Session: Research Posters

Paper Number: 120333

120333 - Computationally Efficient Property Calculation for Mixed Refrigerants Using Weighted Piecewise Polynomial Regression 

Fast, accurate, and stable thermophysical property calculations of refrigerants are crucial to steady-state and transient simulations of refrigeration and air conditioning systems. Property calculation accounts for most of the simulation time, so faster calculations lead to faster simulations. Typically, property calculation resorts to regression models with the data obtained from a standard database such as REFPROP. While polynomial regression has been well received, when dealing with mixed refrigerants, it tends to end up with high-order models to obtain satisfactory accuracy, thus being computationally expensive. In this paper, we propose a weighted piecewise polynomial regression framework to improve the computational efficiency for property calculations of mixed refrigerants. Weighted piecewise polynomial regression is a regression technique that uses piecewise polynomial functions to model the relationship between a dependent variable and one or more independent variables. The method partitions the domain of the independent variable into intervals and fits a separate polynomial function to each interval.  As for the REFPROP-generated datasets, the ranges of independent variables are divided into certain number of intervals, and the polynomial regression is performed for each interval. To achieve better continuity between these intervals, higher weights are applied to the neighborhood of the break points. Also developed is an algorithm to optimize the choice of junction points, which balances the orders of piecewise polynomials and number of intervals. To evaluate the effectiveness of the proposed work, two mixed refrigerants, R410A and R454B, are used. The thermophysical and thermodynamic properties considered for both refrigerants are pressure, temperature, density, specific enthalpy, thermal conductivity, Prandtl number, dynamic viscosity, and surface tension. The refrigerant properties as a function of temperature or pressure are fitted for the saturated liquid and vapor regions, while the properties as a function of pressure and temperature, or specific enthalpy, are considered for the subcooled and superheated regions. The pressure range considered for the R410A refrigerant is 170 to 4269 kPa, and the one for the R454B is 100 to 4000 kPa. Cross-validation was carried out by fitting the standard polynomial regression to the same data range and comparing the outcome with that of the weighted piecewise polynomial regression. The results from the proposed method show good fitting accuracy with relatively low-order polynomials across wide operation range. The piecewise polynomials of 2-5 order result in the average computation time of 1.5~43% of that with a full-range polynomial of 10-20 order, under the same level of fitting accuracy. Such improvement in computational efficiency promises the potential benefit in developing real-time or faster HVAC simulations.

Presenting Author: Abdulmumin Olamilekan Olaoke University of Texas at Dallas

Presenting Author Biography: Abdulmumin Olaoke
University of Texas at Dallas
abdulmumin.olaoke@utdallas.edu
I am interested in energy systems, heat transfer and thermal engineering.

Authors:

Abdulmumin Olamilekan Olaoke University of Texas at Dallas
Baojie Mu Rheem Manufacturing Company
Yaoyu Li University of Texas at Dallas