Solving vapor-liquid flash problems using artificial neural networks

Abstract

Vapor-liquid phase equilibrium (flash) calculations largely contribute to the total computation time of many process simulation models. As a result, process simulations, especially dynamic cases, are limited in the amount of detail that can be included due to time restrictions. In addition, under certain conditions flash calculations can fail to provide acceptable results. In this work, artificial neural networks were investigated as a potentially faster and more robust alternative to conventional flash calculation methods. Classification neural networks were used to determine the phase stability of a given mixture of fluids, while regression networks were used to make predictions of thermodynamic property values. In addition to conventional flash types such as the constant pressure, constant temperature (P T ), and constant pressure, constant entropy (P S) flash, neural networks are used to develop two concept flash types: a constant entropy, constant volume (SV ), and a constant enthalpy, constant volume (HV ) flash. All neural networks were trained on, and compared to, data generated using the P T -flash algorithm from the Thermodynamics for Engineering Applications (TEA) property calculator. Data was generated for mixtures of water and methanol over a wide range of pressures and temperatures. The artificial neural networks showed speed improvements over TEA of up to 35 times for phase classification and 15 times for property predictions. Overall phase classification accuracy scores of around 97% were achieved. Average property value prediction errors range between 0.5% and 7% when compared to the spread in magnitude of the test data, and R 2 scores were in the general order of 0.95 and higher, although classification and property prediction in the two-phase region showed markedly higher errors than properties in the pure liquid or vapor regions. Moreover, thermodynamic consistency and the stability of a system consisting of multiple neural network flash types both still require considerable improvement. Finally, this work shows that artificial neural networks can be used to create unconventional flash types such a the SV -and HV -flash.i

Design, Parameterization, and Implementation of Atomic Force Fields for Adsorption in Nanoporous Materials

“…A few examples of the use of machine learning in the field of classical simulations for materials science are the deep neural network learning of complex binary sorption equilibria from molecular simulation data, solving vapor–liquid flash problems using artificial neural networks, predicting thermodynamic properties of alkanes, charge assignment, prediction of partition functions, simulation of infrared spectra, predicting the mechanical properties of zeolite frameworks, CO 2 capture using MOFs, prediction of methane adsorption performance of MOFs, chemically intuited, large‐scale screening of MOFs, screening for precombustion carbon capture using MOFs, screening of MOF Membranes for the separation of gas mixtures, and screening of MOFs for use as electronic devices . Using ML to combine the accuracy and flexibility of electronic structure calculations with the speed of classical potentials is a very active research field .…”

Thermodynamics and Machine Learning Based Approaches for Vapor–Liquid–Liquid Phase Equilibria in n-Octane/Water, as a Naphtha–Water Surrogate in Water Blends

“…Poort et al [19] studied water/methanol mixtures, using classification neural networks for the phase stability and regression networks to calculate thermodynamic properties. The data for training was generated for 101 feed composition, 500 temperatures (273-700 K), and 500 pressures (1 × 10 4 -3 × 10 7 Pa).…”

Application of Artificial Neural Networks to Chemical and Process Engineering

“…Moreover, ANNs reveal a conceivably faster choice to those property prediction calculations in process simulations, limiting process control applications that require to be conducted in real-time. For this, Poort et al [21] studied the replacement of conventional Equations of State (EoS) for property and phase stability calculations on a binary mixture of methanol-water. They trained ANNs with data generated through the Thermodynamics for Engineering Applications (TEA) to represent four kinds of flash algorithms, leading to an enhancement of 15 times for the predictions of properties and 35 times for classification of the phases.…”