A comprehensive study on indirect evaporative coolers: CFD-based performance analysis, geometric optimization and machine learning models

This thesis presents a comprehensive three-phase investigation into the performance and optimization of dew-point indirect evaporative coolers by combining computational fluid dynamics (CFD), geometric optimization and advanced predictive modeling. In the first phase, five cooler configurations (Type 0 to Type 4) with varying structural and fixed geometrical features were developed and systematically analyzed under a wide range of psychrometric and operating conditions. Key independent variables, including process air inlet temperature, humidity, velocity and working-to-intake air ratio, were varied to assess their influence on four primary performance indicators: dew-point effectiveness, cooling capacity, water consumption and cooling coefficient of performance (COP). Among the tested designs, multi-perforated Type 2 configuration demonstrated superior overall performance. Building upon these findings, the second phase of the thesis focused on the geometric optimization of the Type 2 cooler. Using the desirability function method, a total of 4,200 CFD simulations were performed, of which 2,225 valid cases (excluding reverse flow conditions) were used for optimization. Optimal design parameters, such as channel length, channel gap, aperture number, aperture diameter and working-to-intake air ratio were identified. The multi-objective optimization of the cooler was conducted across five different scenarios, each designed for a specific objective. In the final phase, predictor models were developed to reproduce CFD-level results at reduced computational cost. A total of 20,480 CFD simulations were conducted, and after excluding reverse flow cases, 17,525 valid results were used for model training and testing. Seven multiple linear regression (MLR) models and 32 machine learning (ML) models, including decision trees (DT), support vector machines (SVM) and multilayer perceptrons (MLP), were developed to predict performance parameters. Model accuracy was assessed through coefficient of determination (R²), root mean square error (RMSE) and relative error (RE) distributions. While MLR models offered practicality and ease of use, machine learning models, particularly the models coded as DT-7 and MLP-4, exhibited superior predictive accuracy across most performance parameters. This thesis demonstrates that the systematic integration of CFD analysis, optimization and predictive modeling provides a robust pathway for designing efficient dew-point indirect evaporative coolers and enabling fast, reliable performance predictions for real-world applications.