Date of Award

May 2026

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Chemical Engineering

First Advisor

Gautham Krishnamoorthy

Abstract

In the last 10 years machine learning has become increasingly more used in the field of fluid dynamics and heat transfer. However, with this increase in activity, gaps still exist with respect to supercritical fluids and laminar flows, where laminar flow in general seems understudied. For supercritical fluids, there are many peculiar studies that are isolated to unique systems or they are obscure enough to include everything but the kitchen sink. This strategy provides no real insights into how the field can make further progress in the utilization of the technology available, or rather build-upon previous knowledge. For this reason, we think it inherently makes sense to start with laminar flow as the start of our baseline without the additional uncertainty from turbulent regimes, and to incorporate buoyancy and heat transfer characteristics as they are related by space and thermo-physical properties.For constant q-flux and thermo-physical properties, with full developed velocity in a circular tube the Nu is expected to converge to 4.36 which serves as a reference of comparison, where Nu values less than 4.36 represent heat transfer deterioration (HTD) and Nu values greater than 4.36 as heat transfer enhancement (HTE). In both vertical and horizontal orientations sDecane undergoes heat transfer deterioration with sufficiently large enough nondimensional Dh/z and this allows for accurate prediction of Nu when combining both orientations into a single correlation by symbolic regression. Because the change in Gr* is fixed within a pressure set due to the thermal expansion coefficient then Nu is dependent of inlet conditions represented by the Gz and buoyancy forces can be neglected, which means deterioration is based on the thermo-physical properties.

The sDecane fluid being distinguished as not being affected by buoyancy forces and exhibiting heat transfer deterioration is then combined with two other supercritical fluids, water and carbon dioxide whom undergo HTE. Two machine learning methods are used for the combined data set by Random Forest (RF) and Artificial Neural Network (ANN) for both dimensional and non-dimensional inputs to predict wall temperature and Nu. The combined model is then refined by including the direction of gravity enabling accurate prediction of heat transfer in various orientations.

Download

Share

COinS