Heat exchanger modeling, Kriging metamodel, multi-zone model, vapor compression system simulation
An accurate, fast and robust heat exchanger model is critical for reliable steady state simulation of vapor compression systems. In such simulations, the heat exchanger models are often the most time consuming components and can be plagued by severe non-linearities especially if they are black-box or third-party provided. This paper investigates and compares different approaches for heat exchanger performance approximation, with the distributed parameter approach being the baseline. The methods are: Gaussian kernel based on Kriging, a multi-zone approach, and polynomial regression. Generally, distributed parameter models have the highest level of accuracy but can be time-consuming. Kriging metamodels have relatively low computational cost but has little underlying physics. Multi-zone models have the lowest computation cost due to the lump treatment of heat transfer and pressure drop; however, they also tend to have the least accuracy. To better understand the potential and limitations of those heat exchanger modeling methods, the pressure drop and capacity of the same heat exchangers predicted by the three approximation modeling methods are compared against the baseline approach under the same operating conditions. The comparison between the Kriging metamodel and the distributed parameter model shows that 95.2% out of 10,000 test points have capacity deviation less than 20%, and that 93.9% have pressure drop deviation less than 10%. Large capacity deviations occur at those operating conditions with low inlet pressures, while large pressure drop deviations occur at those with high inlet pressures. The multi-zone model presents relatively larger deviations in terms of both pressure drop and capacity when compared with the distributed parameter model. Thus, regression based techniques are applied to further improve the accuracy of the multi-zone model. The heat exchanger modeling approaches are incorporated to a vapor compression cycle model. Lastly, some ideas on how such an approach can be used to approximate a set of components models, not just heat exchangers, are discussed.