Abstract

This is an alarm management dataset based on the “Tennessee-Eastman-Process” (TEP). The presented dataset aims to provide a suitable benchmark for the development and validation of alarm management methods in complex industrial processes using both quantitative data and qualitative information from different sources. Unlike real industrial processes, the simulation of the TEP allows to design and generate abnormal situations, which can be repeated and varied without risking the loss of equipment or harming the environment. In addition, as the simulation is supervised, all induced disturbances and process normalizations are explicitly known and therefore act as a ground truth, facilitating the utilization of external evaluation metrics, e.g., when using cluster analysis. Further details are described in a supplementary technical report.

Instructions:

This dataset includes a supplementary technical report. The report starts with a brief overview of the facility, implemented control loops and used simulation model. Two types of data, process and alarm data, were collected from the facility, their underlying data structure is included. The process of developing and implementing suitable alarm thresholds and alarm management techniques is described in detail. Furthermore, all initiated abnormal situations and the respective test design is presented thoroughly. The report concludes with a detailed description of the dataset structure and layout.