Dataset Open Access
Zhichao Wu, Li Zhu, Shunqi Zhang, Xirong Xu, Xiaopeng Wei, Xin Yang. ICAD: Rethinking the Role of Inference and Cues for the Anomaly Detection in IIoT. IEEE
Transactions on Knowledge and Data Engineering, 2026.
Introduction:
The Ethylene Oxide dataset is collected from a realworld, continuous chemical production process characterizedby its high-risk and safety-critical nature. The data were recorded in August 2024 at a high sampling frequency of one sample every 5 seconds. It includes measurements from multiple key process units, such as reactors, condensation towers, and distillation towers. To ensure meaningful data representation, we collaborated with experienced frontline operators to select 430 critical variables from thousands of sensors across the plant. These variables include direct physical measurements (e.g., temperature, flow rate, pressure),interlock control variables that trigger automated safety responses, and key indicators used in operational early warning systems. Compared to public benchmarks, EO offers higherdimensionality and richer process semantics, making it a more challenging and realistic testbed.Anomalies within the dataset are labeled by domain expertsand correspond to genuine operational challenges, such as process fluctuations during startup and shutdown phases and abnormal deviations during steady-state operations, providing a reliable ground truth for model evaluation under complex industrial conditions. Due to confidentiality agreements and the proprietary nature of the process, detailed engineering schematics cannot be disclosed. However, we believe that the inclusion of EO as a private, high-fidelity, safety-critical dataset significantly enhances the practical relevance and robustness evaluation of the proposed method.
Please enter your application information for dataset