Now an IChemE‑approved course.
Participants will be introduced to real‑world process data challenges and how to solve them with Industrial AI and data science.
You will learn how GenAI tools (ChatGPT, Microsoft Copilot, Gemini…) support and accelerate the application of machine‑learning techniques—and when to use them.
We cover essential methods for regression (supervised learning) and anomaly detection (unsupervised learning) applied to continuous and batch process data, enabling subject‑matter experts such as process engineers to work independently.
Subject‑matter experts such as process engineers will:
Using Industrial AI, you will learn to:
– Define the problem
– Access relevant knowledge
– Extract and transform process data
– Quantify variability and detect significant process changes
– Quickly identify likely root causes to improve processes
– Plan additional experiments when needed.
The workshop is learn‑by‑doing: each day includes hands‑on exercises where you can work with your own datasets on your own laptop. Basic statistics (e.g., Six Sigma concepts) are helpful but not required. Programming background is not necessary (Python, MATLAB, etc.).
Programme:
Day 1 – Troubleshooting processes with Industrial AI and Data Science
- Industrial AI in manufacturing: what’s useful vs. hype; how GenAI supports problem framing, documentation search, and reporting.
- Core machine‑learning concepts via agentic workflows (prompting → exploratory data analysis → feature importance → quick root‑cause analysis), using the distillation tower case; apply random forest/boosted trees to relate drivers to yield.
- Industrial data landscape: tags & historians, automation pyramid (ISA‑95/88), events/batches, and ERP/LIMS joins.
- Import your own CSV/Excel for analysis; curated datasets are also provided.
Day 2 – Monitoring assets
- Introduction to analytics for batch processes: KPI definition for an industrial dryer (feature engineering) and variability tracking (visual analytics, SPC, robust statistics).
- Batch alignment using phase/event markers (e.g., time warping); FPCA vs. KPI summaries.
- Anomaly detection with PCA, KNN, autoencoders, and decision trees; identify plant changes in datasets such as the Tennessee Eastman process.
Day 3 – Modelling & decision support
- Problem definition for improvement vs. prediction; avoiding the “optimise the wrong thing” trap.
- Variable screening at scale: decision trees, bootstrap forests, boosted trees; controlling overfitting (time‑split/cut‑point).
- Explainable AI with SHAP: ranking drivers and interactions.
- Modelling tactics: missing data, Lasso, neural networks; inferential sensors & digital twins.
- Introduction to Bayesian Optimisation vs. DOE.