An experimental comparative analysis and machine learning prediction on the evaporation characteristics of R1234yf and R290/R13I1 in a plate heat exchanger

The present research attempts to comprehensively compare the evaporation characteristics of a novel R290/R13I1 (35/65 % by mass) with R1234yf in an offset-strip fin embedded plate heat exchanger. The impact of various testing phenomena, namely saturation temperature (T-s) (278 to 288 K), heat flux (q) (4000 to 10,000 W m(-2)), entry vapor quality (x(i)) (0.1 to 0.8), and mass flux (G) (40 to 80 kg m(-2) s(-1)) have been explored. Meanwhile, highly potential machine learning algorithms (MLAs) namely Linear Regression (LR), Multi-Layer Perceptron (MLP), and Extreme Gradient Boost regression (XGB) have been employed to predict the evaporation heat transfer coefficient (EHTC) and evaporation frictional pressure drop (EFPD) of the refrigerants. Findings revealed that the EHTC of R290/R13I1 is significantly lower than that of R1234yf by 7.9-38.8 % in the nucleation boiling or low mean vapor quality (x(m)) domain, whereas it had superior EHTC by up to 18.2 % in the convective boiling domain (high x(m)). Interestingly, there was a dry-out incidence at mid-x(m) ranges (0.35-0.5) for both refrigerants, except for R290/R13I1 at a higher G of 80 kg m(-2) s(-1). In all cases (except at 10000 W m(-2)), the EFPD of R290/R13I1 increased by 0.3-11.1 % compared to that of R1234yf. The evaporation thermo-hydraulic performance (ETHP) factor analysis revealed that utilizing R290/R13I1 could perform satisfactorily in the convective boiling domain (x(m) > 0.5) with an ETHP factor ranging between 0.8 and 1.08, especially at high q, high T-s, and low G conditions. New empirical correlations have been developed based on the experimental dataset for the EHTC and EFPD of the considered refrigerants with an mean absolute error (MAE) of up to 14.7 % and 13.4 %, respectively. Among the three MLAs with different enhancement methods, the EHTC and EFPD predictions using MLP, in combination with principal component analysis and hyperparameter tuning, had superior performance, with MAEs of 0.1119 and 0.0581, respectively, for R1234yf, while they were 0.1726 and 0.0482 for R290/R13I1.