Vilnius Tech DSpace Repository
Not a member yet
52562 research outputs found
Sort by
Skaitinis strypinės armatūros ir betono sukibimo modeliavimas
No full textReinforced concrete is the most widely used building material in the world. Combining materials with fundamentally different mechanical properties creates a composite that benefits from both strengths. The surface interaction, or bonding, between these materials is crucial for the overall performance of reinforced concrete structures, affecting both safety and serviceability.
In engineering calculations, the ideal bond between concrete and reinforcing bars is often assumed, meaning that under load, their deformations and displacements are equal, thus resulting in zero slip value. This simplification works well for assessing the bearing capacity of structures (when reinforcement pull-out failure mode from the concrete is considered). However, it poorly predicts crack widths and distances between cracks in serviceability limit states, leading to significant errors in calculating deflections and deformations.
One approach to modelling the cracking and deformation of structures is the stress transfer mechanism. This mechanism considers the local interaction between reinforcement and concrete, where forces are transmitted through mechanical resistance, internal friction, and initial adhesion due to the concrete matrix bonding to the reinforcement’s rough surface. While the stress transfer mechanism can model creep, concrete failure, and crack distances, it suffers from uncertainties in the bond-slip law. No universal bond stress-slip law currently exists, and existing laws have narrow application limits, leading to large errors in calculating crack distances and widths.
This work examines the stress transfer mechanism numerically, at the microscopic level, to better understand the interaction between concrete and reinforcement. An accurate numerical model, calibrated by double pull-out tests, allows the study of concrete deformation at various levels, propagation of secondary cracks from the tips of reinforcement ribs, and displacement between reinforcement and concrete (or simply slip) along structural elements. This model enables the study of the bond stress–displacement law through virtual experiments. With this new numerical model, it is possible to analyse concrete behaviour under various load levels (serviceability and ultimate limit states), study cracking and deformations in structures, and derive simplified interaction laws between concrete and reinforcing bars for practical engineering calculations.Gelžbetonis yra plačiausiai naudojama statybinė medžiaga pasaulyje. Sujungus dvi iš esmės skirtingas mechanines savybes turinčias medžiagas, gaunamas kompozitas, turintis abiejų medžiagų privalumų. Šių medžiagų paviršinė sąveika (arba sukibimas) turi didelę reikšmę bendram darbui ir lemia gelžbetoninių konstrukcijų elgseną tiek saugos, tiek tinkamumo ribiniame būvyje.
Atliekant inžinerinius skaičiavimus, priimta taikyti idealų sukibimą tarp betono ir armatūros strypų – veikiant apkrovoms medžiagų deformacijos ir poslinkiai bus lygūs, vadinasi, slinktis bus nulinė. Toks supaprastinimas puikiai tinka vertinant konstrukcijų laikomąją galią (kai suirimo pobūdis yra armatūros ištraukimas iš betono), tačiau nėra tinkamas prognozuoti plyšio pločius bei atstumus tarp plyšių tinkamumo ribiniame būvyje. Tai lemia, kad skaičiuojant konstrukcinių elementų įlinkius ir deformacijas gaunamos didžiulės paklaidos.
Vienas iš būdų modeliuoti konstrukcijų pleišėjimą ir deformacijas yra įtempių perdavimo mechanizmas. Jis remiasi vietine armatūros ir betono sąveika, kai jėgos betonui yra perduodamos kompleksiškai per mechaninį armatūros rumbelio įsispyrimą, vidinę trintį tarp skirtingų medžiagų bei pradinę adheziją dėl betono matricos susiklijavimo su šiurkščiu armatūros paviršiumi. Nors įtempių perdavimo mechanizmas leidžia sumodeliuoti slinktį, betono pažaidą, atstumus tarp plyšių, tačiau pagrindinis šio algoritmo trūkumas yra neapibrėžtumas sukibimo ir slinkties dėsnyje. Iki šiol nėra sukurta universalaus sukibimo įtempių ir slinkties dėsnio, kuris gerai aprašytų medžiagų sąveiką. Esami sąveikos dėsniai turi siauras taikymo ribas ir rezultatai skaičiuojant atstumus tarp plyšių bei plyšio pločius gaunami su didžiulėmis paklaidomis.
Šiame darbe skaitiškai nagrinėjamas įtempių perdavimo mechanizmas mikroskopiniu lygmeniu, leidžiantis pažvelgti į betono ir armatūros sąveiką iš arti. Tikslus skaitinis modelis, sukalibruotas pagal dvigubo ištraukimo bandymus, leidžia tyrinėti betono deformaciją įvairiuose lygmenyse, vietinį betono pleišėjimą, prasidedantį nuo rumbelių, tirti slinktį tarp armatūros ir betono išilgai konstrukcinio elemento. Skaitinis modelis įgalina nagrinėti sukibimo įtempių ir slinkties dėsnį vien iš skaitinių eksperimentų. Pasitelkus naująjį skaitinį modelį, galima tirti betono elgseną įvairiuose apkrovos lygmenyse (tiek tinkamumo, tiek saugos ribiniuose būviuose). Naujasis skaitinis modelis leidžia tirti konstrukcijų pleišėjimą ir deformacijas, taip pat išvesti supaprastintus betono ir armatūros strypų sąveikos dėsnius inžinerinio skaičiavimo praktikai atlikti
Deep Learning-Based PID Controller Tuning for Effective Speed Control of DC Shunt Motors
No full textElectric vehicles (EVs) have become essential due to the depletion of fuel energy resources. DC machines and their role in EVs are gaining significant attention. The speed of DC motor-driven wheels in EVs is usually controlled by proportional-integral-derivative (PID) controllers. But, when the EV is running, the mechanical noises, reduction in tire air volume, the corrugated and rugged surface on which it is driven, etc., lessen the robustness of the PID controllers. This continuous disturbance and variation in speed could result in the exertion of EV circuits, which can be fatal for passengers. Thus, this paper proposes artificial neural network (ANN) based control strategies for enhanced speed regulation in DC motor-driven EVs. Initially, different ANN architectures namely radial basis function (RBF) neural network, nonlinear autoregressive network with exogenous inputs (NARX), nonlinear autoregressive (NAR) network, Elman network, recurrent neural network (RNN), feedforward (FF) network, and probabilistic neural network (PNN) are implemented to design the PID. Of these, it is identified that the FF network is the best choice to design the PID based on its superior time-domain performance index. Further, the efficacy of this proposed ANN-PID controller is compared with the conventional Fuzzy-PID controller subjected to various disturbances namely sine, ramp, step, and chirp. The transient response and steady-state response simulation results proved that the proposed ANN-PID controller delivers superior performance compared to the conventional Fuzzy-PID controller.Taip / Ye
Exploring Strategies for Literary Translation Using Large Language Models
No full textLarge language models (LLMs) have made significant progress in processing and generating text across multiple languages. However, translating long literary works remains challenging due to the need for consistent character interactions, specialized vocabulary, and coherence across chapters. This paper explores these difficulties and examines methods to achieve decent-quality LLM-based translation of literary texts. Various approaches are considered, including techniques for improving contextual awareness and integrating domain-specific vocabulary to reduce inconsistencies. The analysis highlights both the strengths and limitations of current methods, suggesting that targeted context management and fine-tuning strategies have the potential to improve translation accuracy in certain cases. These insights contribute to the development of more effective translation systems for literary texts and multilingual content.Taip / Ye
Enhancing Mango Leaf Disease Diagnosis Using Convolutional Neural Networks
No full textMango leaf diseases significantly impact crop yield and quality, necessitating early and accurate detection for effective management. This study explores deep learning-based classification using MobileNetV3Small and EfficientNetB0 to automate mango leaf disease identification. A dataset comprising eight classes of healthy and diseased mango leaves was used to train and evaluate the models. The results show that EfficientNetB0 achieved an average accuracy of 99.33% with a loss of 0.0437, outperforming MobileNetV3Small, which attained an accuracy of 99.22% with a loss of 0.0583. The confusion matrix analysis reveals minimal misclassifications, with EfficientNetB0 demonstrating superior precision in distinguishing visually similar diseases. These findings highlight the effectiveness of deep learning models in plant disease classification, with EfficientNetB0 providing a more reliable solution. The study underscores the potential of AI-driven tools for real-time disease detection, which can significantly enhance precision agriculture and sustainable crop management.Taip / Ye
Industry 4.0 technologies and enterprise architectures: boosters for circular business models
No full textThe transition to circular business models poses significant challenges, particularly for Small and Medium
Enterprises (SMEs). These challenges arise from different perspectives. Strategic alignment and technological barriers
are just two of them. This paper aims to explore how Industry 4.0 technologies and Enterprise Architectures can facilitate
the implementation of circular business models. By analyzing their role in overcoming key obstacles, the study
explores the potential of these technologies in driving sustainable business transformation. The findings indicate that
while integrating circularity into business practices remains complex, Enterprise Architectures, through the adoption
of Industry 4.0 technologies, can mitigate some barriers. Ultimately, the synergy between technological innovation and
circular business models can accelerate the shift towards sustainability.Portuguese Republic’s Recovery and Resilience Plan (PRR) Partnership AgreementTaip / YesEuropean Union (Europos Sąjunga) under the Next Generation EU programAgenda Mobilizadora daFileira das Tecnologias de Produção para a Reindustrializaçã
Biblioteka informuoja, 2025 Nr. 43 (740)
No full textNaujai į Web of Science ir Scopus įtrauktų Vilnius Gedimino technikos darbuotojų publikacijų sąrašai ir kitos bibliotekos aktualijos.43 (740)202
Konteinerizuotų debesų kompiuterijos programų sistemų automatinio masteliavimo algoritmų, pagrįstų susitarimu dėl paslaugos lygio, tyrimas
No full textThe development of cloud-native applications focuses on scalability and loose coupling of containerized microservices to ensure smooth deployment on cloud or container orchestration platforms. An autoscaler is a crucial component responsible for dynamically provisioning compute resources. When dynamically provisioning resources, addressing issues such as timelines and the amount of resources to be provisioned is important. Therefore, most autoscaling algorithms aim to find a balance between avoiding Service Level Agreement (SLA) violations and effectively managing costs or energy. Various rules-based autoscaling approaches were created to address quality of service concerns and minimise the risk of SLA violations. When resources are allocated and adjusted as needed, an autoscaler typically evaluates current service performance by comparing it to a predefined service level indicator (SLI) value. However, this alone may be insufficient to address changes in SLA conformance. To respond appropriately, the autoscaler must also consider the system’s overall SLA fulfillment status.
This research presents two innovative self-adaptive autoscaling solutions for SLA-sensitive applications. The first solution focuses on maintaining the defined Service Level Objective (SLO) to recover from service degradation and achieve the desired service level. The second solution features a novel SLA-aware dynamic CPU threshold adjustment algorithm. The algorithm aims to ensure that the application has sufficient resources to operate at a level that keeps the number of response time violations compliant with the SLO. Additionally, it aims to ensure that the system operates as closely as possible to the defined Service Level Objectives, thus minimising resource wastage. The solution employs exploratory data analysis techniques in conjunction with moving average smoothing to determine the target utilisation threshold.
The Kubernetes Horizontal Pod Autoscaler (HPA) remains the most widely used threshold-based autoscaling due to its simple setup, operation, and seamless integration with other Kubernetes functionalities. For that reason, this research compares the autoscaling solutions proposed here with the Kubernetes Horizontal Pod Autoscaler and evaluates their effectiveness and performance across various real-world workload scenarios. The evaluation methods for algorithms focus on their ability to operate near-defined SLOs and the effectiveness of resource provisioning. The analysis of the experimental results demonstrates that these solutions are successful in SLA fulfillment and SLO restoration goals while providing an adequate amount of resources to achieve these objectives.
The results of the dissertation were published in six scientific publications, two of which were in reviewed scientific journals indexed in Web of Science and presented at five international conferences.Kuriant debesų kompiuterijos taikomąsias programas, daug dėmesio skiriama tam, kad konteinerizuoti mikroservisai būtų lengvai masteliuojami bei turėtų silpną sankibą, kas užtikrina sklandų taikomųjų sistemų diegimą debesų kompiuterijos ar konteinerių orkestravimo platformose. Automatinio masteliavimo komponentas (angl. autoscaler) yra esminis elementas, kai kalbama apie skaičiavimo resursų dinaminį paskirstymą, reaguojant į resursų poreikį. Kai automatinis masteliavimo komponentas paskirsto resursus, jis turi spręsti terminų bei tinkamo resursų kiekio nustatymo uždavinius, kurie daro įtaką paslaugos kokybei. Todėl dauguma automatinio masteliavimo algoritmų siekia rasti pusiausvyrą tarp susitarimo dėl paslaugų teikimo lygio (angl. Service Level Agreement, SLA) sąlygų pažeidimų išvengimo ir efektyvaus išlaidų ar energijos valdymo. Siekiant išspręsti paslaugų kokybės užtikrinimo problemas ir sumažinti SLA pažeidimų riziką, buvo sukurti įvairūs taisyklėmis pagrįsti automatinio mastelio keitimo metodai.
Šiame tyrime pristatomi du novatoriški, savaime prisitaikantis automatinio masteliavimo sprendimai. Pirmas sprendimas skirtas palaikyti paslaugų teikimo lygio tiksluose (angl. Service Level Obectives, SLOs) nurodytą lygį tais atvejais, kai jis pablogėja dėl netinkamo ar uždelsto resursų teikimo. Sprendimu siekiama palaikyti nustatytą paslaugų lygį bei atstatyti jį degradavus. Taip pat šiame tyri- me pristatomas naujas, žiniomis apie SLA pagrįstas dinaminio slenksčio (angl. threshold) koregavimo algoritmas, skirtas CPU apkrovos slenksčiams nustatyti. Algoritmas siekia užtikrinti tokį resursų kiekį, kad taikomosios programos atsako į užklausas laikas neviršytų nustatyto paslaugos lygio tiksluose daugiau kartų nei leidžiama pagal paslaugos susitarimą. Be to, algoritmas siekia užtikrinti, kad teikiamos paslaugos kokybė kuo labiau atitiktų nustatytą paslaugos teikimo lygio tikslą, taip mažindamas resursų švaistymą. Slenksčiui nustatyti naudojami tiriamieji duomenų analizės metodai ir slankiojo vidurkio glodinimas.
HPA yra plačiausiai naudojamas automatinio masteliavimo komponentas. Jis išlieka populiarus dėl pakankamai paprasto valdymo ir integravimo su kitais Kubernetes komponentais. Dėl šios priežasties tyrime siūlomi automatinio masteliavimo sprendimai yra palyginti su HPA. Tyrimo tikslas yra įvertinti siūlomų sprendimų gebėjimą veikti pagal nustatytus SLO, kartu įvertinant jų efektyvumą paskirstyti resursus, esant įvairių tipų apkrovoms. Rezultatų analizė rodo, kad siūlomi sprendimai sėkmingai paskirsto resursus, užtikrindami SLO palaikymą ar SLO atkūrimą.
Disertacijos rezultatai buvo paskelbti 6 moksliniuose leidiniuose, iš kurių 2 – recenzuojamuose mokslo žurnaluose, indeksuotuose Web of Science, ir pristatyti 5 tarptautinėse konferencijose
An Approach for Building IT Support Dataset for Machine Learning Models
No full textThis study investigates the challenges of preparing datasets for machine learning models based on the data of a centralized system for managing IT incidents within an organization. Key challenges include data quality issues, class imbalance, the need for anonymization, and redundancy in the information. Various data preparation techniques are analyzed, such as handling missing values, encoding categorical and textual data, balancing datasets, anonymizing sensitive information, and performing feature selection. The paper highlights its structural complexities and processing difficulties by examining the state enterprise's Service Desk incident data. Furthermore, the impact of data engineering and cleaning techniques on the accuracy and reliability of machine learning models is assessed. Finally, specific techniques to improve data preparation and to optimize model performance are analyzed.Taip / Ye
Low alcohol content excise tax policy management attempts in the Baltic States
No full textThis study evaluates the diverse approaches to implementing alcohol excise tax rates for low-alcohol-content
beverages across the Baltic States. Utilising a comparative case study methodology, the research analyses different
models of low alcohol excise tax groups, taxation rates, and alcohol type policies. Data were gathered from official
government statistics, EU databases, and relevant policy documents spanning the last decade. The analysis involved a
comparative assessment of tax revenue trends and a qualitative examination of the political and economic factors influencing
these models. The principal conclusion is that each Baltic country has developed a unique practice, shaped
by distinct economic, practical, and political considerations.Taip / YesInternal and External Consolidation of the University of Latvia5.2.1.1.i.0/2/24/I/CFLA/00