AI for Chemical Manufacturing

An agentic assistant that finds the root cause of a plant issue. It connects Claude to a historian, DCS alarm logs, an equipment knowledge graph and plant documentation through an MCP server, then writes out every step it took to get there.

Describe a chemical process in plain English and this simulates 90 days of it, with control loops that hide the faults, noisy instruments, historian compression and an answer key for 9 hidden faults.

A root cause analysis benchmark built on plant historian data. 90 days of data from a steam-heated hold tank, noisy and compressed like a real historian, with hidden faults to find and diagnose from first principles. Built to test engineers, ML models and AI agents.

A curated list of public chemical manufacturing data. Five real P&ID sets from permitted facilities and five time-series process datasets, with the source, the size and what each one is actually good for.

Describe a process in plain English and get an editable PFD with equipment, process streams, and control loops. No coding experience required.

Fault detection on the Tennessee Eastman Process, seven models compared head to head. Classical MSPC against four machine learning models, all scored on post-fault samples only.