SSAD Lab analyzes and models small or big data collected from the shape, materials, performance, and usage of mechanical systems — enabling fault diagnosis, prognostics, and design optimization across diverse industrial applications.
Our work bridges fundamental statistical theory with applied engineering. We develop methods in four interconnected areas, each grounded in industrial collaboration.
Big-data collection, preprocessing, and analysis paired with surrogate modeling and digital-twin design. Applied to fault diagnosis, prognostics, and affective design grounded in customer review data.
Parametric & nonparametric statistical modeling, including kernel density estimation, sequential statistical modeling, and copula-based modeling for correlated input variables.
Reliability analysis and reliability-based design optimization (RBDO) under limited information, with confidence-level analysis to handle insufficient experimental data.
Fluid–structure interaction analysis and design optimization using CFD and surrogate models — including pump, fan, and structural component optimization.
Selected case studies demonstrating data-driven methodology applied to real industrial problems — from military vehicles to home appliances, from marine engines to power plants.
군용 차량 내구 시험 최적화
Multi-objective optimization to select road surface types and speeds that minimize endurance damage error between mobility roads and durability test tracks. Pareto-optimal solutions across six objectives covering different chassis and powertrain materials.
세탁 성능 예측 정확도 개선을 위한 데이터 증강
UMAP-based dimensionality reduction forms fuzzy simplicial complexes from high-dimensional washing data; k-NN then synthesizes data in high-error regions to address data scarcity and imbalance — improving prediction accuracy across washing, rinsing, and dehydration metrics.
불확실성 정량화를 활용한 효율적 실험 설계 기법 개발
Sequential sampling pipeline that simultaneously considers prediction uncertainty (NLL loss) and data diversity (normalized distance) — minimizing washing experiments required for system development while improving predictive model generalization.
AI 기반 HRSG 성능설계 추천 알고리즘 개발
Adaptive sampling-based weight optimization for Heat Recovery Steam Generators in combined-cycle power plants, satisfying heat & balance constraints while minimizing component mass.
선박 엔진 고장 진단
Graph Attention Network (GAT) auto-encoder for HiMSEN dual-fuel engine anomaly detection — learning sensor importance via attention, modeling node interactions and connectivity, distinguishing transient signals from real anomalies for early warning.
배관 시스템 PHM 플랫폼 구축
End-to-end PHM lifecycle (design–development–decision) platform tailored to highly generalized piping systems — bridging research methodology and industrial deployment.
데이터 기반 고장 예측 관리 모델 개발
Motor fault classification (normal / demagnetization / inter-turn / bearing defect) using simulation data, plus severity estimation per fault mode — enabling predictive inspection before scheduled maintenance and reducing unnecessary teardowns.
인공지능 기반 공기압축기 이상 탐지 모델 개발
Data-driven fault detection algorithm trained on 18 turbo air compressors deployed in a shipyard — reducing unnecessary inspection and maintenance costs through targeted intervention.
히트펌프 사이클 고장 진단 및 예지 기술 개발
High-accuracy in-field fault diagnosis for dryer heat-pump systems using composite-fault experimental data, plus prognosis methodology — enabling predictive maintenance and reducing total maintenance cost.
에어솔루션 AI 소음진단 모델 MLOps 개발
Robust signal processing extracts only signals critical for noise classification; deviation analysis between normal and faulty data yields numerical and image features used to train high-accuracy noise classifiers — deployed via MLOps pipeline.
데이터 기반 모델 정확도 자동 개선 기술 개발
Tree-structured Parzen Estimator-based hyperparameter tuning for digital-twin air-conditioner cycle models — automatically optimizing target-network parameter update rates and future-reward consideration to maximize learning performance.
냉장고 고내 상태 예측을 위한 CFD-AI 융합 해석
Coupled CFD simulation and machine-learning surrogate framework for accurate prediction of internal refrigerator state — bridging high-fidelity simulation and real-time inference for digital-twin deployment.
Whether you have an industrial challenge or are a prospective student curious about data-driven design under uncertainty, we’d like to hear from you.
View All Projects Get in Touch →