Thermophysical Property Prediction of r32, r1234yf, and r454b Refrigerants Using Artificial Neural Networks

Abstract

Accurate prediction of the thermophysical properties of next-generation refrigerants is essential to improving the energy efficiency and environmental compatibility of cooling systems. Therefore, this paper developed an artificial neural network-based data-driven framework for the prediction of the density and viscosity of R32, R1234yf, and R454B at a wide range of temperatures (Ts) and pressures (Ps). Extensive datasets, validated by high-accuracy experimental measurements, were used to train and validate multilayer feedforward networks developed to provide nonlinear thermodynamic dependency features. The resulting models displayed good quantitative accuracy on all refrigerants. The root mean square errors regarding R32 were found to be 40.70 kg/m3 density and 0.0237 mPa·s viscosity, while the coefficients of determination of 0.98031 and 0.97195 were achieved for density and viscosity, respectively. In the case of R1234yf, the foregoing errors were 51.07 kg/m3 and 0.0256 mPa·s. Meanwhile, the coefficients were given as 0.96488 and 0.97983. The R454B model achieved the Maximum (Max) performance with 22.01 kg/m3 errors concerning density and 0.0044 mPa·s concerning viscosity, while attaining correlation coefficients of 0.99895 and 0.9937, respectively. Relative error analysis showed that all refrigerants had Maximum and mean deviations below 8 % and 25 %, respectively, for density and viscosity. That trend in predictions confirmed that density increased as T gradually increased, remaining nearly P-independent, while viscosity decreased nonlinearly with increasing T. The viscosity sets themselves showed little sensitivity to P. These results could validate the highly accurate and computationally efficient capabilities of artificial neural networks to replicate complex thermophysical behavior, as above, and hence serve as a rigorous alternative to empirical correlations for predictive design of sustainable refrigeration and air-conditioning systems. © 2025 Elsevier Ltd