187037 research outputs found
Sort by
데이터 기반 한국 주택시장 분석과 정부 예산분배 모델 및 주택 추천시스템 개발
No full text학위논문(박사) - 한국과학기술원 : 산업및시스템공학과, 2025.2,[v, 128 p. :]Korean housing market is characterized by its unique characteristics, including the complex Jeonse system and distinct regional disparities, making it challenging to analyze its determinants and understand its socio-economic impacts. This study conducts an empirical analysis based on various housing and text data to examine the effects of sentiment and policies on the housing market. Using econometric models, the research explores how housing market dynamics influence socio-economic indicators such as fertility rates. Furthermore, the study develops a multi-stage stochastic and dynamic programming model to provide an optimized government budget allocation solution that balances short-term market uncertainties with long-term fiscal sustainability. Lastly, the research integrates collaborative filtering and mean-variance optimization to design a housing recommendation system tailored to individual preferences and financial considerations. By combining empirical analysis, financial planning models, and practical tools, this dissertation offers a comprehensive approach to understanding the complexities of the Korean housing market and provides actionable methodologies for policymaking and enhancing housing welfare.한국과학기술원 :산업및시스템공학과
잡 샵 스케줄링 문제를 위한 학습 기반 스케줄링 방법론
No full text학위논문(박사) - 한국과학기술원 : 산업및시스템공학과, 2025.2,[vi, 85 p. :]This paper proposes three learning-based schedulers for minimizing makespan in job shop scheduling problems with different process routes for each job. We achieved performance improvements by incorporating job shop scheduling-specific properties into deep learning-based schedulers. These methodologies are scalable and can handle problems of various sizes, regardless of training instances.
First, we develop a learning-based dispatcher capable of real-time decision-making based on current conditions. The dispatcher defines decision points and performs scheduling by selecting a job-machine assignment as an action from candidate actions at each step. Recently, graph neural network(GNN)-based dispatchers have been proposed leveraging GNN's capabilities to effectively embed node information and handle varying graph sizes without additional learning. Building upon these works, we focused on improving performance through rigorous analysis of whether sufficient information is provided for our dispatcher to learn optimal dispatching strategies, while also examining if any unnecessary information is being utilized. Our dispatcher achieves superior performance compared to existing dispatchers through modifying definitions of state, action, and state transition, and utilizing imitation learning (IL) to learn the dispatching strategies instead of reinforcement learning (RL). Our graph-based state representation utilizes only observable information at the current decision point while excluding information from completed operations, and candidate actions and state transitions ensure the solution space always contains only all active schedules, which have at least one optimal solution.
Second, we propose a learning-based local search method as one of the approaches that can be utilized when real-time decision making is not required and complete schedules need to be generated within relaxed time constraints. As a metaheuristic approach, this local search method remains applicable even when deployed on problems of varying sizes. In this method, determining the neighboring solution to move from the current solution is crucial for performance, which has recently been approached using learning-based methods. Similar to our first methodology, we aimed to improve performance by incorporating job shop scheduling-specific characteristics. We identified conditions for actions that can improve makespan compared to the current schedule and incorporated these conditions into the policy network of the scheduler to utilize the information for selecting a neighboring solution. The scheduler, trained through reinforcement learning, demonstrates superior performance compared to existing learning-based local search methods.
Finally, we propose a depth-limited tree search method that enhances real-time dispatching decisions. While look-ahead tree search methods that simulate future states for decision-making have been used in environments with decision time limits like board games, they may fail to explore any complete solution under extremely tight time constraints. Traditional tree-based search methods like branch and bound aim to explore all nodes, which is time-consuming but guarantees optimality, while beam search is its heuristically modified version for efficiency. Our proposed methodology combines branch and bound with beam search in a look-ahead manner and utilizes a dispatching strategy as a rollout policy which allows fallback to the established dispatching strategy if the time limit prevents construction of trees to the predefined depth or completion of even a single simulation. In static environments, we prove that schedules generated using this method are guaranteed to perform at least as well as or better than schedules created by the underlying dispatching strategy alone.
The proposed three methodologies can be flexibly selected based on decision time constraints. All three methodologies are scalable and can be immediately applied to problems of varying sizes without additional training. The methodologies have been evaluated on job shop scheduling benchmark problems, providing a basis for comparison with other methodologies in the field. Furthermore, we conducted additional performance evaluations of the dispatchers under scenarios with dynamic events to assess their adaptability to unexpected changes in the scheduling environment.한국과학기술원 :산업및시스템공학과
Direct blood cell flow imaging in 3D microvascular networks and quantitative hemodynamic analysis
No full text학위논문(박사) - 한국과학기술원 : 기계공학과, 2025.2,[vii, 87p :]미세혈관 네트워크에서 혈류역학은 조직 및 기관 전체의 건강과 관련되어 있다. 혈류역학을 분석하기 위해 수많은 영상 기법과 시스템이 개발되었지만, 제한된 속도, 외부 조영제의 사용, 간접적인 정량화로 인해 그 사용에 한계가 있다. 본 연구에서는 외부 조영제의 사용 없이 0.69 ms의 시간 분해능으로 0.71 mm x 1.42 mm의 필드에 걸쳐 혈구 각각을 선별하는 직접적 혈구 이미징(DBFI) 기법을 제시한다. DBFI는 단순한 현미경 이미지의 배경 신호에 묻혀 있는 시간에 따라 변하는 혈구에 기반한 신호를 추출하여 혈구만을 선택적으로 시각화 할 수 있다. DBFI는 높은 시간 분해능으로 모세혈관에서 동맥 및 정맥까지 넓은 범위에 걸쳐 양한 혈관의 혈류 속도 및 플럭스에 대한 정밀한 동적 분석을 가능하게 한다. 이 새로운 이미징 기술은 생리학 및 질병 분야뿐 만 아니라 높은 시공간 해상도로 움직이는 미세 물체의 정량적 3차원 이미징이 필요한 다양한 응용 분야에 널리 활용될 수 있다.한국과학기술원 :기계공학과
상간 질량 전달에 의한 다성분 액체의 계면 불안정성에 대한 연구와 그 응용
No full text학위논문(박사) - 한국과학기술원 : 기계공학과, 2025.2,[vi, 71 p. :]In multi-component liquids, the surface tension can be altered by the mass transfer of surface-active substances, such as solutes or surfactants, in contrast to single-component liquids. This alteration can lead to the emergence of interfacial instabilities like Marangoni instability and spreading instability, which play a crucial role in industrial applications of multi-component droplets such as inkjet printing, biological analysis, and the production of membranes used for applications such as seawater desalination, oil recovery, and drug delivery. Clarifying the generation of these instabilities requires an analysis of the mass transfer of solutes and surfactants across the interface. However, it is relatively difficult to directly observe small molecules like solutes or surfactants, so visualization of flow patterns and liquid movements has been used to estimate interphase mass transfer. Thus, this dissertation analyzed the interfacial instability by visualizing flow structures and liquid movements via particle image velocimetry and microscopy, respectively, when the evaporation and absorption of vapor occurs at a liquid-gas interface, and the chemical reaction is induced at an immiscible liquid-liquid interface and liquid-solid interface. Furthermore, effective methods were proposed to utilize interfacial instability by adjusting interphase mass transfer based on the analysis.한국과학기술원 :기계공학과
다층 복합 쉘 유한요소의 개발 및 자체 업데이트 솔리드 유한요소의 개선
No full text학위논문(박사) - 한국과학기술원 : 기계공학과, 2025.2,[vii, 81 p. :]In this thesis, two parts, development of multi-layered composite shell finite element and improvement of self-updated solid finite element, are presented. The first proposed topic is to develop a novel multi-layered shell finite element for linear and nonlinear analyses of multi-layered composite shells with interlayer slips. The main advantage of the proposed shell formulation is that complex geometries of thin-curved multi-layered structures and the complicated partial interactions from interlayer slips can be simply represented in the framework of the shell kinematics by introducing layer degrees of freedom (DOFs) for each layer. The proposed shell element allows simple modeling together regardless of the number of layers and interlayers, and enables the successful prediction of geometric nonlinear behavior of multi-layered shells as well as bi-linear load-slip relations such as fracture mode delamination at interlayers. Furthermore, the proposed shell element model requires fewer DOFs compared to conventional 3D solid finite element models, resulting in significantly reduced computational costs and efficient analysis.
The second proposed topic is to improve the performance of the 4-node 2D solid finite element using a new robust load decomposition method. The previous developed self-updated finite element (SUFE) has been developed to improve the performance of the 4-node 2D solid finite element using deep learning. SUFE reconstructed stiffness matrices for optimal bending modes in the given displacement to minimize deformation energy through the iterative process. SUFE showed excellent bending performance in several numerical examples, even in coarse and highly distorted meshes. However, under combined loads of bending and constant strain modes, another robust method for obtaining accurate solutions is needed. Thus, we propose a new load decomposition method to improve the performance of the 2D solid element under combined loads. We adopt the kinematic mode-based formulation used in the SUFE and modify the iterative process including the load decomposition method. The proposed method passes the patch and zero-energy mode tests. Furthermore, we demonstrated the excellent performance of the first proposed multi-layered shell element and the second proposed method for 2D solid element through the various representative numerical problems.한국과학기술원 :기계공학과
레이저 가공 내부 형판 및 폐쇄형 금형 공정을 이용한 압력 구동 능동 조향 마이크로 가이드와이어 제조
No full text학위논문(박사) - 한국과학기술원 : 기계공학과, 2025.2,[vii, 121 p. :]Vascular intervention is a procedure for diagnosis and treatment by inserting a thin, long, tube-shaped catheter along a vasculature to a target lesion. Active steering catheters and guidewires have been developed for selectively inserting them by steering their distal ends within a patient’s body, to reduce procedure time and risk of radiation exposure. Despite numerous studies, only a few active steering catheters or guidewires are applied clinically, and even these cannot be used in procedures that require micro-scale catheters or guidewires because of their large outer diameter of 1 mm or more.
This dissertation proposes a pressure-driven active steering micro guidewire with high clinical applicability and fabrication method using a laser-patterned core template. The proposed guidewire has a steerable tip of a double bending curvature and a body that is mechanically robust and meets the requirements for hydraulic actuation.
The fabrication method of the steerable tip is developed by improving the previous core template based closed molding method. The core template is machined by engraving micro patterns on the surface of the stainless steel rod using femtosecond laser processing. The steerable tip is fabricated by permanently using the machined core template and the micro mold. The semi-permanent use of the stainless steel core template can reduce process errors due to repetitive fabrication, reduce process time, and satisfy biocompatibility in terms of the manufacturing process. The proposed steerable tip and its endcap are fabricated to possess a barium sulfate mass fraction of 10% and 40%, respectively, so that the distal end of the proposed guidewire can be rendered under X-rays.
The proposed guidewire has a maximum outer diameter of 600 μm to be compatible with a commercial micro catheter. The steerable tip has a steering shape with a double bending curvature suitable for blood vessels with large bifurcation angles. The requirements for steering shape are analyzed based on the statistics of geometrical characteristics of human cerebral arteries. The double bending shape is implemented by placing trapezoidal patterns on the inner surface of the proximal part of an eccentric silicone tube. The presence of patterns causes an increase in the bending radius of the proximal part compared to that of the distal part without the patterns. At an internal pressure of 2.3 atm, the fabricated tip steers with a steering angle of 43.15 degrees at the distal 2.1 mm segment and a total steering distance of 9.20 mm. The experiment results of repeated use under an internal pressure of 2.0 atm show that the fracture occurs at the 111th manipulation due to material degradation. The outer diameter of the steerable tip increases up to 770.36 μm at the 11th steering, which is relatively less affected by material degradation.
The guidewire body is designed to satisfy mechanical requirements while transmitting internal fluid pressure. This dissertation analyzes the mechanical requirements by measuring the flexural rigidity of commercial micro catheter and guidewire through a three-point bending test. The proposed guidewire body consists of a micro spring with variable pitch and outer diameter, tapered wire, stainless steel tube, and Pebax polymer tube, so that the bending stiffness of the proposed guidewire continuously and gradually increases. The fabrication process for the guidewire body is developed to maximize torquability while satisfying biocompatibility.
The selective insertion and catheter guidance performance of the proposed guidewire is evaluated through in-vitro experiments. The proposed guidewire demonstrates successful insertion into a vascular branch with a bifurcation angle of 112 degrees, within two vascular bifurcation models with diameters of 4.2 mm and 1.5 mm, respectively. It is also successfully inserted into two consecutive side branches in a three-dimensional cerebrovascular model constructed from real patient magnetic resonance angiography data. Finally, the proposed guidewire successively guides a commercial micro catheter to the target lesion, positioned 51.8 cm away from the vascular origin, within a two-dimensional vascular model mimicking the geometry of cerebral arteries.한국과학기술원 :기계공학과
저온 구동을 위한 박막 마이크로 고체산화물 연료전지의 성능 및 대면적화에 대한 연구
No full text학위논문(박사) - 한국과학기술원 : 기계공학과, 2025.2,[v, 61 p :]This dissertation investigates the deposition of electrodes and fabrication of thin-film solid oxide fuel cells (SOFCs) for low-temperature operation. A platinum-ceria composite, fabricated in a hierarchical structure with samarium-doped ceria (SDC), demonstrated improved thermal stability and electrochemical performance. A large-scale micro-SOFC with a tapered silicon-edge support was developed to enhance mechanical stability, utilizing circular thin films to ensure uniform stress distribution and eliminate stress concentration. Advanced etching and deposition techniques enabled the fabrication of 100 nm-thick, support-free YSZ thin films with a 5 mm diameter. A scalable thin-film SOFC array with a platinum-ceria cathode was also demonstrated, offering enhanced mechanical stability, adaptability to larger geometries, and improved performance. The results obtained from all chapters were utilized to suggest the application of thin film SOFC for the low temperature operation.한국과학기술원 :기계공학과
자율 수상이동체를 위한 안전성이 향상된 모델예측제어
No full text학위논문(박사) - 한국과학기술원 : 기계공학과, 2025.2,[ix, 137 p. :]Autonomous surface vehicles (ASVs), known for their high safety and efficiency, have been widely utilized in fields such as autonomous transportation and environmental monitoring. The International Maritime Organization classifies autonomous navigation technology into four levels. Current research aims to achieve level 3 autonomy for ocean navigation and level 2 for coastal navigation. Coastal navigation faces greater safety challenges compared to ocean navigation due to its complex, obstacle-laden environments and frequent interactions with other vehicles, leading to slower technological advancements. These challenges are further exacerbated in narrow waterways, where restricted spaces and tighter maneuvering requirements demand precise and reliable control algorithms. To address these issues, this dissertation introduces a model predictive control (MPC) algorithm designed to improve the safety of ASVs operating in such coastal and narrow waterway environments. The research addresses three critical scenarios in these settings. First, in canal environments, an ASV must maintain a safe distance from the side walls while moving toward a target location. To achieve this, a navigable area detection algorithm that uses both cameras and LiDAR, along with an MPC algorithm based on the detected areas, is proposed. Experiments conducted with a cruise boat in the Pohang Canal validated the proposed approach, demonstrating stable navigation over a 1-km course with enhanced safety compared to conventional methods. Second, this thesis tackles a critical aspect of fully ASV operations: berthing algorithms, which face challenges under strong environmental disturbances. To address this, a cascade tube-based MPC algorithm is proposed to enable reliable berthing even in uncertain and disturbed environments. Simulation tests validate the algorithm's performance, and theoretical analysis provides insights into its recursive feasibility and stability. Finally, to enhance the safety and efficiency of path-following and static/dynamic obstacle avoidance, a control barrier function (CBF) incorporating a turning circle is proposed. This overcomes the limitations of conventional Euclidean distance-based CBFs by incorporating the nonholonomic characteristics of ASVs. Experimental results demonstrate that the MPC algorithm, when integrated with the proposed CBF, outperforms conventional approaches in both collision avoidance and path-following tasks, achieving higher efficiency while ensuring safety.한국과학기술원 :기계공학과
금속유기골격체에서 기인한 기능성 소재 및 에너지 저장/변환 응용에 관한 연구
No full text학위논문(박사) - 한국과학기술원 : 신소재공학과, 2025.2,[xii, 196 p. :]Metal-organic framework (MOF)-derived functional materials have emerged in recent decades due to their exceptional electrochemical properties. In this thesis, we propose rational design strategies regarding MOF-derived materials. The carbon-based metals/metal compounds with abundant accessible active sites and heteroatoms-doped matrices can improve electrochemical performance. The additional merits like particle size control of active materials can be achieved when MOF is combined with graphene oxide (GO). In addition, MOFs can be used as precursors for layered double hydroxides (LDHs) and a heterostructure between individual phases can be generated through a further facile pyrolysis process. The formation of heterostructure may tune the electrochemical properties of functional materials. Furthermore, we demonstrate the applications of MOF-derived functional materials in energy storage and conversion.한국과학기술원 :신소재공학과
양극성 정동장애 환자 역분화 줄기세포 유래 별아교세포의 대사 변화
No full text학위논문(박사) - 한국과학기술원 : 의과학대학원, 2025.2,[v, 78 p. :]Metabolic alteration in the brains of patients with bipolar disorder (BD), a neuropsychiatric disorder characterized by biphasic mood episodes of mania and depression, are well-recognized, yet the specific contribution of glial cells to theses metabolic changes in BD patients remain largely unknown and have not been extensively studied. In this study, I investigated the metabolic characteristics of induced astrocytes (iAstrocytes) from BD patients and analyze their responses to lithium treatment. The gene expression profiles of iAstrocytes from BD patients (BD iAstrocytes) indicate dysregulation of metabolic processes compared to iAstrocytes from control subjects (Control iAstrocytes). Furthermore, BD iAstrocytes showed decreased mitochondrial respiration, increased glycolysis, and elevated lactate secretion. These changes in metabolic pathways suggest impaired mitochondrial function. This is further supported by reduced protein expression in the oxidative phosphorylation complex and decreased reactive oxygen species (ROS) production in BD iAstrocytes. Intriguingly, BD iAstrocytes showed an accumulation of lipid droplets (LDs) in both volume and number, possibly reflecting metabolic stress-induced LD accumulation. Lithium, a commonly prescribed treatment for BD, attenuated LD accumulation exclusively in BD iAstrocytes from lithium responders, but failed to restore mitochondrial dysfunction or decrease lactate secretion. These findings provide novel insights into understanding metabolic dysfunction in the brains of BD patients and shed light on the molecular mechanisms underlying lithium’s action, particularly regarding astrocytes. Moreover, the LDs in BD iAstrocytes from lithium non-responder may serve as a useful biomarker for drug screening, aiding in the identification of alternative therapeutic strategies for these patients.한국과학기술원 :의과학대학원