The CSTH Dataset

The CSTH Simulated Benchmark Dataset is designed for time series classification and fault detection and diagnosis (FDD). It is generated using the continuous stirred tank heater (CSTH) simulation model (zenodo.org/records/10093059). The model represents a heating system in which hot and cold water are mixed, heated by steam, and regulated through a closed-loop control system.

The dataset consists of 9,000 multivariate time series samples, each with 200 time steps and three process variables (cold water flow, tank level, and temperature). It includes both normal operating conditions (Y=0) and faulty scenarios (Y=1), where faults are introduced through instrumentation errors. The dataset is split into: i) train.pt (70%, 6,300 samples), ii) val.pt (10%, 900 samples), and iii) test.pt (20%, 1,800 samples).