Theses and Dissertations
Permanent URI for this collection
Browse
Recent Submissions
Item Open Access ADAPTING TO LEARNER’S COGNITIVE DIFFERENCES USING REINFORCEMENT LEARNING(Nazarbayev University School of Engineering and Digital Sciences, 2023) Nurgazy, Symbat; Issa, Ilyas; Kassymbekov, Saparkhan; Kuangaliyev, ZholamanItem Open Access DEVELOPMENT OF BRAIN-BASED SMART-HOME/TYPING SYSTEM(Nazarbayev University School of Engineering and Digital Sciences, 2024-04-19) Yergaliyeva, Aiana; Berikbol, Arnur; Seiilkhan, ArsenOur project combines EEG-EOG signals to develop an efficient Brain-Computer Interface (BCI) spelling system for Virtual Reality (VR) and Mixed Reality (MR) environments. This hybrid speller enables users to spell using brain activity by leveraging multi-modal signals and various classification strategies. Aimed at improving the quality of life for individuals with motor disabilities, such as spinal cord injuries, ALS, locked-in syndrome, and the elderly, our BCI system provides an alternative communication channel. Focusing on the well-established P300 Row-Column (RC) speller paradigm, we incorporate convolutional neural network (CNN) classification techniques for enhanced performance. Additionally, we use mixed reality glasses to improve user comfort and EEG signal quality. Our methodology includes comprehensive experimental procedures, from environment setup to data analysis and iterative refinement. By advancing BCI technology and integrating VR and MR interfaces, our project seeks to promote accessibility and inclusivity, enabling individuals of all abilities to communicate and participate more fully in social, educational, and professional activities.Item Restricted DESIGN OF HIGH-RISE HOTEL BUILDING IN RIVERSIDE, CALIFORNIA, USA(Nazarbayev University School of Engineering and Digital Sciences, 2024-04-12) Aldabergenuly, Magzum; Galymzhankyzy, Anel; Kussainova, Zhanna; Kydyrali, Tolezhan; Temirbekov, Doszhan; Ualiyev, DulatThis capstone project report summarizes our team’s efforts in developing a high-rise hotel in Riverside, California, within a strong wind and seismic zone. Our multidisciplinary approach divided the project into Architectural, Structural & Materials (40%), Geotechnical (30%), Construction Management (15%), and Environmental Engineering (15%) areas. The project's primary challenge was identifying necessary parameters and adhering to professional design procedures. Through a combination of self-study and mentoring, we conducted independent research and literature reviews to set and achieve specific goals in each project area. Our process involved planning the design using knowledge from courses, secondary sources, and faculty guidance, leading to a synthesized and customized design approach. Key accomplishments include load identification, layout design, preliminary member and force-resisting system design, computer modeling (S), suitable foundation type identification and preliminary foundation design (G), project management method establishment, cost analysis, scheduling, risk management (M), and waste generation rate and composition identification (E). Despite challenges in knowledge gaps and sourcing information, we successfully met our objectives, providing a solid foundation for further project development.Item Restricted "MOMENTUM MAYHEM" 3D PUZZLE GAME PROTOTYPE(Nazarbayev University School of Engineering and Digital Sciences, 2024-04-21) Yerzhanuly, Azamat; Kabykesh, Dias; Ussoltsev, Vladimir; Serik, Yernar; Rysbek, Yelnaz"Momentum Mayhem" is a 3D puzzle game prototype that aims to develop physics concepts in the game environment. The project comprises four stages: ideation, pre-production, production, and post-production. The ideation phase involved analyzing popular physics-based games such as "Fall Guys" and "We Were Here" to inspire unique gameplay concepts. In the pre-production phase, the team outlined the game's requirements and designed its architecture. The production phase focused on creating and integrating essential game components, including character controllers, environmental objects, UI/UX, level maps, and multiplayer functionality. "Momentum Mayhem" leverages Unity for its game engine and Photon PUN 2 for multiplayer networking. The game features physics-based puzzles requiring cooperative gameplay, aiming to enhance problem-solving skills and teamwork. This project highlights the growing game development scene in Kazakhstan by showcasing creative gameplay and technical skills. The document further elaborates on the background, related work, project approach, execution, and evaluation of the system.Item Restricted MOTION PLANNING WITH OBSTACLE AVOIDANCE FOR ROBOT MANIPULATORS VIA DEEP REINFORCEMENT LEARNING(Nazarbayev University School of Engineering and Digital Sciences, 2024-05-03) Sadykov, Zhengisbek; Khussainov, TamerlanThe integration of Deep Reinforcement Learning (DRL) in robotic motion planning represents a cutting-edge approach to enhancing the adaptability and efficiency of robotic manipulators in complex environments. In this project we trained a UR5 manipulator for autonomous navigation within a 2D environment. Our methodology hinges on the Stable Baselines 3 library and Proximal Policy Optimization (PPO) algorithms, grounded within the PyBullet and Gym simulation platforms. The culmination of our research affirms the thesis that it is indeed feasible to train a manipulator to proficiently navigate a 2D environment using DRL. The implications of this work not only bolster the potential for practical applications in various domains but also pave the way for further advancements in the field of robotics.Item Open Access COMPUTATIONAL ANALYSIS OF FLUID STRUCTURE INTERACTION (FSI) IN HORIZONTAL AXIS WIND TURBINES (HAWTS)(Nazarbayev University School of Engineering and Digital Sciences, 2024-05-02) Makshatov, Madi; Nurzhanova, Diana; Sataibekova, Aruzhan; Kussinov, ZhanibekWind power plays a crucial role in the worldwide shift towards sustainable and renewable sources of energy. Wind turbine power generation performances made them a widely adopted method for electricity production, playing a crucial role in the world’s energy resources. Accordingly, optimizing the wind turbine blade’s design is essential for increasing wind turbine performance and reducing expenses. The main aim of this capstone project is to analyze the Fluid Structure Interaction of the HAWTs and to achieve the most effective design of the turbine, in terms of power generation performance and resource requirement by optimizing the blades using low fidelity methods. In engineering and scientific studies, low-fidelity and high-fidelity simulation and optimization have become common concepts, especially in the field of wind turbine design and analysis. These concepts are essential in order to study computational structure and controlling the resource demand, as well as improving the operation of the turbines. This paper focuses on applying low-fidelity optimization techniques with QBlade, which is a commonly used open-source software for creating aerodynamic simulations of horizontal axis wind turbines. A low-fidelity simulation can involve simplified fluid dynamics calculations and simplified structural models, in the context of wind turbine design, in order to forecast the turbine’s effectiveness. Compared to the high-fidelity simulations, low fidelity simulations are economically and computationally reasonable. It means that, in the process of optimization of design parameters of the wind turbine more design alternatives are available in order to reach the most effective parameters. Computational Analysis of Fluid-Structure Interaction (FSI) within Horizontal Axis Wind Turbines (HAWTs) will be studied on the NREL 5MW and NREL Phase VI turbine. In order to get the optimization results of these wind turbine blades, low fidelity optimization methods, such as Betz and Schmitz theories will be used.Item Embargo COMPUTATIONAL FLUID DYNAMICS IN PANCREATICOBILIARY JUNCTION(Nazarbayev University School of Engineering and Digital Sciences, 2024-04) Turgaliyev, AlimPancreaticobiliary maljunction without biliary dilation is associated with pancreaticobiliary reflux, a pathophysiologic factor underlying a wide range of diseases, including gallbladder cancer. This work used computational flow simulations to examine the mechanisms and relevant geometric features that influence pancreaticobiliary reflux and assess the impact of surgical interventions. The results suggest that the refilling phase is the primary mechanism driving pancreaticobiliary reflux. Moreover, the cystic duct diameter was the most critical factor determining the reflux dynamics. Furthermore, the configuration of the baffle system (the baffle height ratio and the number of baffles) affected the dynamics of pancreaticobiliary reflux. Also, the study underscores the potential therapeutic efficacy of cholecystectomy and endoscopic retrograde cholangiopancreatography in managing pancreaticobiliary reflux in cases of pancreaticobiliary maljunction without biliary dilatation. These interventions offer promising avenues for reducing reflux-related complications and mitigating the progression of associated diseases.Item Open Access TOWARDS MORE RELIABLE DRUG TOXICITY PREDICTION: AN ENSEMBLE APPROACH(Nazarbayev University School of Engineering and Digital Sciences, 2024) Yarovenko, VladislavThe development of a single pharmaceutical drug is a time- and resource-consuming process with a high likelihood of rejection. In recent years, the cost-effectiveness of a single drug has decreased drastically, as the criteria for passing has become more rigorous. A huge fraction of attrition rates is caused by the toxicity of chemical compounds. Recent findings in Machine Learning (ML) have revolutionized the drug toxicity prediction field, developing many model architectures and data representations. The faced challenges are different ways of representing the molecules’ chemical structure, as well as many different toxicity types. This study proposes a novel drug toxicity prediction framework. It uses several classification models, based on different data representations and different ways of combining their features. The evaluation of six different datasets with different toxicity types shows that choosing majority voting across all models can improve the ROC AUC score and accuracy. Using a single classification model to combine these datasets demonstrates that it is possible to achieve 84% accuracy on data with various toxicity types. The findings of this research provide insights into the application of ML in pharmaceutical research. Improving current methods of toxicity assessment can have a positive effect on the efficiency and cost-effectiveness of drug development.Item Restricted ENHANCING EMERGENCY RESPONSE: THE ROLE OF INTEGRATED VISION-LANGUAGE MODELS IN IN-HOME HEALTHCARE AND EFFICIENT MULTIMEDIA RETRIEVAL(Nazarbayev University School of Engineering and Digital Sciences, 2024-06) Abdrakhmanov, RakhatIncidents of in-home injuries and sudden critical health conditions are relatively common and necessitate swift medical expertise. This study introduces an innovative use of vision-language models (VLMs) to elevate human healthcare through improved emergency recognition and efficient multimedia search capabilities. By harnessing the combined strengths of large language models (LLMs) and vision transformers (ViTs), this study enhances the analysis of both visual and textual information. We propose a framework that utilizes the PrismerZ VLM in both its Base and Large forms, along with a key frame selection (KFS) algorithm, to pinpoint and exam- ine pertinent images within video streams. This allows for the creation of enriched datasets, filled with images that are paired with descriptive narratives and insights gained from visual question answering (VQA). Through the integration of the CLIP- ViT-L-14 model and the MongoDB Atlas cloud database, we developed a multimodal retrieval system that achieves complex query handling and improved user experience. Additionally, this research undertakes data collection to assess the system’s adaptabil- ity, providing proof of concept and refining the framework. The results showcase the system’s robustness, evidenced by high accuracy rates—86.5% in image captioning and 92.5% in VQA tasks—on the kinetics dataset. When tested with human subject data, the PrismerZ Large model achieved 85.8% accuracy in image captioning and 87.5% in VQA tasks. This performance was further enhanced through fine-tuning with the GPT-4 based Chat GPT, one of the largest language assistants, leading to a 20% improvement in semantic text similarity as measured by the BERT model. The PrismerZ models also stand out for their speed, with the Base and Large versions processing image captioning and VQA tasks in just seconds, even on the NVidia Jet- son Orin NX edge device. These findings confirm the system’s reliability in real-life scenarios. The multimodal retrieval system achieved top performance with a mean average precision at k (MAP@k) of 93% and mean reciprocal rank (MRR) of 94.79% on the kinetics dataset, maintaining an average search latency of merely 0.33 seconds for text queries. This research significantly propels the fields of human activity recognition (HAR) and emergency detection forward, carving out new paths for anomaly detection and enriched multimedia understanding. Our objective in integrating the VLM with multimedia information retrieval is to establish new benchmarks for hu- man care, improving its timeliness, comprehensiveness, and efficiency in accessing multimedia dataItem Restricted PREDICTING SOLAR FLARE OCCURRENCES USING TIME-SERIES DATA FROM NASA’S API FOR SPACE WEATHER FORECASTING(Nazarbayev University School of Engineering and Digital Sciences, 2024) Seit, DamirThis thesis investigates the complicated area of solar flare and Coronal Mass Ejection (CME) prediction using data science approaches and NASA’s Application Programming Interface (API) for the purpose of comparing time-series prediction models in terms of accuracy and interpretability. Quick, intense solar eruptions labeled as CMEs and solar flares have significant impacts on space weather and Earth’s technological systems. The data for this research is taken from the open NASA API service. To understand the underlying relationships governing these solar events, we utilize both state-of-the-art LSTM and well-known models such as Autoregressive Integrated Moving Average (ARIMA), Linear Regression, and Autoregressive (AR) model. The dataset was gathered using a specifically developed Python application with the help of RESTful API. A key part of our research is the dual assessment of these models, which evaluates in terms of both forecast accuracy and interpretability. The interpretability of the tested models is measured using the SOC (simulatability operations count) score metric, which counts the number of arithmetic operations performed to predict a single test case. This study shows that simpler approaches in univariate time-series forecasting perform better than more complex ones in terms of accuracy and interpretability with small dataset. However, only Linear Regression made a reasonable predictions of extreme peak events.Item Restricted ASSESSMENT OF MICROPLASMA SPRAYING OF BIOACTIVE COATINGS ON TITANIUM IMPLANTS FOR BIOCOMPATIBILITY IN VITRO(Nazarbayev University School of Engineering and Digital Sciences, 2024-04) Nwachukwu, China JesseThis study presents an assessment of microplasma spraying (MPS) as a technique for applying bioactive coatings on titanium implants to enhance biocompatibility in vitro. The research aims to evaluate the effectiveness of MPS in promoting osteogenic differentiation and antimicrobial activity of titanium implants through in vitro assays. Surface-modified titanium implants with zirconium, hydroxyapatite, tantalum, and titanium coatings of varying roughness were fabricated using MPS. Osteogenic differentiation potential was assessed via alkaline phosphatase (ALP) activity and alizarin red staining, revealing significant variations in mineralization among the different coatings. Antimicrobial studies utilizing Escherichia coli as a model organism demonstrated differential growth kinetics in the presence of implant material extracts, suggesting potential antimicrobial properties associated with specific coatings. The results underscore the importance of surface modifications in influencing osteogenic differentiation and antimicrobial activity of titanium implants. Zirconium, hydroxyapatite, and tantalum coatings exhibited distinct mineralization patterns and antimicrobial effects compared to uncoated titanium, highlighting the potential of MPS in enhancing the biocompatibility of titanium implants. These findings contribute to the understanding of surface engineering techniques for implant materials and pave the way for further research aimed at optimizing coating formulations and deposition parameters to improve implant performance and clinical outcomesItem Restricted PHOTOTOUCH: SOFTWARE AND HARDWARE METHODS FOR PHOTOELASTIC TACTILE SENSOR(Nazarbayev University School of Engineering and Digital Sciences, 2024-04) Zhuzbay, NurdauletThis thesis introduces a novel approach for predicting force at the fingertips of tendon driven robotic fingers using the photoelastic effect. The core of this technique lies in the detection of stress-induced birefringence in silicone, which becomes visible as distinctive fringe patterns under polarized light. These patterns emerge when tendons, embedded within a silicone matrix, apply forces that compress the material. They are not merely visual markers, but also contain valuable data about the applied forces and their distribution within the silicone. To extract and utilize this data effectively, Convolutional Neural Network (CNN) was employed, specially designed to analyze and interpret the intricate fringe patterns. Thousands of these images were captured in various force application states, resulting in a substantial dataset for the CNN to learn from. The finger was used to show the position control and force control capabilities of photoealastic tactile sensor. It was successful in following the sine wave during force control mode, with RMSE of 0.59 N at a frequency of 0.05 Hz. This document will further delve into relevant literature, elaborate on the research methodology, describe the experimental setup, and present preliminary findings. Col lectively, these components unveil the immense potential of the proposed system in augmenting the tactile capabilities of robotic appendages, with far-reaching implica tions for the fields of robotics and automated systems.Item Restricted FIBER OPTIC SHAPE SENSORS FOR DEVICES(Nazarbayev University School of Engineering and Digital Sciences, 2024-04) Karzhaubayeva, DanaThe usage of fiber optics for sensing purposes become popular in recent years due to its advantages such as compatibility, flexibility, and high sensitivity. The literature review observed different types of fiber optics sensors, their applications, and strengths and weaknesses. The setup, which provides real-time shape and temperature sensing, was constructed during this study. The purpose of this work is to conduct laboratory experiments to design a setup that will sense strain and temperature values by fiber optical fibers in real time. Insertion into phantom results illustrate how the shape of the inserted needle changes through different layers of it. Radiofrequency thermal ablation experiment results demonstrate how the temperature was sensed by fibers based on their positions in a heated phantom.Item Open Access DEVELOPMENT OF INTERFEROMETRIC OPTICAL FIBER BIOSENSOR FOR HCC1806 BREAST CANCER CELL DETECTION(Nazarbayev University School of Engineering and Digital Sciences, 2024-04) Meirambekuly, TimurThis thesis explores the use of semi-distributed interferometric (SDI) optical fiber biosensors for detecting breast cancer cells, with a focus on assessing the potential and limitations of this technology. Given the critical importance of early detection in improving breast cancer outcomes, this research aims to advance diagnostic methodologies by leveraging the sensitivity and specificity advantages of optical fiber technology, combined with targeted antibody functionalization. The study commenced with the fabrication of optical fiber sensors, incorporating interferometric tips to single-mode optic fiber pigtails. Calibration was performed using specialized software and equipment, selecting sensors with optimal sensitivity based on their performance in standardized tests. These sensors were then functionalized with CD44 antibodies, targeting the detection of HCC1806 breast cancer cells across a range of concentrations. Experimental trials were conducted to evaluate the sensors' ability to detect cell concentrations from sterilized PBS up to 10^6 cells per ml, with data collection at regular intervals.Item Embargo PREPARATION OF LIFEPO4 -BASED ELECTRODES WITH MAGNETO-SENSITIVE IRON OXIDE (FE2O3 ) NANOPARTICLES UNDER MAGNETIC FIELD(Nazarbayev University School of Engineering and Digital Sciences, 2024) Emmanuel, Iheonu NdukaLithium iron phosphate (LiFePO4) is one of the most widely utilized cathode materials in lithium-ion batteries (LIBs) due to its low cost and environmental safety. Despite this attribute, they still face significant challenges, including high rates and prolonged cycling. However, research has shown that utilizing a magnetic field (MF) can help tackle these issues, thereby improving the ion conductivity and inhibiting polarization of the LiFePO4 cathode. In this study, LiFePO4 cathodes were produced by subjecting them to MF (LFP-MF), while others were optimized using magnetic-sensitive Fe2O3 nanoparticle additives in two concentrations (LFP+1 wt% Fe2O3-MF and LFP+3 wt% Fe2O3-MF). The cathodes were compared to conventional LiFePO4 electrodes prepared under the same conditions but without applying MF (LFP-WMF, LFP+1% Fe2O3-WMF, and LFP+3% Fe2O3-WMF). Application of MF improved the materials' electrochemical properties, contributing to the battery's superior electrochemical performance. The lithium-ion diffusion coefficients of LFP-MF (4.1376 × 10-4 cm2 S-1), LFP+1% Fe2O3-MF (4.0614 × 10-4 cm2 S-1), and LFP+3% Fe2O3-MF (2.3021 × 10-4 cm2 S-1) were more significant than those of LiFePO4 cathodes without subjecting to the magnetic field. Furthermore, the cathodes subjected to MF had higher reversible capacity and a reduced capacity decay than those without MF at an enhanced rate capability greater than 1 C. This study discovered that an MF improves the high-rate performance of LiFePO4 cathodes.Item Restricted IMPROVING THE SPECIFICITY AND SENSITIVITY OF OPTICAL SENSORS FOR BIOLOGICAL PURPOSES(Nazarbayev University School of Engineering and Digital Sciences, 2024-04) Adilzhankyzy, AinaThe development of antibody-detecting biosensors using interferometric devices is a promising area in biomedical engineering. In this study, a semi-distributed interferometer (SDI) was employed as the biosensor device, utilizing a simplified Fabry-Perot structure. The sensor was functionalized using a combination of surface functionalization and silanization, with nanostructuring to enhance the surface area through the application of gold nanorods. The focus of the research was on the detection of two key biological analytes: the CD44 cancer biomarker and the A33 glycoprotein of the vaccinia poxvirus. The fabrication and functionalization processes were optimized for sensitivity and specificity, with the aim of achieving real-time, label-free detection of these important biomarkersItem Restricted CLASSIFICATION OF BABY CRIES INTO DISTINCT CATEGORIES USING CONVOLUTIONAL NEURAL NETWORKS(CNN) WITH SOUND AND SPECTROGRAM ANALYSIS(Nazarbayev University School of Engineering and Digital Sciences, 2024-03) Tulegenov, MaxatThe act of a baby crying is a complex form of communication that reflects various physical, medical, and emotional states. Understanding the nuances within baby cries is essential, as it provides valuable insights into the baby’s needs and can assist in the early detection of developmental disorders and medical conditions. Machine Learning (ML) and Deep Learning (DL) techniques, specifically Convolutional Neural Networks (CNNs), coupled with sound processing and data augmentation, play a pivotal role in this endeavor. This research explores methods encompassing data preprocessing, feature extraction, postprocessing, and classification. A primary focus is acoustic analysis and CNN for automatic feature extraction.Item Restricted BENCHMARKING LARGE LANGUAGE MODELS IN KAZAKH LANGUAGE(Nazarbayev University School of Engineering and Digital Sciences, 2024-04) Maxutov, AkylbekItem Open Access ADVANCED CIRCUIT CONFIGURATIONS FOR RF WIRELESS POWER TRANSFER(Nazarbayev University School of Engineering and Digital Sciences, 2024-04-26) Kudaibergenova, ZhanelTechnology for wireless power transfer (WPT) has gained more importance in the contemporary world. The upsurge spike can be a result of the WPT system's ability to power devices without the use of traditional connections. In particular, near-field WPTs have a wide range of applications, including wireless sensors, IoT, biomedical implants, RFID, and consumer electronics. It is essential to emphasize that the WPT system can be realized in a number of ways, one of which is a defected ground structure technique. This approach is well-established for its simple design process and compact system. Despite this recently developed DGS-based WPTs demonstrate poor performance, in other words, low power transfer efficiency in practical validations. The inevitable factors, such as imperfections of lumped elements, the in-house fabrication, and energy losses during transfer, have an impact on the experimental results. Therefore, various performance enhancement strategies have to be considered to realize the compact and efficient WPT system. In this regard, one of the promising methods for improving WPT operation is the use of metamaterial, which is an artificial material with unique electromagnetic features. As a result, this thesis work focuses on the development of compact and efficient WPTs applicable to various fields and on performance enhancement strategies based on metamaterial utilizationItem Restricted VISION-LANGUAGE MODELS ON THE EDGE: AN ASSISTIVE TECHNOLOGY FOR THE VISUALLY IMPAIRED(Nazarbayev University School of Engineering and Digital Sciences, 2024-06) Arystanbekov, BatyrVision-Language Models, or VLMs, are deep learning models at the intersection of Computer Vision and Natural Language Processing. They effectively combine image understanding and language generation capabilities and are widely used in various as sistive tasks today. Nevertheless, the application of VLMs to assist visually impaired and blind people remains an underexplored area in the field. Existing approaches to developing assistive technology for the visually impaired have a substantial limitation: the computation is usually performed on the cloud, which makes the systems heavily dependent on an internet connection and the state of the remote server. This makes the systems unreliable, which limits their practical usage in everyday tasks. In our work, to address the issues of the previous approaches, we propose utilizing VLMs on embedded systems, ensuring real-time efficiency and autonomy of the assistive module. We present an end-to-end workflow for developing the system, extensively covering hardware and software architecture and integration with speech recogni tion and text-to-speech technologies. The developed system possesses comprehensive scene interpretation and user navigation capabilities necessary for visually impaired individuals to enhance their day-to-day activities. Moreover, we confirm the prac tical application of the wearable assistive module by conducting experiments with actual human participants and provide subjective as well as objective results from the system’s assessment.