성균관대학교 스마트팩토리융합학과

본문 바로가기
주메뉴 바로가기
서브메뉴 바로가기

일반대학원 스마트팩토리융합학과
(석사 30명, 박사 5명)

홈페이지 방문을 환영합니다.

Read More

2016년 세계최초 스마트팩토리융합학과 신설

글로벌 선도 미래인재 양성, 지역수요 맞춤형 전문인력 양성, 산학협력 친화형 인재양성

Read More

지산학 연계·협력을 통한 선순환 체계 구축

기업수요기반 캡스톤디자인, 현장실습 운영

Read More

미래성장동력 육성을 통한 글로벌 미래변화 선도

2025 英QS 세계 대학순위 국내 톱5

Read More

제조분야 R&D 활성화로 기술혁신 선도

국내 최고수준 연구실적, 글로벌 연구역량 확보

Read More

Notice

News

[Best Paper Award] ICCSA 2026 - 김정훈 석사과정
2026-07-03
우수논문상(Best Paper Award) 수상 - 김정훈 석사과정 ICCSA 2026 The 26th International Conference on Computational Science and Its Applications University of Minho, Braga, Portugal (June 30 - July 3, 2026) "DeOP-Light: Analyzing the Fine-tuning Paradox in CLIP Visual Encoder Compression for Open-Vocabulary Semantic Segmentation" doi: https://doi.org/10.1007/978-3-032-30488-9_12 Abstract. Open-vocabulary semantic segmentation (OVSS) relies on contrastive language-image pre-training (CLIP) visual encoders to clas- sify unseen categories at inference time. Compressing these encoders for efficient deployment is practically important, yet the impact on the crit- ical seen/unseen class distinction has not been studied. We present a diagnostic empirical study rather than a new architectural proposal of CLIP visual encoder compression within the decoupled one-pass (DeOP), applying floating-point (FP)16 quantization and low-rank adaptation (LoRA) on COCO-Stuff and Pascal VOC. We uncover a consistent asym- metric degradation: fine-tuning (FT) the CLIP visual encoder recovers seen class performance but simultaneously degrades unseen class perfor- mance, emerging within the first 1,000 iterations. By comparing against an uncompressed FT baseline, we show that FT itself not compression is the primary driver of this degradation. FT-free compression best pre- serves unseen performance, and post-hoc weight restoration partially re- covers it without retraining. Our findings provide practical guidelines for model compression in resource-constrained deployment and highlight the need for seen/unseen-aware evaluation in OVSS. 우수논문상을 수상한 논문 외에 3편의 논문이 발표되었습니다. 논문발표는 이지은 박사연구원, 김소연 석사연구원이 진행하였습니다. 1. C-MTSS: Compact Multi-teacher Single-Student Framework for Real-Time Multimodal Industrial Anomaly Detection (2026-06-30) doi: https://doi.org/10.1007/978-3-032-30488-9_21 2. DepthGate: Confidence-Gated Depth Verification for Multimodal Industrial Anomaly Detection (2026-06-30) doi: https://doi.org/10.1007/978-3-032-30488-9_7 3. GPS: GlobalCLIP-PatchCore-SAM Based Zero-Shot Anomaly Detection and Localization in Smart Manufacturing (2026-06-30) doi: https://doi.org/10.1007/978-3-032-30488-9_28

석사수료 윤태휘, ESCI 논문지(Elsevier - Results in Engineering/Q1, 96.9%) 게재
2026-06-03
석사수료 윤태휘(지도교수: 정종필)의 연구(Bio–SFDA: Physics–Guided Multimodal Source–Free Domain Adaptation for Reliable Bearing Fault Diagnosis)가 Elsevier - Results in Engineering Jounrnal(Impact Factor: 7.9 (2024), 5-Year JIF: 7.4)에 게재됐다. https://www.sciencedirect.com/science/article/pii/S259012302602133X / https://doi.org/10.1016/j.rineng.2026.111105 논문요약 - We propose Bio–SFDA, a physics-guided, bio-inspired multimodal framework for strict source-free domain adaptation in bearing fault diagnosis. A dual 1D vibration / 2D spectrogram encoder is adapted via an EMA teacher–student pipeline equipped with three lightweight but crucial modules: WBC, a signal-hygiene guard that filters corrupted or low-quality target windows; EAGLE, a physics-guided reliability functional that exploits BPFO/BPFI/BSF and impulsive-band activity to weight and gate pseudo-labels; and SPIDER, a bounded multi-metric controller that maps online trends in accuracy, macro-F1, mAP, loss, and ECE to conservative updates of learning rate, momentum, augmentation strength, and acceptance thresholds. These bio-inspired operators white-blood-cell filtering, eagle-like spectral hunting, and spider-like tension control—form a closed reliability–control loop that improves pseudo-label quality, mitigates confirmation bias, and preserves fault mechanics under drift. On strict CWRU → PU, Bio–SFDA attains 99.0% accuracy, 98.5% macro-F1, and 99.0% mAP with markedly lower NLL and ECE than source-only and recent SFDA/TTA baselines, while running in real time on a single commodity GPU. Cross-modal (1D↔2D) experiments confirm robustness to sensing mismatch, and ablations highlight EAGLE as the main driver of performance with SPIDER ensuring stable convergence, yielding reliable and deployment-ready adaptation for industrial condition monitoring.

석사과정 이근호, SCIE 논문지(MDPI Sensors/Q2) 게재
2026-05-13
석사과정 이근호 학생(지도교수 : 정종필)의 연구(An Improved YOLOv9 Based Object Detection with Attention Mechanism for Personal Protective Equipment)가 MDPI Sensors(Impact Factor: 3.5 (2024); 5-Year Impact Factor: 3.7 (2024))에 게재됐다. https://www.mdpi.com/1424-8220/26/10/3058 or https://doi.org/10.3390/s26103058 논문요약 - Industrial sites pose numerous hazards where unexpected accidents can occur at any time, and personal protective equipment (PPE) is a primary safeguard for worker safety. In this study, PPE specifically refers to safety helmets, safety shoes, and safety gloves. Manual verification of PPE usage is infeasible in environments with many workers, motivating automated detection. This study proposes a method that integrates the Convolutional Block Attention Module (CBAM) exclusively into the training-only auxiliary reversible branch of YOLOv9’s Programmable Gradient Information (PGI) architecture. The proposed CBAMLinear module enhances gradient information during training while introducing zero additional computational overhead at inference, as the entire auxiliary branch is removed. The proposed YOLOv9 with CBAMLinear achieved consistent mAP@0.5:0.95 gains of 0.005–0.007 over the baseline for the three larger model variants, while maintaining identical inference-time parameters and FLOPs. In industrial safety, even modest performance gains can directly contribute to accident prevention by reducing false positives and false negatives, making this approach well suited for real-time safety management systems in industrial settings.

석사과정 임재훈, SCIE 논문지(Journal of King Saud University - Computer and Information Sciences/Q1, 88.2%) 게재
2026-04-29
석사과정 임재훈 학생(지도교수: 정종필)의 연구(mLLM-CRD: multi-modal LLM based context-aware rule detection using LoRA-integrated attention for worker safety monitoring)가 Journal of King Saud University - Computer and Information Sciences (Impact Factor: 6.1 (2024); 5-year Journal Impact Factor: 5.7 (2024))에 게재됐다. https://link.springer.com/article/10.1007/s44443-026-00572-2 / https://doi.org/10.1007/s44443-026-00572-2 논문요약 - Large Vision-Language Models (VLMs), which jointly process visual and textual information, have demonstrated strong performance in general scene understanding; however, their application to safety-critical industrial environments remains limited by the need for precise semantic reasoning, interpretability, and computational efficiency. Conventional worker safety monitoring systems typically rely on rigid object detection pipelines, which are prone to error propagation and provide limited explanatory capability. This paper presents mLLM-CRD, a multimodal framework that formulates safety compliance assessment as an image-to-language reasoning task. Built upon Bootstrapping Language-Image Pre-training (BLIP-2), the proposed approach applies a joint parameter-efficient fine-tuning strategy using Low-Rank Adaptation (LoRA), which is integrated into both the Querying Transformer (Q-Former) and the encoder-decoder-based language model while keeping the vision encoder frozen, thereby enabling task adaptation with minimal additional parameters. To support rule-aware semantic reasoning, we construct a scenario-based question-answer (QA) dataset that encodes safety compliance conditions as structured language queries derived from industrial safety rules. Experimental results show that mLLM-CRD consistently outperforms zero-shot VLM baselines in both classification accuracy and natural language generation quality, particularly under class-imbalanced conditions. These findings suggest that unified multimodal reasoning can provide interpretable, language-based assessments of worker safety compliance, offering a practical alternative to conventional multi-stage detection pipelines in real-world industrial settings.

석사과정 김은호, SCIE 논문지(Scientific Reports/Q1, 82%) 게재
2026-04-13
석사과정 김은호 학생(지도교수: 정종필)의 연구(CLARIS: Control-based language-guided realistic imperfection synthesis)가 Scientific Reports (Impact Factor: 3.9 (2024); 5-Year Eigenfactor: 4.3)에 게재됐다. https://www.nature.com/articles/s41598-026-46066-z / https://doi.org/10.1038/s41598-026-46066-z 논문요약 - Automated vision inspection is vital in modern manufacturing, but advanced processes with high yield rates cause a severe data imbalance: abundant normal data and scarce defective data. To overcome this, we propose CLARIS (Control-based Language-guided Realistic Imperfection Synthesis), a novel framework combining natural language semantic flexibility with 3D structural constraints to generate physically consistent, high-quality defect images. CLARIS utilizes a Vision-Language Model (VLM) to interpret user instructions and input images, dynamically generating tailored text prompts and defect masks. Subsequently, ControlNet ensures the synthesized defects adhere to the object’s physical shape and surface curvature by explicitly applying normal maps as constraints. Furthermore, Textual Inversion (TI) and Low-Rank Adaptation (LoRA) are employed to efficiently learn and reflect the unique visual characteristics of specific defects using minimal parameters. Evaluated on the 15 categories of the MVTec Anomaly Detection (MVTec AD) dataset, the framework achieved an average Kernel Inception Distance (KID) of 11.07, Inception Score (IS) of 1.63, and intra-cluster pairwise LPIPS distance (IC-LPIPS) of 0.27.

석사과정 신동하, SCIE 논문지(IEEE Access/Q2, 64.1%) 게재
2026-04-10
석사과정 신동하 학생(지도교수: 정종필)의 연구(Precision Design Change Detection and Manufacturing Cost Prediction for Plastic Products Using a Hybrid Intelligent Agent Based on B-Rep Parametric Sampling and LightGBM-RAG)가 IEEE Access (Impact Factor: 3.6 (2024); 5-Year Eigenfactor: 0.34597)에 게재됐다. https://ieeexplore.ieee.org/document/11478273 / https://doi.org/10.1109/ACCESS.2026.3682507 논문요약 - To maximize intelligent manufacturing efficiency under the Industry 4.0 paradigm, this study addresses the critical challenge of precision design change detection and manufacturing cost prediction for plastic products by leveraging exact Boundary Representation (B-Rep) geometric information and data-driven learning models. Unlike conventional mesh-based approaches that suffer from tessellation approximation errors, the proposed method performs direct parametric sampling on NURBS surfaces from STP (CAD) files. This enables the noise-free, micrometer-level detection of fine shape displacements and the rigorous extraction of 12 key engineering feature vectors relevant to machining decision-making. Based on the extracted features, a hierarchical LightGBM-based learning structure utilizing a leaf-wise growth strategy is proposed to classify seven core manufacturing processes, including NC machining and EDM, and to predict the associated processing time and costs. Furthermore, an LLM- and RAG-based explainable AI (XAI) framework is incorporated to provide natural language interpretations of the engineering rationale underlying the numerical predictions. To ensure model robustness and prevent overfitting, the framework was evaluated on a dataset of 15,120 change patches using 5-fold cross-validation, with cost labels validated against real-world Enterprise Resource Planning (ERP) data. Experimental results demonstrate a process classification accuracy of 94.2% and a Mean Absolute Percentage Error (MAPE) of 7.8% in cost prediction, significantly outperforming conventional baselines such as Linear Regression, Random Forest, and XGBoost. The proposed approach reduces the design change analysis lead time from several days to under 3 minutes, demonstrating its effectiveness for real-time decision support in intelligent manufacturing environments.

RISE 사업으로 지역 제조 혁신 이끌 미래 인재 양성... 학부-대학원 융합 연구 결실
2026-01-22
RISE 사업으로 지역 제조 혁신 이끌 미래 인재 양성... 학부-대학원 융합 연구 결실 - 캡스톤디자인 연계로 기업 난제 해결, 학부생 참여 논문 세계적 권위지 게재 - ‘설명하는 AI’ 기술 개발로 지산학 협력 모델의 새로운 이정표 제시 스마트팩토리융합학과 정종필 교수 연구팀이 지역 혁신 중심 대학지원체계(RISE) 사업의 일환으로 수행된 융합 연구를 통해, 산업 현장의 제조 결함을 자연어로 설명하는 차세대 멀티모달 인공지능(AI) 기술을 개발했다. 이번 성과는 단순한 기술 개발을 넘어, 학부생과 대학원생이 머리를 맞대어 지역 기업의 실제 고민을 해결한 교육적 혁신 사례로 주목받고 있다. 이번 연구는 ‘2025 스마트팩토리 캡스톤디자인’ 수업에서 시작되었다. 캡스톤디자인이란 학생들이 배운 이론을 바탕으로 실제 산업 현장의 문제를 해결하는 작품을 기획·제작하는 실무형 교육 과정이다. 정종필 교수의 지도 아래 하태원 석사과정생과 황채선 학부생은 팀을 이루어, 대학원과 학부의 경계를 허문 긴밀한 협력을 진행했다. 이는 성균관대학교가 지향하는 ‘지산학(지자체·산업체·대학) 협력’의 대표적인 성공 모델로 평가받는다.

석사과정 하태원, SCIE 논문지(Springer - Journal of Intelligent Manufacturing/Q1, 86.6%) 게재
2026-01-19
석사과정 하태원 학생(지도교수: 정종필)의 연구(CLIP-MDC: CLIP encoder based multimodal defect classification with synthetic anomaly generation for real-time surface defect detection)가 Springer - Journal of Intelligent Manufacturing (Impact Factor: 7.4 (2024))에 게재됐다. https://doi.org/10.1007/s10845-025-02773-4 논문요약 - In this study, using various text prompts that combine objects and defect types, we establish a semantic space linking images and texts, enabling explainable defect predictions using natural language. We introduce contrastive language–image pre-training-based multimodal defect classification (CLIP-MDC), a framework designed for multimodal defect detection and classification in smart manufacturing. The model integrates a lightweight backbone network with contrastive language–image pre-training (CLIP) encoders to perform both pixel-level anomaly segmentation and image-level defect classification effectively in supervised and weakly supervised settings. Additionally, we incorporate a Perlin noise-based synthetic anomaly generation technique to facilitate learning in environments with limited labeled data, and the dual prediction architecture enables accurate simultaneous inference of defect location and type. In experiments on the MVTec AD and KSDD2 datasets, the model achieved outstanding performance with an area under the receiver operating characteristic curve (AUROC) of 99.9%, an area under the per-region overlap curve (AUPRO) of 98.6%, a pixel-level AUROC (P-AUROC) of 99.9%, and an average precision for localization of 87.6%. It also demonstrated real-time capability, registering an average inference speed of 6.6ms on an A100 GPU. CLIP–MDC uses a semantic-based multimodal learning framework that combines visual and linguistic information to deliver accuracy, explainability, generalization, and real-time efficiency in defect detection, making it a practical and scalable solution for industrial defect analysis in real-world manufacturing environments.

[Best Paper Award] ISAAC & ICACT & ICKAI 2025 - 강건욱 박사과정
2025-12-23
우수논문상(Best Paper Award) 수상 - 강건욱 박사과정 ISAAC & ICACT & ICKAI 2025 -The 13th International Symposium on Advanced and Applied Convergence (ISAAC) -The 13th International Conference on Advanced Culture Technology (ICACT) -The 5th International Conference on Knowledge and Artificial Intelligence (ICKAI) Incheon, Korea (Nov 13 - 15, 2025) "Korea Virtual Railway Connection (KVRC): Immersive XR Experience and Physical AI for Inter-Korean Railway Reconnection"

성균관대, ‘산업 현장 밀착형 AI 전문가’ 배출…K-Digital Training 성료
2025-12-16
‘Industrial AI Expert Projects with Deep Learning’ 과정, 수료율 100% 달성 제조·의료·농업 등 실제 산업 데이터 기반 프로젝트 수행으로 실무 역량 입증 성균관대학교(총장 유지범)는 지난 13일 경기도 수원시 자연과학캠퍼스 제1공학관에서 ‘K-Digital Training(KDT): Industrial AI Expert Projects with Deep Learning’ 과정의 최종발표회 및 수료식을 성황리에 개최했다고 밝혔다. 이번 행사는 4차 산업혁명 시대를 이끌어갈 실무형 인공지능(AI) 인재를 양성하고, 그간의 교육 성과를 공유하기 위해 마련됐다. 이번 교육 과정은 성균관대 ICT명품인재양성사업단과 스마트팩토리융합학과, 산학협력단, 그리고 RISE사업단이 공동 주관했으며, 과학기술정보통신부 정보통신기획평가원(IITP)과 고용노동부 직업능력평가원이 지원하는 다부처 협력 사업으로 진행됐다. 이날 행사에는 고용노동부 및 직업능력심사평가원 주요 관계자를 비롯 최승진, 추현승, 정종필, 이채규 교수 등 성균관대의 AI 분야 주요 교수진과 교육 수료생들이 참석해 자리를 빛냈다. 본 과정은 지난 7월부터 약 5개월간 진행된 장기 교육 프로그램으로, 제조, 의료, 농업 등 다양한 산업 현장에서 급증하는 디지털 전환(DX) 수요에 대응하기 위해 기획됐다. 교육의 핵심 목표는 단순한 이론 학습을 넘어, AI 및 딥러닝(Deep Learning) 기술을 활용해 실제 산업 현장의 데이터를 분석하고 구체적인 문제를 해결하는 ‘실무형 고급 인재’를 양성하는 데 있다. 특히 이번 과정은 재직자들을 대상으로 진행됐음에도 불구하고, 놀라운 참여 열기를 보였다. 전체 수강생 30명 전원이 출석률 80% 이상 및 최종 프로젝트 발표 참여라는 까다로운 수료 기준을 모두 충족해 ‘수료율 100%’라는 괄목할 만한 성과를 달성했다. 또한, 업무와 학업을 병행하는 어려움 속에서도 평균 출석률 98%를 기록하며 교육에 대한 높은 만족도와 몰입도를 입증했다. 이날 최종발표회에서는 총 10개 팀이 지난 5개월간 수행한 프로젝트 결과물을 발표했다. 수료생들은 기업이 실제로 직면한 과제를 주제로 선정하고, 문제 정의부터 데이터 전처리, 모델 구현에 이르는 전 과정을 주도적으로 수행하는 프로젝트 기반 학습(PBL)을 통해 현장 적응력을 극대화했다. 참가 팀들은 △이미지 분석에 특화된 CNN △시계열 데이터 처리에 강한 RNN △최신 자연어 처리 모델인 Transformer 등 딥러닝 핵심 기술을 적재적소에 활용했으며, 고성능 GPU 클라우드 환경을 통해 수준 높은 결과물을 도출해 심사위원들의 이목을 집중시켰다. 심사위원단은 각 팀의 발표뿐만 아니라 데모 시연을 직접 참관하며 기술의 완성도와 실무 투입 가능성을 면밀히 검증했다. 우수한 성과를 보인 프로젝트 팀에게는 시상이 이루어졌으며, 과정을 성공적으로 마친 수료생 전원에게는 성균관대학교 명의의 수료증이 수여됐다. 행사를 총괄한 정종필 성균관대 교수는 “이번 K-Digital Training 과정은 단순한 지식 전달을 넘어, 산업 현장에서 즉시 활용 가능한 실전 문제 해결 역량을 확보하는 데 중점을 다”며 “앞으로도 정부 및 유관기관과의 긴밀한 협력을 통해 AI 전문 인재 생태계를 탄탄히 구축하고, 수료생들의 커리어 개발과 고용 연계까지 적극적으로 지원하겠다”고 포부를 밝혔다. □ 한편, 성균관대는 이번 교육 과정의 성공적인 운영 성과를 바탕으로 스마트팩토리 분야를 넘어 다양한 산업군으로 교육 및 협력 범위를 확대해 나갈 예정이다. 이를 통해 국가 산업 경쟁력을 강화하고 미래 사회를 선도하는 산학협력의 모범 사례를 지속적으로 창출해 나갈 계획이다.

prev next

Schedule

이달의 행사 오늘의 행사

Read More

read more

SKKU Movie

SKKU News & Notice

News
Notice
Press
Research

Link