Xiaoyu Wang

Research

My research sits at the intersection of industrial AI and agentic manufacturing: how autonomous, tool-using AI systems can be embedded into the value chains of real factories — planning and scheduling, process control, quality management — in ways that are designable, verifiable, and economically accountable.

Twenty years in automotive Tier-1 and battery manufacturing shape both the questions I ask and the data I can reach. I am interested in the systems-theoretic foundations of this problem: modeling the business and the AI in one framework, so that “where AI creates value” is an engineering question rather than a bet.

Current projects

  • AutoTier1-Bench (concept): a benchmark for agentic AI on realistic automotive Tier-1 manufacturing decision tasks.
  • LLM × classical OR: natural-language business rules compiled into solvable optimization models, with human-in-the-loop review (deployed in production; see Practice).
  • AI agents for process control: first-principles models + MPC + LLM agents, published at IEEE ICAIM 2025.

The core model

The digital space drives physical operations through an OODA loop; the L1–L4 framework grades how deeply digitalization reaches into decisions and execution.

FIG.01 — Intelligent data flow ⇄ business value stream: the digital space drives physical operations through an OODA loop

L3 Predictive prevention — Key question: what will happen?

DIGITAL SPACE · Data flow (intelligence)ObserveSENSEOrientANALYZEDecideOPTIMIZEActEXECUTEL1 Real-time visibility — Key question: what is happening now?L1Historical recordsDescriptive: what happenedManagement actionsManual executionL2 Rapid root-cause — Key question: why did it happen?L2Process data, multi-sourceDiagnostic: why it happenedImprovement actionsTargeted responseL3 Predictive prevention — Key question: what will happen?L3Real-time + full historyPredictive: what will happenPreventive actionsEarly interventionL4 Algorithmic decision & execution — Key question: what should be done — decided and executed by algorithmsL4Full-domain, closed-loopPrescriptive: what to doAuto / assisted decisionFully automatedPHYSICAL SPACE · Value stream (value)EquipmentProcessLinePlantprecise execution ↓↑ real-time feedback

Click any data-flow layer (L1–L4) to switch level; click a value-stream object (equipment/process/line/plant) to re-anchor the loop

Publications

Synced from Google Scholar

Book

Papers (12)

  • Optimizing the Emulsion Polymerization Reaction Process Using Artificial Intelligence (AI) Agents — The International Conference on Artificial Intelligence and Materials (ICAIM) / IEEE, 2025
  • Frequency characteristics of the synchronous-frame based DQ methods for active power filters — Journal of Power Electronics, 2008
  • Frequency characteristics of the dq synchronous-frame current reference generation methods for active power filter — 7th International Conference on Power Electronics (ICPE), 2007
  • A family of control methods for parallel active power filters based on current detection — IEEE Applied Power Electronics Conference (APEC), 2007
  • On the control of active power filters — 7th International Conference on Power Electronics (ICPE), 2007
  • Study of DC-side Voltage Boost of Series Power Quality Controller — Power Electronics(《电力电子技术》), 2007
  • Frequency-domain transformation approaches to develop the dq synchronous-frame models of three-phase symmetrical networks and applications in modeling of PWM power converters — 7th International Conference on Power Electronics (ICPE), 2007
  • Comparative analysis of popular control schemes for parallel active power filter and experimental verification — IEEE 37th Power Electronics Specialists Conference (PESC), 2006
  • Generalized control approach for active power filters — CES/IEEE 5th International Power Electronics and Motion Control Conference (IPEMC), 2006
  • A comparative study on voltage-source control and current-source control of series active power filter — IEEE Applied Power Electronics Conference (APEC), 2006
  • A novel voltage-mode control of parallel active power filter — IEEE Industry Applications Conference (IAS), 2006
  • An unified scheme and respective algorithms for the control of DC-linked double converters in an universal power quality controller — IEEE 34th Power Electronics Specialist Conference (PESC'03), 2003

Patents (7)

  • Power generation performance assessment method and device for generator sets — CN104915747B, 2019
  • Wind farm primary electrical system and topology configuration method and device — CN106558886B, 2017
  • Wind generating set and work mode switching method — CN103138669A, 2016
  • Frequency converter test platform and its debugging method — CN103616584B, 2016
  • Ice melting device and method for wind generating set — CN105443325A, 2016
  • Power generation performance evaluation method and apparatus for power generator set — US20160223600A1, 2016
  • Wind generating set (utility model) — CN203151423U, 2013