Modeling of refrigerant flow through adiabatic capillary tubes using neural network and response surface methodology.

This paper presents a new Response Surface Methodology based neural network approach to model the refrigerant flow through adiabatic capillary tube. Experimental data of ten different refrigerants in the literatures covering subcooled, two-phase and supercritical inlet conditions are collected as the database, which plays as an experiment rig. Box-Behnken design (BBD) and Central Composite design (CCD) are applied to determine a small dataset for neural network training. With BBD, 25 sets of data are selected for neural network training and the average deviation (A.D.), standard deviation (S.D.) and coefficient of determination (R2) of trained neural network for all data are 2.6%, 9.6% and 0.948, respectively. With CCD, 22 sets of data are selected and the A.D., S.D. and R2 for all data are 0.05%, 10.2% and 0.934, respectively. In addition, the results show that the proposed model is superior than classical polynomial response surface model in such a nonlinear problem.