Pre-Congress Workshop · IFAC World Congress 2026

LLMs and Agents for Process Industries

Foundations, Applications, and Open Challenges

Date
Sunday, August 23, 2026
Time
9:00 AM – 5:00 PM (KST)
Location
Busan, Republic of Korea
  Register via IFAC 2026   See full schedule
Large Language ModelsAutonomous AgentsProcess ControlArtificial IntelligenceIndustrial AutomationMachine LearningFault Diagnosis
01 · Overview

Why we are running this workshop

Large Language Models have evolved from text generators to agents that reason, use tools, and take actions. Researchers are now applying these agents to process industries: interpreting flowsheets, predicting control structures, diagnosing faults, even acting as autonomous operators. The results so far are intriguing but preliminary, and the gap between laboratory demonstrations and plant deployment remains substantial.

This full-day workshop brings together researchers from three continents to examine what LLM agents can actually do today, where they fail, and what problems deserve our attention. The morning covers the fundamentals of LLMs and agents and surveys applications in process design and process simulation. The afternoon turns to autonomous control and fault diagnosis, followed by a hands-on session in which participants build simple agents and probe their limitations. We close with an open discussion aimed at identifying research priorities.

Participants will leave with practical experience, a grounded assessment of the technology, and ideas for their own work.

  Why now?

LLMs have moved past the initial hype phase, and the process control community now has enough published work and practical experience to evaluate what actually works. At the same time, the technology is still immature enough that we can shape its development rather than simply adopt tools designed for other domains.

Plants generate vast amounts of unstructured text — maintenance logs, alarm records, operating procedures, incident reports — that traditional analytics cannot easily exploit. LLM agents can work with this data directly. As experienced operators retire, these agents offer a mechanism for encoding and applying organizational knowledge. Natural language interfaces could change how engineers interact with process data and systems, invoking simulations, querying historians, and executing calculations. But the limitations are real: hallucination, weak numerical reasoning, no safety guarantees by default, and the challenge of validating agent behavior in autonomous settings.

02 · Learning outcomes

What you'll take away

After attending, participants will be able to:

1
Explain how LLM agents work and critically evaluate claims about their capabilities.
2
Build simple LLM agents for process engineering tasks using standard tools.
3
Recognize current LLM applications in process design, control, and simulation.
4
Identify the key challenges — safety validation, hallucination, integration — that must be addressed for industrial deployment.
5
Articulate open research problems and opportunities in this emerging field.
03 · Schedule

Sunday, August 23, 2026 · 9:00 AM – 5:00 PM (KST)

A full day balanced across foundations, hands-on exercises, applications, and open discussion.

Morning — Foundations & Applications

9:00 – 9:15
Welcome and Framing
Prof. Bhushan Gopaluni, Prof. Mehmet Mercangöz, Prof. Chunhui Zhao
Overview of the day, learning outcomes, and an introduction to where LLM agent technology stands in the process industries.
9:15 – 10:15
LLM and Agent Fundamentals
Prof. Bhushan Gopaluni
Attention mechanism and transformer architecture; introduction to simple agent architectures via process control examples (alarm lists, operator procedures, simplified P&ID encodings).
10:15 – 10:30
☕ Coffee Break
10:30 – 11:30
Application I — Process Design and Engineering
Prof. Bhushan Gopaluni
P&ID interpretation, control structure prediction, and automated design review. Discussion of Hirtreiter et al. (2024), Balhorn et al. (2024), and graph-based approaches for capturing process topology.
11:30 – 12:30
Application II — Process Simulation and Optimization
Prof. Bhushan Gopaluni & Prof. Chunhui Zhao
Automated simulation configuration, multimodal approaches combining process measurements with text and images, and natural language interfaces for querying process historians and generating analysis reports.
12:30 – 13:30
🥢 Lunch

Afternoon — Applications, Open Problems & Hands-On

13:30 – 14:30
Application III — Autonomous Control and Fault Diagnosis
Prof. Mehmet Mercangöz & Prof. Chunhui Zhao
Agentic frameworks for industrial automation: LLM-based fault diagnosis integrating sensor data with maintenance history, autonomous agents that reason about process upsets, integration of LLM agents with digital twins, and industrial control system integration — delivered jointly by Prof. Mehmet Mercangöz and Prof. Chunhui Zhao.
14:30 – 14:45
☕ Coffee Break
14:45 – 15:45
Hands-On Session — Build, Break & Debug Simple LLM Agents
Prof. Bhushan Gopaluni, Prof. Mehmet Mercangöz, Prof. Chunhui Zhao
Participants work directly with LLM APIs on process control tasks using provided Python notebooks. Three exercises progressing from prompt engineering through agent construction to failure analysis.
15:45 – 16:45
Open Problems, Research Directions, and Q&A
Prof. Bhushan Gopaluni, Prof. Mehmet Mercangöz, Prof. Chunhui Zhao — with all participants
Structured discussion: What works now? What's tractable? What's hard? Followed by an open floor for questions about content, research directions, industrial deployment, and collaboration.
16:45 – 17:00
Wrap-up and Resources
Prof. Bhushan Gopaluni, Prof. Mehmet Mercangöz, Prof. Chunhui Zhao
Synthesis of the day's themes plus pointers to resources for continued learning — key papers, open-source tools, benchmark datasets, and community forums.
04 · Hands-on session

Build, break, and debug simple LLM agents

Bring a laptop. We'll give you everything else. Observers are welcome — code if you like, watch if you prefer.

  What to bring

  •   A laptop (required — you will work directly with LLM APIs on Python notebooks during the session).
  •   Ability to connect to the workshop Wi-Fi.

  What we provide

  •   Jupyter notebooks and starter code.
  •   Access to LLMs for the duration of the session.
  •   Sample datasets: simplified P&IDs, alarm lists, maintenance logs.
  •   Offline backup — pre-recorded demos and local notebooks using open-source models — in case of connectivity issues.

Exercises

01

Prompt engineering basics

Prompt a model to interpret a simplified P&ID and critique its control structure suggestions. Experiment with different prompting strategies.

02

Building a simple agent

Construct a minimal fault diagnosis agent that combines an LLM with tool calls to query sensor data and search maintenance logs.

03

Failure analysis

Probe agent limitations — numerical reasoning errors, hallucinated equipment names, overconfident recommendations — to develop intuition for where these systems fail.

  Prerequisites

Participants should understand basic process control concepts (feedback loops, PIDs, cascade control) and have some familiarity with machine learning ideas (training, inference, overfitting). Prior experience with LLMs or Python programming is helpful but not required — the morning sessions provide necessary background and hands-on exercises can be completed collaboratively. Attendees who prefer to observe rather than code are also welcome.

05 · Speakers

Three continents, complementary expertise

Our speakers are also the workshop organizers — spanning academic research in North America, Europe, and Asia, with extended industrial practice represented alongside.

Prof. Bhushan Gopaluni
Primary Contact & Speaker

Prof. Bhushan Gopaluni

University of British Columbia
Professor, Chemical & Biological Engineering
Vice-Provost & Associate Vice-President, Faculty Planning
  Vancouver, Canada
  Sessions: Welcome · LLM and Agent Fundamentals · Application I (Process Design) · Application II (Process Simulation, jointly with Prof. Zhao) · Hands-On · Wrap-up

Directs the Data Analytics and Intelligent Systems (DAIS) Lab, focusing on machine learning, reinforcement learning, and data analytics for process industries. Author of over 200 refereed publications. Recipient of the UBC Killam Teaching Prize and the D.G. Fisher Award from the Canadian Society for Chemical Engineering. Extensive experience organizing IFAC workshops, including sessions at ADCHEM and DYCOPS.

Prof. Mehmet Mercangöz
Speaker & Organizer

Prof. Mehmet Mercangöz

Imperial College London
ABB Reader in Autonomous Industrial Systems
Sargent Centre for Process Systems Engineering
  London, United Kingdom
  Sessions: Application III (Autonomous Control & Fault Diagnosis, jointly with Prof. Zhao) · Hands-On · Open Problems

Spent 14 years at ABB Corporate Research, most recently as coordinator for industrial artificial intelligence, before joining Imperial College. His combination of academic research and industrial R&D experience uniquely positions him to address the practical challenges of deploying LLM agents in industrial settings. Current research focuses on agentic frameworks that leverage LLMs for autonomous industrial control with appropriate safety guarantees.

Prof. Chunhui Zhao
Speaker & Organizer

Prof. Chunhui Zhao

Zhejiang University
Professor, College of Control Science and Engineering
  Hangzhou, China
  Sessions: Application II (Process Simulation, jointly with Prof. Gopaluni) · Application III (Autonomous Control, jointly with Prof. Mercangöz) · Hands-On · Open Problems

Leading researcher in AI for process systems engineering. Work on process monitoring, fault diagnosis, and multimodal data fusion — combining process measurements, text, and images — is directly relevant to LLM applications in process industries. Brings perspective on developments in the rapidly growing AI and process systems research community in Asia.

06 · Target audience

Who this is for

  •   Academic researchers and graduate students in process control and process systems engineering exploring AI applications.
  •   Industry practitioners — control engineers, automation specialists, and technology developers — evaluating LLM tools.
  •   Research managers and technical leaders assessing the technology for their organizations.

Expected attendance: 25–40 participants across academia and industry.

07 · Relevance to IFAC

Aligned with IFAC technical areas

This workshop aligns directly with IFAC's mission to promote automatic control and its applications. The topics span multiple IFAC technical areas:

TC 6.1
Process Control
Techniques for modeling, analysis, and control in chemical and bioprocess systems.
TC 6.2
Mining, Mineral and Metal Processing
Process control and digital technologies in mining and metals production.
TC 6.4
Fault Detection, Supervision and Safety
Methodologies for detection, supervision, and safety of technical processes.
08 · Code & materials

Notebooks, slides, and reading list

  Participant materials

Notebooks, slides, datasets, and a curated reading list live on a password-protected page. The passcode is shared with registered participants by email. On the day of the workshop, head here first.

  Access materials

Selected references

Recent papers that ground the workshop content. A fuller reading list will be shared on the day.

  1. Balhorn, L. S., Caballero, M., Schweidtmann, A. M. (2024). Toward autocorrection of chemical process flowsheets using large language models. Computer Aided Chemical Engineering, Vol. 53, 3109–3114. Elsevier. Preprint: arXiv:2312.02873
  2. Du, W., Yang, S. (2025). The potential and challenges of large language model agent systems in chemical process simulation. Frontiers of Chemical Science and Engineering, 19(10), 99.
  3. Gill, M. S., Vyas, J., Markaj, A., Gehlhoff, F., Mercangöz, M. (2025). Leveraging LLM agents and digital twins for fault handling in process plants. arXiv:2505.02076
  4. Hirtreiter, E., Schulze Balhorn, L., Schweidtmann, A. M. (2024). Toward automatic generation of control structures for process flow diagrams with large language models. AIChE Journal, 70(1), e18259.
  5. Qaid, H. A. et al. (2024). FD-LLM: Large language model for fault diagnosis of machines. arXiv:2412.01218
  6. Schulze Balhorn, L. et al. (2024). Graph-to-SFILES: Control structure prediction from process topologies using generative artificial intelligence. arXiv:2412.00508
  7. Venkatasubramanian, V. (2019). The promise of artificial intelligence in chemical engineering: Is it here, finally? AIChE Journal, 65(2), 466–478.
  8. Vyas, J., Mercangöz, M. (2024). Autonomous industrial control using an agentic framework with large language models. arXiv:2411.05904
  9. Woo, T., Kim, S., Tariq, S. et al. (2025). Leveraging generative AI and large language models for process systems engineering: a state-of-the-art review. Korean Journal of Chemical Engineering, 42, 2787–2808.
  10. Xia, Y., Dittler, D., Jazdi, N., Chen, H., Weyrich, M. (2024). LLM experiments with simulation: Large language model multi-agent system for process simulation parametrization in digital twins. IEEE ETFA (Best Paper Award).
  11. Xia, Y., Jazdi, N., Zhang, J., Shah, C., Weyrich, M. (2024). Control industrial automation system with large language model agents. arXiv:2409.18009

Related workshops organized by Prof. Gopaluni

2023 FOPAM — Machine learning for process industries
2022 DYCOPS — State and parameter estimation
2021 ADCHEM — Data analytics and machine learning
09 · Register

Join us in Busan on August 23, 2026

Workshop registration is handled through the official IFAC 2026 World Congress portal. Fees and logistics are set by the congress.

  Sunday, August 23, 2026
  9:00 AM – 5:00 PM (KST)
  Busan, Republic of Korea
  Venue room TBA
  Register via IFAC 2026   Ask the organizers

Questions? Email bhushan.gopaluni@ubc.ca

Supported by the UBC DAIS Lab, Imperial College London, and Zhejiang University. Held as a pre-congress workshop at the IFAC World Congress 2026, Busan, Republic of Korea.