01
High-dimensional spatio-temporal modeling
Deep representation learning and dynamic analysis of multi-source industrial time-series signals.
SUN YAT-SEN UNIVERSITY · SAM LAB PRESENTS
Jianshe Feng冯建设
Industrial AI · Smart Manufacturing · Shenzhen, China
From factory floors to the lectern — building Industrial AI that works in real production lines.
Associate Professor & Doctoral Supervisor, School of Advanced Manufacturing, Sun Yat-sen University; Director of SAM Lab (AI³). Previously CTO at Xinrun Fulian Digital Technology, Industrial AI product lead at Foxconn Industrial Internet, and research scientist collaborating with General Motors R&D.
CH-06 · ANOMALY DETECTED
2025 1st
PHM Data Challenge — World Champion (team)
100 +
Invention patents (US / DE / CN)
30 +
Peer-reviewed papers · 1600+ citations
7 +
Industry partners: GM, Xiaomi, Applied Materials…
CHAPTER 01 — RESEARCH
Three directions, one thread: AI that survives the factory.
02
Multimodal industrial foundation models
Time-series–vision–signal fusion for domain foundation models and embodied closed-loop decisions.
03
Intelligent process monitoring & optimization
Multimodal sensing and process optimization for semiconductors, EV, and high-end equipment.
All three directions live as 30+ funded projects at SAM Lab — real production lines, real data.
Explore them at SAM Lab ↗CHAPTER 02 — PATH
Industry first. Academia by choice.
- 2024 —Associate Professor, Sun Yat-sen University School of Advanced Manufacturing · Director, SAM Lab (AI³)
- 2020 – 2024Chief Technology Officer Xinrun Fulian Digital Technology
- 2019 – 2021Industrial AI Product Lead Foxconn Industrial Internet
- 2015 – 2020Ph.D., Mechanical Engineering IMS Center, University of Cincinnati · Advisor: Prof. Jay Lee · Research collaboration with GM R&D
- 2012 – 2015M.Sc., Mechatronics Zhejiang University
- 2008 – 2012B.Sc., Mechanical Engineering Tongji University
- RESS · 2026Reliability Engineering & System Safety
Uncertainty-informed cascaded diagnosis of compound faults via a physics-informed hypergraph framework
J. Feng, F. Qi, F. Zhu, J. Zhang, P. Gao, Y. Zhu - IEEE TII · 2023IEEE Transactions on Industrial Informatics
Profile abstract: optimization-based subset selection and summarization for profile data mining
F. Zhu, J. Feng*, M. Xie, L. Li, J. Lee - IEEE TSM · 2019IEEE Transactions on Semiconductor Manufacturing
An online virtual metrology model with sample selection for dynamic manufacturing processes with slow drift
J. Feng, X. Jia, F. Zhu, J. Moyne, J. Iskandar, J. Lee - MPCE · 2020Journal of Modern Power Systems & Clean Energy
Cross trajectory Gaussian process regression model for battery health prediction
J. Feng, X. Jia, H. Cai, F. Zhu, X. Li, J. Lee - Comput. Ind. · 2020Computers in Industry
A virtual metrology method with prediction uncertainty based on Gaussian process for CMP
H. Cai, J. Feng*, Q. Yang, W. Li, X. Li, J. Lee
CHAPTER 04 — PROJECTS
Where the instincts come from: six Ph.D.-era builds.
Current projects at SAM Lab ↗2016 – 2019
Online virtual metrology for semiconductor manufacturing ↗
Sequential Bayesian ARX with time-variant parameters tracks slow process drift; a sample-importance test screens online updates. Beat JIT and deep-learning baselines on PHM 2016 CMP data.
2016 – 2020
Battery capacity prognosis across cells
Cross-trajectory Gaussian process regression exploits correlations across cells for efficient, adaptive online capacity prediction — without inflating GPR complexity.
2017 – 2019
PHM-enabled maintenance scheduling optimization ↗
Offshore wind farm maintenance scheduling & routing that prices turbine production loss from forecasted wind; a GA solver outputs schedules, vessel routes and cost breakdowns. Validated on real 4 MW farm data.
2019 – 2020
Semi-automated FDC for semiconductor processes
Fault detection that keeps the engineer in the loop: feature-based UVA extended with parameter correlations, drift rejection, and automatic limit setting — visualized for SME decision-making.
2018 – 2019
CPS-enabled rehabilitation system
A cyber-physical gait-training system: physiological + machine data fused at the edge to quantify recovery, personalize treatment, and tune machine parameters — toward a "virtual doctor".
2019 – 2020
Adaptive PHM against stream data
A systematic methodology for online evolving PHM with an adaptive sampling mechanism, robust to continuous stream data and in-process changes.
- Survey Industrial time-series across 11 top venues: the full map →
- Genealogy A genealogy of industrial time-series research, 2024–2026 →
- Survey NeurIPS 2025 industrial TS: 281 papers across 9 families →
- Lectures Bayesian methods & reliability engineering: lecture series →
- Resource Industrial TS datasets catalog →
- Commentary Reading China's decade of smart manufacturing →