This study explores the performance augmentation of a regenerative counterflow indirect evaporative cooler (RCFIEC) both experimentally and numerically. A counter flow heat/mass exchanger between the dry channel and the wet channel is effectively designed using a silica gel wet material. The RCFIECS is examined under different operational scenarios and thoroughly assessed in terms of wet-point effectiveness, and dew-point effectiveness, energy efficiency ratio, and cooling capacity. More so, leveraging machine learning models including Categorical Boosting Regressor (CatBR), Extra Tree Regressor (ETR), Random Forest Regressor (RFR), Voting Hybrid Regressor (VHR), Support Vector Regressor (SVR), and SVR coupled with Particle Swarm Optimization (PSO) are also developed to predict the performance parameters of the RCFIECS. Furthermore, the robustness of the six models is assessed using eight different statistical measures. The experimental findings indicated that the wet-bulb effectiveness of the proposed RCFIEC varied from 65 % to 110 %, while the dew-point effectiveness varied from 45 % to 80 %, at inlet air velocity varied from 0.15 to 0.90 m/s, respectively. Additionally, the system maintained improved cooling performance in which the averaged values of dew effectiveness, energy efficiency ratio, cooling capacity, and average cold air temperature are 0.60, 0.788, 31.6 W, and 25.0 °C, respectively. Furthermore, the statistical analyses prove that the optimal prediction accuracy achieved by ETR, followed by SVR-PSO, CatBR, VHR, and RFR, respectively, while the least accuracy yielded by the standalone SVR method. Specifically, the deterministic coefficient and RMSE of the predictive wet Effectiveness are 0.9978 and 0.0103 for ETR, and 0.9977 and 0.0128 for SVR-PSO, and 0.5515 and 0.1509 for typical SVR, respectively. Conclusively, the ETR approach can be deemed a potent optimization tool not only for forecasting the energetic performance of building cooling systems but is also valuable for various real-time applications.
Read full abstract