Machine Learning Informed Digital Twin for Chemical Flow Processes

Abstract

Digital Twin (DT) is a virtual representation that is parameterized based on the real process data to model, simulate, monitor, analyze, and optimize the physical systems they represent. DT have been predominantly used in the mechanical engineering field and have yet to be extensively used in chemical flow processes particularly for the challenge of scale-up which is very important particularly when moving from lab experiment to industrial scales. It is a challenge to maintain various process parameters while increasing the scale of reactors geometry. The parameters might not show a predictable linear co-relationship due to concurrent chemical conversion processes behaving differently on different scale. We apply and compare various machine learning methodologies such as Radial Basis Function Neural Networks, Gaussian Process Regression and Polynomial Regression to the development of chemical flow process DT for scale-up. We show that these methodologies can be used to predict the product yield of a chemical flow process during scale-up.