Gulzat Ziyatbekova
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Gulzat Ziyatbekova has a total of 87 co-authors. Here are the top co-authors by number of shared publications:
- Talgat Zh. Mazakov
Al-Farabi Kazakh National University
10 shared publications - Aigerim T. Mazakova
Al-Farabi Kazakh National University
8 shared publications - Sholpan Jomartova
Al-Farabi Kazakh National University
6 shared publications - Maksat Kalimoldayev
Higher School of Information Technology and Engineering, Astana International University, Astana 010000, Kazakhstan
3 shared publications Ш.а. Джомартова
Казахский национальный университет имени аль-Фараби 2 Казахская национальная академия искусств имени Темирбека Жургенова 3 Институт информационных и вычислительных технологий КН МНВО РК Алматы Алматы Алматы Казахстан Казахстан Казахстан
3 shared publications- Akyltai D. Burgegulov
Al-Farabi Kazakh National University
2 shared publications
Latest Publications
- Open Access
- 30 September 2026
- Published by D. Serikbayev East Kazakhstan Technical University in EKTU Journal of Information and Communication Sciences
- Vol. 1 (3)
ENHANCING INTRUSION DETECTION SYSTEMS WITH AI-BASED NETWORK TRAFFIC ANALYSIS: A FOCUS ON ANOMALY DETECTION
In an era of sophisticated cyber threats and the growing complexity of network infrastructures, intrusion detection systems (IDS) play a critical role in ensuring the security of digital assets. Traditional IDS methods based on predefined rules and signatures are becoming less effective against sophisticated attacks. The study covers the stages of data collection, preprocessing, model training, and evaluation. This study examines current security solutions, such as next-generation firewalls (NGFWs) and intrusion detection and prevention systems (IDS/IPS). In the experimental study, linear regression, Ridge regression, Lasso regression and Random Forest models were used. Particular attention is paid to the methodological problem of data leakage. Special attention was paid to the methodological problem of data leakage, and a leakage-free modeling scenario based only on previously known variables was created. Taking into account the pronounced right-hand asymmetry and heavy-tailed distribution of the dependent variable, the effect of using the log1p transformation on the generalization ability of the models was investigated. The obtained results demonstrate the advantages of regularized linear models. The use of log1p transformation in combination with Ridge regression allowed us to improve the quality of forecasting, reduce the variance of errors and increase the value of the determination coefficient R². Additional validation based on time-based data partitioning confirmed the methodological correctness of the conclusions drawn. The Random Forest model did not show any improvement in forecasting accuracy, but was used to analyze the structure of factors influencing the engagement rate. In addition, issues such as model interpretability, false positives, and the need for continuous learning are addressed. An intelligent approach based on event log analysis and machine learning methods is proposed to identify signs of potential information leaks.
- 30 September 2026
- Published by D. Serikbayev East Kazakhstan Technical University in EKTU Journal of Information and Communication Sciences
- Vol. 1 (3)
Open AccessENHANCING INTRUSION DETECTION SYSTEMS WITH AI-BASED NETWORK TRAFFIC ANALYSIS: A FOCUS ON ANOMALY DETECTION
In an era of sophisticated cyber threats and the growing complexity of network infrastructures, intrusion detection systems (IDS) play a critical role in ensuring the security of digital assets. Traditional IDS methods based on predefined rules and signatures are becoming less effective against sophisticated attacks. The study covers the stages of data collection, preprocessing, model training, and evaluation. This study examines current security solutions, such as next-generation firewalls (NGFWs) and intrusion detection and prevention systems (IDS/IPS). In the experimental study, linear regression, Ridge regression, Lasso regression and Random Forest models were used. Particular attention is paid to the methodological problem of data leakage. Special attention was paid to the methodological problem of data leakage, and a leakage-free modeling scenario based only on previously known variables was created. Taking into account the pronounced right-hand asymmetry and heavy-tailed distribution of the dependent variable, the effect of using the log1p transformation on the generalization ability of the models was investigated. The obtained results demonstrate the advantages of regularized linear models. The use of log1p transformation in combination with Ridge regression allowed us to improve the quality of forecasting, reduce the variance of errors and increase the value of the determination coefficient R². Additional validation based on time-based data partitioning confirmed the methodological correctness of the conclusions drawn. The Random Forest model did not show any improvement in forecasting accuracy, but was used to analyze the structure of factors influencing the engagement rate. In addition, issues such as model interpretability, false positives, and the need for continuous learning are addressed. An intelligent approach based on event log analysis and machine learning methods is proposed to identify signs of potential information leaks.
- Open Access
- 1 June 2026
- Published by Institute of Advanced Engineering and Science in Bulletin of Electrical Engineering and Informatics
- Vol. 15 (3) , 2537-2552
Hybrid deep learning architecture for autism spectrum disorder detection from gait kinematic data
Early detection of autism spectrum disorder (ASD) is essential for enabling timely interventions that significantly impact developmental outcomes. Traditional diagnostic procedures often rely on clinical observations and subjective assessments, which can reduce objectivity and delay diagnosis. This study presents an automated classification approach for detecting ASD in children using gait kinematic data obtained through video-based skeletal tracking. We propose a hybrid deep learning architecture that combines one-dimensional convolutional layers (Conv1D), bidirectional long short-term memory (BiLSTM) networks, and a multi-head attention mechanism to capture complex spatiotemporal patterns in motion data. The dataset includes 100 participants—50 diagnosed with ASD and 50 typically developing (TD) peers. Preprocessing steps included Euclidean norm transformation, logarithmic scaling, Z-score normalization, and sliding window segmentation. The proposed model achieved 94.9% accuracy, 0.91 recall, 0.86 precision, and an area under the curve (AUC) of 0.97, outperforming a baseline long short-term memory (LSTM) architecture. These findings demonstrate the potential of gait-based kinematic features and hybrid neural networks for objective and reproducible ASD screening. The developed system can contribute to AI-assisted decision support tools in clinical and educational environments, enhancing diagnostic accuracy, and supporting inclusive developmental care.
- 1 June 2026
- Published by Institute of Advanced Engineering and Science in Bulletin of Electrical Engineering and Informatics
- Vol. 15 (3) , 2537-2552
Open AccessHybrid deep learning architecture for autism spectrum disorder detection from gait kinematic data
Early detection of autism spectrum disorder (ASD) is essential for enabling timely interventions that significantly impact developmental outcomes. Traditional diagnostic procedures often rely on clinical observations and subjective assessments, which can reduce objectivity and delay diagnosis. This study presents an automated classification approach for detecting ASD in children using gait kinematic data obtained through video-based skeletal tracking. We propose a hybrid deep learning architecture that combines one-dimensional convolutional layers (Conv1D), bidirectional long short-term memory (BiLSTM) networks, and a multi-head attention mechanism to capture complex spatiotemporal patterns in motion data. The dataset includes 100 participants—50 diagnosed with ASD and 50 typically developing (TD) peers. Preprocessing steps included Euclidean norm transformation, logarithmic scaling, Z-score normalization, and sliding window segmentation. The proposed model achieved 94.9% accuracy, 0.91 recall, 0.86 precision, and an area under the curve (AUC) of 0.97, outperforming a baseline long short-term memory (LSTM) architecture. These findings demonstrate the potential of gait-based kinematic features and hybrid neural networks for objective and reproducible ASD screening. The developed system can contribute to AI-assisted decision support tools in clinical and educational environments, enhancing diagnostic accuracy, and supporting inclusive developmental care.
- Open Access
Performance Evaluation of zk-SNARK Protocols for Privacy-Preserving Sensor Data Verification: A Systematic Benchmarking Study
The proliferation of sensor networks in critical infrastructure, healthcare monitoring, and smart city applications demands robust privacy-preserving mechanisms for data verification. Zero-knowledge succinct non-interactive arguments of knowledge (zk-SNARKs) offer a promising cryptographic primitive that enables data integrity verification without revealing sensitive sensor readings. However, the practical feasibility of deploying zk-SNARKs in resource-constrained sensor network environments remains insufficiently characterized. This paper presents a systematic benchmarking study of the Groth16 zk-SNARK protocol across eight representative circuit types spanning six orders of magnitude in computational complexity, from basic arithmetic operations (1 constraint) to ECDSA signature verification (1,510,185 constraints). Using an automated open-source benchmarking framework built on the Circom-snarkjs toolchain, we conducted 160 statistically controlled measurements (20 iterations per circuit) with cold/warm separation, collecting proof generation time, verification time, proof size, memory consumption, and witness generation overhead. Our results demonstrate that Groth16 proofs maintain a constant size of bytes and near-constant verification time of s regardless of circuit complexity, with coefficients of variation below 5% across all circuit types. Proof generation time exhibits sub-linear scaling (, ), with statistically significant differences between circuit categories confirmed by one-way ANOVA (, , ). We identify three operational deployment tiers for sensor network architectures and estimate energy budgets for battery-powered devices. These findings provide actionable guidance for the design of privacy-preserving data verification systems in next-generation sensor networks.Open AccessPerformance Evaluation of zk-SNARK Protocols for Privacy-Preserving Sensor Data Verification: A Systematic Benchmarking Study
The proliferation of sensor networks in critical infrastructure, healthcare monitoring, and smart city applications demands robust privacy-preserving mechanisms for data verification. Zero-knowledge succinct non-interactive arguments of knowledge (zk-SNARKs) offer a promising cryptographic primitive that enables data integrity verification without revealing sensitive sensor readings. However, the practical feasibility of deploying zk-SNARKs in resource-constrained sensor network environments remains insufficiently characterized. This paper presents a systematic benchmarking study of the Groth16 zk-SNARK protocol across eight representative circuit types spanning six orders of magnitude in computational complexity, from basic arithmetic operations (1 constraint) to ECDSA signature verification (1,510,185 constraints). Using an automated open-source benchmarking framework built on the Circom-snarkjs toolchain, we conducted 160 statistically controlled measurements (20 iterations per circuit) with cold/warm separation, collecting proof generation time, verification time, proof size, memory consumption, and witness generation overhead. Our results demonstrate that Groth16 proofs maintain a constant size of bytes and near-constant verification time of s regardless of circuit complexity, with coefficients of variation below 5% across all circuit types. Proof generation time exhibits sub-linear scaling (, ), with statistically significant differences between circuit categories confirmed by one-way ANOVA (, , ). We identify three operational deployment tiers for sensor network architectures and estimate energy budgets for battery-powered devices. These findings provide actionable guidance for the design of privacy-preserving data verification systems in next-generation sensor networks. - Open Access
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 16245-16254
New Approaches to Solving Flood and Breakthrough Wave Modeling Problems
The article deals with modern methodological approaches to mathematical modeling of flood processes and breakthrough waves, taking into account the geometric characteristics of reservoirs, hydraulic structures, and terrain. The limitations of classical one-dimensional models are critically analyzed, and an improved hybrid approach is proposed that combines empirical dependencies with numerical solutions of hydrodynamic equations. Particular attention is paid to the three-dimensional description of the breakthrough wave propagation process using simplified forms of the Navier-Stokes equations in the plane, which allows estimating the flood front in a mode close to real time.The proposed hybrid model demonstrates a minimum improvement of 33% and a maximum improvement of 77% in prediction accuracy compared to baseline one-dimensional and purely empirical models, depending on reservoir geometry and boundary conditions. The expediency of integrating mathematical models into software and hardware monitoring systems, including forecasting algorithms, sensor systems, and operational warning systems, is also substantiated. Numerical simulation results obtained for different types of reservoirs indicate a reduction in computational time by 40-60% while maintaining acceptable accuracy for real-time applications.The presented results of numerical modeling for different types of reservoirs demonstrate the effectiveness of the proposed methods for civil protection, evacuation planning, and reducing the risks of catastrophic flooding. The main scientific and practical problem is the insufficient accuracy and limited applicability of traditional one-dimensional models in cases of complex reservoir geometry and non-stationary hydrodynamic conditions, which makes it impossible to use them for operational forecasting. In order to overcome these limitations, a hybrid mathematical model has been developed and analysed, combining a simplified three-dimensional description of the break wave with computationally efficient numerical algorithms designed for integration into real-time monitoring systems.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 16245-16254
Open AccessNew Approaches to Solving Flood and Breakthrough Wave Modeling Problems
The article deals with modern methodological approaches to mathematical modeling of flood processes and breakthrough waves, taking into account the geometric characteristics of reservoirs, hydraulic structures, and terrain. The limitations of classical one-dimensional models are critically analyzed, and an improved hybrid approach is proposed that combines empirical dependencies with numerical solutions of hydrodynamic equations. Particular attention is paid to the three-dimensional description of the breakthrough wave propagation process using simplified forms of the Navier-Stokes equations in the plane, which allows estimating the flood front in a mode close to real time.The proposed hybrid model demonstrates a minimum improvement of 33% and a maximum improvement of 77% in prediction accuracy compared to baseline one-dimensional and purely empirical models, depending on reservoir geometry and boundary conditions. The expediency of integrating mathematical models into software and hardware monitoring systems, including forecasting algorithms, sensor systems, and operational warning systems, is also substantiated. Numerical simulation results obtained for different types of reservoirs indicate a reduction in computational time by 40-60% while maintaining acceptable accuracy for real-time applications.The presented results of numerical modeling for different types of reservoirs demonstrate the effectiveness of the proposed methods for civil protection, evacuation planning, and reducing the risks of catastrophic flooding. The main scientific and practical problem is the insufficient accuracy and limited applicability of traditional one-dimensional models in cases of complex reservoir geometry and non-stationary hydrodynamic conditions, which makes it impossible to use them for operational forecasting. In order to overcome these limitations, a hybrid mathematical model has been developed and analysed, combining a simplified three-dimensional description of the break wave with computationally efficient numerical algorithms designed for integration into real-time monitoring systems.
- 30 December 2025
- Published by National Engineering Academy of the Republic of Kazakhstan in Bulletin of the National Engineering Academy of the Republic of Kazakhstan
- Vol. 98 (4) , 227-236
Bipolar disorder states and markov decision processes
Bipolar disorder has been under-researched in Kazakhstan, although many individuals are susceptible to this disorder in a latent form. In this article, we applied Markov processes to study the bipolar individual’s states. Different reward results can be obtained by employing Markov decision processes for analyzing bipolar individual’s mood states. After employing test scenarios, we conclude that BD individual needs social support more, than other people. The practical benefit of decision-making theory is to find different ways of making decisions in order to help bipolar individual.
- 30 December 2025
- Published by National Engineering Academy of the Republic of Kazakhstan in Bulletin of the National Engineering Academy of the Republic of Kazakhstan
- Vol. 98 (4) , 227-236
Bipolar disorder states and markov decision processes
Bipolar disorder has been under-researched in Kazakhstan, although many individuals are susceptible to this disorder in a latent form. In this article, we applied Markov processes to study the bipolar individual’s states. Different reward results can be obtained by employing Markov decision processes for analyzing bipolar individual’s mood states. After employing test scenarios, we conclude that BD individual needs social support more, than other people. The practical benefit of decision-making theory is to find different ways of making decisions in order to help bipolar individual.
- 12 December 2025
- Published by Natural and Engineering Sciences
- Vol. 10 (3) , 879-889
Molecular Evolution of Invasive Species and their Ecological Impacts on Global Fisheries and Aquaculture Systems
The increasing rate of international trading and global warming has supported the fatality of introducing non-native species in large numbers, which is a major threat to aquatic biodiversity and world food security. In this paper, consider the molecular evolution of the invasive species and the implications that it has had on the global fisheries and aquaculture systems. Recent discoveries in genomics, such as epigenetic plasticity, hybridization, and adaptive evolution at high rates, are combined to describe how the invasive population of a species can break the founder effect and initial genetic bottleneck to conquer new environments. Nature is found to be unstable through molecular shifts in invasive taxa that cause much ecological disturbance, including genetic pollution of wild stocks by introgression and introduction of new pathogens that annihilate commercial aquaculture. It also elaborates on the effect of the Trojan horse, where the genomic strength of invaders enables them to act as unrelenting vectors of diseases in warming oceans. The shortcomings of conventional ecological modeling are shown through analysis of case studies, including the Lessepsian migration and the invasion of Atlantic lionfish, which are in favor of merging the environmental surveillance systems of environmental DNA (eDNA) and genomic biosecurity. As a possible way to reduce the invasive effects, the potential and ethical aspects of biotechnological interventions, including CRISPR-based gene drives, are considered. Conclusion: It is determined that to have a sustainable management of the global blue economy, knowledge of the molecular basis of invasiveness is a prerequisite. To protect the sustainability of fisheries and the socio-economic well-being of coastal societies all over the globe, the integration of genomic insights into international policy is needed.
- 12 December 2025
- Published by Natural and Engineering Sciences
- Vol. 10 (3) , 879-889
Molecular Evolution of Invasive Species and their Ecological Impacts on Global Fisheries and Aquaculture Systems
The increasing rate of international trading and global warming has supported the fatality of introducing non-native species in large numbers, which is a major threat to aquatic biodiversity and world food security. In this paper, consider the molecular evolution of the invasive species and the implications that it has had on the global fisheries and aquaculture systems. Recent discoveries in genomics, such as epigenetic plasticity, hybridization, and adaptive evolution at high rates, are combined to describe how the invasive population of a species can break the founder effect and initial genetic bottleneck to conquer new environments. Nature is found to be unstable through molecular shifts in invasive taxa that cause much ecological disturbance, including genetic pollution of wild stocks by introgression and introduction of new pathogens that annihilate commercial aquaculture. It also elaborates on the effect of the Trojan horse, where the genomic strength of invaders enables them to act as unrelenting vectors of diseases in warming oceans. The shortcomings of conventional ecological modeling are shown through analysis of case studies, including the Lessepsian migration and the invasion of Atlantic lionfish, which are in favor of merging the environmental surveillance systems of environmental DNA (eDNA) and genomic biosecurity. As a possible way to reduce the invasive effects, the potential and ethical aspects of biotechnological interventions, including CRISPR-based gene drives, are considered. Conclusion: It is determined that to have a sustainable management of the global blue economy, knowledge of the molecular basis of invasiveness is a prerequisite. To protect the sustainability of fisheries and the socio-economic well-being of coastal societies all over the globe, the integration of genomic insights into international policy is needed.
- 22 November 2025
- Published by Springer Nature
- p. 54-66
Using a Liquid Time-Constant Network with Physics-Informed Neural Network to Predict Aftershocks
- 22 November 2025
- Published by Springer Nature
- p. 54-66
Using a Liquid Time-Constant Network with Physics-Informed Neural Network to Predict Aftershocks
- 17 October 2025
- Published by Wiley in Internet Technology Letters
- Vol. 8 (6)
Leveraging AI in 5G Networks for Estimating Signal‐to‐Interference‐and‐Noise Ratio
A key role for artificial intelligence (AI) is anticipated in the era of 5G networks. Effective radio resource management has become more and more important for network operators. But when new technologies, network topologies, and sophisticated equipment are integrated more quickly, it's getting harder to allocate enough radio resources for precise channel condition assessment in mobile networks. Predicting channel conditions automatically helps to make effective use of resources. Research presents the mutated gray wolf‐driven spiking neural network (MGW‐SNN) model, an ML‐based method for signal‐to‐interference‐and‐noise ratio (SINR) estimation based on the cyber‐physical system (CPS) location. An MGW‐SNN model uses the present position of the CPS to forecast the SINR. Initially, obtain the data to validate suggested algorithms. Min–max normalization and noise reduction were used to preprocess after collecting data. The spiking neural network (SNN) structure's parameters are optimized with the usage of the gray wolf optimization (GWO) method to improve the network's performance. The research was implemented using MATLAB. The proposed strategy in terms of accuracy (93%), RMSE (1.35), (0.99), MAE (0.75), and average SINR (10 dB) is validated. It shows that the proposal is effective in predicting the SINR. The approach stabilizes the computational economy while simultaneously increasing accuracy, which makes it suitable for real‐time applications in dynamic network environments.
- 17 October 2025
- Published by Wiley in Internet Technology Letters
- Vol. 8 (6)
Leveraging AI in 5G Networks for Estimating Signal‐to‐Interference‐and‐Noise Ratio
A key role for artificial intelligence (AI) is anticipated in the era of 5G networks. Effective radio resource management has become more and more important for network operators. But when new technologies, network topologies, and sophisticated equipment are integrated more quickly, it's getting harder to allocate enough radio resources for precise channel condition assessment in mobile networks. Predicting channel conditions automatically helps to make effective use of resources. Research presents the mutated gray wolf‐driven spiking neural network (MGW‐SNN) model, an ML‐based method for signal‐to‐interference‐and‐noise ratio (SINR) estimation based on the cyber‐physical system (CPS) location. An MGW‐SNN model uses the present position of the CPS to forecast the SINR. Initially, obtain the data to validate suggested algorithms. Min–max normalization and noise reduction were used to preprocess after collecting data. The spiking neural network (SNN) structure's parameters are optimized with the usage of the gray wolf optimization (GWO) method to improve the network's performance. The research was implemented using MATLAB. The proposed strategy in terms of accuracy (93%), RMSE (1.35), (0.99), MAE (0.75), and average SINR (10 dB) is validated. It shows that the proposal is effective in predicting the SINR. The approach stabilizes the computational economy while simultaneously increasing accuracy, which makes it suitable for real‐time applications in dynamic network environments.
- 7 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Integrated Early Fire Detection and Evacution System Based on Arduino and MQ Seriessensors: Development and Implementation
This article presents the development and implementation of an advanced intelligent fire safety system This system is designed to enhance public safety by enabling rapid fire detection and streamlined evacuation procedures. The system integrates an Arduino microcontroller with MQ-series gas sensors to monitor environmental conditions in real time. A key innovation is the use of a graph theory-based algorithm to dynamically generate optimal evacuation routes, ensuring rapid and safe occupant egress during emergencies. The article details the system architecture, including sensor placement, data acquisition, and decision-making processes. When critical sensor readings are detected, the system activates the evacuation algorithm, guiding occupants through the safest available paths while avoiding hazardous areas. Additionally, the developed software processes sensor inputs, assesses risk levels, and coordinates emergency responses. The paper includes numerical simulations validating the system's effectiveness in various fire scenarios, demonstrating improvements in evacuation times and safety. Future research directions are discussed, such as integrating machine learning for predictive analytics, expanding IoT connectivity for smart city applications, and enhancing multi-sensor fusion for improved accuracy. This work contributes to the advancement of intelligent fire safety solutions, offering scalable and adaptive protection for modern urban infrastructures.
- 7 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Integrated Early Fire Detection and Evacution System Based on Arduino and MQ Seriessensors: Development and Implementation
This article presents the development and implementation of an advanced intelligent fire safety system This system is designed to enhance public safety by enabling rapid fire detection and streamlined evacuation procedures. The system integrates an Arduino microcontroller with MQ-series gas sensors to monitor environmental conditions in real time. A key innovation is the use of a graph theory-based algorithm to dynamically generate optimal evacuation routes, ensuring rapid and safe occupant egress during emergencies. The article details the system architecture, including sensor placement, data acquisition, and decision-making processes. When critical sensor readings are detected, the system activates the evacuation algorithm, guiding occupants through the safest available paths while avoiding hazardous areas. Additionally, the developed software processes sensor inputs, assesses risk levels, and coordinates emergency responses. The paper includes numerical simulations validating the system's effectiveness in various fire scenarios, demonstrating improvements in evacuation times and safety. Future research directions are discussed, such as integrating machine learning for predictive analytics, expanding IoT connectivity for smart city applications, and enhancing multi-sensor fusion for improved accuracy. This work contributes to the advancement of intelligent fire safety solutions, offering scalable and adaptive protection for modern urban infrastructures.
- 7 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Implementation and Adaptation of Machine Learning Approaches for the Diagnosis of Pumping Systems
This article discusses methods for monitoring and predicting the remaining service life of pumps used in irrigation systems. The relevance of this research is due to the need to improve the reliability and efficiency of pumping equipment, reduce the likelihood of unexpected breakdowns, and optimize maintenance costs. This study aims to create and apply a highly accurate method for predicting the residual service life (RUL) of an irrigation system. The method is based on the use of a convolution neural network. Research Objectives: Study of existing methods for monitoring and predicting the technical condition of pumping equipment. Development of convolutional neural network architecture with two parallel data processing channels: temporal and time-frequency. Creation of the software complex “AITUM” for collection, processing and analysis of data coming from vibration and temperature sensors. Development of recommendations on application of the method in pumping equipment maintenance systems. The constructed method on the basis of deep learning shows high accuracy of prediction of RUL of pumps, which significantly exceeds the generally accepted approaches. A series of experiments on real data of pumping systems were conducted, which helped to discover the degradation patterns of the equipment. Application of this technology will reduce the risk of emergency failures, reduce operating costs and increase the efficiency of irrigation systems through timely maintenance of equipment.
- 7 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Implementation and Adaptation of Machine Learning Approaches for the Diagnosis of Pumping Systems
This article discusses methods for monitoring and predicting the remaining service life of pumps used in irrigation systems. The relevance of this research is due to the need to improve the reliability and efficiency of pumping equipment, reduce the likelihood of unexpected breakdowns, and optimize maintenance costs. This study aims to create and apply a highly accurate method for predicting the residual service life (RUL) of an irrigation system. The method is based on the use of a convolution neural network. Research Objectives: Study of existing methods for monitoring and predicting the technical condition of pumping equipment. Development of convolutional neural network architecture with two parallel data processing channels: temporal and time-frequency. Creation of the software complex “AITUM” for collection, processing and analysis of data coming from vibration and temperature sensors. Development of recommendations on application of the method in pumping equipment maintenance systems. The constructed method on the basis of deep learning shows high accuracy of prediction of RUL of pumps, which significantly exceeds the generally accepted approaches. A series of experiments on real data of pumping systems were conducted, which helped to discover the degradation patterns of the equipment. Application of this technology will reduce the risk of emergency failures, reduce operating costs and increase the efficiency of irrigation systems through timely maintenance of equipment.