Development of a surrogate model of an amine scrubbing digital twin using machine learning methods. (June 2023)

• Record Type: Journal Article
• Authors: Galeazzi, Andrea; Prifti, Kristiano; Cortellini, Carlo; Di Pretoro, Alessandro; Gallo, Francesco; Manenti, Flavio
• Is Part Of: Computers & chemical engineering. Volume 174(2023)
• Publication Date: 2023-06
• Subjects: Machine-learning -- Surrogate modeling -- Digital twin -- Amine scrubbing -- Design of experiments -- Latin hypercube; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2023.108252
• Languages: English
• ISSNs: 0098-1354

Abstract

Abstract: Advancements in the process industry require building more complex simulations and performing computationally intensive operations like optimization. To overcome the numerical limit of conventional process simulations a surrogate model is a viable strategy. In this work, a surrogate model of an industrial amine scrubbing digital twin has been developed. The surrogate model has been built based on the process simulation created in Aspen HYSYS and validated as a digital twin against real process data collected during a steady-state operation. The surrogate relies on an accurate Design of Experiments procedure. In this case, the Latin-Hypercube method has been chosen and several nested domains have been defined in ranges around the nominal steady state operative condition. Several machine learning models have been trained using cross-validation, and the most accurate has been selected to predict each target. The resulting surrogate model showed a satisfactory performance, given the data available. Highlights: The method proposed can be applied to automatically surrogate any digital twin. Data-driven machine learning algorithms can be applied to metamodel a digital twin. Surrogate performance is greatly affected by the design of experiments training data.