Siberian Aerospace Journal

The aerospace industry often deals with complex black-box problems, where neural networks offer a potential solution. However, a major challenge with this approach is finding the optimal neural network architecture. Designing an overly large network wastes computational resources and requires significant training time or more powerful hardware. A more efficient solution is to design a compact network that performs the task effectively. The field of neural evolution addresses this by using evolutionary algorithms to automatically design network structures through natural selection. Genetic Programming is one such algorithm; it builds functions as trees and then transforms them into neural networks. These trees contain functional nodes with various operators. Currently, the standard encoding method uses only two operators: one adds hidden layers, and the other increases the number of neurons in a layer. These operators cannot create recurrent connections, which could enhance solution diversity and model performance. Our research proposes a new unary functional operator that generates recurrent connections by analyzing neurons in the hidden layer. We studied the effectiveness of this approach on several tasks from the Feynman Symbolic Regression Database and compare the results with previous versions of the algorithm. The study yields the following key findings: on average, the new algorithm performs worse than its predecessors. However, in several independent runs, it discovers better solutions to the given problems. We propose several hypotheses to explain the low average effectiveness of this recurrent connection mechanism, which aims to guide future improvements to the algorithm.

Fuzzy logic systems are widely used in classification problems due to their ability to handle uncertainty, imprecision, and subjectivity in data. Unlike traditional “crisp” methods, fuzzy logic allows input features and output classes to be described in terms of linguistic variables – such as “high”, “medium”, and “low” – making models more interpretable and aligned with human reasoning.

A key limitation of this approach is the need to construct a rule base using expert knowledge, as well as the ambiguity in selecting appropriate membership function shapes. To address these challenges, various optimization algorithms have been developed. One such method is genetic programming, which purpose is to evolve a rule base capable of accurately capturing underlying patterns for correct classification while preserving interpretability through structural adaptation. This article explores the theoretical framework of self-configuring genetic programming combined with differential evolution for constructing fuzzy logic rule bases. It also presents their practical implementation and application to classification tasks.

The selection mechanism in genetic programming has attracted attention of many researchers, as a result lexicase selection and its variants have been proposed. However, the replacement schemes are often out of the scope of most studies on genetic programming. In this paper, the effects of different selection mechanisms are studied, when they are applied to both parents’ selection and replacement. The aim of the study is to evaluate the efficiency of different parent and environment selection schemes in genetic programming. For this in the paper the four-parent selection, seven replacement are compared and two methods for selecting first parent in a pair, giving a total of 56 combinations. The experiments are performed on a set of 10 regression problems from the PMLB dataset collection, and the determination coefficient is used as a performance metric. The Mann-Whitney statistical tests and Friedman ranking is used for the analysis of the performance values. The analysis of the results show that the down-sampled lexicase selection can is an efficient mechanism, selecting current individual as one of the parents could be beneficial, and that either pair-wise comparison or selection of best of parents and offspring is an efficient replacement strategy. The results obtained in this study can be applied to improve the efficiency of solving symbolic regression problems, which often arise in the aerospace industry, for example for predicting solar panels degradation parameters, or antenna radiation patterns.

This article deals with the analysis of the deformed shape of local stability loss of a reinforced flexible beam, occurring due to constrained expansion during heating. The multiple elastic supports of the beam modeled as an elastic medium, which provides constant resistance to both longitudinal and transverse displacements of the rod. The infinitely long beam divided into a buckling region and an adjacent region under compression. The lengths of these regions are unknown and should be determined during the solution process. A part of the potential energy accumulated during compression is expended on the work of internal forces during bending deformation following the loss of stability. This leads to a reduction in the magnitude of the compressive force in the buckled region. The problem of determining the displacement functions and the critical value of the safe heating temperature is formulated as a system of nonlinear differential equations concerning the deformations in the regions of the flexible beam. The solution obtained using the finite difference method, which transforms a system of differential equations into a system of linear algebraic equations. This system takes the closed form with boundary conditions and transversality conditions. A sufficient number of grid nodes for constructing the difference scheme determined through an iterative procedure that compares two adjacent solutions. The criterion for comparison of solutions is a tuple of areas under the graphs of the sought functions, which are calculated through numerical integration using the trapezoidal rule. The obtained final solution compared with the classical solution to the stability problem of an evenly loaded beam, which does not take into account longitudinal displacements. Additionally, it is contrasted with the known solution in the field of operation of continuously welded railway tracks, which also disregards resistance to longitudinal displacements in the buckled region. The refined results obtained by the proposed modified method for calculation of the parameters of the deformed shape of a flexible beam is important for monitoring the pre-critical state of the modelling system.

This study analyzes the design features of small, aircraft-type unmanned aerial vehicles. The first stage of the study examined the chronology of the evolution of the concept of “small”. It was shown that this concept should be understood as a combination of two factors: small size and low flight speeds. Because of this, the boundary layer flow occurs in a laminar-turbulent transition regime. Due to boundary layer instability, a “laminar bubble” forms above the wing, reducing the lift properties of the vehicle. Reynolds numbers in the range of Re = 10⁵…10⁶ were proposed as an aerodynamic criterion for smallness. At speeds of 70–150 km/h, the mean aerodynamic chord size is in the range of b_average = 0.1–0.9 m. The second stage analyzed the influence of the “square-cube” law on the energy efficiency of the vehicle during the miniaturization of its design. According to this law, the specific wing loading and the (weight)1.5/power ratio are reduced by the magnitude of the reduction factor for the linear dimensions of the aircraft. This is interpreted in the article as a reduction in the efficiency of the aircraft, which converts the kinetic energy of the airflow into lift. When transporting a unit of weight over a unit of distance, a small aircraft consumes several times more fuel than a full-size aircraft. A potential way to increase energy efficiency is to reduce or eliminate the trim losses of lift. This reduction is achieved by reducing the area of the rear-mounted horizontal tail, and the elimination of this loss is achieved by replacing the standard aerodynamic configuration with alternatives: canard or tailless configurations. These configurations do not use a rear-mounted horizontal tail, which generates negative lift, requiring additional energy to compensate. At the 3rd stage of the study, recommendations were formulated for the developers of the considered class of devices: 1) small-sized devices have a naturally low aerodynamic and energy efficiency, increasing which to the level of full-size devices is practically impossible; 2) improving aerodynamics is possible through the use of special profiles, morphing wing technology and boundary layer control methods; 3) increasing the duration and range of flight is possible due to the reduction of balancing losses of lift, using a beam tail in a normal configuration or using alternative aerodynamic configurations; 4) the article presents a block diagram of the iterative process for determining the wing area, the value of which will satisfy the conditions of the balance of forces and moments for various values of the ratio of the wing and horizontal tail areas.

Increasing the efficiency of liquid rocket engines (LRE) is inextricably linked to the intensification of cooling processes in heat-critical elements such as combustion chambers and nozzles. Additive Manufacturing (AM) technologies, in particular selective laser fusion (SLM), make it possible to create cooling paths with complex internal channels and heat transfer intensifiers. However, the surface microrelief formed in this case has a significant effect on hydrodynamic resistance and heat transfer, which is not always taken into account in existing design methods.

An experimental assessment of the effect of surface roughness, characteristic of additive manufacturing, on the formation of a dynamic boundary layer under conditions simulating flows in the cooling channels of liquid propellant is considered.

The sample plates made of AlSi10Mg aluminum alloy were made using the SLM method: with controlled roughness and flow intensifiers. The surface roughness profile was evaluated on a TR 110 profilometer. The research was carried out on a specially designed aerodynamic installation that simulates the flow in a rectangular channel. The velocity fields in the wall area were measured by the pneumometric method at a fixed air flow rate. The errors were estimated and the data approximated.

It is established that the roughness of SLM surfaces leads to a significant restructuring of the velocity profile in the boundary layer compared to the hydraulically smooth case: a decrease in its thickness and a change in the shape of the velocity diagram are observed. The largest deviations were recorded for the plate with intensifiers. Power-law approximating dependences are obtained for each type of surface, where the exponent varies depending on the roughness parameters.

It is shown that the use of boundary layer models for smooth surfaces in liquid propellant calculations in relation to additive parts can lead to incorrect estimation of hydraulic losses and, as a result, to errors in predicting the temperature state of the wall. The obtained dependences are necessary for the development of adjusted calculation methods that take into account the actual topography of the surfaces formed by AM methods, which will increase the reliability and efficiency of the design of liquid propellant cooling systems.

During the assessment of rocket components vibroacoustic loading using the finite element modeling, it is not feasible to analyze the entire rocket due to the design's complexity and its large spatial scales compared to high-frequency wavelengths. Therefore, in practice, individual structural groups (compartments, rocket fuel tanks, and the warhead) must be considered. This limited model for calculations under rigid constraint can produce distorted vibroacoustic parameter characteristics; these errors will be greater the shorter the considered compartment. To address this issue, it is proposed to use beam finite elements for modeling the missing components of the rocket. This should allow to consider the missing vibration modes propagating along its entire body and affecting the compartment under consideration, as well as the dynamics of its end frames. In this regard, in the proposed article, a finite element model of the stiffened shell of a launch vehicle's dry compartment was developed. The model is constructed from flat and solid finite elements, with brackets containing instruments positioned within it. These brackets represent concentrated masses with specified centering and inertial characteristics. Such structural elements, described by a system of oscillator equations, interact with acoustic (fluid) elements in the model, where the acoustic wave equations are solved. The missing parts of the launch vehicle were modeled using beam elements, considering the existing design of tunnel pipelines, fuel tanks, propulsion systems, etc. The rocket's beam structure is connected to the compartment frames by elastic contact elements. Solving the resulting systems of equations allowed us to determine the displacements of the structural finite elements and pressure pulsations in the acoustic domain. The calculations are presented under various boundary conditions, which take into account low-frequency vibrations that occur during engine operation and their influence on the internal acoustic pressure and the dynamics of the structure, with the beam part of the model and without it.

The study aims to analyze the influence of structural changes in thylakoid membranes, induced by foliar fertilization, on energy transfer in the photosynthetic system using fluorescence spectroscopy methods.

Control and experimental plant groups were established for the research. Structural analysis was performed using transmission electron microscopy (TEM), which provided images of ultra-thin leaf sections followed by quantitative assessment of the morphometric parameters of chloroplast grana. Micrograph processing and statistical analysis were carried out using artificial intelligence-based software (GRANA). The functional state of the photosystems was evaluated based on chlorophyll fluorescence spectra were obtained using a spectrofluorometer in a 90-degree geometry mode upon excitation with light at a wavelength of 425 nm.

Results. It was found that the application of foliar fertilization leads to a statistically significant structural reorganization of the thylakoid membranes. The experimental samples showed a reduction in the median grana area by ~ 38 % and their perimeter by ~ 16 %, indicating denser and more ordered packing. Simultaneously, fluorescence spectroscopy revealed a narrowing of the peak half-width corresponding to photosystem II by 1.8 nm (8%) in the treated plants. This change indicates a reduction in the energetic heterogeneity of the antenna complexes and an increase in the efficiency of excitation energy transfer.

Conclusion. Based on the comprehensive studies, it was concluded that the primary mechanism of action of foliar fertilizers is the optimization of the thylakoid membrane nanostructure, which, within the concept of photonic crystals, creates more favorable conditions for light energy conversion. An increase in the regularity of the ultrastructural organization was demonstrated. The obtained results confirm the potential of using fluorescence spectroscopy as a rapid, non-invasive method for diagnosing the physiological state of plants in closed space systems.