scope:

INTRODUCTION

Two commonly used system models are HPSIM (Domanski and Didion 1983) and PUREZ (Rice and Jackson 1994). These models have been found to underpredict charge inventory. Several issues have been identified as sources of error in the modeling of charge inventory, including incomplete internal volume accounting, neglecting refrigerant-oil diffusion effects, and void fraction modeling assumptions (Damasceno et al. 1991; Marques and Melo 1993). Perhaps the most challenging of these is void fraction determination.

At a given cross section of a tube, the void fraction is defined as the fraction of area occupied by vapor. While mass quality can be determined using conservation equations, in general, void fraction cannot be directly calculated and must be modeled in some manner. Rice (1987) presented a comprehensive review of the available void fraction models. The void fraction correlations of Hughmark (1962), Premoli et al. (1971), Tandon et al. (1985), and Baroczy (1965) were recommended, since they yield the highest charge predictions for condensers and the best overall agreement with experimental data. Rice stated that there are insufficient data to recommend one over the others. He noted that the Hughmark method may overpredict charge in the condenser yet still yield good agreement with the total charge by way of error cancellation with respect to unaccounted charge elsewhere in the system.

Other studies have compared predictive models with experiments. Damasceno et al. (1991) used HPSIM to compare predicted and measured capacity for various charge levels in a residential air-to-air heat pump. They found that they needed to modify HPSIM to include the Hughmark (1962) correlation to get acceptable results. They did not consider refrigerant-oil diffusion. Marques and Melo (1993) compared the predicted and measured charge inventory of a room air conditioner using HPSIM. They also found that the Hughmark correlation provided the best results. Furthermore, they found that the addition of a refrigerant-oil diffusion calculation was a necessary modification to HPSIM. LeRoy et al. (2000) used PUREZ to compare the predicted and measured performance of ten unitary air conditioners. They sought to reduce modeling errors through the use of two tuning methods. The recommended tuning method is to adjust the heat transfer coefficients to match the cooling capacity. They found that all of the untuned charge inventory results were less than the measured charge. Even the tuned results tended to underpredict charge. Of the eight available void fraction correlations considered, the Hughmark model led to the best agreement between the measured and calculated results. PUREZ does not account for refrigerant-oil diffusion effects. For three of the systems they considered a wider range of tuning parameters, including air-side and refrigerant heat transfer coefficients, pressure drop multipliers, refrigerant charge, superheat, subcooling, refrigerant mass flow rate multipliers, and compressor power multipliers. The tuned parameters were adjusted to match the predicted results with the measured data. The relative importance of each parameter was not considered.

PUREZ and HPSIM differ substantially in complexity. However, these models cannot be directly compared for the purpose of understanding the effects of model complexity because of their fundamental differences. For example, these models use different correlations for heat transfer and pressure drop in the heat exchangers. The literature suggests that the total charge inventory is routinely underpredicted. If the void fraction has been underpredicted in previous studies and void fraction correlations were selected to compensate for unaccounted charge, then the most accurate void fraction correlation available has yet to be identified. The literature also suggests that the effect of the uncertainty in two-phase flow parameters on system model predictions is unclear.

This paper presents a comparison of two system modeling approaches to determine the level of complexity required to obtain accurate system performance predictions. Furthermore, the impacts of various two-phase modeling parameters on performance predictions were carefully investigated. Each parameter under consideration was systematically varied, and the resulting predictions are compared. The relative importance of the uncertainty associated with each parameter was evaluated in this manner. At each step the models were validated against experimental data.