This study uses computational fluid dynamics (CFD) with the k-ε turbulence model and an artificial neural network (ANN) to analyze cyclone flow. The results show that the pressure drop rises from 0.90 kPa to 6.54 kPa for inlet velocities of 7 m/s to 20 m/s. The ANN predicts the pressure drop with a 4.3 % error. The CFD-ANN approach improves insight into cyclone design.
This study introduces a hybrid optimization framework for the multi-drop testing of a lithium-ion battery enclosure. The framework integrates a Quantum-Inspired Evolutionary Algorithm (QEA) with surrogate modeling techniques. Three types of metamodels were applied—Artificial Neural Networks (ANN), Kriging, and Polynomial Regression (PNR)—using datasets generated via Latin Hypercube Sampling and from prior QEA iterations. The hyperparameters tuning methods are the main part of the paper. Two fitness functions were analyzed, including a logarithmically scaled variant designed to compress the output range for damaged cases and enhance classification accuracy near the damage/no-damage boundary. A dual-model strategy was employed for ANN with model switching determined by a plastic strain threshold. Across datasets, ANN more consistently identified superior individuals compared to Kriging, while PNR occasionally exhibited instability. Two hybrid schemes were implemented: HYBRID1, which enforces finite element model re-evaluation of the best candidate in each iteration, resulting in the lowest minima but with increased variability; and HYBRID2, which minimizes mandatory Finite Element Method (FEM) evaluations and retraining cycles, thereby improving runtime and stability at a slight cost to solution quality. Overall, the combination of QEA with ANN and the proper objective function reduced FEM computational time by an order of magnitude while maintaining decision-making effectiveness, supporting its feasibility for application in industrial design workflows.
Topology optimization is a valuable tool in engineering, facilitating the design of optimized structures. However, topological changes often require a remeshing step, which can become challenging. In this work, we propose an isogeometric approach to topology optimization driven by topological derivatives. The combination of a level-set method together with an immersed isogeometric framework allows seamless geometry updates without the necessity of remeshing. At the same time, topological derivatives provide topological modifications without the need to define initial holes [7]. We investigate the influence of higher-degree basis functions in both the level-set representation and the approximation of the solution. Two numerical examples demonstrate the proposed approach, showing that employing higher-degree basis functions for approximating the solution improves accuracy, while linear basis functions remain sufficient for the level-set function representation.
This paper presents a novel framework for goal-oriented optimal static sensor placement and dynamic sensor steering in PDE-constrained inverse problems, utilizing a Bayesian approach accelerated by low-rank approximations. The framework is applied to airborne contaminant tracking, extending recent dynamic sensor steering methods to complex geometries for computational efficiency. A C-optimal design criterion is employed to strategically place sensors, minimizing uncertainty in predictions. Numerical experiments validate the approach's effectiveness for source identification and monitoring, highlighting its potential for real-time decision-making in crisis management scenarios.
This article presents a comparative analysis of AR technologies: Vuforia, Immerse, Multiset, and Geospatial API, in terms of performance, accuracy, and interference tolerance in indoor and outdoor positioning and navigation systems. Two test environments were performed: one was an indoor (laboratory) environment enabling detailed module testing, and the other was a hybrid environment implemented on the CUT campus to illustrate the feasibility of implementing virtual presence in AR and navigation in diverse environmental conditions. The research was conducted according to six scenarios. One involved outdoor GPS navigation, while the others concerned indoor navigation. Based on the research, recommendations were made for the use of the considered AR environments for mixed navigation. As a part of the detailed testing, an AR navigation system was implemented on the CUT campus as a combination of indoor and outdoor approaches. The final implementation was developed in the Unity environment. Software tests were carried out with particular emphasis on transitions between internal and external navigation.
Earth’s natural resources are finite, which is why engineers and scientists are increasingly directing their attention to the extraction of materials from celestial bodies. Beyond the extraction itself, however, a major challenge remains the localization of valuable substances and the assessment of their quality in situ. In this article, we present a flexible and robust method for estimating the content of selected components in heterogeneous mixtures using RGB image processing. The proposed deep learning architecture achieves high prediction accuracy with root mean squared error (RMSE) of (0.190 ±0.024) %. The framework supports a variety of backbone architectures, including lightweight models, making it suitable for deployment on edge devices such as planetary rovers. Furthermore, the method is flexible, allowing for easy adaptation to other tasks, for example, the analysis of more complex mixtures or inference based on multi- or hyperspectral imagery.
In two-moment radiation transport, the closure is the constitutive relation that maps the energy and momentum to the radiation-pressure tensor. Among available closures, the maximum-entropy (ME) approach is the most reliable. However, it is associated with a high computational cost. In this paper, we propose a machine-learning approach for the rapid evaluation of the ME closure for bosonic, classical and fermionic radiation. We generate ME reference data using Gauss–Legendre angular quadrature combined with a robust bisection-based inversion of the moment constraints. Next, we train a small physics-constrained multilayer perceptron (MLP) with output restricted to the physically admissible range (between one third and one). Monotonicity in the reduced flux is enforced, and the derivative is matched to the ME reference. The neural network (NN)-based closure achieves a mean absolute error (MAE) of 9.0×10−4 over the range ϕ ∈ [0, 0.98], which yields a latency reduction of about ∼103× per closure evaluation. In the Marshak wave benchmark the full simulation runs about 247 × faster while the hyperbolicity indicator remains strictly positive (minimum 5.1 · 10−2). Compared with the analytic Kershaw closure for bosonic and classical radiation, our model is substantially more accurate and faster. For practical adoption, we also provide a lightweight rational approximation (MAE 1.01× ∼10−3), and in the bosonic and classical cases we confirm positivity of the hyperbolicity indicator for degeneracy parameters between −5 and 5.
The present paper discusses mathematical barriers in the development of software for preprocessing of atomistic models of dislocation networks. As a matter of fact, as yet, there are neither analytical nor numerical methods nor programs available which can be used for atomistic reconstruction of complex dislocation networks. Some of the problems to overcome are discussed in this paper. In the previous papers discussed below it was shown that a direct superposition of analytic formulae for displacements of atoms induced by single dislocations does not give possibility to hold the essential geometric properties of the resultant atomistic models. Namely, after the input of first dislocation, the lattice symmetry required to input the next dislocations is usually broken. These inaccuracies compose the mathematical barrier for atomistic reconstruction of advanced dislocation nets. A method developed here has been applied to reconstruction of the dislocation nodes localized in the copper/shaffire interface. In the present case, the partial dislocations are inserted by slips. For comparison, the junction corresponding to the stacking faults obtained by the rigid shifts of copper on the Burgers vector $\frac 1 6 \langle 112 \rangle$ are discussed.
Advances in high-content microscopy and artificial intelligence (AI) are transforming the quantitative study of infection biology. Automated imaging platforms now enable rapid, large-scale acquisition of host-pathogen interactions across thousands of cells and multiple experimental conditions. When combined with AI-based segmentation, these workflows extract infection-relevant features such as pathogen load, intracellular localization, and host response markers at single-cell resolution. Deep-learning models have proven especially powerful, outperforming classical threshold-based methods under different imaging conditions, reducing reliance on manual annotation, and detecting rare infection outcomes. Beyond robust image analysis, these approaches generate scalable and reproducible datasets that can be integrated with computational modelling and systems biology, providing predictive insight into infection dynamics. This review highlights recent progress in AI-assisted microscopy for bacterial infection and outlines future directions toward multimodal integration, clinical translation, and open-source tool development.
An equivalent viscous damping ratio is introduced to characterize the energy dissipation caused by the plastic impact in the periodic responses of piecewise linear elastic systems.
This study focuses on a numerical analysis of heat transfer in biological tissue. The proposed model is formulated using the Pennes equation under transient conditions within a two-dimensional (2D) cylindrical domain. The tissue undergoes laser irradiation, with internal heat sources determined based on the Beer–Lambert law. Moreover, key parameters, including the perfusion rate and effective scattering coefficient, are modeled as functions dependent on tissue damage. Numerical computations are performed using the finite pointset method (FPM). The findings, discussed in the final section, indicate that the FPM approach is a viable and effective tool for analyzing thermal processes in biological tissues.
A systematic approach to the macroscopic damage analysis of bone-like cellular materials is presented in which damage conditions are expressed as tabularized functions of microstructure geometry parameters. Based on three different strain-based microscopic damage criteria, a large number of cellular microstructures, characterized by different values of geometric parameters, are analysed by the finite element method to determine damage factor values for a number of macroscopic strain states. As a result, an exhaustive database is prepared in which macroscopic damage conditions for a variety of microstructures are presented as tabularized parametric functions of both geometric parameters and strain states. A numerical procedure of data interpolation is proposed as a tool to predict parameterized damage surfaces for any bone-like microstructure. The results are made publicly available in an open data repository to enable further research on their characterization and analytical approximation.
In this article, a numerical analysis of thermal processes occurring in biological tissue during laser irradiation is presented. The mathematical model is based on the two-dimensional Pennes equation. The analyzed tissue is exposed to the laser irradiation of a moving beam with constant velocity along the tissue surface. The upper face of the skin tissue is subjected to the vertical laser beam, and it is assumed that heat dissipation through convection and radiation from the surface is negligible compared to the heat delivered by the laser beam. Thus, the surface is treated as thermally insulated surface. The effect of the laser beam’s transitional speed and power on the temperature distribution within the skin tissue are investigated. Moreover, the perfusion rate and the effective scattering coefficient are treated as variables dependent on tissue damage. In the computational part of this study, the finite pointset method (FPM) is applied. The temperature distribution computed with FPM is compared with an analytical one obtained for a three-dimensional problem by analyzing a relevant cross-section under the same conditions. This modeling of the dynamic thermal processes within biological tissue subjected to laser irradiation supports the evaluation of biological tissue damage and provides a basis for determining the time and intensity of laser irradiation. In the last part of the article, numerical examples and conclusions are presented.
Large language models (LLMs) excel at various natural language tasks, even those beyond their explicit training. Fine-tuning these models on smaller datasets enhances their performance for specific tasks but it can also lead to risk of training data memorization, raising privacy concerns. This study explores the extraction of private training data from fine-tuned LLMs through a series of experiments. The focus is on assessing the ease of data extraction using various techniques and examining how factors such as the size of training data, number of epochs, training sample length and content, and fine-tuning parameters influence this process. Our results indicate that data extraction is relatively straightforward with direct model access, especially when training loss is computed over entire prompts. Models with higher precision (8-bit and 16-bit) demonstrate increased memorization capabilities compared to 4-bit quantized models. Even without direct access, insights into training data can be obtained by comparing output probability scores across multiple queries. Furthermore, the study also reveals that the proportion of extractable data increases with training dataset size, given a fixed number of epochs. These findings highlight the privacy risks faced by individuals whose data is used in fine-tuning, as well as for organizations deploying fine-tuned models in public applications.
This publication presents the results of a study on text similarity between Belarusian and Ukrainian, utilizing a matrix-based analysis method grounded in edit distance. A distinctive feature of this approach is the absence of language-specific vocabulary rules, highlighting the algorithm’s linguistic universality in similarity analysis. The analyzed texts were sourced from excerpts of online encyclopedias, translated using AI-powered online translation services provided by well-known companies. The primary objective of this study is to determine whether it is possible to compare texts written in these languages without prior translation into a common language. Additionally, it aims to assess whether a method that does not belong to the large language model (LLM) family or the broader category of AI-based approaches can effectively compare languages within the same linguistic group. Furthermore, the study provides insights into the degree of similarity between Belarusian and Ukrainian, investigating the extent to which speakers of one language might partially understand the other.
In the era of Industry 4.0, one of the key challenges facing underground mines is the real-time tracking of both the production process and machinery movements. Significant emphasis is placed on comprehensive monitoring to achieve situational awareness to ensure informational continuity of operations in dispersed organizations. This knowledge is fundamental for safe and efficient extraction, current production reconciliation, and all operational and planning activities, particularly when considering specialized simulation environments for production optimization. So far, implementations of such solutions on an industrial scale have primarily been encountered in open-pit mines or smaller underground mines. This article presents a solution for machine monitoring and tracking based on data from a collision avoidance system, specifically designed for multi-site underground mining enterprises, where the scale of implementation is incomparably more challenging. This anti-collision system was originally designed for detecting machine-to-machine or machine-to-worker collisions. Consequently, the development of validation algorithms, including error correction and adaptive filtering, was imperative. This also required integration with enterprise resource planning (ERP) systems. Moreover, it was also essential to enhance the system infrastructure with additional sensors to enable the registration of machine localization in specified mining zones (e.g., heavy machinery chamber, mining area, loading and unloading point). As part of this study, several analytical models (enhanced by machine learning techniques) were developed to identify movement patterns and cooperation among wheeled transport machinery, as well as the entire course of ore logistics within the mining area. Finally, the process of implementing the system in the target environment is presented, along with a description of the user interface, which features manager dashboards for production visualization.
This study investigates a self-referencing method for damage detection and localization using guided waves (GW) sensed by fiber Bragg grating (FBG) sensors. The research integrates advanced numerical simulations with an innovative configuration of sensors to enhance structural health monitoring (SHM). A self-referencing setup, employing FBG sensors with edge filtering method and remote bonding, enables a baseline-free damage detection approach. The methodology is validated as a proof-of-concept numerical model. The simulation framework incorporates a three-dimensional spectral element method for precise and efficient modelling of GW propagation and interactions with structural anomalies. Three different machine learning (ML) techniques are employed to detect and localize damages, demonstrating effectiveness of ML methods compared to traditional methods. The three techniques employed are decision tree, logistic model tree and random forest. Key findings highlight the effectiveness of random forest models in classifying damage states with a 98.67% accuracy. Different feature selection methods, are used to identify critical features. The proposed methodology reduces sensor requirements, lowers system complexity and cost, and enables efficient SHM solutions in extreme or large-scale environments. This work underscores the potential of ML techniques to perform detection and localization where traditional techniques fail.
This research primarily focuses on evaluating the effectiveness of various methodologies for the topological and geometrical optimization of steel building structures through parametric descriptions. The study specifically addresses steel trusses, frames, and beams, emphasizing their integration within the broader structural system. Initial investigations have highlighted the benefits of an innovative pattern-based approach that segments the structure into distinct patterns, namely groups of structural elements subjected to localized optimization. This method effectively overcome the challenges of global parameter optimization, by providing enhanced control over local criteria and enabling a more detailed assessment of each pattern’s contribution to global optimization objectives. Building on these insights, the research seeks to advance and refine the concept of patterns, aiming to further enhance their applicability and efficiency in structural optimization.
This paper describes the application of particle swarm optimization (PSO) for the hyperparameter optimization problem of multi-layered perceptron (MLP) model. Several PSO algorithms are presented by many researchers; basic PSO, PSO with inertia weight (PSO-w), PSO with constriction factor (PSO-cf), local PSO-w, local PSO-cf, union of local and global PSOs (UPSO), PSO with second global best particle (SG-PSO), and PSO with second local best particle (SP-PSO). The wine dataset is taken as a numerical example and hyperparameters of MLP the model are determined by the above-mentioned PSO algorithms. The sets of hyperparameters determined by these PSO algorithms are compared with the results of the traditional algorithms for hyperparameter optimization such as random search, tree-structured Parzen estimator (TPE), and covariance matrix adaptation evolution strategy (CMA-ES). Numerical results indicate that PSO-cf is the best-performing and local PSO-w is the second best among the PSO algorithms. The sets of hyperparameters determined by the PSO algorithms were relatively similar. An important finding from the numerical results is that PSO algorithms could find better hyperparameters than random search, TPE, and CMA-ES. This demonstrates that PSO is suitable for the hyperparameter optimization problem in MLP models.
This article explores innovative approaches to the design of reinforced concrete bubble deck slabs. The primary objective is to achieve weight minimization while ensuring compliance with both ultimate limit state (ULS) and serviceability limit state (SLS) requirements. Advanced numerical homogenization techniques and a general nonlinear constitutive law (GNCL), within a finite element method (FEM) framework are employed to perform rapid and precise structural analysis. The study addresses the environmental impacts of traditional construction methods, emphasizing the need for sustainable design practices. By introducing voids into the structural elements of the deck slab, the research aims to reduce material consumption without compromising structural integrity. The optimization process involves identifying optimal design parameters, including the size of the bubble deck unit and the dimensions of the bubbles, to balance material efficiency and structural performance. Computational verification demonstrates that the proposed method accurately predicts displacements and stresses when compared to full 3D models. The results highlight the potential for significant material and cost savings, as well as a reduced environmental impact. The study concludes that the combination of numerical homogenization and GNCL offers a robust and flexible tool for the optimal design of reinforced concrete bubble deck slabs, offering a sustainable alternative to traditional construction methods.
