Projects

Industrial Physical AI for Manufacturing

I develop and validate intelligent manufacturing systems spanning shop-floor Industrial Physical AI and network-level manufacturing decision-making.

CompletedJune 2021 – May 2023

Digital-Twin Reinforcement Learning Platform and Modular Robot for Nuclear Power Plants

Led development of a digital-twin and reinforcement-learning platform for robotic execution of high-risk nuclear maintenance tasks.

Role: Project Lead
Sponsor: Korea Hydro & Nuclear Power Co., Ltd.

Selected technical outcome

Technical publication

Challenge: Nozzle-dam replacement exposes workers to radiation and heavy loads inside a steam-generator chamber whose narrow access makes robot development and validation difficult.
Technical contribution: Developed an end-to-end high-fidelity digital twin for perception, planning, and control in confined hazardous workspaces.
Introduced DOPE–ICP pose estimation and IRM-guided PPO with a rewarded centroidal waypoint, then demonstrated the integrated system in simulation and on hardware.
My contribution: Led the system architecture and built the NVIDIA Omniverse and Isaac Sim digital-twin workflow.
Integrated perception, positioning, grasping, and motion-generation modules and conducted digital-twin and physical-robot experiments.

92%Digital-twin task success

74%Physical-system task success

CompletedSeptember 2019 – February 2024

Korea–Germany Intelligent Manufacturing Systems Laboratory

Led digital-twin and intelligent-control research for manufacturing equipment and robotic systems within a Korea–Germany collaboration.

Role: Project Lead
Sponsor: Ministry of Science and ICT, Republic of Korea

Selected technical outcome

Technical publication

Challenge: Hardware degradation causes inaccurate and oscillatory robot motion, while repair, replacement, and precise dynamics modeling are costly and time-consuming.
Technical contribution: Introduced a Real2Sim–Sim2Real pipeline combining physics-parameter tuning with domain-randomized PPO.
Transferred the learned policy to the physical manipulator without additional fine-tuning and improved control over built-in and PID methods.
My contribution: Implemented the degraded robot digital twin and physics-parameter calibration in NVIDIA Omniverse.
Conducted PPO and domain-randomization training in Isaac Sim and deployed and evaluated the policy on the physical robot.

100%Success rate (75% with PID; +25 pp)

85.2%Shorter execution time vs. PID (14.2 s vs. 96.3 s)

ActiveJune 2024 – Present

Data-Driven Methods for Future Cyber Manufacturing as a Service

Conducting foundational research on data-driven manufacturing capability representation, multi-tier supplier retrieval, and capability-aware decision methods that establish the methodological basis for resilient production networks.

Role: Postdoctoral Researcher
Sponsor: National Science Foundation · Future Manufacturing Research Grant · Award #2229260

Selected technical outcome

Technical publications

Cyber Manufacturing Mesh Network: A Manufacturing Capability-Aware Supplier Decision Support Framework Deep Unsupervised Learning-Based Supplier Selection and Ranking for Assembly Manufacturing

Challenge: Manufacturers lack fine-grained visibility into supplier capabilities, making it difficult to identify feasible suppliers, rank alternatives, and reallocate production during disruptions.
Technical contribution: Developed multimodal capability representations combining geometry, topology, materials, tolerances, and production attributes for feasible supplier matching.
Integrated supplier identification, ranking, production allocation, and disruption-driven reassignment into a capability-aware cyber manufacturing decision framework.
My contribution: Led development of the capability representation and supplier decision framework, including matching, ranking, and allocation methods.
Designed the evaluation scenarios, implemented experiments, and analyzed supplier-selection and disruption-response performance.

>95%Supplier-selection accuracy with complete requirements

>0.99Mean nDCG@k for supplier ranking

CompletedAugust 2025 – June 2026

Follow-on phase under review

OpenWerks: AI-Powered Marketplace for On-Demand Manufacturing

Completed Phase IA by developing multimodal manufacturing capability representations and LLM/RAG-based supplier-discovery methods. A follow-on proposal has been submitted and is currently under review to advance the framework toward an AI-powered marketplace for on-demand manufacturing.

Role: Postdoctoral Researcher
Sponsor: Georgia Research Alliance · Phase IA

CompletedSeptember 2020 – August 2023

Safe Smart Factory Demonstration Programs

Led two on-site deployments at legacy manufacturing facilities operated by SMEs in Gyeonggi Province, Korea: an IoT retrofit monitoring system for an aluminum stamping press and a vision-based worker hand-monitoring system for safer equipment interaction.

Role: Team Lead
Sponsor: Gyeonggi Industry–University Convergence Center

Selected technical outcome

Factory workstation, low-cost edge monitoring module, detected worker hands, and model activation map

Technical publication

Challenge: SME factories need reliable hand monitoring despite scarce labeled data, changing site conditions, constrained installation space, and limited safety-system budgets.
Technical contribution: Proposed CutMix-based synthetic augmentation using limited factory images and diverse public hand images.
Developed a modular edge safety system with legacy-alarm and MQTT integration and validated it on operating factory equipment.
My contribution: Collected and labeled site data and implemented the synthetic-augmentation and YOLOv7-tiny training pipeline.
Built the webcam–Jetson–GPIO/MQTT monitoring module and led on-site deployment and validation.

98.5%On-site detection from one minute of data (906/920; +28.7 pp vs. GA)

<$200Complete edge-monitoring hardware cost

CompletedSeptember 2019 – February 2024

Korea–Tanzania Innovative Technology and Energy Center Project

Led project administration, research execution, and external coordination for the development and field validation of an e-motorcycle and micro off-grid energy system in rural Tanzania.

Role: Project Lead
Sponsor: Ministry of Science and ICT and National Research Foundation of Korea

Selected technical outcome

Mobile energy management system connecting an electric motorcycle with rural micro off-grid villages in Tanzania

Technical publication

Challenge: Rural mobility electrification must operate with sparse charging infrastructure and the limited, variable generation capacity of village micro off-grids.
Technical contribution: Developed a mobile energy-management framework coupling e-motorcycle operation with measured micro off-grid supply and demand.
Implemented routing and charging decisions that preserve at least 30% arrival SOC and validated the model against Tanzanian field data.
My contribution: Designed the field-validation protocol and coordinated experiments with local partners.
Conducted testing and data acquisition in Tanzania and verified multi-source data integrity.

~85.5%Lower energy consumption vs. ICE (37 vs. 256 Wh/km)

~6%Simulation-to-field energy-use difference

ActiveSeptember 2025 – August 2027

Acoustic Digital Twin for Assembly Anomaly Detection

Co-developed the proposal and currently leads development of an acoustic digital twin and anomaly-detection framework for assembly operations.

Role: Co-Principal Investigator
Sponsor: Siemens Corporation

UpcomingSeptember 2026 – August 2028

Automated Inspection System for Semiconductor Manufacturing Equipment

Co-developed the proposal and will lead the technical development of an automated inspection system for semiconductor manufacturing equipment.

Role: Co-Principal Investigator
Sponsor: TSMC Arizona