Pub Date : 2026-03-01Epub Date: 2026-02-02DOI: 10.1016/j.jestch.2026.102291
Jia Yang, Jixiang Zhang, Deguang Wang, Ming Yang, Chengbin Liang
Accurate classification of power quality disturbances (PQDs) is essential for improving the stability of power systems, ensuring reliable integration of renewable energy sources and advancing smart grid technologies. To address the challenges posed by complex PQDs, this study introduces a novel integrated model, IST-MSCNN-OXGBoost, which combines advanced signal processing, deep learning-based feature extraction, and an optimized classifier. The improved S-transform (IST) enables adaptive time–frequency resolution, facilitating precise detection and localization of transient events and signal variations across different frequency ranges. The multi-scale convolutional neural network (MSCNN) employs pyramid convolution operations to extract multi-scale features from time–frequency representations, effectively capturing intricate patterns and complex relationships within the data. Classification accuracy is further enhanced by optimized XGBoost (OXGBoost), which utilizes the duck swarm algorithm for automated hyperparameter tuning, ensuring robust and efficient performance. Comprehensive evaluations underscore the contributions of each component. IST delivers superior time–frequency analysis and improves classification accuracy by 3.33% compared with the conventional ST when integrated with MSCNN-OXGBoost. MSCNN excels in automated and multi-scale feature extraction, and OXGBoost achieves high classification accuracy with improved generalization. The final IST-MSCNN-OXGBoost achieves a classification accuracy of 99.86% and maintains robust performance under adverse noise conditions, preserving an accuracy of 96.67% at a signal-to-noise ratio of 20 dB. Additional analyses across varying dataset sizes, training ratios, image resolutions, noise levels, parameter configurations, and computational loads further validate its suitability for real-time industrial applications. These findings confirm the potential of IST-MSCNN-OXGBoost as robust and reliable solution for the accurate classification of complex PQDs, paving the way for smarter and more resilient power systems.
{"title":"IST-MSCNN-OXGBoost: An integrated model for accurate classification of complex power quality disturbances","authors":"Jia Yang, Jixiang Zhang, Deguang Wang, Ming Yang, Chengbin Liang","doi":"10.1016/j.jestch.2026.102291","DOIUrl":"10.1016/j.jestch.2026.102291","url":null,"abstract":"<div><div>Accurate classification of power quality disturbances (PQDs) is essential for improving the stability of power systems, ensuring reliable integration of renewable energy sources and advancing smart grid technologies. To address the challenges posed by complex PQDs, this study introduces a novel integrated model, IST-MSCNN-OXGBoost, which combines advanced signal processing, deep learning-based feature extraction, and an optimized classifier. The improved S-transform (IST) enables adaptive time–frequency resolution, facilitating precise detection and localization of transient events and signal variations across different frequency ranges. The multi-scale convolutional neural network (MSCNN) employs pyramid convolution operations to extract multi-scale features from time–frequency representations, effectively capturing intricate patterns and complex relationships within the data. Classification accuracy is further enhanced by optimized XGBoost (OXGBoost), which utilizes the duck swarm algorithm for automated hyperparameter tuning, ensuring robust and efficient performance. Comprehensive evaluations underscore the contributions of each component. IST delivers superior time–frequency analysis and improves classification accuracy by 3.33% compared with the conventional ST when integrated with MSCNN-OXGBoost. MSCNN excels in automated and multi-scale feature extraction, and OXGBoost achieves high classification accuracy with improved generalization. The final IST-MSCNN-OXGBoost achieves a classification accuracy of 99.86% and maintains robust performance under adverse noise conditions, preserving an accuracy of 96.67% at a signal-to-noise ratio of 20 dB. Additional analyses across varying dataset sizes, training ratios, image resolutions, noise levels, parameter configurations, and computational loads further validate its suitability for real-time industrial applications. These findings confirm the potential of IST-MSCNN-OXGBoost as robust and reliable solution for the accurate classification of complex PQDs, paving the way for smarter and more resilient power systems.</div></div>","PeriodicalId":48609,"journal":{"name":"Engineering Science and Technology-An International Journal-Jestech","volume":"75 ","pages":"Article 102291"},"PeriodicalIF":5.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174607","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-18DOI: 10.1016/S2215-0986(26)00041-8
{"title":"Front Matter 1 - Full Title Page (regular issues)/Special Issue Title page (special issues)","authors":"","doi":"10.1016/S2215-0986(26)00041-8","DOIUrl":"10.1016/S2215-0986(26)00041-8","url":null,"abstract":"","PeriodicalId":48609,"journal":{"name":"Engineering Science and Technology-An International Journal-Jestech","volume":"75 ","pages":"Article 102315"},"PeriodicalIF":5.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147385579","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-09DOI: 10.1016/j.jestch.2026.102290
Onur Polat , Ömer Durmuş , Ferdi Doğan , Muammer Türkoğlu , Hüseyin Şeker , Ferhat Atasoy , Enes Algül
Software-Defined Networking (SDN) offers significant advantages over traditional network architectures by providing flexibility, programmability and centralized control in network management. However, the centralized nature of this architecture brings new vulnerabilities, especially against security threats such as Distributed Denial of Service (DDoS) attacks. In this context, Machine Learning (ML) based methods offer effective and innovative solutions for detecting DDoS attacks in SDN environments.
This paper presents a comprehensive review of machine learning techniques for DDoS attack detection in SDN-based networks. The most remarkable aspect is that, unlike many existing works in the literature, it does not only focus on general detection methods, but also examines in detail various scenarios in different application areas of SDN, such as Internet of Things (IoT), SCADA systems, 5G and mobile networks, and vehicular ad-hoc networks (VANET). This provides a holistic perspective on the security dynamics of SDN architecture in different contexts and comparatively evaluates current threats and solution approaches in these areas.
In the study, the success, usage areas and limitations of different machine learning algorithms (supervised, unsupervised and deep learning methods) in detecting DDoS attacks are analyzed and conclusions are made to guide researchers. In this respect, the study contributes to the literature on SDN security in terms of both technical depth and application diversity.
{"title":"Supervised and deep learning techniques for DDoS detection in software-defined network architectures: a systematic review","authors":"Onur Polat , Ömer Durmuş , Ferdi Doğan , Muammer Türkoğlu , Hüseyin Şeker , Ferhat Atasoy , Enes Algül","doi":"10.1016/j.jestch.2026.102290","DOIUrl":"10.1016/j.jestch.2026.102290","url":null,"abstract":"<div><div>Software-Defined Networking (SDN) offers significant advantages over traditional network architectures by providing flexibility, programmability and centralized control in network management. However, the centralized nature of this architecture brings new vulnerabilities, especially against security threats such as Distributed Denial of Service (DDoS) attacks. In this context, Machine Learning (ML) based methods offer effective and innovative solutions for detecting DDoS attacks in SDN environments.</div><div>This paper presents a comprehensive review of machine learning techniques for DDoS attack detection in SDN-based networks. The most remarkable aspect is that, unlike many existing works in the literature, it does not only focus on general detection methods, but also examines in detail various scenarios in different application areas of SDN, such as Internet of Things (IoT), SCADA systems, 5G and mobile networks, and vehicular ad-hoc networks (VANET). This provides a holistic perspective on the security dynamics of SDN architecture in different contexts and comparatively evaluates current threats and solution approaches in these areas.</div><div>In the study, the success, usage areas and limitations of different machine learning algorithms (supervised, unsupervised and deep learning methods) in detecting DDoS attacks are analyzed and conclusions are made to guide researchers. In this respect, the study contributes to the literature on SDN security in terms of both technical depth and application diversity.</div></div>","PeriodicalId":48609,"journal":{"name":"Engineering Science and Technology-An International Journal-Jestech","volume":"75 ","pages":"Article 102290"},"PeriodicalIF":5.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174608","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-12DOI: 10.1016/j.jestch.2026.102303
Kamran Dawood
Leakage reactance is a fundamental electrical parameter in transformers that directly affects voltage regulation, efficiency, and short-circuit performance. In dry-type transformers, the physical configuration of the windings plays a critical role in determining leakage reactance. Spacers placed between turns of the low-voltage winding to improve cooling also influence the electromagnetic behaviour of the winding. However, the extent to which the position of these spacers affects reactance remains underexplored. This study presents a detailed analysis of the impact of spacer positioning on leakage reactance in the low-voltage winding of a dry-type transformer. Using the finite element method, seven distinct cases are examined, each representing a different location of a single spacer placed between the winding turns. The simulation results provide data on how leakage reactance varies with spacer position, revealing the complex interaction between winding geometry and leakage reactance. The results show that spacer placement has a non-linear effect on leakage reactance; it even leads to a slight decrease. Overall, leakage reactance varied by about 10% across the different spacer positions, with the highest increase observed when the spacer was nearest to the high-voltage winding side. Additionally, one of the simulated cases is verified through experimental measurements to validate the simulation model.
{"title":"Leakage reactance variation in dry-type transformers due to spacer positioning: A FEM and experimental study","authors":"Kamran Dawood","doi":"10.1016/j.jestch.2026.102303","DOIUrl":"10.1016/j.jestch.2026.102303","url":null,"abstract":"<div><div>Leakage reactance is a fundamental electrical parameter in transformers that directly affects voltage regulation, efficiency, and short-circuit performance. In dry-type transformers, the physical configuration of the windings plays a critical role in determining leakage reactance. Spacers placed between turns of the low-voltage winding to improve cooling also influence the electromagnetic behaviour of the winding. However, the extent to which the position of these spacers affects reactance remains underexplored. This study presents a detailed analysis of the impact of spacer positioning on leakage reactance in the low-voltage winding of a dry-type transformer. Using the finite element method, seven distinct cases are examined, each representing a different location of a single spacer placed between the winding turns. The simulation results provide data on how leakage reactance varies with spacer position, revealing the complex interaction between winding geometry and leakage reactance. The results show that spacer placement has a non-linear effect on leakage reactance; it even leads to a slight decrease. Overall, leakage reactance varied by about 10% across the different spacer positions, with the highest increase observed when the spacer was nearest to the high-voltage winding side. Additionally, one of the simulated cases is verified through experimental measurements to validate the simulation model.</div></div>","PeriodicalId":48609,"journal":{"name":"Engineering Science and Technology-An International Journal-Jestech","volume":"75 ","pages":"Article 102303"},"PeriodicalIF":5.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174675","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-06DOI: 10.1016/j.jestch.2026.102288
Muhammad Auwal Shehu , Kulash Talapiden , Tin Trung Chau , Auwal Haruna , Mokhtar Aly , Vijayakumar Gali , Ton Duc Do , Ahmad Bala Alhassan
While standalone microgrids are an essential means of electrifying remote communities, high renewable penetration poses significant problems with power sharing, voltage/frequency stability, and optimal dispatch in low-inertia, communication-constrained scenarios. Using structured analysis across control methodologies, optimization techniques, and validation platforms, this paper synthesizes emerging paradigms in hierarchical control and energy management systems (EMS) through a systematic review of studies conducted in 2025. The following key findings show clear shifts: (i) adaptive droop and event-triggered consensus reduce communication overhead by 80% while maintaining voltage accuracy within 2%; (ii) super-twisting sliding mode control shows chattering-free operation with 98% cyber-attack detection capability; (iii) hybrid model predictive control frameworks enable real-time execution on embedded hardware with 25%–40% cost reduction; and (iv) deep reinforcement learning-based EMS shows 12% cost improvement and 97.8% reduction in computational load. There are still significant gaps: 68% of studies do not have hardware validation, 78% do not integrate cyber-security, power-sharing errors surpass 5% when there is an impedance mismatch, and there are no standardized benchmarking protocols. The review offers practical suggestions covering lifecycle-aware battery management, distributionally robust optimization (DRO) for renewable uncertainty, edge-computing architectures for communication-light operation, and cooperative cyber–physical testbeds for field validation. This synthesis provides a well-organized road map for developing technically demanding, financially feasible, and operationally robust microgrids that can provide sustainable access to electricity in underserved areas.
{"title":"Control and energy management of standalone microgrids in remote areas: A review of recent advances, challenges, and opportunities for future research","authors":"Muhammad Auwal Shehu , Kulash Talapiden , Tin Trung Chau , Auwal Haruna , Mokhtar Aly , Vijayakumar Gali , Ton Duc Do , Ahmad Bala Alhassan","doi":"10.1016/j.jestch.2026.102288","DOIUrl":"10.1016/j.jestch.2026.102288","url":null,"abstract":"<div><div>While standalone microgrids are an essential means of electrifying remote communities, high renewable penetration poses significant problems with power sharing, voltage/frequency stability, and optimal dispatch in low-inertia, communication-constrained scenarios. Using structured analysis across control methodologies, optimization techniques, and validation platforms, this paper synthesizes emerging paradigms in hierarchical control and energy management systems (EMS) through a systematic review of studies conducted in 2025. The following key findings show clear shifts: (i) adaptive droop and event-triggered consensus reduce communication overhead by 80% while maintaining voltage accuracy within <span><math><mo>±</mo></math></span>2%; (ii) super-twisting sliding mode control shows chattering-free operation with 98% cyber-attack detection capability; (iii) hybrid model predictive control frameworks enable real-time execution on embedded hardware with 25%–40% cost reduction; and (iv) deep reinforcement learning-based EMS shows 12% cost improvement and 97.8% reduction in computational load. There are still significant gaps: 68% of studies do not have hardware validation, 78% do not integrate cyber-security, power-sharing errors surpass 5% when there is an impedance mismatch, and there are no standardized benchmarking protocols. The review offers practical suggestions covering lifecycle-aware battery management, distributionally robust optimization (DRO) for renewable uncertainty, edge-computing architectures for communication-light operation, and cooperative cyber–physical testbeds for field validation. This synthesis provides a well-organized road map for developing technically demanding, financially feasible, and operationally robust microgrids that can provide sustainable access to electricity in underserved areas.</div></div>","PeriodicalId":48609,"journal":{"name":"Engineering Science and Technology-An International Journal-Jestech","volume":"75 ","pages":"Article 102288"},"PeriodicalIF":5.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174610","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-08DOI: 10.1016/j.jestch.2026.102300
Wei Wu , Chaochao Wang , Xiaotian Pan , Wensen Yu , Abdulilah Mohammad Mayet , Yisu Ge , Guodao Zhang
This paper presents a high-performance, resource-efficient digital implementation of the Dressed Neuron Model (DNM), a biologically inspired system that captures bidirectional interactions between neurons and astrocytes. Unlike classical neuron models, the DNM incorporates astrocyte-mediated calcium and IP signaling, forming a closed-loop feedback system capable of exhibiting spontaneous, seizure-like oscillations. To address the high computational complexity of this model on hardware, we introduce a Hybrid Model of DNM (HMoDNM), which approximates all major nonlinearities using a combination of dual-sinusoidal expressions and ROM-based lookup tables. These approximations achieve high numerical fidelity with root mean square error (RMSE) below for most functions, while ensuring hardware efficiency. The full system is implemented on a Xilinx Zynq-7000 XC7Z010 FPGA using a shift-and-add architecture with zero DSP slice utilization. All signals are represented in a fixed-point format , with dynamic range coverage up to 800 M for IP and 600 M for calcium. The design includes pipelined neuron and astrocyte cores, clock-gated nonlinear units, and shared computation modules, achieving a maximum clock frequency of 305 MHz and throughput of 21.7 million Euler steps per second. Overall resource usage is 6690 LUTs (38.0%), 2640 FFs (7.5%), and 4 BRAMs (6.7%), with a low dynamic power consumption of 167 mW and operating temperature of 35.1 °C at room ambient. To validate the model’s functional accuracy, we compare the HMoDNM outputs against the original DNM across two dynamic regimes, achieving correlation coefficients above 94% and NRMSE values below 0.06 for membrane voltage, calcium, and IP. Designed specifically for epilepsy modeling, this architecture provides a robust foundation for real-time tracking and control of astrocyte-influenced seizure dynamics. The proposed HMoDNM architecture offers a versatile foundation for hardware-based neuromorphic applications, including real-time seizure detection, closed-loop neurostimulation systems, and low-power embedded platforms for modeling neuron–glia interactions in brain-inspired computing.
{"title":"Nonlinear hardware realization and fast digital approximation of the dressed neuron model for astrocyte–neuron coupling dynamics","authors":"Wei Wu , Chaochao Wang , Xiaotian Pan , Wensen Yu , Abdulilah Mohammad Mayet , Yisu Ge , Guodao Zhang","doi":"10.1016/j.jestch.2026.102300","DOIUrl":"10.1016/j.jestch.2026.102300","url":null,"abstract":"<div><div>This paper presents a high-performance, resource-efficient digital implementation of the Dressed Neuron Model (DNM), a biologically inspired system that captures bidirectional interactions between neurons and astrocytes. Unlike classical neuron models, the DNM incorporates astrocyte-mediated calcium and IP<span><math><msub><mrow></mrow><mrow><mn>3</mn></mrow></msub></math></span> signaling, forming a closed-loop feedback system capable of exhibiting spontaneous, seizure-like oscillations. To address the high computational complexity of this model on hardware, we introduce a Hybrid Model of DNM (HMoDNM), which approximates all major nonlinearities using a combination of dual-sinusoidal expressions and ROM-based lookup tables. These approximations achieve high numerical fidelity with root mean square error (RMSE) below <span><math><mrow><mn>1</mn><msup><mrow><mn>0</mn></mrow><mrow><mo>−</mo><mn>2</mn></mrow></msup></mrow></math></span> for most functions, while ensuring hardware efficiency. The full system is implemented on a Xilinx Zynq-7000 XC7Z010 FPGA using a shift-and-add architecture with zero DSP slice utilization. All signals are represented in a fixed-point format <span><math><mrow><mo>〈</mo><mn>1</mn><mo>,</mo><mn>10</mn><mo>,</mo><mn>27</mn><mo>〉</mo></mrow></math></span>, with dynamic range coverage up to 800 <span><math><mi>μ</mi></math></span>M for IP<span><math><msub><mrow></mrow><mrow><mn>3</mn></mrow></msub></math></span> and 600 <span><math><mi>μ</mi></math></span>M for calcium. The design includes pipelined neuron and astrocyte cores, clock-gated nonlinear units, and shared computation modules, achieving a maximum clock frequency of 305 MHz and throughput of 21.7 million Euler steps per second. Overall resource usage is 6690 LUTs (38.0%), 2640 FFs (7.5%), and 4 BRAMs (6.7%), with a low dynamic power consumption of 167 mW and operating temperature of 35.1 °C at room ambient. To validate the model’s functional accuracy, we compare the HMoDNM outputs against the original DNM across two dynamic regimes, achieving correlation coefficients above 94% and NRMSE values below 0.06 for membrane voltage, calcium, and IP<span><math><msub><mrow></mrow><mrow><mn>3</mn></mrow></msub></math></span>. Designed specifically for epilepsy modeling, this architecture provides a robust foundation for real-time tracking and control of astrocyte-influenced seizure dynamics. The proposed HMoDNM architecture offers a versatile foundation for hardware-based neuromorphic applications, including real-time seizure detection, closed-loop neurostimulation systems, and low-power embedded platforms for modeling neuron–glia interactions in brain-inspired computing.</div></div>","PeriodicalId":48609,"journal":{"name":"Engineering Science and Technology-An International Journal-Jestech","volume":"75 ","pages":"Article 102300"},"PeriodicalIF":5.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174673","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-11DOI: 10.1016/j.jestch.2026.102301
B.N. Al Sameera, Vilas H. Gaidhane
According to recent studies, the second largest cause of cancer-related fatalities among women is breast cancer. However, the earlier detection might remarkably increase the survival rates of the patients. Therefore, in this paper, an efficient and robust triangular geometry-based pectoral muscle removal approach is proposed. The motivation of the proposed approach is to improve detection and classification accuracy in two aspects: (i) the pre-processing methods associated with the segmentation and localisation of the affected area, and (ii) the accuracy of the features extracted to categorise as the normal, benign and malignant classes. The variance-weighted average filter-based image denoising and pixel-level image self-fusion method performs robust pre-processing for varying breast densities and preserves fine details. Moreover, a novel angle-based triangular geometry pectoral muscle removal approach with an automatic optimal step length-based multi-adaptive Otsu thresholding is used for improved segmentation. Feature extraction and hybrid optimal feature selection using an adaptive weighted objective function are also introduced. Further, the classification is performed with a hybrid ensemble classifier using a majority voting rule and Bayesian optimisation technique. The experimentations show the classification accuracy of 91.61%, 94.1%, sensitivity 90.77%, 94.87% and specificity of 81.58%, 94.25% for multiclass classification for MIAS and DDSM datasets, respectively. Moreover, an AUC of 0.99 on the ROC curve demonstrate an excellent performance and good diagnostic accuracy in differentiating between benign, malignant, and normal cases of breast cancer.
{"title":"A novel triangular geometry-based automatic pectoral muscle removal approach for breast cancer detection and classification","authors":"B.N. Al Sameera, Vilas H. Gaidhane","doi":"10.1016/j.jestch.2026.102301","DOIUrl":"10.1016/j.jestch.2026.102301","url":null,"abstract":"<div><div>According to recent studies, the second largest cause of cancer-related fatalities among women is breast cancer. However, the earlier detection might remarkably<!--> <!-->increase<!--> <!-->the survival rates of the patients. Therefore, in this paper, an efficient and robust triangular geometry-based pectoral muscle removal approach is proposed. The motivation of the proposed approach is to improve detection and classification accuracy in two aspects: (i) the pre-processing methods associated with the segmentation and localisation of the affected area, and (ii) the accuracy of the features extracted to categorise as the normal, benign and malignant classes. The variance-weighted average filter-based image denoising and pixel-level image self-fusion method performs robust pre-processing for varying breast densities and preserves fine details. Moreover, a novel angle-based triangular geometry pectoral muscle removal approach with an automatic optimal step length-based multi-adaptive Otsu thresholding is used for improved segmentation. Feature extraction and hybrid optimal feature selection using an adaptive weighted objective function are also introduced. Further, the classification is performed with a hybrid ensemble classifier using a majority voting rule and Bayesian optimisation technique. The experimentations show the classification accuracy of 91.61%, 94.1%, sensitivity 90.77%, 94.87% and specificity of 81.58%, 94.25% for multiclass classification for MIAS and DDSM datasets, respectively. Moreover, an AUC of 0.99 on the ROC curve demonstrate an excellent performance and good diagnostic accuracy in differentiating between benign, malignant, and normal cases of breast cancer.</div></div>","PeriodicalId":48609,"journal":{"name":"Engineering Science and Technology-An International Journal-Jestech","volume":"75 ","pages":"Article 102301"},"PeriodicalIF":5.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174674","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-06DOI: 10.1016/j.jestch.2026.102275
Ebru Erdem , Tolga Aydın , Burak Erkayman
<div><div>Last-mile logistics operations in urban environments are becoming increasingly complex due to fragmented customer demands, multiple depots, vehicle capacity constraints, and the need for split deliveries across multiple trips. Classical optimization approaches often fail to address these challenges, as they typically rely on static heuristics or do not integrate real-time data and adaptive learning. Addressing the computational complexity of Multi-Depot Vehicle Routing Problems (MDVRPs) with last-mile split deliveries and multiple trips requires algorithmic innovation and system-level efficiency. To tackle this challenge, we propose a hybrid Artificial Intelligence (AI)-based framework that integrates list-based scheduling heuristics—As Soon As Possible (ASAP), As Late As Possible (ALAP), and List Scheduling—with Transformer networks, Deep Reinforcement Learning (DRL), NeuroEvolution of Augmented Topologies (NEAT), and Model-Agnostic <em>Meta</em>-Learning (MAML). Among the models evaluated, the List Scheduling + Transformer (LST-Former) configuration achieved the best performance regarding route accuracy, resource utilization, and robustness under varying demand conditions. While DRL-based models demonstrated strong adaptability to dynamic logistics, they incurred higher computational costs. This trade-off was mitigated by designing the proposed architecture with High-Level Synthesis (HLS) compatibility, enabling future deployment on low-latency, energy-efficient hardware platforms.</div><div>The framework was validated using a real-world case involving a distribution company based in Istanbul, Türkiye. The scenario captures realistic daily last-mile operations with dynamic orders, multi-depot routing, and high-volume palletized deliveries. In addition to real-world data, five widely used Cordeau MDVRP benchmark instances (p01, p07, p11, p17, p22) were used to assess generalizability and solution competitiveness against best-known solutions (BKS). Experimental validation was conducted through K-Fold cross-validation and a suite of performance metrics, including MSE, MAE, RMSE, DTW, PAP10, POFP, and Coverage Score. Furthermore, comparative analyses with classical algorithms – List Scheduling (LS), Nearest Neighbor (NN), Genetic Algorithm (GA), and Ant Colony Optimization (ACO)—showed that while traditional heuristics offered simplicity or stability, the proposed LST-Former consistently achieved lower route costs and more balanced travel times across datasets. This explicit integration of split delivery, multi-depot coordination, and hardware-aware optimization distinguishes the proposed study from prior VRP research and underscores its practical relevance for urban last-mile logistics. The results confirm the effectiveness of combining learning-based optimization with hardware-aware design to support scalable, real-time routing in logistics. This integrated approach enhances solution quality under complex constraints and facilitates dep
{"title":"Hybrid AI models for multi-depot vehicle routing with split deliveries and multiple trips","authors":"Ebru Erdem , Tolga Aydın , Burak Erkayman","doi":"10.1016/j.jestch.2026.102275","DOIUrl":"10.1016/j.jestch.2026.102275","url":null,"abstract":"<div><div>Last-mile logistics operations in urban environments are becoming increasingly complex due to fragmented customer demands, multiple depots, vehicle capacity constraints, and the need for split deliveries across multiple trips. Classical optimization approaches often fail to address these challenges, as they typically rely on static heuristics or do not integrate real-time data and adaptive learning. Addressing the computational complexity of Multi-Depot Vehicle Routing Problems (MDVRPs) with last-mile split deliveries and multiple trips requires algorithmic innovation and system-level efficiency. To tackle this challenge, we propose a hybrid Artificial Intelligence (AI)-based framework that integrates list-based scheduling heuristics—As Soon As Possible (ASAP), As Late As Possible (ALAP), and List Scheduling—with Transformer networks, Deep Reinforcement Learning (DRL), NeuroEvolution of Augmented Topologies (NEAT), and Model-Agnostic <em>Meta</em>-Learning (MAML). Among the models evaluated, the List Scheduling + Transformer (LST-Former) configuration achieved the best performance regarding route accuracy, resource utilization, and robustness under varying demand conditions. While DRL-based models demonstrated strong adaptability to dynamic logistics, they incurred higher computational costs. This trade-off was mitigated by designing the proposed architecture with High-Level Synthesis (HLS) compatibility, enabling future deployment on low-latency, energy-efficient hardware platforms.</div><div>The framework was validated using a real-world case involving a distribution company based in Istanbul, Türkiye. The scenario captures realistic daily last-mile operations with dynamic orders, multi-depot routing, and high-volume palletized deliveries. In addition to real-world data, five widely used Cordeau MDVRP benchmark instances (p01, p07, p11, p17, p22) were used to assess generalizability and solution competitiveness against best-known solutions (BKS). Experimental validation was conducted through K-Fold cross-validation and a suite of performance metrics, including MSE, MAE, RMSE, DTW, PAP10, POFP, and Coverage Score. Furthermore, comparative analyses with classical algorithms – List Scheduling (LS), Nearest Neighbor (NN), Genetic Algorithm (GA), and Ant Colony Optimization (ACO)—showed that while traditional heuristics offered simplicity or stability, the proposed LST-Former consistently achieved lower route costs and more balanced travel times across datasets. This explicit integration of split delivery, multi-depot coordination, and hardware-aware optimization distinguishes the proposed study from prior VRP research and underscores its practical relevance for urban last-mile logistics. The results confirm the effectiveness of combining learning-based optimization with hardware-aware design to support scalable, real-time routing in logistics. This integrated approach enhances solution quality under complex constraints and facilitates dep","PeriodicalId":48609,"journal":{"name":"Engineering Science and Technology-An International Journal-Jestech","volume":"75 ","pages":"Article 102275"},"PeriodicalIF":5.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174672","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-02DOI: 10.1016/j.jestch.2026.102276
Muhammad Ismail , Muhammad Habib Ullah Khan , Mushtaq K. Abdalrahem , Waqar Azeem Khan , Zohaib Arshad , Taseer Muhammad
<div><div>The current study aims to investigate entropy generation in a two-dimensional magnetic Williamson hybrid nanofluid flow that contains titanium oxide and cobalt ferrite nanoparticles and is subjected to surface-catalyzed reactions via a thin vertical needle by using Levenberg-Marquardt backpropagated neural networks. The properties of heat transport are elaborated by considering the effects of viscous dissipation and joule heating. Additionally, the effects of homogeneous-heterogeneous response, thermal radiation, and thermal stratification are considered. The system of coupled ordinary differential equations is dimensionless by the use of suitable similarity variables. By using “ND-solve” method in Mathematica software the graphical results with matrix data set is generated for <span><math><mrow><msup><mi>f</mi><mo>′</mo></msup><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> , <span><math><mrow><mi>θ</mi><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span>, <span><math><mrow><msub><mi>g</mi><mn>1</mn></msub><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> and <span><math><mrow><msub><mi>N</mi><mi>G</mi></msub><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span>. Further, the obtained matrix data set from Mathematica software is used in MATLAB software to achieve the required graphical for <span><math><mrow><msup><mi>f</mi><mo>′</mo></msup><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> , <span><math><mrow><mi>θ</mi><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span>, <span><math><mrow><msub><mi>g</mi><mn>1</mn></msub><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> and <span><math><mrow><msub><mi>N</mi><mi>G</mi></msub><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span>. The 86 samples are obtained by using artificial intelligence neural networks on Williamson hybrid nanofluid. The total 86 samples are divided into three types of data with 60 samples are used for training, 13 samples for testing and 13 samples for validation. The increase in the <span><math><mrow><msup><mrow><mi>f</mi></mrow><mo>′</mo></msup><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> profile with rising values of <span><math><mi>λ</mi></math></span> is attributed to enhanced stretching or surface tension effects, which increase the momentum gradient near the boundary, and the moderate absolute error values reflect the artificial intelligence neural networks’ ability to handle such sharp gradients. The observed decrease in <span><math><mrow><msup><mrow><mi>f</mi></mrow><mo>′</mo></msup><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> with increasing <span><math><msub><mi>P</mi><mi>m</mi></msub></math></span> is due to the influence of magnetic fields, which introduce Lorentz forces that resist fluid motion, and the consistently low absolute error shows that the model accurately captures this Magnetohydrodynamics behavior. The dec
{"title":"Artificial intelligence driven heuristics approach to analyze entropy optimized MHD flow of non-linear radiative hybrid nanofluids considering vertical thin needle","authors":"Muhammad Ismail , Muhammad Habib Ullah Khan , Mushtaq K. Abdalrahem , Waqar Azeem Khan , Zohaib Arshad , Taseer Muhammad","doi":"10.1016/j.jestch.2026.102276","DOIUrl":"10.1016/j.jestch.2026.102276","url":null,"abstract":"<div><div>The current study aims to investigate entropy generation in a two-dimensional magnetic Williamson hybrid nanofluid flow that contains titanium oxide and cobalt ferrite nanoparticles and is subjected to surface-catalyzed reactions via a thin vertical needle by using Levenberg-Marquardt backpropagated neural networks. The properties of heat transport are elaborated by considering the effects of viscous dissipation and joule heating. Additionally, the effects of homogeneous-heterogeneous response, thermal radiation, and thermal stratification are considered. The system of coupled ordinary differential equations is dimensionless by the use of suitable similarity variables. By using “ND-solve” method in Mathematica software the graphical results with matrix data set is generated for <span><math><mrow><msup><mi>f</mi><mo>′</mo></msup><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> , <span><math><mrow><mi>θ</mi><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span>, <span><math><mrow><msub><mi>g</mi><mn>1</mn></msub><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> and <span><math><mrow><msub><mi>N</mi><mi>G</mi></msub><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span>. Further, the obtained matrix data set from Mathematica software is used in MATLAB software to achieve the required graphical for <span><math><mrow><msup><mi>f</mi><mo>′</mo></msup><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> , <span><math><mrow><mi>θ</mi><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span>, <span><math><mrow><msub><mi>g</mi><mn>1</mn></msub><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> and <span><math><mrow><msub><mi>N</mi><mi>G</mi></msub><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span>. The 86 samples are obtained by using artificial intelligence neural networks on Williamson hybrid nanofluid. The total 86 samples are divided into three types of data with 60 samples are used for training, 13 samples for testing and 13 samples for validation. The increase in the <span><math><mrow><msup><mrow><mi>f</mi></mrow><mo>′</mo></msup><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> profile with rising values of <span><math><mi>λ</mi></math></span> is attributed to enhanced stretching or surface tension effects, which increase the momentum gradient near the boundary, and the moderate absolute error values reflect the artificial intelligence neural networks’ ability to handle such sharp gradients. The observed decrease in <span><math><mrow><msup><mrow><mi>f</mi></mrow><mo>′</mo></msup><mfenced><mrow><mi>η</mi></mrow></mfenced></mrow></math></span> with increasing <span><math><msub><mi>P</mi><mi>m</mi></msub></math></span> is due to the influence of magnetic fields, which introduce Lorentz forces that resist fluid motion, and the consistently low absolute error shows that the model accurately captures this Magnetohydrodynamics behavior. The dec","PeriodicalId":48609,"journal":{"name":"Engineering Science and Technology-An International Journal-Jestech","volume":"75 ","pages":"Article 102276"},"PeriodicalIF":5.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174670","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-31DOI: 10.1016/j.jestch.2026.102289
Hamedalneel BA Hamid , Ahmed Mohamed Ishag , Jamal Hassan , Gomaa Haroun Ali , Tianjun Ma , Adeel Abbas
Proton exchange membrane fuel cell (PEMFC) system is a promising renewable energy source for power system grid integration due to their high energy efficiency. Nevertheless, PEMFC system is highly sensitive to the operating conditions, which could degrade their output performance over time during operation. This article proposes a robust control strategy for a two-stage single-phase grid-connected PEMFC system with an LCL filter to ensure that a sinusoidal current is injected into the utility grid. A robust control strategy includes a reinforcement learning-based maximum power point tracking (RL-MPPT) algorithm and an adaptive current predictive control (ACPC) scheme. The synthesis of RL into an MPPT algorithm simplifies the control problem, eliminates the need for the system model, and prevents deviations in the PEMFC’s maximum power point (MPP) during dynamic variations in temperature and membrane water content (MWC) by simultaneously tuning the boost converter duty cycle. Furthermore, an (ACPC scheme comprises an outer-loop dc-link voltage controller using a PI controller augmented with a notch filter (NF) to prevent double-line frequency dc-link voltage ripple from affecting the grid current reference amplitude and an inner-loop current controller to generate the predicted grid current. To achieve high-accuracy current predictions, a real-time parameter estimator based on the Kalman filter (KF) is integrated into the controller framework. Lastly, findings show that the RL-MPPT algorithm achieves faster settling time and 95.5% MPP average tracking efficiency compared to INC and FLC MPPT algorithms. Additionally, an ACPC scheme shows good sinusoidal reference tracking and minimum THD in the presences of the large LCL filter parameter variations and model uncertainties.
{"title":"A robust control strategy for two-stage single-phase grid-connected proton-exchange membrane fuel cell system with an LCL filter","authors":"Hamedalneel BA Hamid , Ahmed Mohamed Ishag , Jamal Hassan , Gomaa Haroun Ali , Tianjun Ma , Adeel Abbas","doi":"10.1016/j.jestch.2026.102289","DOIUrl":"10.1016/j.jestch.2026.102289","url":null,"abstract":"<div><div>Proton exchange membrane fuel cell (PEMFC) system is a promising renewable energy source for power system grid integration due to their high energy efficiency. Nevertheless, PEMFC system is highly sensitive to the operating conditions, which could degrade their output performance over time during operation. This article proposes a robust control strategy for a two-stage single-phase grid-connected PEMFC system with an LCL filter to ensure that a sinusoidal current is injected into the utility grid. A robust control strategy includes a reinforcement learning-based maximum power point tracking (RL-MPPT) algorithm and an adaptive current predictive control (ACPC) scheme. The synthesis of RL into an MPPT algorithm simplifies the control problem, eliminates the need for the system model, and prevents deviations in the PEMFC’s maximum power point (MPP) during dynamic variations in temperature and membrane water content (MWC) by simultaneously tuning the boost converter duty cycle. Furthermore, an (ACPC scheme comprises an outer-loop dc-link voltage controller using a PI controller augmented with a notch filter (NF) to prevent double-line frequency dc-link voltage ripple from affecting the grid current reference amplitude and an inner-loop current controller to generate the predicted grid current. To achieve high-accuracy current predictions, a real-time parameter estimator based on the Kalman filter (KF) is integrated into the controller framework. Lastly, findings show that the RL-MPPT algorithm achieves faster settling time and 95.5% MPP average tracking efficiency compared to INC and FLC MPPT algorithms. Additionally, an ACPC scheme shows good sinusoidal reference tracking and minimum THD in the presences of the large LCL filter parameter variations and model uncertainties.</div></div>","PeriodicalId":48609,"journal":{"name":"Engineering Science and Technology-An International Journal-Jestech","volume":"75 ","pages":"Article 102289"},"PeriodicalIF":5.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081426","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号