https://scholarworks.unist.ac.kr/handle/201301/65 2026-04-08T21:36:49Z https://scholarworks.unist.ac.kr/handle/201301/91027 Title: Adaptive Point-Wise Error Metric Design for LiDAR Scan Matching Using Deep Neural Networks Author(s): Lee, Jinwoo Abstract: Iterative Closest Point (ICP) is one of the most widely used and powerful algorithms for LiDAR scan registration, yet its performance is highly dependent on the design of the underlying error metric. Con- ventional choices such as point-to-point or point-to-plane distances are often based on hand-crafted heuristics, which limits their robustness and generalization across diverse environments. To overcome this limitation, we propose a deep learning–based framework that adaptively learns point-wise error metrics for ICP. Inspired by the probabilistic formulation of Generalized ICP (GICP), which models uncertainty through point-wise covariance matrices, we directly learn anisotropic covariance represen- tations using a neural network. The network interprets geometric structures of LiDAR scan and predicts point-wise covariance matrices that parameterize the registration error metric. To enable stable training, we introduce mathematical approximations for direct supervision of loss and propose prospective drift loss that balances scale between rotation and translation component in transformation matrix. We eval- uate our approach on LiDAR odometry tasks and demonstrate that the learned error metric significantly improves registration accuracy compared to traditional ICP variants that rely on hand-crafted metrics. Furthermore, we provide empirical analysis on proposed method. Our results highlight the potential of data-driven metric design for scan matching and open new directions for integrating deep learning with classical LiDAR scan matching algorithms. Major: Department of Mechanical Engineering 2026-01-31T15:00:00Z https://scholarworks.unist.ac.kr/handle/201301/91026 Title: Numerical analysis of thermal stratification in cryogenic storage tanks Author(s): Lee, Kyungwon Abstract: To numerically investigate transient natural convection inside a cryogenic storage tank, a numerical model combining a 2-D axisymmetric model for the liquid region with a 1-D model for the ullage region is developed and validated against experimental data. A cylindrical tank subjected to a uniform heat flux along the side wall is considered, which induces thermal stratification—the phenomenon in which warmer liquid near the wall rises, flows towards the colder core of the liquid bulk, and accumulates near the liquid–vapor interface. The model accurately predicted stratified-layer thickness, i.e. the thickness of the accumulated hot-liquid layer, to within ±5% of experimental measurements. Using this model, the temporal evolution of stratified layer is examined for tanks with various liquid aspect ratios—defined as the product of tank aspect ratio and liquid-fill ratio—ranging from 0.125 to 1.5. The results show that the ratio of stratified-layer thickness to liquid height increases over time and increases more rapidly as the liquid aspect ratio decreases at a fixed radius. A quantitative analysis of stratified-layer growth over time is conducted and compared with results obtained from a correlation based on Sparrow’s 1-D natural-convection solution. It is found that the 1-D correlation is useful if an adequate criterion is defined to distinguish the stratified layer from the colder bulk liquid. Therefore, a criterion that enables the 1-D correlation to be applicable to a 2-D domain is proposed, and shown to predict numerical results for various liquid aspect ratios with an error of ±8%. These findings offer valuable guidance in choosing between 1-D and 2-D modelling approaches for cryogenic-tank analyses. Major: Department of Mechanical Engineering 2026-01-31T15:00:00Z https://scholarworks.unist.ac.kr/handle/201301/91025 Title: AIRBORNE VIRUS COLLECTION USING THE BIOSAMPLER AT VARIOUS SAMPLING CONDITIONS Author(s): Jung, Chaewoon Abstract: For several decades, humans have faced challenges from various respiratory viruses, such as influenza viruses and coronaviruses. These viruses can spread through multiple transmission routes, including droplets, direct and indirect contact, and aerosols. Among these routes, aerosol transmission is particularly critical because airborne virus particles (or viral aerosols) can remain suspended in the air for hours, depending on their size. This allows the virus to spread and infect people through inhalation. To proactively prevent and mitigate airborne virus transmission, it is important to collect airborne viruses using an appropriate air sampler before measuring their concentrations. The purpose of this study is to characterize the widely used impinger, named the SKC BioSampler. The BioSampler has been used in many previous studies to sample various kinds of bioaerosols. Although airborne virus sampling using the BioSampler has been characterized in many studies, only a few studies investigated the effect of flow rates including its maximum sampling flow rates. Thus, the performance of the BioSampler was evaluated under a range of sampling conditions in this study: flow rates of 4.0, 8.0, 10.0, 12.5, and 13.3 standard L/min (SLPM) (4.0, 8.1, 11.3, 17.6, and 23.5 L/min at the BioSampler outlet, respectively); sampling periods of 10, 60, and 360 min; three airborne virus concentrations of MS2 bacteriophages (108, 106, and 104 PFU/m3) and two airborne virus concentrations of influenza A viruses (106, 105 gene copies/m3) at the BioSampler inlet; collection liquid volumes of 20 and 13 mL. Virus concentrations were determined through plaque assay for MS2 bacteriophages, and reverse transcription quantitative polymerase chain reaction (RT-qPCR) assay for influenza A viruses. For viable MS2 viruses, both the relative infectious virus concentration (RIVC) and the intrinsic collection efficiency (ICE) increased with sampling flow rates during 10- and 60- min sampling periods. At the lowest virus concentration—similar to field-level concentrations, viruses could be detected after 360 min of sampling. Under all sampling conditions, the performance of the BioSampler at a flow rate of 13.3 SLPM was better than that at the manufacturer-recommended flow rate of 12.5 SLPM when sampling airborne viable MS2 viruses. A similar trend was observed for influenza A viruses. Both the relative total virus concentration (RTVC), as determined by RT-qPCR, and the ICE increased with the sampling flow rate. Again, 13.3 SLPM proved to be more efficient for collecting influenza virus nucleic acids compared to 12.5 L/min. Additionally, to evaluate the change in performance during 120- min sampling, sampling was conducted with initial 13-mL collection liquid volume. Between the two collection liquid volumes of 20 mL and 13 mL, RTVC did not differ significantly, but the collection efficiencies were lower when using 13 mL compared to 20 mL. This comprehensive evaluation of the BioSampler under various conditions— including sampling flow rates, sampling periods, airborne virus types, airborne virus concentrations, and collection liquid volumes—is expected to benefit future studies on airborne virus sampling. Major: Department of Mechanical Engineering 2026-01-31T15:00:00Z https://scholarworks.unist.ac.kr/handle/201301/91024 Title: A voxel-based automated life cycle assessment framework for additive manufacturing: a case study of laser powder bed fusion Author(s): Lee, Juchan Abstract: As global regulations on industrial greenhouse gas emissions become increasingly stringent, manufacturers are required to quantify the environmental consequences of their process chains with finer resolution and shorter turnaround times. Additive manufacturing (AM), with its high geometric design freedom and tunable process parameters, further amplifies this need because relatively small modifications in part geometry or process settings can substantially alter energy use, material consumption, and, ultimately, environmental impacts. Conventional life cycle assessment (LCA) provides a rigorous, ISO 14040/14044-compliant framework for such evaluations, but it is typically time-consuming, case-specific, and heavily dependent on expert modelling and software operation. Recent rapid LCA (RLCA) approaches partially alleviate these burdens by automating inventory construction and employing data-driven models. However, many of these methods remain tied to specific geometries or training datasets and provide limited transparency regarding how design and process changes propagate to environmental impact results. To address these limitations, this study develops a voxel-based automated life cycle assessment (VLCA) framework that adopts a small volumetric element (voxel) as the functional unit and analytically scales its environmental impact to arbitrary parts. The framework is conceptually applicable to a broad range of AM processes, but is instantiated and validated here for laser powder bed fusion (LPBF) solid parts under a gate-to-gate system boundary. First, voxel-level operating times are predicted from LPBF process parameters such as laser power, scan speed, hatch distance, and layer thickness, and the corresponding voxel-level environmental impacts are quantified using process energy use and gas consumption. Second, physics-informed regression models are constructed to link process parameters to voxel-level laser exposure and recoating times, enabling data-efficient prediction from a limited set of simulated voxel cases rather than extensive experimental campaigns. Finally, closed-form voxel-to-part scaling relations based on part volume and build height are used to compute part-level environmental impacts, defining a geometry-agnostic voxel-to-part architecture that does not require retraining when new solid geometries are introduced. In the LPBF case study, 225 voxel cases are generated and used to train and validate the voxel-level models, and six solid validation parts with different base areas and heights are employed to evaluate part-level prediction performance. The resulting mean absolute percentage error in part-level environmental impact ranges from 0.04% to 2.32% across variations in part geometry, and remains within 1.66–2.05% when the scan speed is systematically varied while other parameters are fixed. These results demonstrate that the proposed VLCA framework can provide accurate, interpretable, and data-efficient environmental impact predictions suitable for early-stage LPBF design and process parameter exploration. In addition, the voxel-based structure offers a generalizable foundation for extending the framework to additional impact indicators and other AM processes, supporting more responsive and scalable environmental decision-making in additive manufacturing. Major: Department of Mechanical Engineering 2026-01-31T15:00:00Z