Kernel Regression for the Approximation of Heat Transfer Coefficients

Experimentally-based correlations and other parametric methods for approximating heat transfer coefficients, while popular, have a number of shortcomings that are manifest when they are used in dynamic simulations of thermofluid systems. This paper studies the application of a nonparametric statistical learning technique, known as kernel regression, to the problem of approximating heat transfer coefficients for single-phase and boiling flows for the use in dynamic simulation. This method is demonstrated to accurately predict heat transfer coefficents for subcooled, two-phase, and superheated flows for a finite volume model of a refrigerant pipe, as compared to results obtained from established correlations drawn from the literature.