IChemE - The Three Cs of Industrial AI: Connectivity, Contextualization, and Cognition

Can you dump raw plant data and documentation into Claude or GPT and expect accurate analysis? No, and not because the models aren’t smart enough.

This session breaks the path to industrial AI into three steps: connectivity, contextualization, and cognition. Connectivity is largely solved. Most plants already extract telemetry from equipment through PLCs, SCADA, and historians. Cognition is ready too: frontier models reason carefully over real data rather than inventing numbers.

Context is the bottleneck. Give a model a few hundred tags streaming from mixed PLCs and it will produce trends and correlations, not diagnosis. It has no way to know which tags sit upstream of which, which signals are causes and which are downstream symptoms, or which correlations actually matter for throughput. That knowledge lives in control system configuration, plant documentation, and operators’ heads.

We’ll cover why manual tag mapping doesn’t scale, why documentation is always stale relative to the systems that actually run the plant, and how a hybrid of deterministic code and configuration analysis plus LLM parsing of unstructured documents can build a context layer that stays accurate without manual labor. We’ll finish with live examples of agents built on that foundation: root cause analysis, failure prediction, and throughput optimization.