This work presents the results of experimental approbation of the ability to fuse telecommunication graded index multimode optical fibers of ISO/IEC Category OM2+/OM3 and silica microstructured optical fibers by using a commercially available field fusion splicer kit Fujikura-36S (Fujikura Ltd., Japan). We tested silica microstructured optical fiber with a hexagonal configuration that provides quasi-singlemode operation over an extended spectral band up to short wavelengths (λ = 800–1700 nm) in comparison with conventional singlemode optical fiber of ratified ITU-T Recommendations. The potential possibilities were explored for reducing of attenuation by selection and combination of fusion program parameters (arc current and arc time) to find a compromise between quality fusion splicing without any defects and minimization of insertion loss. About 1.5 m long short segment of researched microstructured optical fiber was spliced between two spools with multimode fibers, while each one of them was also spliced with two pigtailed spools with singlemode optical fibers. Fiber length of each spool was at the least 1 km long in order to provide insertion loss measurements by using optical time domain reflectometer EXFO AXS-110 (EXFO, Canada). Based on a series of experiment, we have identified an optimal combination of arc current (STD-30 a.u.) and arc time (1100 ms) which provides desired compromise between reducing the fusion program parameters to prevent the MOF air hole collapsing under the quality splice without defects, and decreased insertion loss. As a result, the total insertion loss has been reduced to 0.8–1.0 dB and more in comparison with the reference value. For instance, mentioned above program settings gave the loss 5.721 dB under the reference value 6.722 dB. Therefore, the experimental results confirm ability of fusion splicing microstructured and multimode optical fibers with an acceptable insertion loss by using commercially available field fusion splicer kits. This provides to utilize proposed microstructured optical fibers with extended singlemode operation wavelength range in various applications of laser technology, optical devices and short-range high bit rate optical networks.

We present a defocus-dependent, quantitatively validated framework for nanosecond fiber-laser irradiation of thermally oxidized silicon. By systematically varying the beam defocus, the on-axis peak power density i and spot diameter d are correlated with distinct surface morphologies on SiO2 (150 nm)/Si (111). Controlled scanning experiments made with nanosecond fiber laser (IRE-Polyus ILI-1-50, wave length λ = 1062 nm, pulse width τ ≈ 120 ns, Pulse Repetition Frequency 50 kHz, and scan velocity v = 100 mm·s–1) reveal a continuous morphological sequence — wrinkling, blistering, and premelting — governed by thermo-mechanical stress and interfacial adhesion. Surface features were quantified using two-dimensional Fourier analysis for wrinkle wavelength (λ ≈ 8–10 µm) and automated image segmentation for blister size distributions (1–3 µm). Point-exposure measurements yield a blister-onset energy of 0.09 J, corresponding to a per-pulse fluence of 0.17 J·cm–2. Logistic regression with 95 % bootstrap confidence intervals with bootstrap replications B = 2000 defines the wrinkle-to-premelting transition, producing a statistically bounded (I, d) phase diagram. Control experiments confirm that both dynamic scanning and the oxide cap are indispensable: point irradiation on SiO2/Si produces only shallow ⟨110⟩ slip lines from substrate relaxation, while bare Si remains featureless under identical conditions. Hydrofluoric (HF) etching verifies that the observed blisters are true interfacial delaminations. This framework extends classical buckling–delamination mechanics to cyclic nanosecond heating and provides predictive guidelines for precision, sub-melting laser microfabrication of dielectric–semiconductor systems.

The main image quality characteristics used in optical system design and fabrication for testing produced devices are the Modulation Transfer Function (MTF) and the Encircled Energy Function (EEF) which are calculated based on pinhole test objects images captured by the testing equipment. For each inspected lens, taking into account the testing equipment parameters, the pinhole diameter, which can be considered infinitely small, can be determined. Such pinholes exhibit the properties of ideal theoretical models and do not cause additional errors. However, pinholes with a diameter greater than infinitely small are most often used in practice. The paper proposes pinhole diameter influence correction methods for EEF calculation and for increasing the maximum calculable spatial frequency in MTF calculation during lens testing based on the captured images. Inverse filtering is used for the correction. The Fourier image of a pinhole with a given diameter is used as a filter. This method has limitations due to the possibility of undefined values appearing in the corrected function when dividing by values close to zero in the pinhole spectrum. To eliminate this effect, the linear interpolation of the MTF is used at frequencies where the test object spectrum function is close to zero. This approach is necessary when correcting the pinhole diameter for EEF. For this purpose, the corrected point spread function, which is the inverse Fourier transform of the MTF corrected over the entire frequency range, is calculated. The proposed method was tested on various-sized test objects generated images. As a result of the research, a divisor function threshold value that does not cause a strong increase in the corrected MTF values due to division by values close to zero was selected for MTF correction. An expression that allows the determination of the maximum test object diameter suitable for correction using the proposed method was obtained. It is shown that the corrected MTF values interpolation allows calculating the MTF and EEF using apertures 2.5 times larger in diameter than with the standard inverse filtering method, and 8 times larger than the pinhole diameter, which can be considered infinitely small. At the same time, the maximum absolute error of the corrected MTF and EEF does not exceed 0.01 relative units. The proposed method reduces the lens testing time by eliminating the need for strict pinhole diameter selection. Furthermore, the use of pinholes with a diameter 8 times larger than infinitely small ones increases the light flow during EEF and MTF testing, which is particularly important when testing lenses with low optical efficiency (small apertures).

Automation of optical system design is one of the key directions in modern optical engineering. The combination of physically grounded simulators, numerical optimization techniques, and machine-learning algorithms enables the development of compact, energy-efficient, and manufacturable optical systems. However, the high dimensionality of parameter spaces, significant computational costs, and the lack of unified verification criteria necessitate a systematic analysis of the applicability of different methods. The article presents a systematization of current approaches to optical system design automation, an analysis of their key characteristics, and an assessment of prospects for further development. Five main classes of methods are examined: differentiable physical models, deep-learning algorithms, evolutionary and metaheuristic optimizers, hybrid schemes combining machine learning methods and physics-based modeling, and fully forward (optical) training approaches. Consideration is given to the formation of a unified comparison framework that enables objective evaluation of speed, accuracy, reliability, robustness, generalization capability, computational complexity, and energy efficiency across different algorithms. A classification of automated optical design methods is proposed. The analysis includes physical models, neural-network architectures, and optimization algorithms. A comparative evaluation based on a unified set of metrics is provided, including qualitative and quantitative assessments derived from peer-reviewed publications from 2019 to 2025. The study demonstrates that differentiable physical methods provide the highest level of physical fidelity and accuracy. Deep-learning methods ensure maximal speed of solution generation. Evolutionary algorithms exhibit robustness against local minima. Hybrid approaches offer an effective balance between speed and physical correctness. Fully forward training methods (FFM approaches) and optical neural networks deliver high energy efficiency and show potential for hardware acceleration of the design process. The results can serve as a basis for selecting design strategies for optical systems across tasks of varying complexity — from preliminary configuration search to high-precision optimization. Progress in automated optical system design is linked to further standardization of datasets, integration of hybrid approaches, and the advancement of optical neural networks.

Printed Circuit Board manufacturing is a key sector of modern electronics industry where improving the throughput of microvia drilling operations is of paramount importance. One effective solution is maskless laser technology which provides high accuracy and processing flexibility. However, its bottleneck remains the beam positioning speed, limited by the inertia of galvanometer-based scanners. This work proposes a hybrid control method for microvia laser drilling that combines a galvanometer-based scanner and an acousto-optic deflector. The set of vias is pre-partitioned into clusters so that all vias within each cluster can be processed by the acousto-optic deflector inside its deflection field without involving the galvanometer. Cluster centers are then connected by a minimal trajectory computed using a combination of a greedy algorithm and the pairwise exchange method (2-opt) which minimizes the total travel length of the galvanometer and the overall drilling cycle time. This approach enables coordinated use of the high-speed acousto-optic deflector for local processing and the long-range galvanometer for movements between clusters. Implementation of the proposed method reduced the galvanometer travel length from 3,097.05 mm to 1,674.19 mm and decreased the total processing time by more than a factor of 3.3 compared with traditional approaches. The effect is achieved by minimizing the number of large inertial moves and shifting a portion of the motion tasks to the high-speed acousto-optic deflector. Unlike known approaches that optimize a single traveling salesman problem route over all vias, the proposed method realizes a hierarchical routing scheme. Classical methods minimize route length but do not account for the dynamic limitations of the galvanometer which leads to excessive inertial moves. Pure acousto-optic deflector based systems provide very high speed but are limited by a small deflection field. The hybrid approach combines advantages of both technologies: the acousto-optic deflector delivers high-speed processing within clusters, while the galvanometer performs efficient transitions between them. The method requires no substantial hardware modifications, can be integrated into existing control systems, and is adaptable to microprocessing of glass substrates (for through glass vias) for 2.5D and 3D packaging architectures.

This paper presents an output-feedback control algorithm for linear systems with input delays in the presence of external disturbances. The proposed control law consists of two main components. The first component is an output predictor employed for synthesizing the control input that guarantees the stability of the closed-loop system. The second component is based on disturbance prediction and is designed to compensate for external disturbances. A synthesis procedure for the output and disturbance predictors is provided, along with a method for selecting an auxiliary loop for disturbance estimation. The obtained results are validated through strict mathematical analysis. A key advantage of the proposed approach is that the output predictor is constructed directly from the measured system output, which enables controller synthesis without relying on the system state, unlike many traditional methods. The effectiveness of the developed algorithm is demonstrated through computer simulations in MATLAB/Simulink.

In the era of rapidly evolving digital infrastructures, ensuring the scalability and efficiency of technological transactions has become a critical challenge. Traditional blockchain models often suffer from limitations, such as high latency, restricted throughput, and network congestion, particularly under high transaction volumes. This paper proposes a novel dual-layer blockchain architecture designed to address these limitations by segregating transaction processing and consensus mechanisms into two distinct but interoperable layers. The first layer, a lightweight transactional layer, handles real-time data exchange and verification with minimal computational overhead, while the second layer focuses on robust consensus, security, and long-term data immutability. By decoupling these functions, the proposed model significantly improves scalability, reduces latency, and enhances system responsiveness. Experimental simulations demonstrate that the dual-layer approach outperforms conventional single-chain systems in terms of transaction throughput, confirmation time, and scalability under varying loads. This architecture holds promising potential for deployment in sectors requiring high-performance, secure, and decentralized transaction systems, such as finance, supply chain, and smart industry ecosystems.

Interpretability of machine learning models is a key requirement for robustness in the implementation of artificial intelligence technologies. Traditional explanation methods provide formal interpretation and produce results that are fragmented and hard to interpret, reducing the transparency of decision making. To address this problem, a modern approach is proposed based on the use of large language models combined with augmented search, which ensures the involvement of external knowledge and increases the coherence of sequences. A distinctive feature of this approach is the focus on semantic consistency, the stability of interpretations, and user comprehensibility. In the presented approach, a large language model acts as an interpreter, generating explanations based on the model input data and external knowledge obtained through augmented search. The approach includes a mechanism for generating human-readable observations, verifying interpretable features (tokens), and assessing the robustness of inferences. As a result, inferences are generated that are consistent with the context of the subject domain. The effectiveness of the proposed approach is tested using MITRE ATT&CK data, which provides standardized information on cyber threats. A comparison with the SHapley Additive exPlanations method showed that the proposed approach provides higher semantic consistency of consequences and greater robustness of feature importance assessment while maintaining advanced level accuracy. Experimental results obtained by the interpreter based on large language models are in terms of human comprehensibility and perception. The developed approach, when considered with conservative methods, provides understandable and context-based explanations. The obtained results make the proposed method promising for application in economics where critical explanations of decision-making are essential. Future research includes integrating the solutions into real systems, investigating automatic validation of results, and adapting them to various large language model architectures.

Modern methods for processing medical images are mostly based on convolutional neural networks and transformer architectures. A major issue is the large number of parameters in existing neural network models, which leads to high computational resource requirements. Kolmogorov-Arnold Networks, built upon spline functions, demonstrate comparable accuracy to traditional architectures while having significantly fewer parameters. In the convolutional layers of Kolmogorov-Arnold Networks, various polynomial basis functions can be used instead of splines, which substantially affect segmentation quality. The objective of this study is to compare the effectiveness of 22 different polynomial functions within Kolmogorov-Arnold Networks for segmenting two-dimensional medical images on small datasets. Experimental results show that the polynomial functions studied achieve performance comparable to state-of-the-art transformer models while requiring significantly fewer parameters. The obtained results confirm the promising potential of further research into polynomial functions within the convolutional blocks of Kolmogorov-Arnold Networks, especially under limited computational resources. Further studies of polynomial functions in the convolutional layers of Kolmogorov-Arnold Networks are necessary for processing advanced medical imaging data.

In contemporary engineering and scientific practice, multi-objective optimization often facilitates the search for compromise solutions without prescribing weight coefficients or bounds, forming a Pareto front via heuristic approximation based on genetic algorithms. However, even an approximated Pareto front consists of a large set of points, which complicates analysis and selection of solutions. To organize and structure the obtained results, clustering can be employed to identify representative groups of trade-offs. The scientific novelty of the proposed clustering method lies in the combination of Ordering Points to Identify the Clustering Structure and k-means algorithms with the introduction of medoids identification, which ensures automatic noise removal and a compact representation of representative strategies. A two-stage clustering approach is proposed. At the first stage, Ordering Points to Identify the Clustering Structure algorithm is used to construct an ordered density profile and to automatically filter out noise points based on the reachability threshold. At the second stage, the k-means algorithm is applied to the filtered Pareto front core to partition it into clusters, compute the centroids, and then determine the medoids — real representative data points. Two experiments were conducted on three-dimensional Pareto front datasets (1226 and 2514 core points after filtering). As a result of applying the proposed approach, a partition into 10 clusters was achieved. It was found that after filtering, the proportion of noise points was less than 1 % of the total number of solutions. The filtering step significantly reduced the metric assessing the quality of cluster centers, with only a moderate increase in the total clustering time. A small discrepancy between centroids and their corresponding medoids indicates the high representativeness of the resulting clusters. The proposed hybrid method, combining Ordering Points to Identify the Clustering Structure and k-means algorithms, requires the adjustment of only two parameters and automatically adapts to nonlinear densities and input data scales. The scope of this method can be extended to any multi-objective optimization problems solved through the construction and analysis of the Pareto front, including engineering optimization, logistics, energy systems, and financial modeling. In the future, the approach may be enhanced by integrating adaptive mechanisms for automatic determination of optimal algorithm parameters, as well as dynamically changing multi-objective problem settings.

The spread of vulnerable Internet of Things devices leads to an increase in the number of attacks on them, which requires the development of accurate and resource-efficient detection methods. Existing Intrusion Detection System models adapt poorly to different datasets. This paper proposes a solution to this problem based on the Edge-Mamba architecture — a “lightweight model” (distilled models) built on a linear-time selective State Space architecture. An evaluation is provided of the ability to transfer models across heterogeneous datasets and ensure their operation on end devices in real time. The proposed model is based on a selective State Space architecture and provides linear complexity for sequence processing. Adaptation of the model for network traffic analysis is achieved through the encoding of 74 features and the application of two State Space Model blocks. This design reduces computational costs while maintaining high accuracy in attack classification. Experiments were conducted on modern datasets CICIDS-2017 and TII-SSRC-23. The results demonstrate that Edge-Mamba achieves an accuracy of 99 % with a latency of 0.15 ms on the TII-SSRC-23 dataset, and an accuracy of 98 % with a latency of 2.4 ms on the CICIDS-2017 dataset. When transferring the model from one dataset to another without additional training, the classification accuracy drops to 65 %; however, fine-tuning on 10 % of the target dataset increases the accuracy to 99 % without any increase in classification latency. Thus, the proposed model demonstrates comparable or superior accuracy relative to existing approaches. In multiclass classification, the Edge-Mamba model outperforms CNN-BiLSTM and Transformer by 1–3 % in terms of macro-F1 score while maintaining lower latency. The model preserves its efficiency on resource-constrained devices. Therefore, the proposed approach combines high accuracy with transferability across datasets, making it applicable for Intrusion Detection System deployment on network gateways, Internet of Things hubs, and containerized infrastructures.

Object detection in video surveillance systems, medical diagnostics, autonomous vehicles, and other technical systems incorporating computer vision components is closely tied to selecting the optimal model for a specific dataset. This task remains labor-intensive and requires substantial computational and time resources. In this work, the authors propose a method for automated generation of recommendations for selecting an object detection model based on the analysis of image meta-features and a knowledge base of results from previous studies. The proposed method involves forming meta-feature vectors from a dataset (images and annotations) and searching for similar datasets within a collected knowledge base using the CatBoostClassifier algorithm (F1-score = 0.82). After identifying analogous datasets, object detection models are ranked by accuracy (mAP50) and performance (FPS) based on previously accumulated experimental data, enabling the recommendation of the most suitable architectures. The method is implemented in the open-source recommendation framework Object Detection Recommendation System (ODRS). Experimental evaluation confirmed that applying the proposed algorithm reduces model selection time by 40 % compared to traditional approaches that require manual testing. Validation on 12 test datasets demonstrated high efficiency of the method in tasks with limited computational resources and diverse image characteristics. Due to its integration into ODRS, the method is accessible to specialists without deep expertise in machine learning or computer vision. It is particularly useful in scenarios with constrained time and computational resources, where a quick, well-founded initial approximation for model selection is needed. Future work includes expanding the knowledge base and adapting the method for semantic segmentation and object tracking tasks.

Computed tomography is used for diagnostic purposes in various fields: oncology, traumatology, dentistry, etc. Additionally, it is implemented to provide the information on the correct patient positioning relative to therapeutic equipment for example in brachytherapy or in image guided radiation therapy complexes. As a rule, such installations include an electric power drive system for gantry — moving part of the apparatus holding medical equipment: X-ray tube and detector, radiation head et al. Improving the event synchronization between the components of medical equipment and those to the motion control system opens up new abilities for control system architecture design. Implementing a single real time fieldbus such as EtherCAT for data exchange between the subsystems can enable high level of synchronization expediting the therapy procedure and improving safety. A platform for deploying EtherCAT in medical devices is proposed. It includes EtherCAT master and slave implementations. The PC-deployed EtherCAT master is based on open-source software: Linux, IgH EtherCAT master stack. The slave devices are implemented with market available components. The stability of the EtherCAT cycle time at 1 kHz and the slave devices events synchronization abilities are being investigated. The experimental results obtained from a designed setup with such an EtherCAT master controlling two slave devices developed showed the ability to synchronize events between the two slave devices within a sub 100 ns range. The stability of the proposed EtherCAT platform was proved with the measured 1000 us EtherCAT cycle time jitter of a few microseconds. It is possible to improve the performance of medical devices with computed tomography by implementing the EtherCAT industrial network. The proposed solution, based on open-source software and market-available components, provides a high level of production safety.

Traditionally, multithreaded data structures have been designed and tested for use with Operating System (OS) threads. However, in recent years, programming languages have introduced an alternative — lightweight threads (also known as asynchronous calls or coroutines). These threads can be divided into two types: stackful and stackless; this paper focuses on stackful coroutines. The main advantage of lightweight threads lies in their reduced overhead. Unlike OS threads, the C++ Standard Library lacks a native implementation of stackful lightweight threads. They are instead provided by several third-party libraries which differ in scheduling policies, memory management approaches, and interfaces. Moreover, reusing existing code is often infeasible due to the explicit marking of context-switching points required by coroutines, their absence may lead to deadlocks between lightweight threads. To narrow the research scope, this work focuses on a specific synchronization primitive — the mutex (mutual exclusion). Several lightweight threading libraries were examined, including Argobots, Boost Fibers, and Userver. Based on an analysis of their functionality, a three-phase waiting mechanism was developed that combines active spinning, context switching, and coroutine suspension and resumption techniques. The description of the mechanism includes a general approach for integrating it into existing mutex algorithms. This mechanism acts as an additional abstraction layer between lightweight threading libraries and mutex implementations, enabling a unified design. Due to the absence of existing solutions, a dedicated tool was developed for testing and performance evaluation of the modified mutexes. Using the proposed method, several well-known mutexes were adapted for the usage from lightweight threads. To test these primitives, a tool was created that allows integration with arbitrary coroutine implementations and accounts for their specific execution models. The throughput of the adapted mutexes was measured using this tool under a specialized workload scenario, reflecting the characteristics of cooperative multitasking. In future work, the aim is to develop a comprehensive library encompassing all known mutex implementations adapted for lightweight thread environments. This will enable detailed performance comparisons across coroutine libraries, scheduling algorithms, and processor architectures. It is expected that a thorough analysis of existing algorithms behavior will lead to the development of a new mutex design optimized for lightweight thread execution.

The evaluation of Large Language Models (LLMs) for code generation tasks presents unique challenges, because conventional Natural Language Processing (NLP) methods might be not applicable for assessing the code. Traditional text similarity metrics may fail to capture the functional correctness of generated code. This study investigates the effectiveness of various evaluation metrics by comparing a LLM-generated code with the mutated versions of the original code snippets. Using state-of-the-art models and benchmarks, the generated and the mutated codes were evaluated using some widely used NLP metrics, including code-oriented CodeBLEU and Ruby, and the neural network-based BERTScore and CodeBERTScore. Results demonstrated that text-oriented metrics tend to have inferior relevance in assessing programming tasks, particularly when functional accuracy is crucial. Code-specific and neural metrics show higher correlation with test pass rates, although their limitations highlight the need for a further refinement. The findings underscore the importance of developing functionality-aware evaluation methods for LLM-driven code generation. This research suggests insights into metrics selection to assess the quality of AI-generated code.

In the context of rapid development of the Internet of Things, energy-efficient mobile sensor networks with moving nodes are becoming increasingly relevant. This work considers a kinematic sensor system with a Control and Information Processing Center (CIPC) where mobile nodes transmit data in cyclically organized time slots. It is assumed that communication sessions during data transmission from mobile nodes to the CIPC are arranged cyclically. The transmission cycle is divided into equal time segments (time slots) such that each slot is dedicated to data transfer from a specific mobile node to the CIPC. A method is proposed for constructing an optimal schedule for node-CIPC interactions based on the criterion of the system total energy consumption. The method enables the CIPC, at each new cycle, to select an order of slot distribution among nodes that ensures minimal energy expenditure. The proposed method optimizes the schedule of communication sessions to minimize the overall energy consumption of the system. It integrates a kinematic model of node movement based on Dubins differential equations, a radio physics model of signal propagation (Friis formula), and an assignment problem for scheduling. Developed are: an energy consumption model considering predicted distances to the CIPC based on Dubins trajectories; an algorithm for constructing the optimal data transmission schedule; and a software implementation of the method. Numerical experiments with a network of 10 nodes demonstrated a reduction in total energy consumption by 29.8 % compared to uniform slot allocation. The proposed approach complements existing research in mobile sensor networks where, as a rule, realistic kinematic constraints are either not considered or global schedule optimization is absent. The method is especially effective in scenarios with controlled mobility (drones, ground robots, autonomous platforms).

This paper addresses the issues related to the implementation of the interaction of automaton objects formalized using specialized state transition graphs. This representation approach, similar to state machine diagrams in Unified Modeling Language, significantly simplifies the development and subsequent maintenance of software. Each automaton object manages specific behavioral aspects of the system while their interaction through the corresponding interfaces ensures the achievement of common goals. Visualization of these objects is implemented using the CIAO (Cooperative Interaction of Automaton Objects) v.3 automata-based programming language. The implementation of the interaction mechanism involves developing a software system that supports the joint execution and interaction of automaton objects. To implement the proposed automaton objects interaction, the bootstrapping technique, known since the mid-1960s, is used. This method involves creating a compiler or interpreter in the same language for which it is being developed. The stepwise refinement method is used to construct the initial interpreter. Subsequently, using transformation patterns from imperative to automata-based constructs, the interpreter is modified into a system of interacting automaton objects, thus achieving the result of the bootstrapping process. This research yielded data structures for representing CIAO v.3 programs. The interpreter’s structure was described in pseudocode using the stepwise refinement method. A set of patterns is proposed to implement imperative constructs through automata-based programming techniques. The structure of the CIAO v.3 language interpreter is presented using CIAO v.3 itself. A Python-based interpreter prototype was realized. The conducted study demonstrates the successful software self-implementation of the CIAO v.3 using the bootstrapping method. The CIAO v.3 language provides efficient design and implementation of software solutions, and also guarantees fault-tolerant component interaction due to the ability to automatically check the properties of CIAO v.3 programs. The proposed approach can be utilized for implementing domain-specific languages in multi-agent systems and human-machine interaction interfaces.

With the development of digital interaction, there is a need to control the information distributed by users so that it complies with legal requirements. The study proposes a technique for classifying Russian-language texts by considering both the biological sex of the authors (male and female) and gender-related distinctions, including heterosexual and homosexual groups as well as specific LGBT categories such as gay men, lesbian women, bisexual, and transgender authors. The article proposes a technique for identifying the sex and gender of authors of Russian-language texts by utilizing an ensemble of methods, including Support Vector Machines (SVM), Convolutional Neural Networks (CNN), and the Russian Bidirectional Encoder Representations from Transformers (RuBERT). This research is introducing a technique for gender identification in Russian texts, as prior studies have only addressed the automatic determination of an author’s sex. To conduct the study, custom datasets of user comments from social networks were created and annotated with both sex and gender labels. In tasks related to identifying the author’s sex, the proposed technique achieved an accuracy exceeding 90 %. This includes the classification of heterosexual men and women as well as a mixed dataset comprising both heterosexual and homosexual individuals. When distinguishing individuals of the same sex based on sexual orientation, the results showed that homosexual and heterosexual women display uncommon writing styles in contrast to men of different orientations, with accuracy rates of 93 % for women and 85 % for men. Additionally, an experiment focused on identifying LGBT individuals and their gender identities based on writing style achieved an accuracy of 93 %. The key takeaway from this study is that combining CNN, RuBERT, and SVM leads to a more robust model for gender classification. The method has been tested on varying numbers of samples. In the basic experiments for determining sex, the method achieves an accuracy of 92 %. When addressing the task of detecting the author’s LGBT affiliation, the method achieves 93 % accuracy. The proposed technique can be applied for automated monitoring of social networks to detect and analyze the gender identity of text authors. It is also promising for use in content moderation systems as well as in sociological and linguistic research.

Agent-Based Models (ABMs) have proven to be an effective tool for describing and predicting the dynamics of respiratory infections and forecasting future outbreaks and have helped health organizations control the disease by developing effective intervention strategy. The use of ABMs is accompanied by very high computational cost, which limits their use in real time. Replacing ABMs with machine learning-based models that can replicate the output or couple the two models together is a solution to the computational cost problem. This paper proposes a machine learning-based surrogate model to simulate an ABM simulating the spread of respiratory infection in Saint Petersburg to reduce simulation time and maintain equivalent accuracy in estimates. The research was based on evaluating the performance of a set of machine learning models under different approaches as surrogate models to use in place of ABM. Methods for generating ABM output chains were compared and evaluated through experiments using single-model approaches or ensemble approaches as a predictive model for each time step in the output (independent multi-output and regression chaining) or hybrid models between agent-based and machine learning. The results indicated that there are several models capable of replicating the simulation output sequence of the ABM with a slight superiority of eXtreme Gradient Boosting within the regression chaining approach. In the hybrid approach, the Long Short Term Memory model with the first values of the output sequence within the feature space outperformed the other models in obtaining more accurate results and achieved the lowest Mean Absolute Error and Root Mean Square Error.

Switched reluctance motors (SRMs) have been very popular in academia for the last few decades since the invention of the IGBT. However, the drawbacks of SRMs, such as high torque ripple and the cost of the power electronic converter, limit their application in industry and traction. To address these issues, a traction drive topology with an input multi-winding transformer and active rectifiers, based on a multi-phase switched reluctance motor with switched windings, is proposed. This topology which utilizes a multi-phase motor provides significantly lower torque ripple. The use of a multi-phase configuration increases fault tolerance, and in the event of a failure in one of the drive elements (converter or winding), allows for continued operation at a reduced load. The inductance and magnetization profiles of the proposed motor are analyzed from the perspective of torque production and control. A feed-forward vector control system is presented. The values of currents and torque of the motor in both motoring and braking modes were obtained, and the torque ripple was determined. Data on the operation of the active rectifier were also obtained. A start-up model with a propeller was implemented. The simulation results of the developed system confirm the feasibility and advantages of the proposed configuration and its use in industry.

Currently, the development of new control approaches for asynchronous electric drives with stringent requirements for vibration-acoustic performance and spectral composition of autonomous inverter output currents represents a highly relevant research challenge. The key challenges in designing this class of electric drives stem from the relatively low effectiveness of existing technical solutions. This limitation arises either from constraints in current controller synthesis methods or from rigorous demands regarding power-to-weight and dimensional parameters. This paper presents an original method for generating control signals in an alternating current electric drive autonomous inverter. The proposed approach utilizes regulation based on the deviation of the generalized output voltage vector amplitude in the autonomous inverter. The synthesis procedure for such a controller begins with defining the desired closed-loop system transfer function. The system dynamic processes are determined by a characteristic polynomial that can be of arbitrary type. For comparative analysis, two controller types are examined: one based on a Butterworth filter and another utilizing a Newton polynomial. The study proposes employing bilinear transformation to implement the derived continuous functions in discrete form, enabling software implementation in Simulink and subsequent microprocessor-based execution. The developed model, which accounts for discrete control signal generation, has yielded the spectral composition of the drive converter output currents and voltage-frequency characteristics under parametric disturbances introduced by the control object. Results demonstrate that the Butterworth filter-based controller shows superior efficiency compared to both open-loop systems and closed-loop systems with Newton polynomial-based controllers. The obtained results can be effectively applied in the development of low-noise electric drives for specialized applications.

This article examines the geography-aware class of telecommunication networks with mobile nodes. Their defining feature is that each network device knows the geographic coordinates of all other devices and, consequently, is fully aware of the network graph. The aim of this work is to develop a technique for constructing a set of message transmission paths with subsequent placement of messages across these paths. A multipath routing technique is introduced, featuring a procedure for finding a set of paths within the network that connect the message source to the target node. This procedure is based on a modified Dijkstra’s algorithm for determining the shortest path in a directed graph. A method for constructing additional paths is described, which utilizes both the minimum path length criterion and the criterion of minimum intersections with the original shortest path. To accelerate message delivery, the proposed routing technique involves a preliminary ordering of the output message queue based on optimal scheduling rules. The optimality criterion is the minimization of total message delivery time. These rules are formulated for various scenarios involving the presence of pre-partially ordered and unordered message groups within the queue. The process concludes with a procedure for placing the ordered queue onto the set of information transmission paths. The proposed technique is exemplified by transmitting a message queue containing three pre-ordered groups. The difference in the resulting orderings is demonstrated for cases where preemptions are prohibited and where they are allowed. The feasibility of applying the proposed technique is determined by the performance of the device onboard processor, which is assumed to be a priori sufficient in the case of an autonomous unmanned underwater vehicle.

This paper proposes a new algorithm for estimating the carrier frequency of an amplitude modulated signal in the presence of measurement noise. An estimate is provided for the maximum amplitude of the disturbance for which the problem of estimating the unknown frequency can be solved using the proposed approach. The problem of frequency estimation is solved in several stages: parameterization of the initial measured signal to the form of linear regression; application of a nonlinear transformation of the coordinates of the initial regression model; estimation of the unknown frequency. The results of the paper can be used to solve practical problems in the areas of processing and evaluating sinusoidal signals subject to disturbances.

Antisymmetric forms (A-A) of the stability loss of a highly elastic rectangular plate in which two parallel faces are pinched, and the other two are free (CFCF), under the influence of a compressive load on the pinched faces, are investigated. The desired shapes are represented by two odd hyperbolic-trigonometric series with coefficients which should ensure the exact fulfillment of all the conditions of the problem. The problem was reduced to solving a homogeneous infinite system of linear algebraic equations with respect to a single sequence of coefficients containing as a parameter the desired critical load which was found by “firing” during the iterative process. The first three critical loads for a square plate are found and their 3D images are presented. The results obtained can be used in calculations of sensitive elements of various sensors in microelectronics, biology, and medicine.