Data Generation and Knowledge-based Augmentation: Batch Distillation

间歇精馏是化学工业中最重要的工艺之一。然而，在公开文献中缺乏可用于开发和训练机器学习方法的间歇蒸馏装置操作的实验数据。因此，在本项目中，这些数据将在实验室规模的间歇精馏塔中产生，该精馏塔配备了先进的传感器，包括在线核磁共振光谱仪和相机。附加数据将通过基于动态物理模型的间歇蒸馏过程模拟生成。在广泛的操作策略和条件下，两种方法将研究不同异常情况和无异常情况；许多不同的流体混合物的分离将被研究，包括不明确的混合物。生成的数据还将包括关于不确定因素的资料。实验设计领域的方法将用于规划实验室蒸馏和模拟。完整的数据集将向公众提供。详细考虑实验数据和仿真数据之间的关系，并将两类数据合并为混合数据集。该项目还将提供本研究股A研究区项目所需的间歇蒸馏过程的物理知识。我们的目标是以一种最适合化学过程异常检测的方式提供项目中生成的复杂和异构数据。在该项目中，将探索一种新的批量蒸馏整体方法，其成果可能远远超出异常检测。

Batch distillation is one of the most important processes in the chemical industry. Nevertheless, experimental data on the operation of batch distillation plants that could be used to develop and train machine learning methods is lacking in the open literature. Therefore, such data will be generated in the present project in a laboratory scale batch distillation column, which is equipped with advanced sensors, including an online NMR spectrometer and cameras. Additional data will be generated by simulations of batch distillation processes based on a dynamic physical model. Cases with different anomalies as well as anomaly-free cases will be studied with both methods for a wide range of operating strategies and conditions; and separations of many different fluid mixtures will be investigated, including poorly specified mixtures. The generated data will also comprise information on the uncertainties. Methods from the field of design of experiments will be used to plan the laboratory distillations as well as the simulations. The full data sets will be made publicly available. The relations between the experimental data and the simulation data will be considered in detail and the two types of data will be merged to hybrid data sets. This project will also provide the physical knowledge on batch distillation processes needed in the projects of Research Area A of this Research Unit. Our ambition is to supply the complex and heterogeneous data generated in the project in a way that is optimal for anomaly detection in chemical processes. In the project, a new holistic approach to batch distillation will be explored that could be fruitful far beyond anomaly detection.