Autoencoder is a neural network. It is trained on design operation data and is able to detect anomalies by the value of the reconstruction error, which increases with the deviation of the signal from its design condition. Comparative analysis of various autoencoder training algorithms, such as stochastic gradient descent (SGD), Adam, and RMSProp was undertaken, its results are presented. The experiments conducted on model data, simulating the design operation of flame sensors in gas preheaters, are described. Experimental results demonstrate the high efficiency of the proposed approach along with its potential for a variety of industrial monitoring and diagnostic applications.

Keywords: autoencoder, flame sensor, data recovery, machine learning, time series, false positives, gas preheater, neural networks, recovery error, anomalies

Livshits I.I.  The effect of present-day artificial intelligence technologies on the safety of industrial automation systems

The paper discusses the influence of present-day technologies, including the artificial intelligence (AI), on the safety of industrial automation systems. The study focuses on the set of standards, which could underlie the development of an estimation system for the application of AI technologies for ensuring the specified safety level of industrial automation components. In a certain sense, the study can be considered as a specific task within a broader problem, namely, the development of a trusted system of AI technologies application using the known requirements to technical system validation and verification.

Keywords: standardization, validation, verification, industrial controller, audit, risk, artificial intelligence, safety.

Vladova A.Yu.  Steel data mining: algorithms against subsidence and skewing

A railroad track is an extended technical object whose sections are exposed to various external impacts. Statistical analysis of the data from monitoring systems has revealed the deviations from targets in types, years, and seasons, weak relationship between parameters, as well as the lack of any records of ideal railway track condition. The paper offers and identification method for improving the quality of railway track health estimate an identification. The method’s key feature is the aggregation of deviations with respect to types with reference to their physical nature. The aggregates are balanced with respect to the number of overall and seasonal deviations; the length of rail track segment and the deviation parameter are selected, and the deviation parameter’s standard deviation is further estimated for each aggregate. Eventually, the sum of standard deviations over all aggregates is used for the identification of railway track segment’s health.

Keywords: railroad track, monitoring system, aggregated indicators, machine learning, statistical analysis.

Zakharov N.A.  Machine learning and digital twins

The paper argues for the application of machine learning combined with the implementation of digital twins of industrial objects and processes. Case studies from food industry (fruit storage and high-temperature sterilization) and petrochemical industry.

Keywords: machine learning, digital twin, Internet of things, thermal imaging system, petrochemical industry.

Nezamaev S.B., Bobkov V.I., Bykov A.A.  Information system for environmental safety decision support in chemical and energy ore processing systems

The information system for calculating and simulation of treatment facilities of phosphate ore processing plants is presented. Its application enables the design of state-of-the-art treatment facilities for gas emissions and waste water meeting both domestic and international environmental safety requirements. The information system includes a set of mathematical models of key process facilities; the models allow for effluent gas properties and purification efficiency specifications. Based on the calculated parameters, the system, which includes the equipment database, automatically selects the necessary components and develops a graphical functional diagram of a treatment system. Numerical experiments and production tests were performed for estimating the efficiency of the modeled treatment systems.

Keywords: information system, chemical and energy systems, modeling, environmental safety, treatment facilities, ore feedstock.

Terentieva O.A.  Optimization of client/server system reliability by movable redundancy method

Key reliability indicators affecting the selection of optimal redundancy of information systems with client/server architecture are discussed. Theoretical background of reliability investigation and a Markovian reliability model are presented. Key reliability characteristics of a redundant information system are calculated, a software application is implemented in Python environment for controlling client/server systems. The study results can be used for developing fault-tolerant systems.

Keywords: information system, computer simulation, reliability characteristics, movable redundancy, Markovian process.

Tyutin S.S., Anokhin A.I., Bugreev A.V., Ivanchenko D.S., Tikhonov A.M., Kamnev M.A.  Automatic taking of pressure-energy characteristics of electric pumps during their testing

The paper describes an automated control system for taking pressure-energy characteristics during electric pump testing. It describes a hard-/software system for automatic water flowrate maintenance, which comprises a real-time controller and a software based PID-controller.

Keywords: PID controller, automation, pressure-energy characteristics, electric pump.

Journal’s Club

Kravchenko V.N.  Russian market of magnetosrictive linear displacement transducers after the leave of western vendors: three years in new conditions

The leave of western vendors resulted in drastic changes in the Russian market of magnetosrictive linear displacement transducers. The three recent years have seen the development of the approaches to filling up the shortfall of high-precision reliable instruments. The key trend is bringing domestic brands to market in cooperation with Chinese leaders. A good example is KTSL linear displacement transducers form Russia-based K&T Sensors.

Keywords: magnetosrictive linear displacement transducers, import replacement, domestic consumers.