An open-source dataset of heat-integrated side-stream extractive distillation process to support development of fault diagnosis algorithms

Abstract

Datasets are crucial for the development of fault diagnosis and detection (FDD) methods. This study presents the development and application of an open-source dataset for a heat-integrated side-stream extractive distillation (HSSED) process. The HSSED process is a multi-column complex system combining heat recovery and side-stream extraction. Firstly, a steady-state simulation for the benzene–isopropanol–water azeotropic system was built, and multi-objective optimization was carried out to minimize total annual cost, CO₂ emissions, and the process route index. The integrated digital workflow, which combines optimal design, control, and data acquisition, produced an open-access HSSED dataset containing normal operating conditions and twenty-five faults, each simulated at four severity levels for a total of one hundred fault scenarios. Each scenario is accompanied by time series data for 39 measurement variables, including liquid level, flow rate, temperature, pressure, and purity variables. The HSSED dataset offers a new benchmark for FDD and can be downloaded at: https://github.com/wsbai321/HSSED. For this dataset, we have developed a bidirectional hybrid model with the architecture GRU(64) – BN – BiLSTM(64) – BN – GRU(128) – FC. The accuracy and layer-wise visualization show the effectiveness of the hybrid model.