Data sets from CPS for ML applications
The Tennessee Eastman process (TEP) is a widely used benchmark process for plant-wide control, statistical process monitoring, and fault detection and diagnosis research. This dataset extends the classic TEP simulation data with repeated simulations of both healthy and faulty process data, additional process measurements, and multiple magnitudes for each process disturbance. All six production modes of the TEP are simulated, including mode transitions and different operating points.
In total, the dataset contains simulations of 28 process faults across 6 operating modes; each fault is simulated 500 times using a different random-number-generator seed. Every simulation covers 100 hours of process operation with a sampling interval of 3 minutes, resulting in an extensive reference dataset for developing and benchmarking fault-detection, diagnosis, and decision-support methods.
The dataset was developed at the Technical University of Denmark as part of a PhD project on automating causal analysis for online diagnosis and planning in complex industrial processes. The accompanying thesis uses the dataset as an open-source testing framework, combining it with causal (Multilevel Flow Modeling) approaches for root-cause diagnosis and reconfiguration planning.
Tennessee Eastman Reference Data for Fault-Detection and Decision Support Systems
| Title | Authors | Year |
|---|---|---|
| An extended Tennessee Eastman simulation dataset for fault-detection and decision support systems | Reinartz, C., Kulahci, M., Ravn, O. | 2021 |
| Automating causal analysis for online diagnosis and planning in complex industrial processes (PhD thesis) | Reinartz, C. | 2021 |