https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/262 2026-07-24T13:56:19Z 2026-07-24T13:56:19Z Hwang, Hyoseong Kang, Jonggu Joe, Inwhee https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/218032 2026-07-06T04:30:19Z 2026-11-01T00:00:00Z

Title: Integrating domain knowledge and large language models for automatic generation of function block-based PLC logic in maritime systems Authors: Hwang, Hyoseong; Kang, Jonggu; Joe, Inwhee Abstract: The growing complexity of shipboard automation and the expanding scale of IEC 61131-3 control logic have increased the demand for systematic engineering workflows. Traditional practices rely on manual interpretation of Function Design Specifications and expert-driven mapping of field I/O signals to vendor-specific Function Blocks (FBs), resulting in variability in logic quality and significant development effort. While recent advances in large language models (LLMs) have improved general code generation, existing approaches remain insufficient for industrial control systems because they overlook FB semantics, hardware constraints, and ship-specific structural relationships. This study presents a domain-aware framework that integrates LLMs with retrieval-augmented domain knowledge to automatically generate FB-based PLC logic for maritime systems. The framework embeds FB manuals and normalized I/O descriptions into vector stores and incorporates graph-based relational knowledge capturing FB selection rules, cooperation patterns, and symmetry and grouping structures in I/O configurations. Using these structured representations, the pipeline performs FB type selection, instance estimation, parameter assignment, and I/O mapping, after which a rule-based generator produces IEC 61131-3 compliant Structured Text to ensure syntactic and structural correctness. Experiments using control narratives, FB manuals, and I/O datasets from operational ship projects show that relational knowledge significantly improves planning accuracy, enhancing FB identification, instance estimation, and I/O mapping across model scales. The framework produces deployable FB-based logic without model fine-tuning, relying solely on in-context knowledge injection. These results demonstrate that domain-grounded LLM workflows can reduce engineering variability and provide a practical, scalable approach for automating control-logic development in maritime systems.

2026-11-01T00:00:00Z Lim, Hyojin Oh, Jaehoon Kim, Tae Hyun Lee, Juncheol Chung, Jae Ho Lee, Dong Keon https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/218680 2026-07-09T02:30:26Z 2026-07-01T00:00:00Z

Title: Deep Learning-Based Anatomical Segmentation of the Foot and Ankle: A Multi-View Radiograph Approach Authors: Lim, Hyojin; Oh, Jaehoon; Kim, Tae Hyun; Lee, Juncheol; Chung, Jae Ho; Lee, Dong Keon Abstract: Purpose: We aimed to develop deep learning models that can identify and separate the shapes of 14 bones in the foot and ankle using multi-view radiographs and predict their masks. Materials and Methods: We retrospectively collected 273 radiographs from 99 patients with anatomically normal feet, including anteroposterior (AP), oblique (OBL), and lateral (LAT) views, obtained between January 2020 and December 2021. In each view, 14 bones were segmented using AP and OBL radiographs and 6 bones using LAT radiographs. Ground truth masks were manually annotated by two radiology technologists and reviewed by an emergency medicine physician. Two deep learning models, a fully convolutional network (FCN) with ResNet-50 and DeepLabv3 with ResNet-50, were independently fine-tuned for semantic segmentation and evaluated using five-fold cross-validation. Results: In the AP and OBL views, both models attained mean intersection over union (mIoU) values ranging from 0.899 to 0.975 and from 0.875 to 0.978, respectively. In the LAT view, mIoU values varied from 0.926 to 0.976 for FCN-ResNet-50 and from 0.872 to 0.961 for DeepLabv3. FCN-ResNet-50 achieved slightly higher mIoU values than DeepLabv3, with statistically significant differences identified between the two models in the overall OBL view and across the 14 specific bones (p<0.05). Conclusion: The FCN-ResNet-50 and DeepLabv3 models could be effective in automatically segmenting foot and ankle bone structures using multi-view radiographs.

2026-07-01T00:00:00Z 김태욱 https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/218888 2026-07-23T16:33:22Z 2026-06-24T00:00:00Z

Title: 한국어 LLM-as-a-Judge를 위한 루브릭 전략 비교 Authors: 김태욱

2026-06-24T00:00:00Z 김태욱 https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/218886 2026-07-23T16:33:19Z 2026-06-24T00:00:00Z

Title: LLM 내부 지식 분석을 위한 표현 단위 비교: Neurons vs. SAEs Authors: 김태욱

2026-06-24T00:00:00Z