https://scholarworks.unist.ac.kr/handle/201301/66 2026-04-08T00:32:34Z https://scholarworks.unist.ac.kr/handle/201301/91014 Title: Development of a Confidence Estimation based Domain-Adaptive Industrial AI Framework Author(s): Seo, Junyoung Abstract: This study presents CONDA-AI, a confidence-estimation–driven domain-adaptive industrial AI framework designed to maintain reliable decision-making under complex distribution shifts. Unlike conventional industrial predictors that treat confidence as a post-hoc statistic, CONDA-AI formulates confidence as a control variable that determines whether the system should act autonomously, defer decisions, or trigger conservative mitigation under risk. The framework separates industrial shifts into long time-constant domain drift and short time-constant environmental volatility and addresses them through two complementary modules: Module 1 constructs a domain-stable feature space with structured confidence to prevent confident-but-wrong behavior under drift, while Module 2 produces robust, physically meaningful indicators and prioritized control targets via physics-guided mapping and constrained residual learning. CONDA-AI is validated across three industrially grounded domains, including high-throughput continuous casting defect forecasting for selective surface treatment, large scale NCM precursor co-precipitation monitoring under long-horizon drift, and rapid-change acoustic environments requiring real-time robustness. Across these case studies, the proposed framework improves stability of confidence–correctness alignment, supports risk–coverage operating policies, and enhances operational readiness through integrated evaluation and governance protocols. The results suggest that reliability in industrial AI is best achieved by co-designing data handling, shift-aware learning, and decision policies, enabling scalable deployment of trustworthy AI systems across heterogeneous plants and evolving operational conditions. Major: Department of Mechanical Engineering 2026-01-31T15:00:00Z https://scholarworks.unist.ac.kr/handle/201301/91013 Title: Control system of cryogenic substances for rapid temperature control and transdermal drug delivery Author(s): Son, Hyunjoon Abstract: Medical cooling has evolved beyond its conventional role in cryoablation to encompass a broad range of therapeutic applications, including anti-inflammatory treatment, pain relief, neural blockade, cryo- lipolysis, and surface protection during energy-based thermal procedures. As these applications expand, precise control of cooling temperature has become increasingly critical, since inadequate or excessive cooling can result in unintended cellular damage, treatment failure, tissue discoloration, or irreversible functional impairment. Despite their widespread use, existing medical cooling technologies suffer from fundamental limitations. Many rely on mechanically actuated modulation strategies, which inherently constrain system responsiveness and stability. As a result, cooling is often operated under open-loop control schemes, with jet temperature predominantly regulated indirectly through pressure adjustment. These constraints hinder reliable and reproducible temperature control, thereby limiting the safe expansion of cooling-based therapies into emerging clinical domains that demand higher precision and adaptability. In this doctoral dissertation, an enthalpy control–based cooling technology utilizing a carbon dioxide (CO₂) two-phase jet is proposed to achieve both high cooling performance and precise temperature regulation. By directly controlling the thermodynamic state of the cryogen prior to expansion, the proposed system overcomes the limitations of pressure-dependent and mechanically modulated cooling approaches. The cooling mechanism and controllability of the CO₂ two-phase jet are systematically investigated through high-speed imaging and quantitative heat transfer characterization. These analyses elucidate the relationship between pre-expansion enthalpy, two-phase jet composition, and cooling performance, enabling the development of a robust and precise cooling control strategy. Building upon this capability, foundational studies on subcutaneous temperature modulation are conducted to demonstrate the feasibility of extending the proposed control framework toward medical cooling applications requiring spatially and temporally regulated thermal management. Furthermore, leveraging the precise temperature control and supersonic characteristics of the CO₂ jet, a novel transdermal drug delivery approach is explored. Particle dynamics and penetration behavior are analyzed through combined theoretical, in vitro, and in vivo investigations, confirming enhanced delivery trends and therapeutic efficacy. Collectively, this dissertation establishes a unified enthalpy- based CO₂ two-phase jet control framework, providing both the physical understanding and practical control strategies required for next-generation medical cooling and cooling-assisted therapeutic technologies. Major: Department of Mechanical Engineering 2026-01-31T15:00:00Z https://scholarworks.unist.ac.kr/handle/201301/91012 Title: Smart Integration of Process Monitoring in Composite Manufacturing Using Electromechanical Behavior and Artificial Intelligence Author(s): Lee, Dahun Abstract: As carbon neutrality becomes a global imperative, the demand for advanced technologies that mitigate greenhouse gas emissions has increased significantly. Among various strategies, reducing structural weight in key sectors such as transportation, aerospace, and renewable energy has emerged as a highly effective method to reduce carbon dioxide (CO₂) emissions. This shift has accelerated the adoption of lightweight materials with high specific mechanical properties. In this context, carbon fiber- reinforced plastics (CFRPs) have attracted considerable attention due to their superior strength -to- weight ratio, excellent corrosion resistance, and design flexibility—qualities that make them highly suitable for high-performance structural applications. With growing environmental and performance demands, CFRPs are increasingly used in the manufacturing of aircraft fuselages, automotive body panels, wind turbine blades, and marine structures. However, as applications scale up, so does the complexity of their production, particularly in ensuring uniform resin distribution and consistent quality. Among the various manufacturing techniques, resin transfer molding (RTM) stands out as a widely adopted process due to its capability to produce large, geometrically complex components with high fiber volume fraction, low void content, and fine surface finishes. Nonetheless, RTM faces persistent challenges, including the formation of internal defects such as dry spots, voids, and incomplete resin impregnation. These defects can significantly degrade the mechanical properties and long-term reliability of the final products. Traditionally, ensuring product quality in manufacturing has relied on destructive testing or post- process inspections, both of which are time-consuming, labor-intensive, and costly. These methods also limit real-time corrective actions. As a result, in-situ process monitoring techniques have been studied using several types of sensors. Among these, self-sensing approaches, based on the inherent electromechanical behavior of carbon fibers, have been gaining popularity due to their cost- effectiveness and ease of sensor installation. However, most previous studies remain limited to small- scale specimens and focus primarily on identifying the current status of resin flow (e.g., flow-front tracking or dry-spot detection), without extending toward predictive or quantitative quality evaluation. With the growing trend of automation and active control in manufacturing, quality control strategies that provide real-time feedback and construct digital twins are becoming increasingly important. Recent advances in sensor technologies, combined with artificial intelligence (AI) and machine learning (ML), have opened new possibilities for intelligent, adaptive manufacturing frameworks capable of monitoring, predicting, and optimizing composite production processes in real time . This dissertation proposes a comprehensive, data-driven approach to improve RTM-based CFRP manufacturing through the integration of electrical resistance-based sensing and AI-driven analysis. The research is structured around three key areas: (1) real-time process parameter decoupling, (2) flow-front identification and prediction, and (3) mechanical property estimation using process signals. Each stage is interconnected, forming a robust framework for real-time quality assurance and intelligent process control. The first part of the study presents the development of a multi-stage monitoring system using electrical resistance measurements. Electrodes embedded on the fiber surface capture resistance variations during resin infusion, enabling accurate, in-situ detection of resin arrival and flow progression. The data were analyzed using two types of decoupled strategies involving classification and quantitative comparison. The system demonstrated the ability to detect resin arrival within 10 seconds and achieved high agreement with simulation results, with flow front prediction errors maintained within 4%. The approach allowed for visualization of resin movement in both the in -plane and through-thickness directions, enabling early detection of flow irregularities and potential defects. Building upon the monitored data, the second phase applies artificial intelligence to predict flow- front evolution and configurations under different scenarios. The electromechanical behavior captured by electrical resistance data was used to classify flow patterns and generate flow-front configurations. A robust framework was developed, combining a decision tree algorithm for real-time scenario classification and a generative adversarial network (GAN) to produce spatially accurate flow-front images. The model successfully identified 17 distinct scenarios across three mold regions and effectively captured complex resin behaviors such as racetracking—without requiring explicit permeability data. This approach significantly reduced computational time and improved prediction robustness under varying process conditions. In the third part, the study investigates the correlation between real-time monitoring data and the mechanical properties of the final CFRP products. Full-scale CFRP panels (400 mm × 600 mm) were manufactured and segmented into 26 test specimens. Mechanical performance was assessed through three-point bending tests, and results were correlated with a novel monitoring index derived from electrical resistance data. This index accounted for impregnation quality based on increase in electrical resistance during the infusion and curing. The model achieved low root-mean-square error (RMSE) and mean absolute error (MAE), demonstrating high accuracy in estimating mechanical properties directly from process data. Microscopic analysis of specimens validated the influence of internal defects on mechanical behavior and supported the reliability of the monitoring index. In conclusion, this study introduces an integrated framework that combines real-time sensing, data interpretation, and artificial intelligence to enhance the quality, efficiency, and sustainability of RTM- based CFRP manufacturing. Three key contributions of this research are as follows: the development of a real-time electrical resistance monitoring system for full-scale RTM processes, AI-based flow front prediction using hybrid models for enhanced spatial and temporal accuracy, a quantitative monitoring index for in-situ mechanical property estimation, enabling non-destructive quality control. Collectively, these innovations contribute to the advancement of intelligent composite manufacturing and offer a practical pathway toward digitalized, real-time process optimization. The findings have broad implications for industries such as aerospace, automotive, and renewable energy, where high-performance composites are essential. This work supports the transition to smarter, more sustainable manufacturing systems and underscores the importance of data-driven methodologies in next-generation composite production. Major: Department of Mechanical Engineering 2026-01-31T15:00:00Z https://scholarworks.unist.ac.kr/handle/201301/91011 Title: Holistic Predictive Maintenance for CFRPs under Multi-modal Loading Using Electromechanical Characterization Author(s): Oh, So Young Abstract: This study aims to develop a holistic predictive approach for carbon-fiber-reinforced polymers (CFRPs). This includes real-time structural health monitoring (SHM) and prognostics and health management (PHM) measures for securing structural integrity and saving maintenance expenses. SHM involves the real-time identification of current health status of the material, and PHM enables prediction of health indicators and remaining useful life (RUL). Based on the idea that the conductive network inherent in CFRPs can be modeled as electrical circuits, structural deformation and damages can be analyzed by means of electrical resistance of the material, which can be monitored in real-time. By integrating the signal simultaneously measured under multi-modal loadings with optimal data processing approaches, the health status of CFRPs have been characterized. The keynotes of the thesis are shown below: 1. Demonstrate quantitative relationships between CFRP damage and electromechanical responses 2. Establish general decision-making SHM/PHM algorithms for enhanced performance and output practicality 3. Predict RUL of CFRP structures with a high level of reliability Starting from the basic electrical circuit model, damages in CFRPs were quantitatively analyzed. The electromechanically represented damages were digitized and presented as health indices, which can estimate the degradation in mechanical properties. Existing self-sensing studies only exploit the rough relationship between electrical resistance and mechanical strain which results in poor reproducibility and confined to abstract damage information. To address these intrinsic limitations, machine learning approaches were involved to extract detailed outcomes, such as location, type, and severity of random damages. The algorithm has also been extended into 3-dimensional structures under multi-modal loading scenarios targeting industrial in-service applications. One of the most crucial parts to complete predictive maintenance is reliable RUL prediction. A hybrid algorithm was newly proposed by integrating statistics-based and machine-learning approaches to overcome existing time-consuming and knowledge-dependent tasks. In addition, a material degradation model which can precisely explain electromechanical responses during fatigue fracture was suggested. The ultimate goal of this dissertation is ‘extracting maximal information from minimal signal.’ Towards efficient operation and economic maintenance of CFRP structures, electromechanical characterization with a tight collaboration with mechanical analyses has been performed throughout the study. Major: Department of Mechanical Engineering 2026-01-31T15:00:00Z