Tennessee Eastman Process Simulation Data

Summary

The Rieth et al. Tennessee Eastman Process simulation data is a synthetic industrial process-control dataset for anomaly detection and fault diagnosis. It is based on the classic Tennessee Eastman Process benchmark and exposes multivariate measured and manipulated process variables over simulated runs.

Official Artifacts

Dataset Shape

The dataset has fault-free and faulty training/testing runs. Each record includes a fault label, simulation-run ID, sample index, and 52 process variables sampled every 3 minutes.

Role In The Wiki

This entity belongs in the industrial-control branch of action-conditioned time-series datasets. It is relevant because manipulated variables can be modeled as control inputs, but it remains primarily an anomaly/fault benchmark rather than a clean policy-learning dataset.

Relation To Foundation TSFM Agenda

Use the source-level agenda mapping in tennessee-eastman-process-2017 rather than duplicating verdict rows here.

At the entity level, the main boundary is terminology: manipulated variables are control-input-like channels; fault labels and fault injections are benchmark conditions or exogenous disturbance events.