A Framework for Computationally Efficient Process Simulation with Machine Learning Aided Flash Calculations

Debasis Maity

doi:10.1016/j.cherd.2026.05.041

A Framework for Computationally Efficient Process Simulation with Machine Learning Aided Flash Calculations

2026-05-23

0

PRE

AI

D

Debasis Maity

H

Hariprasad Kodamana *

M

Manojkumar Ramteke *

DOI：10.1016/j.cherd.2026.05.041

Original

Original request for help

Save

Abstract

En 中文

• Physics-constrained NN surrogate for PT-flash with embedded heat/material balances • Satisfies thermodynamic constraints by design, ensuring physical consistency • Vectorized bisection algorithm extends surrogate to PH and Pβ-flash problems • Achieves up to 500× speedup over Aspen HYSYS for large-batch flash evaluations • Generalized flowsheet framework validated against Aspen HYSYS with close agreement

Keywords:

Physics-constrained neural network

Flash calculation

Thermodynamic consistency

Vectorized bisection algorithm

Process simulation

A Framework for Computationally Efficient Process Simulation with Machine Learning Aided Flash Calculations

Abstract

Journal

Organization