A Fluid Mixing Benchmark for Anomaly Detection in CPS With Real and Simulated Data

Ensuring the safety and reliability of Cyber-Physical Systems requires effective anomaly detection. However, research in this field is often limited by the lack of publicly available benchmark datasets that accurately capture real-world system behavior and provide sufficient documentation. This paper addresses this gap by defining requirements for anomaly detection datasets, evaluating existing benchmarks, and introducing a novel dataset collected from a real-world fluid mixing system augmented with a simulation model. The dataset captures diverse operational states, fault scenarios, and non-linear system dynamics, while the simulation model enables further system understanding. To demonstrate its utility, we apply six state-of-the-art unsupervised anomaly detection models on the dataset. The results confirm the dataset’s suitability for anomaly detection research. To promote reproducibility, we publicly provide the software framework for model implementation, along with the datasets, simulation model, and evaluation results, accompanied by comprehensive documentation.