Russian Journal of Cybernetics

Visualization and Configuration Subsystem for Integration Components in Process Control Systems: Architecture and Prototype

R. D. Gimranov — 2026-09-30

we developed a subsystem for visualizing and configuring integration components in automated process control systems (PCS). We compared several architectural approaches to desktop application development, including layered, clean, and hexagonal architectures, as well as the modular monolith and plug-in architectures. Based on this comparison, we selected the modular monolith architecture for the subsystem. We developed a prototype, the Integration Component Configurator, which supports visual configuration of integration components, seven component types, three interaction protocols, and several export formats. We also developed a data storage system based on PostgreSQL and an API based on FastAPI, and implemented mechanisms for inter-component integration. The results demonstrate the feasibility of the proposed architecture and prototype for configuring and integrating components in PCS.

Physics-Informed Neural Network for Modeling One-Dimensional Transient Two-Phase Flow with Capillary Pressure at Varying Injection Rates

V. A. Galkin — 2026-09-30

we developed a physics-informed neural network parameterized by injection rate for modeling two-phase flow. We trained the network using the saturation transport equation, initial and boundary conditions, and the integral water balance, without fitting it to observed curves. For validation, we used data generated by the tNavigator reservoir simulator and results obtained with an independently developed finite-volume code. We compared six integral characteristics for three injection regimes. The maximum root-mean-square error of the neural network relative to the tNavigator data, normalized by the range of the corresponding reference series, was 1.5%. At the intermediate injection rate, the neural network reproduced the integral displacement characteristics obtained with tNavigator without retraining.

Logical Foundation of Neural Networks as Categorical Splices, Axiomatics, Dualities, and Applications

G. K. Tolokonnikov — 2026-09-30

we developed a formal first-order axiomatic theory of neural networks with arbitrary topologies within classical logic. The theory is a special case of the author’s axiomatic framework for categorical systems, which generalizes and formalizes P. K. Anokhin’s theory of functional systems, developed in the 1930s and, as N. Wiener acknowledged after meeting P. K. Anokhin in Moscow in 1961, incorporating the fundamental principles of cybernetics, including feedback. We identified and described three types of dualities and proved the corresponding duality principles. We applied these results to justify and generalize S. Osowski’s well-known formulas for error backpropagation to neural networks with arbitrary topologies. A scientific theory reaches a mature stage of development when its fundamental principles can be expressed as a system of postulates and axioms. The proposed axiomatic theory provides such a formal foundation for neural networks and, through its systemic basis, contributes to the formalization of the foundations of cybernetics and artificial intelligence methods based on neural networks.

A New Approach to Determining Rainbow Scattering Cross Sections for Atoms and Ions

V. P. Koshcheev — 2026-09-30

we calculated the differential cross section for rainbow scattering of atoms and ions using first-order perturbation theory and accounting for the collective oscillations of atomic electrons and nuclei. We found that the differential cross section for atom–atom rainbow scattering is zero at zero scattering angle, whereas it remains nonzero for ion–atom scattering. We also found that, at zero scattering angle, the rainbow-scattering cross section exhibits a discrete set of values proportional to the square of the ion’s charge state.

Hybrid Approach to Compression of Multichannel Ultrasonic Array Data Using Dynamically Adjustable Bit Depth and Multiresolution Methods

A. Yu. Babkin — 2026-09-30

we studied redundancy and excessive data volume generated by multichannel ultrasonic arrays in diagnostic applications. We justified the need for a hybrid compression method that combines structural compression based on multiresolution methods with algorithmic encoding approaches, including techniques derived from MP4. We described the characteristics of the A-scan data format and outlined the parameters that affect data volume. We proposed a conceptual approach to a compression model that uses dynamically adjustable bit depth. The study is theoretical and focuses on potential directions for developing efficient compression algorithms without presenting experimental results.

Information System for Software Development Process Analysis Using DORA and SPACE Metrics

S. N. Yatsuk — 2026-09-30

we developed an information system for automated diagnostics of software development processes based on version-control data. We analyzed limitations of existing engineering analytics tools, which often focus on managerial monitoring of individual performance and may create incentives to optimize measured indicators rather than underlying processes, consistent with Goodhart’s law. We designed an architecture in which privacy protection and resistance to such metric-driven behavior are treated as requirements of the application programming interface rather than as user-interface features. Role-based access to data is enforced at the REST API level according to the defense-in-depth principle, with unauthorized requests rejected at the service layer rather than hidden only in the user interface. The methodological framework combines the five-metric DORA model with the SPACE framework for multidimensional assessment. We defined algorithms for computing lead time for changes, deployment frequency, change failure rate, decomposed merge-request cycle time, merge-request size distribution, and the Bus Factor for code modules from incrementally collected GitLab data. We implemented a minimum viable product with three role-based dashboards, paired visualization of speed and quality metrics, transparent calculation algorithms, report export, and Docker Compose deployment suitable for data-localization requirements. The system can be used to diagnose software development processes in teams working with GitLab.

Reality of Dynamic Chaos in Biomechanics

V. V. Poluhin — 2026-09-30

in the late 20th century, Nobel laureate I. R. Prigogine published his final monograph, the result of two decades of work, in which he argued that dynamical systems theory and determinism cannot be applied to the study of biological systems. At the same time, Prigogine, as well as M. GellMann and R. Penrose, accepted the existence of deterministic Lorenz chaos as a mechanism underlying biological processes. We show that this assumption is incorrect for the organization of human movement. We examined the principal criteria of Lorenz chaos and demonstrated that they are not satisfied in postural movements, i.e., involuntary human movements. At the same time, our results support N. A. Bernstein’s principle of “repetition without repetition” in biomechanics. Bernstein first raised the problem of movement uniqueness in 1947, but this line of research received little attention for more than five decades.

Parallel Numerical Simulation of Preseismic Deformation in Earthquake-Prone Areas

M. I. Gapeev — 2026-09-30

we studied the parallel numerical simulation of preseismic deformation processes in earthquake-prone areas. We developed a parallel algorithm based on a numerical scheme for solving the static elasticity problem using the finite element method. The software implementation uses the Message Passing Interface (MPI) and partitions the computational domain into subdomains, with each process solving a corresponding part of the resulting distributed system of linear algebraic equations. Regression analysis showed that the algorithm has nearly linear time complexity, amounting to O(N1.17). Its memory complexity over the range of meshes with ≈ 105 to ≈ 108 elements was almost strictly linear, at O(N0.985). Statistical analysis of parallel efficiency showed that the optimal number of computational processes was between 6 and 8. Under these conditions, the computation time was approximately 5 min, and the speedup was about 2.9 for a mesh containing 18·107 elements. Parallel efficiency decreased monotonically and fell below 0.3 when more than 9 processes were used. The cost increased almost linearly with the number of processes. The algorithm was therefore not cost-optimal over the considered ranges of mesh sizes and numbers of computational processes.

Numerical Simulation of Steady Thermocapillary Flow in a Flat Channel Using Linear Approximation

K. V. Gubareva — 2026-09-30

we performed a numerical study of steady thermocapillary flow of a viscous incompressible fluid in a flat channel with a stationary solid bottom and a free surface. We considered the creeping-flow regime, in which the only driving force was temperature-dependent surface tension. We used a quasi-one-dimensional expansion of the velocity, temperature, and pressure fields in the transverse coordinate, retaining terms up to second order in the longitudinal coordinate. We solved the resulting boundary-value problem for a system of ordinary differential equations using a fourth-order collocation method implemented in MATLAB with the bvp4c solver. We applied the computational algorithm for five values of the Marangoni number. We obtained profiles of the vertical and horizontal velocity components and the modal temperature and pressure, and constructed two-dimensional distributions of the stream function, pressure, and temperature. We found that two counter-flowing streams formed over the range of parameters considered, separated by a level of zero horizontal velocity, and determined the positions of the corresponding zeros. Analysis of the residuals of the governing equations confirmed the high accuracy of the numerical solution. We investigated the dependence of the flow structure on the governing parameter. The results provide a benchmark for testing numerical schemes for thermocapillary convection.

Simulation of a Virtual IoT System for Remote Engine Monitoring and Control Using Temperature Hysteresis and Manual Control

E. M. Ismailov — 2026-09-30

we conducted a virtual experiment using the educational versions of nanoCAD and Tinkercad to study electric drive control systems for a ventilation system. We developed and tested simplified virtual models of direct motor starting, remote starting through a controller, and remote starting with automatic activation based on a temperature sensor signal as an example of an IoT-based control system. We compared these control architectures to trace the transition from relay switching, direct-start circuits to software- and hardware-based systems capable of storing data in cloud services. We focused on the interaction logic between system components and evaluated the use of modern development and simulation tools for virtual testing. The results provide a basis for developing digital twins of industrial systems and for their subsequent integration into artificial intelligence of things (AIoT) applications.

Lossless Compression Algorithm for Medical Image Sequences with Selective Inter-Frame Redundancy Reduction

D. V. Krayushkin — 2026-09-30

the increasing number and resolution of computed tomography examinations place a growing burden on medical image archives, while lossless compression is required for diagnostic images. The lossless compression schemes specified by the medical image interchange standard are intra-frame and do not exploit correlations between adjacent slices. Existing inter-frame methods, in turn, typically select a prediction mode for an entire frame and therefore do not account for the distribution of homogeneous and edge regions within the frame. We developed a lossless coding algorithm for medical image series that extends JPEG-LS with selective inter-frame decorrelation of prediction errors. We partitioned each frame into non-overlapping 8 × 8-pixel blocks and identified blocks containing anatomical boundaries using the Canny edge detector. For each selected block, we searched the preceding frame for a spatially corresponding block using normalized cross-correlation. We then encoded the difference between the prediction errors of the current and reference blocks, thereby reducing the variance of the values supplied to the entropy coder. Experiments on 18 computed tomography series comprising 6,213 frames showed mean compression-ratio gains of 18.44% over JPEG-LS and 22.81% over JPEG2000.

Interpolation of Normals Along a 3D Curve

A. L. Legotin — 2026-09-30

we solved the problem of interpolating the normal along the edge of a curvilinear and discusses other related problems.

Computing Cluster Load Prediction from Time-Series Data

D. A. Mikhalevich — 2026-09-30

we studied short-term computing cluster load prediction using the Alibaba GPU Trace 2020 production trace. We considered hourly processor, graphics accelerator, and random-access memory (RAM) loads, with a one-hour prediction horizon and a 168-hour test window. We aggregated task events into hourly time series by averaging the concurrent load over time. We compared five prediction models: a naive last-value model, an integrated moving-average model, a long short-term memory (LSTM) recurrent neural network, a Bayesian dynamic linear model, and a multivariate vector autoregression model with lag selection up to 168 hours. We evaluated prediction accuracy, 99% prediction-interval coverage, accuracy in predicting the direction of load changes, and the potential savings in reserved resources. The Bayesian dynamic linear model achieved the lowest errors, with mean absolute errors 6–9% lower than those of the naive model. The LSTM model was competitive for RAM, whereas the multivariate model performed well for processor and accelerator loads. Replacing reservation based on the historical maximum with reservation based on the upper bound of the prediction interval reduced the reserve by 22–30% while maintaining 96–99% coverage.

Structural Integrity of the TUK-27/1 Packaging System in Simulated Transportation Accidents Involving Uranium Hexafluoride Traps

K. K. Novichkov — 2026-09-30

we assessed the structural integrity and leak tightness of a new protective case for uranium hexafluoride traps used in the TUK-27/1 transport package under dynamic loads simulating the accident conditions specified for Type B packages by NP-053-16. The study forms part of the overall safety assessment of the TUK-27/1 for transporting uranium hexafluoride traps. Results from full-scale tests of the TUK-27 transport package conducted by an interdepartmental commission were used as a basis for assessing the structural integrity of the new protective case. We performed a stress-strain analysis of its metal frame by solving the equations of continuum mechanics in Lagrangian coordinates. The equations were integrated using an explicit time-stepping finite-element method implemented in the certified ANSYS LS-DYNA software package. The physical and mechanical properties of the case materials were conservatively taken from PNAE G-7-002-86. The material models used a bilinear approximation of the stress-strain curves for the respective metals. The calculations demonstrated that the new protective cases retain their structural integrity and safety under the specified accident loads during transport of the TUK-27/1 transport package in accordance with the applicable Russian regulations.

An Epsilon-Net-Based Algorithm for Minimizing Functionals on Complexes

A. N. Rozhkov — 2026-09-30

we developed an ε-net algorithm for minimizing functionals defined on complexes with a prescribed rank structure. A major challenge in analyzing highly detailed complexes is the rapid growth in computational complexity as the state space increases. To address this problem, we proposed a method for partitioning the state space into ε-net classes and replacing the original complex with a coarsened graph. We grouped rank-zero elements into classes according to a specified metric and constructed classes of higher-rank elements from the classes of their constituent elements. We used lexicographic optimization of the objective functional: we first minimized the number of transitions between ε-classes and then minimized the route length in the original complex. We demonstrated the algorithm on a polyhedral complex representing an actual building and applied it to routing problems. The model included the building’s nodes, edges, faces, and three-dimensional cells. Our software implementation supports the construction of the original and factorized graphs, the computation of conventional, ε-class-based, and lexicographically optimal routes, and their visual comparison. The proposed approach can also be applied to other functionals defined on complexes, provided that a suitable proximity measure and rules for forming classes are specified.

Comparison of Audio Signal Encoding Methods for Classification with Multilayer Spiking Neural Networks

R. B. Rybka — 2026-09-30

we compared three computationally simple methods for encoding short speech recordings as spike sequences for real-time classification with a multilayer spiking neural network. The methods included integral coding of spectral features, stream spectral encoding, and direct encoding of local signal dynamics. All three methods used the same network of Leaky Integrate-and-Fire neurons trained by surrogate-gradient backpropagation. Experiments on the Google Speech Commands v0.01 (GSC) and Heidelberg Digits (HD) datasets showed that integral rate coding achieved the highest accuracy on GSC (83.1%), whereas all three methods performed comparably on HD. Stream spectral encoding required only four simulation time steps per spectrogram frame, substantially reducing response latency. Speech2Spikes achieved 80.1% accuracy on GSC, only 3% below integral rate coding, while using a more compact input representation. The results provide a basis for selecting spike encoding methods under different computational and latency constraints.

Typed Canonical Pattern Dictionary for Static Comparison of Structured Text PLC Programs

S. E. Shchigal — 2026-09-30

we investigated a method for comparing numerical-control (NC) programs that reduces the sensitivity of literal comparison to variable names while retaining syntactic and semantic information relevant to program structure. We formalized a dictionary of typed canonical Structured Text patterns and evaluated its use for static program comparison. An industrial ventilation project comprising 189 program units was used to construct the dictionary. Statements were converted into signatures that retained statement types, selected expression structures, variable roles and data types, and literal classes. Frequent signatures were included as individual patterns, whereas rare signatures were grouped into generic categories. After correcting syntactic analysis errors, the dataset contained 7,794 statement occurrences. At a pattern depth of one and a minimum frequency of five, the dictionary contained 238 patterns covering 54.0% of all occurrences. Replacing lexical identifiers in 188 modified program units did not change either the full or thresholded pattern sequences. Across six procedures, comparison distances were zero for renamed variants and ranged from one to three for controlled modifications. In an external archive containing 110 program variants, full representations differed from the modal reference in 65 cases at depth one and 75 cases at depth two. The method does not establish functional correctness. Its limitations include loss of variable identity, coarse representation of literals, incomplete syntax coverage, and the absence of independent expert validation of the generated programs.

Calculation of the Total Energy of a Carbon Molecule Using First-Order Perturbation Theory and Machine Learning

Yu. N. Shtanov — 2026-09-30

we developed a machine-learning approach for calculating the total energy of a carbon molecule. We compared its performance with numerical integration and two established machine-learning methods: a Kolmogorov–Arnold network and a random forest. We evaluated the accuracy of the methods using the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE). We compared the methods in terms of accuracy, computational efficiency, and applicability to different calculation tasks.

Simulation of Transient Processes in Pneumohydraulic Suspensions Using a Coupled Hydrodynamic and Thermodynamic Model

A. V. Titova — 2026-09-30

we developed a simulation model and performed a numerical simulation of transient processes in pneumohydraulic devices used in vehicle suspension systems. The model coupled a zerodimensional thermodynamic model of the gas chamber and the piston equation of motion with axisymmetric unsteady Navier–Stokes equations for a viscous incompressible fluid filling the working chambers. We solved the hydrodynamic equations using the finite volume method with an implicit time-stepping scheme and treated the pressure–velocity coupling using an implicit operator-splitting procedure. We validated the model against experimental data and the results of a zero-dimensional model. We then performed a parametric study to assess the effects of the throttle-orifice radius, initial gas-chamber volume, and fluid viscosity and temperature on the transient response to step loading. We found that the throttle-orifice radius is the main damping parameter: small orifices produce an aperiodic response with a short-lived pressure peak similar to a local hydraulic shock, whereas large orifices produce weakly damped oscillations. The initial gas volume determines the effective stiffness of the gas spring and the final piston displacement, while changes in fluid temperature over the operating range alter the device’s stiffness–damping characteristics because of changes in fluid viscosity spanning several orders of magnitude. Based on these results, we formulated practical recommendations for selecting throttle-unit parameters and working fluids for suspension-system design.

A Review of Neural Network Accuracy Improvement Using Averaged Outputs

T. S. Akobyan — 2026-09-30

we studied the application of Richardson extrapolation to improving the accuracy of neural network computations. We analyzed the mathematical framework of the Richardson method, including the derivation of the relevant formulas, asymptotic error expansion, recursive extrapolation scheme, and convergence estimates. We developed a mathematical model of a neural network, formalized the main sources of computational error, and implemented an extrapolation-based procedure for reducing these errors. We then conducted a series of computational experiments on test problems involving approximation, numerical integration, and classification in the presence of noise. The results showed that Richardson extrapolation improved computational accuracy by a factor of 3.5 compared with conventional approaches at the same computational cost and increased the convergence order. We also compared Richardson extrapolation with alternative accuracy-enhancement methods. The comparison showed that Richardson extrapolation provided a more favorable trade-off between accuracy and computational cost.