This paper analyzes contemporary methods of end-to-end speech-to-speech translation that employ discrete representations of the speech signal as an intermediate representation. The relevance of the topic is driven by the growing demand for real-time machine speech translation systems capable of preserving the unique vocal characteristics of the speaker. The study examines approaches that perform cross-lingual speech-to-speech conversion without an intermediate transition to text. Particular attention is paid to the role of discrete units as carriers of semantic and paralinguistic information as well as their significance in end-to-end speech translation architectures. A systematic review and comparative analysis of the literature covering the period 2012–2025, encompassing more than 50 publications, was conducted. Direct speech translation architectures, self-supervised representation learning methods, neural audio codecs, and multilingual scaling systems were analyzed. For each class of methods, the design principles, advantages, and limitations were identified. The main data corpora and translation quality evaluation metrics were reviewed. It is shown that the transition from cascaded machine translation architectures to end-to-end methods reduces error accumulation and decreases translation latency. It is further demonstrated that, when pre-training is employed, the speech synthesis quality of end-to-end systems approaches that of cascaded ones. Discrete acoustic units obtained via self-supervised learning methods and neural codecs are capable of preserving the information necessary for recovering semantic content and individual speech characteristics. Key research gaps have been identified: in the majority of works, discrete representations are used as a technical tool, while their internal structure, linguistic interpretability, and information capacity remains insufficiently studied. The findings of the review may be applied in the design of simultaneous machine translation systems that preserve the vocal characteristics of the speaker. Directions for further research have been formulated, including analysis of the internal structure of discrete representations, development of methods for controlled cross-lingual transfer of paralinguistic characteristics, and the creation of metrics for separately evaluating semantic accuracy and prosody preservation quality in translation. The conclusions obtained are of interest both for fundamental research and for applied problems in the development of machine translation systems.

Thermal hardening of metals under laser irradiation with an auxiliary absorbing layer occurs due to the formation and propagation of a plasma plume in confined conditions. Most post factum studies are focused on the hardness values achieved by processing, while the characteristics of the plasma plume in confined conditions stay deficiently investigated. However, it can reasonably be assumed that the contribution of the plasma plume determines energy reallocation, the development of shock wave processes, and the formation of surface structures with improved functional properties. The aim of this work is to identify the relationship between the properties of laser-induced microplasma during the laser hardening of titanium and the laser irradiation parameters as well as the target configuration. Patterns of plasma plume formation were investigated in the work using the method of optical emission spectroscopy with variable parameters of nanosecond laser radiation with a wavelength of 1064 nm. Spectral lines were recorded in the range of 380–740 nm using an “Avantes” fiber-optic spectrometer. Technically pure titanium plates were used as the object of the study and were examined in three target configurations: titanium plate, a titanium plate under glass, and a titanium plate under a layer of graphite powder and glass. The processing of the recorded emission spectra of the plasma plume included the background signal subtraction, correction for the curve of spectral sensitivity of the spectrometer and the curve of transmission spectrum of the glass, and analysis of spectral lines using the NIST Atomic Spectra Database. Numerical values of surface hardness were determined by performing Vickers microhardness measurements under a low load. It was demonstrated that during laser processing with an impulse power density above 4.1∙1011 W/m2, neutral atom and singly ionized atom lines can be identified for all processing configurations. This indicates that the ablation limit of the titanium target is reached, and a partially ionized plasma plume is formed. The introduction of absorbing (graphite) and confining (glass) layers leads to a visible decrease in the registered absolute emission intensity while keeping the characteristic spectral lines. At the same time, the surface hardness values increase from 185 HV to 520 HV. It was demonstrated that the target configuration significantly influences the formation patterns of the plasma plume and the hardness of the titanium surface layer within the target assembly. The identified trends may indicate the implementation of a combined thermomechanical (shock) hardening mechanism, in which efficient conversion of laser energy absorbed by the graphite layer located beneath the glass plate results in heating and strengthening of titanium to a greater extent than in heating of graphite particles ablated into the plasma plume. The results of this work can be used to optimize laser processing regimes of titanium for controlled formation of surfaces with improved hardness values.

The problem of deep neural network compression is discussed using Convolutional Neural Networks (CNNs) as an example. The size of deep neural networks is an obstacle to their practical application under conditions of limited computing resources, energy, and inference latency requirements. One of the developing approaches to compressing deep neural network models is thinning-removing some of the parameters or structural elements of the neural network model. It is shown that thinning is a tradeoff between classification accuracy and computational efficiency. The problem of compressing CNN models was formulated by thinning parameters while maintaining classification quality at the level of the original model, reducing the number of parameters, computational complexity, and inference latency. Standard Top-1 and Top-5 classification accuracy metrics were used to evaluate classification quality. The degree of compression and inference latency were assessed using metrics, such as the number of model parameters, model size, computational complexity, and inference latency. To ensure a fair comparison of the proposed method with others, inference latency measurements were conducted under identical conditions with a fixed input data size. The concept of representation geometry is introduced to control the stability of the internal feature space. The change in the interclass similarity matrix, calculated from the class centroids in the feature space, is used as a metric for preserving the structure of the feature space. The main result of this work is the development of a new compression method for CNNs based on geometrically controlled thinning of the CNN model. In this method, thinning is performed greedily across CNN blocks, and the admissibility of each step is determined not only by the local importance of channels but also by a global constraint on changing the geometry of the representations. The results of the experiment demonstrate that the proposed method consistently maintains classification quality while reducing computational complexity and the number of parameters compared to the baseline method without thinning and the depth-based thinning method of the CNN model. The proposed method for geometrically controlled thinning of a CNN model can be applied to problems requiring real-time decision making as well as on mobile and embedded devices.

This article addresses the task of machine translation from a spoken-language to Russian Sign Language in the form of an intermediate textual representation — a gloss sequence. The goal of this work is to develop a data preparation method and an automatic translation method for mapping verbal text to a sign-language gloss sequence. A manually annotated parallel corpus of “spoken-language text–gloss sequence” pairs is constructed. A gloss vocabulary is defined based on examples from a sign-language corpus and is used to constrain the set of admissible output tokens. Two model classes are compared: Transformers with an encoder-decoder architecture, adapted to the target task on the parallel corpus; and Large Language Models with a decoder-only architecture applied via In-Context Learning with a few examples and a prompt that includes instructions, the gloss vocabulary, and output-format constraints. Translation quality is evaluated using the BLEU metric on the test split of the parallel corpus. Experimental results show that Transformer-based models provide higher machine translation quality than Large Language Models; the best Transformer result is achieved by mT5-small (0.84). Among Large Language Models, the best value of 0.60 is obtained for GPT-5.2. The proposed method can be applied as part of a system for enabling digital bidirectional communication between sign-language users and spoken-language users. The method translates spoken-language text into a gloss sequence which can subsequently be synthesized using digital avatars to allow sign-language users to understand information that is spoken or written by spoken-language users. Source materials, the parallel corpus, and instructions for reproducing the experiments are available in the public repository dedicated to the method of automatic machine translation of texts from verbal language into a sequence of glosses.

The role of big data technology in the processes of ensuring the security of objects of the critical information infrastructure (CII) of the Russian Federation in the context of increasing cyber-attacks and digital transformation is investigated. The task of moving from a reactive to a proactive protection model is being actualized. The scientific novelty of the research lies in the development of original directions for integrating Big Data into CII security systems: modeling cascading cyber-physical incidents on intersystem dependency graphs, analyzing the “dark data” of industrial equipment as a source of compromise indicators, and a hybrid architecture of federated learning for interdisciplinary cooperation of CII subjects without transferring confidential data beyond the organization perimeter. A mathematical apparatus for constructing resilient security systems is proposed, based on the representation of a cyber-physical infrastructure in the form of a dynamic weighted oriented time multigraph with Bayesian updating of weights in real time. The dynamics of impact propagation is modeled by continuous Markov chains with absorption, and optimization of the placement of protective equipment is formalized as a stochastic defender–attacker game with the search for an e-approximate Nash equilibrium using reinforcement learning methods. A two-stage iteratively refining algorithm for maximizing sustainability under budgetary, resource, and regulatory constraints has been developed. A closed data-driven loop of proactive security management of CII facilities has been formed, in which big data flows ensure continuous updating of model parameters and adaptation of protective mechanisms. Approach verification on a distribution grid digital twin demonstrated 94.3 % incident classification accuracy. Unlike traditional Security Information and Event Management and User and Entity Behavior Analytics solutions focused on the correlation of known events, the proposed approach provides predictive modeling of the cyber-physical consequences of attacks, taking into account interlayer dependencies. The architecture of federated learning allows for the formation of cross-industry threat models without violating the requirements of information protection legislation. The prospects for application are related to the regulatory consolidation of the developed mechanisms as mandatory elements of security systems at significant CII facilities and the creation of a national platform for the exchange of impersonal threat models.

Container orchestration platforms such as Kubernetes use automatic scaling to maintain performance under dynamic workloads. At the same time, the choice of load balancing disciplines is carried out using the queuing theory apparatus. However, there is currently a lack of queue models and studies reflecting the impact of heterogeneity of request flows on service delays in the currently actively implemented container virtualization cluster systems. The purpose of this study is to quantify and determine the impact of workload heterogeneity on cluster system performance. An improved approach to modeling a Kubernetes cluster is proposed, taking into account the mixed nature of the workload, focused on both computing and information input and output (I/O). The proposed approach is based on a combination of analytical modeling and experimental research. The experiments were conducted on a dedicated physical server with a two-node Kubernetes cluster, which provides increased reliability of the results compared to fully virtualized environments. It has been experimentally confirmed that the heterogeneity parameter fundamentally determines the nature of the impact of the number of deployed containers (scaling) on cluster performance. Workloads with a high proportion of requests requiring significant consumption of computing resources have shown a sharp decrease in performance when the optimal number of containers is exceeded, while workloads focused on I/O are characterized by a plateau effect with a slight decrease in performance when resources are overallocated. The results obtained show a V-shaped dependence of service time on the number of containers deployed in the cluster, demonstrating different sensitivity to scaling depending on the type of workload. The results obtained are the basis for a differentiated approach to container scaling depending on the heterogeneity of the request flow with intensive CPU usage or I/O. A promising area of further research is the development of adaptive scaling algorithms that dynamically take into account the changing nature of the workload in real time.

Identifying the technologies used by web applications is crucial for vulnerability assessment, resource inventory, and research analysis of the digital environment. Traditional tools that rely on static signature rules suffer from fundamental limitations: they require constant manual maintenance, are sensitive to code obfuscation and dynamic loading, and perform poorly when distinguishing features are subtle, distorted, or noisy. The novelty of this work lies in applying a weak supervision method based on behavioral characteristics to create a robust and scalable classifier. This work presents a hybrid approach combining weakly supervised machine learning with rule-based post-processing. Two independent signature-based systems, Wappalyzer and BuiltWith, serve as sources of weakly labeled data; a final positive label for a technology is assigned only when their detections agree. A multi-label Random Forest model is trained on this labeled data. The feature space is constructed from behavioral patterns extracted during an automated browser session, including Hypertext Transfer Protocol headers, cookies, network requests, and structural features of the Hypertext Markup Language and the Document Object Model. To correct model predictions by accounting for typical technology co-occurrences, a post-processing step based on association rules mined from the training data is applied. The experimental evaluation was conducted on a dataset of 8,594 websites for identifying 122 technologies. The Random Forest model demonstrated consistently high precision and recall for common web technologies, such as web servers, content management systems, and front-end libraries. The hybrid approach with post-processing provided a statistically significant improvement in both micro-averaged and weighted harmonic mean of precision and recall, while preserving the results interpretability. On a fixed test set (20 % of the data), the Random Forest model achieved a micro-averaged harmonic mean of precision and recall of 0.750, while the hybrid approach reached 0.763. Association rules affected 23.8 % of the predictions, moderately improving recall. For widespread technologies (Apache, Nginx, WordPress, jQuery), the harmonic mean of precision and recall exceeded 0.840, and for several components (Drupal, Google Cloud) it reached values of 0.960–1.000. The obtained results show that the approach based on weak supervision and behavioral analysis can provide a scalable method for web technology identification. It effectively complements traditional signature-based methods in cases of deliberate code concealment or dynamic execution. This method is promising for automated auditing, enhancing security tool capabilities, and scalable research of the web technology landscape. Future work may focus on improving the feature set and adapting the system for detecting novel and unique technologies.

Microservice architecture has become a standard approach to building highly loaded and evolving information systems. In digital educational platforms, compatibility of external interfaces and low delays of internal inter-service calls are important at the same time. The purpose of the work is to compare the Google Remote Procedure Calling (gRPC) and Representational State Transfer Application Programming Interface (REST API) in terms of response time and bandwidth metrics and formulate recommendations on choosing an approach for the analytics services of educational platforms. The object of the research is a mini-analytics service of the educational platform which, upon request, returns aggregated indicators of user actions for a given time interval. Two functionally equivalent versions of the service are implemented in Python with the same data model and identical business logic: a REST service based on Hypertext Transfer Protocol (HTTP) version 1.1, with data transmission in JavaScript Object Notation format and a gRPC service based on HTTP/2 with data transmission in Protocol Buffers format. The measurements were performed by load scripts on the client side with measurement of the full request execution time in the basic filtering scenario by user ID and interval. The tests were performed with 100, 200, and 500 simultaneous clients with a measuring window duration of 20 seconds. The average response time, 95th percentile latency, and throughput (number of requests per second) were used as metrics. According to the measurement results, with 100 simultaneous clients, gRPC provides lower average latency, while with 500 clients, REST demonstrates higher throughput and lower average response time. As the load increases, the 95th percentile of latency in gRPC remains lower than in REST, which indicates a more stable behavior of the “tail” of delays in conditions of high query competition. It is shown that the choice between REST and gRPC is not mutually exclusive: REST is convenient for public interfaces and integrations with external clients, and gRPC is rational to use for internal calls between microservices where overhead delays are critical and strict contractual specification is required. Recommendations on the choice of technology are formulated taking into account the load profile and integration requirements.

This study presents a method for registering data lifecycle events in cloud storage under an untrusted cloud provider infrastructure. According to the adopted threat model, if the cloud provider is compromised, an adversary is capable of modifying, reordering, rolling back, and deleting stored data as well as the corresponding lifecycle event records. The novelty of the proposed method lies in binding each data version to a dual-signed lifecycle record anchored in a blockchain which enables detection of unauthorized modifications independently of the provider after transaction confirmation. The method is intended for deployment in systems that include cloud storage operated by an untrusted provider and client devices acting as nodes of a private blockchain network. Each create, modify, or delete operation is accompanied by the formation of a registration record containing the hash of the current data version, the hash of the previous version, the operation type, a timestamp, and a user identifier. The record is signed by both the client and the provider, after which a corresponding transaction is formed and submitted to a blockchain using the Tendermint consensus protocol. To accelerate verification procedures, registration records and transaction confirmation statuses are additionally stored on the provider side. The confirmation status is synchronized during subsequent accesses; data versions without confirmed transactions are available in read-only mode. Integrity verification includes matching the payload hash, verifying both signatures, and confirming the presence of the transaction in the blockchain. An experimental evaluation of the method was conducted in a private blockchain network using the Tendermint consensus protocol with 1,000 clients and 10 validators. The average transaction confirmation time achieved in the experiment was 1.3 s, with a peak value of 1.8 s. Client-side cryptographic operations required by the proposed logging method take less than 3 ms per data operation. Compared to commercial solutions (Amazon Web Services CloudTrail and Google Cloud Audit Logs), the proposed method enables audit execution immediately after transaction confirmation but introduces additional latency in the data operation path. Issues related to availability and prevention of physical data deletion is beyond the scope of this study.

Multi-variant collective decision-making in multi-agent robotic systems is a complex problem of swarm intelligence. The architectural features of homogeneous robot groups with decentralized control and the limited capabilities of individual agents provide an opportunity for attackers to introduce and use malicious robots within the swarm. Malicious robots employ various behavioral strategies and create conditions for erroneous decision-making during consensus formation. A method is proposed to detect malicious robots while the swarm performs the task of mapping the original scene. The proposed approach solves the problem of detecting malicious robots within a swarm, regardless of the behavioral strategy they use. The method is based on the hypothesis that the statistical characteristics of local maps obtained by benign robots correspond to the statistical characteristics of the original scene and do not match the characteristics of local maps from malicious robots. Frequency histograms of attributes of the examined object, generated during the analysis of local maps, are proposed as recognition features. The recognition problem is solved using a naive Bayes classifier. This approach ensures high recognition quality by identifying statistically significant differences in the histograms of local maps from benign and malicious robots. The performance metrics of the developed algorithm are evaluated for various scene types. A series of experiments is conducted in which the probability of correctly identifying malicious robots using different behavioral strategies is assessed for the same initial data. It is shown that when the Bayesian classifier is pre-trained on a scene with identical statistical characteristics, the Type II error is reduced to minimal values. Identifying the Type II error is critically important to prevent malicious agents from being admitted to corrective decision-making discussions. The proposed algorithm features a high degree of abstraction, allowing it to be considered for use in a wide range of collective perception tasks involving deliberate malicious information attacks.

We hypothesize and experimentally validate that maximizing the entropy of embedding distributions under specific constraints enhances the generalization ability of machine learning models. To test this hypothesis, we propose a novel regularization method — Hyperspherical Entropy Maximization with Class Separation — designed for noisy and imbalanced classification tasks (e.g., Orthogonal Frequency Division Multiplexing (OFDM) signal recognition). The method is based on extending the loss function of the classical model with three specialized components: the first maximizes entropy by enforcing uniformity of distribution on the hypersphere, the second controls the covariance of features, and the third ensures inter-class separability. Dynamic regularization allows adaptively balancing the influence of the entropy term during training. We establish a link between maximizing the entropy of the embedding distribution and a reduction in the model generalization error bound. In experiments on an OFDM signal classification task, our method shapes a highly uniform feature space: mean cosine similarity decreases to 0.064, coverage reaches 1.71, and the entropy of the similarity distribution attains 10.34 nats, while preserving strong discriminative power, achieving 0.97 accuracy and a Macro F1-score of 0.98, outperforming standard regularization techniques. The approach effectively mitigates overfitting and improves robustness on minority classes. Our approach, which explicitly maximizes representation entropy under task-specific constraints, establishes a promising principle for enhancing generalization in conditions of data scarcity or high noise. Future work will focus on relaxing theoretical assumptions and applying the method to other domains where classification reliability critically depends on representation quality.

The paper presents an applied approach to the automated markup of user communications in the social network VKontakte (VK) and the construction of a compact neural network model for simultaneously determining the communicative class of a comment communicative class of a comment (constructive/destructive/neutral) and its tonality (positive/negative/neutral). The empirical database contains 996 “post-comment” pairs collected from popular regional communities with an emphasis on urban and news topics. Architecturally, a Russian-language BERT-compatible encoder with a multi-purpose formulation is used. Tokenization is implemented according to the WordPiece scheme, optimization is carried out using weighted cross-entropy with class weights to compensate for the imbalance, validation is stratified. The quality is assessed using the macro-F1 metric for both tasks. For the stability of the model on a small case, the encoder is “frozen” for the first epoch followed by a short adjustment as well as iterative retraining on new marked-up batches via the web interface. The explanatory elements of the neural network are provided by highlighting the contributions of tokens (based on the “attention” convolution) and searching for the nearest example from the training sample which connects the model solution with interpreted language markers. Experimental testing on the processor infrastructure has shown that it is achievable to work without a graphics accelerator and external services with limited context. The initial results on the base case demonstrate the expected low values of macro-F1 due to the imbalance and volume of data (about 0.31 for constructiveness and 0.20 for tonality), however, the inclusion of class weights and short cycles of retraining lead to a steady increase in metrics without degradation on rare classes. The practical result of the work is to combine the context-sensitive “post-comment” classification, two-headed multitasking training, and a minimalistic deployment scheme using the FastAPI and Gradio software packages into a reproducible network communications monitoring tool. The scientific novelty is reflected in the joint contextual classification of the tonality of the analyzed text as well as in the application of the “markup”–“retraining”–“validation” cycle with elements of explainability, which distinguishes the approach from single-purpose models of sentiment analysis and behavioral detection schemes for online communication.

Modern companies that incorporate complex software systems as part of their market products are increasingly seeking technological independence in the development and maintenance of these systems. Such companies typically manage very large codebases, often comprising millions of lines of code. Testing, maintaining, and transforming these codebases requires multi-functional analysis and development tools, which cannot simply be obtained off the shelf. This is largely due to the inherent difficulty of many fundamental program analysis problems — indeed, several of these problems are formally undecidable (for example, pointer analysis and garbage collection for C programs). At the same time, large industrial codebases often use only a limited subset of the features available in their respective programming languages. Consequently, program analysis problems that are undecidable in the general case may become tractable for these restricted, domain-specific subsets. This observation creates opportunities for specialized approaches. This article proposes an approach for developing domain-specific program analysis solutions tailored to particular large industrial codebases. Such solutions are in high demand for tasks including fuzzing, symbolic execution, automated test generation, static analysis, code transformation and optimization, and vulnerability detection. The development of such custom domain-specific analysis tools becomes feasible because code-owning companies typically possess substantial engineering resources. These resources can be leveraged in combination with open-source frameworks and ecosystems such as Eclipse, Low Level Virtual Machine, and Microsoft Visual Studio Code. Furthermore, the economic impact is often significant: effective analysis solutions lead to substantial cost savings and quality improvements, which justifies the investment in specialized tool development. The proposed method consists of the following key stages: problem and idea analysis, solution design, requirements development, implementation, testing and validation, deployment and transfer. A central element of this method is the development of code templates that define the subset of the basic programming language used within the codebase. These templates provide a formal foundation for analysis and constrain the variability that tools must support. Notably, requirements development, tool implementation, and validation proceed in parallel, with testing tightly integrated into the development process. Using this method, several tools have been successfully developed, including a static analysis tool for C applications tailored for fuzzing, a clone detection and refactoring tool for a large network device software codebase, and other domain-specific solutions. Each project typically involved between one and three developers and lasted one to two years. While effective, the proposed method is resource-intensive. Analysis of completed projects shows that when a fully specified technical assignment is available from the outset, the required resources for implementing program analysis solutions can be reduced by several times. However, the strength of the method lies in situations where no detailed technical specification can be produced in advance. In such cases, the method enables iterative elicitation of requirements, reducing the risk of developing tools that do not meet real project needs.

Perforated structural elements are widely used in the chemical and other process industries due to their low cost and ease of manufacture, fulfilling various specific functions as parts of apparatuses, machines, and equipment. One of the most common functions is the uniform distribution of gas throughout the working volume. This aspect is crucial, for example, in fluidized-bed apparatuses, because the flow pattern and the efficiency of the ongoing process depend directly on it. Therefore, the design of a perforated element at the engineering stage must be clearly justified in terms of compliance with the specified process requirements for a particular application. One of the key parameters governing the behavior of a continuous-medium flow in such systems is the discharge coefficient which characterizes the deviation of the actual volumetric flow rate from the theoretically possible one. Its accurate determination makes it possible to establish a quantitative relationship between geometry and the hydrodynamic characteristics of the flow which is especially important for engineering calculations and design optimization. Six configurations of single orifices and one type of perforated element with simple cylindrical holes arranged in an equilateral-triangular pattern were investigated. Parametric computational fluid dynamics modeling was performed in the finite-element environment COMSOL Multiphysics 6.1. Dimensionless correlations were formulated in accordance with similarity theory and the Buckingham Pi theorem. The simulation results were processed using multivariate nonlinear regression methods (Levenberg–Marquardt algorithm with multi-start initialization of the parameter vector). The selection of specific correlation forms was carried out using the Bayesian and Akaike information criteria. The accuracy of the resulting models was assessed using correlation and determination coefficients as well as the mean absolute percentage error. A set of dimensionless correlations for the discharge coefficient was obtained for six single-orifice configurations and for a perforated element with cylindrical holes. It was found that within the studied ranges of dimensionless parameters, the discharge coefficient is predominantly described by exponential relationships with saturation with respect to the Reynolds number. For single orifices, correlation and determination coefficients of 0.97–0.99 were achieved with a mean absolute percentage error of 3–8 %. For the perforated plate the coefficient of determination reaches 0.99 with a mean absolute percentage error of 5.48 %. The derived discharge coefficient correlations have practical value as part of broader design methodologies for flat perforated gas-distribution devices, since they provide a quantitative link between geometry, the overall hydraulic resistance of the perforated element, and the actual linear velocity in the channels. The approach proposed in this work can also be used independently to study gas-distribution devices with both regular and more complex configurations.

The subject of this study, which is the focus of this article, is the three-dimensional reconstruction of the trajectory of a free jet of fire-extinguishing agent based on full-scale tests of firefighting monitor systems. The resulting geometric shapes are used in the synthesis and debugging of computer vision systems, control algorithms for firefighting robots, and the construction and validation of mathematical and neural network models of the studied flow processes. Key features of free jets in this applied field include their large scale, dynamically changing boundary shapes, variable optical density across different regions, and more, which significantly complicates the application of existing 3D reconstruction methods. The aim of this work is to develop a technology for modeling the trajectories of fire-extinguishing agents that can address these challenges. This study proposes a methodology for conducting full-scale tests of monitor systems on a pre-marked experimental site with synchronized video recording of the free jet in three planes, including the use of an onboard camera from an unmanned aerial vehicle. Additionally, an algorithm for 3D trajectory reconstruction based on the obtained digital images was developed, taking into account the geometric features of the studied flow processes. The article presents the results of testing the developed technology in studying the motion of a fire-extinguishing agent stream from a fire monitor nozzle under the influence of crosswinds which can significantly deflect the jet laterally. A 3D model of the investigated trajectory was constructed using digital images captured during the experiment. A comparative analysis of the jet geometric characteristics (height, range, and lateral deflection) — calculated from the reconstruction results and measured during full-scale tests — showed a maximum relative error of 3.7 %, indicating the high accuracy of the proposed method. The practical significance of this technology lies in its ability to obtain high-spatial-resolution empirical data in free-jet studies. These data can be used to validate models describing jet motion based on computational fluid dynamics and machine learning methods, particularly in the development of computer vision systems and fire-extinguishing agent targeting algorithms that account for external disturbances in robotic firefighting systems.

The study presents optimization results aimed at improving fuel–oxidizer mixing while preserving the operational characteristics of a liquid rocket engine combustion chamber. Traditional chamber design methods described in classic textbooks are based on semi-empirical procedures intended primarily for high-thrust engines delivering several tens of tons of thrust. There is now a growing demand for commercial launch vehicles of the light and ultralight classes. Given the tight size, mass, and energy budgets of small liquid engines, particular attention is paid to the compactness and reliability of injector assemblies. The work addresses the design and optimization of the injector head to achieve optimal mixing at a standoff from the injector faceplate sufficient to minimize its thermal load. Computational fluid dynamics is applied with combustion, heat transfer, species transport, and radiation. To account for the liquid phases of oxygen and kerosene and to represent their velocities in the injectors correctly, a pseudo-gas equation of state is used. Injector throttling characteristics are computed in ANSYS. Global parametric optimization (particle swarm method) is performed over injector angles, diameters, and layout. “NASA CEA” equilibrium calculations are used to validate output parameters. The developed optimization technique reduces combustion chamber dimensions by nearly a factor of two. Combining computational fluid dynamics of mixing and combustion with optimization algorithms enables preliminary design optimization before prototype fabrication, thereby lowering development and manufacturing costs. Compared with common approaches — parametric sweeps, gradient optimization based on simplified correlations, design of experiments, and manual tuning through test campaigns — the method relies on coupled flow and heat-transfer calculations with a fast pseudo-gas model and automatic selection of injector geometry by criteria of mixture uniformity and combustion stability. This reduces the number of physical iterations, improves spray uniformity, and lowers thermal stresses on components. Application areas of the suggested method: injector heads for small- and medium-thrust liquid engines, gas generators, and ignition chambers of test stands. Prospects of the method: accounting for unsteady oscillations and acoustics, optimization under uncertainties, integration of additive manufacturing constraints, and automated channel synthesis.

Biodegradation of medical devices is a complex process that depends on many factors, ranging from the methods of polymer synthesis to the concentration of active enzymes in the medium. Mathematical modeling of the biodegradation of polyhydroxyalkanoates (PHAs) is of great practical importance as it allows the degradation of the material to be analyzed with different parameters and different product forms to be considered without the need for many lengthy experiments. Considering the kinetics and mechanisms of enzymatic degradation of PHA, two main processes can be distinguished: adsorption and hydrolysis. The first step is the adsorption of the enzyme onto the polymer surface, and the second step is the hydrolysis of the polymer chain. In finite element modeling, adsorption can be described by surface homogeneous diffusion and hydrolysis by a reaction equation, similar to models for polylactides. To find the unknown model parameters, a nonlinear optimization problem is solved using Sparse Nonlinear OPTimizer (SNOPT) and experimental data on the biodegradation of poly(3HB) and poly(3HB-co-3HV) polymer fibers in three biological environments (human blood, blood serum, and in vivo). Based on the simulation results, numerical and experimental results were compared, and the influence of model parameters on the biodegradation profile was analyzed. The model was also applied to assess the biodegradation of other polymeric medical devices such as a vascular stent and a bone implant. The proposed finite element model allows for assessing the biodegradation of polymeric medical devices made from PHA in various biological environments. The biodegradation simulation results are in good agreement with experimental data. This model is the first step towards modeling the biodegradation of PHA and does not include the change in crystallinity, the characteristics of the enzyme during degradation, which may be a further aim of the work.

Cryptographic primitives implemented on Field-Programmable Gate Array (FPGA) have a number of advantages over their software counterparts; they are more efficient in terms of performance and energy consumption. At the same time, for implementation on FPGAs, it is important to have a compact representation of cryptographic primitive elements such as substitution tables (S-boxes). This paper presents a cryptographically strong S-box that is compact in its implementation on FPGAs. Based on the algorithm for generating bent functions using wavelet transform, an algorithm for constructing S-boxes has been developed. The nonlinearity, differential uniformity, avalanche effect, and Bit Independence Criterion (BIC) of the S-boxes generated by the developed algorithm are calculated. Diagrams have been constructed showing the distribution of S-boxes by nonlinearity and differential uniformity values for different values of the algorithm input data. A comparison was made of the cryptographic properties of the generated S-boxes and the S-boxes used in the American Advanced Encryption Standard symmetric encryption standard and the Russian Kuznechik cipher. The avalanche effect of the Streebog hash function was analyzed when using its S-box and the S-box generated by the developed algorithm, and a comparison of FPGA resource utilization was performed. The best results in terms of nonlinearity, differential uniformity, avalanche effect, and BIC are achieved by S-boxes consisting of bent functions that have no terms higher than degree two (quadratic bent functions) in algebraic normal form. In terms of cryptographic properties, the S-boxes generated by the developed algorithm are superior to or not inferior to existing ones, but they are not bijective, which narrows their field of application. The avalanche effect of the hash function with its original S-box and the S-box generated by the developed algorithm do not differ. Analysis of the results of the synthesis of hash function implementations showed that the use of the developed S-box allows saving about 1000 Look Up Tables on the FPGA. The S-boxes presented in this work are not inferior to existing ones in terms of cryptographic properties, or surpass them. The developed S-boxes are more compact than existing ones in hardware implementation on FPGAs, but they are not bijective, which makes their use most appropriate and natural in irreversible cryptographic primitives implemented on resource-limited embedded systems.

Currently, one of the most common types of heat exchangers is plate heat exchangers, including microchannel heat exchangers. To improve the efficiency of heat exchangers, optimization of the plate geometry is necessary which is achieved through numerical flow and heat transfer modeling. This requires validation of the results not only for integral characteristics such as pressure drop or heat transfer coefficient, but also for the flow structure. In this paper, we experimentally and numerically investigate the flow structure in slotted channels with a one-sided smooth expansion and contraction with a minimum (33 % of the slot height), average (53 % of the slot height), and maximum (200 % of the slot height) expansion at a Reynolds number of approximately 300. For numerical modeling, we used a steady-state three-dimensional incompressible formulation based on the Reynolds-averaged Navier-Stokes equations with their closure using the k-ε Realizable turbulence model. During the experimental studies, we employed the optical method of particle image velocimetry which consists of cross-correlation analysis of successive images of the flow seeded with tracer particles, which allows us to determine their average displacement over the time between frames, and, consequently, the two-dimensional velocity field. To implement this method, a setup was created, including the development, manufacture, and testing of a droplet generator based on a Laskin nozzle. Using numerical modeling and experiments, instantaneous and averaged velocity fields in the mid-longitudinal cross-section of the channels were obtained. The numerical modeling results show good qualitative and quantitative agreement with the experimental data, with deviations not exceeding 8 %. It was also established that, in the case of significant expansion, a vortex flow can occur in the wide section of the channel. The vortex is not stationary, despite the low Reynolds number, but moves both left and right and up and down. This, when averaged, leads to differences in the velocity vector fields between the experimental and modeled results. The models and approaches used in this study were found to provide acceptable accuracy in reproducing flow structures. Due to the presence of vortex structures at significant expansion, which can be formed by fins with significant fin heights, low- and medium-expansion channels are recommended for use in heat-exchange channels as they ensure a flow without separation.

Microreactors require intensive computational modeling to analyze complex multiphase chemical processes, but existing software solutions either have limited functionality or are expensive. To date, there is no comprehensive architectural platform that combines modeling, parameter identification, sensitivity analysis, and hydrodynamic calculations in a single software package. Such a platform is necessary for optimization and increasing the efficiency of the microreactors design process. The study uses the C4 software architectural design methodology based on which client-server architecture with a modular division into subsystems was developed. JavaScript for the client part and Python with Django REST Framework for the server part were used as a technology stack, which ensures the cross-platform and scalable nature of the implementation. A scalable architecture of a software package has been developed which is able to solve direct and inverse kinetic problems of arbitrary chemical reactions, to identify and to optimize process parameters, and to make sensitivity analysis for identification of key factors of the system. Particular attention is paid to the hydrodynamic modeling subsystem the architecture of which includes an input data parser, a generator of dynamic systems of equations, and solvers for multiphase flows and chemical reactions at the phase boundary. The modular structure ensures independent development of components and the ability to expand the functionality of the software package by introducing new or modifying existing containers. The software package was tested using a three-stage transesterification simulation for biodiesel fuel production in a T-shaped micromixer. The results of the computational study confirmed the correct operation of the hydrodynamic subsystem. The proposed architecture forms the basis for creating an open software package available to microreactor design engineers, chemistry researchers, and students of chemical engineering specialties. The use of the C4 model ensures a clear division of responsibilities between the system components and simplifies its scaling. The client-server organization allows for efficient distribution of the computational load and ensures the availability of the system for users with different hardware capabilities. The developed architecture can serve as a basis for further development of specialized software tools in the field of microreactor system design and integration with computer-aided design systems.

Algorithms and data structures are important in information technology (particularly in 6G terahertz wireless communications), logistics, transportation, and route planning in medicine. One of the fundamental algorithms that find application in many practical problems is the shortest path algorithm. The shortest path algorithm is used to find the optimal route between two vertices while minimizing the length or number of intermediate nodes. This problem is one of the 21 NP-hard problems formulated by Richard Karp in 1972. It still poses a serious challenge to science, since there is no polynomial algorithm that can provide an exact solution to this problem. The article considers approaches to solving the problem of finding the shortest path on a graph, and discusses their advantages and disadvantages. Various algorithms, such as the Bellman-Ford algorithm, Dijkstra, and the A* algorithm, are considered; their efficiency and applicability in various situations are analyzed. Issues of optimization of shortest path algorithms are discussed, such as the use of heuristics, graph preprocessing, and other methods that can improve the performance and accuracy of algorithms. The article proposes an optimization of Dijkstra’s algorithm with quantum acceleration. The stability of the quantum algorithm and the magnitude of distortions in a circuit with a different number of qubits are estimated. It is shown that with an increase in the intensity of distortions, the probability of detecting the desired element decreases, and the time spent on its search increases. When using Grover’s algorithm to optimize a path in a quantum network, these interferences cause an increase in the time it takes to find the optimal path and an increase in the probability of choosing an incorrect route. It is shown that an increase in the number of qubits and, accordingly, the number of iterations makes the algorithm more stable. It is demonstrated that the stability of Grover’s algorithm for circuits with a large number of qubits has a similar nature. The work of quantum circuits on 9, 10 and 11 qubits, the difference in the probabilities of correct operation differs by less than 10–3. Thus, from the obtained dependencies, it is possible to predict the nature of such an influence on circuits with an even larger number of qubits, the stability of which cannot be studied due to the insufficient computing power of classical computers. The most efficient way to modify Dijkstra’s algorithm is to replace the priority queue in the classical version of the algorithm with some quantum data structure that finds the minimum element for the next iteration of the algorithm. Despite the fact that quantum search does not provide a potential gain in asymptotics, another quantum modification of Dijkstra’s algorithm can be used. Namely, instead of maintaining and searching for a minimum in the priority queue, use the quantum algorithm immediately to search for a vertex that needs to be added to the shortest path tree, which will additionally acceleration the graph traversal by Dijkstra’s algorithm.

The problem of optimal control involving two opposing players is considered where optimality is understood in the minimax sense of achieving the best guaranteed outcome, and the control strategy is constructed with respect to the worst case admissible under the available measurements. The differential game problem is reduced to an optimal control synthesis problem by means of an abstract maximum principle using the method of Lagrange multipliers. A procedure is presented for applying the abstract maximum principle to the maximin formulation of the differential game problem within the Bellman framework in terms of dynamic programming. It is shown how the abstract maximum principle leads to the fundamental relations of Bellman’s optimization method for the differential game under consideration. The developed methodology for deriving optimality conditions in an antagonistic differential game using the abstract maximum principle can be applied to the analysis and design of nonlinear controlled dynamical systems with internally conflicting objectives.