Advanced Machine Learning Driven Optimization of Radiator Heat Exchanger Performance for Automotive Applications

In this research, response surface methodology (RSM) based experimental investigation coupled with three machine learning techniques namely, Bagging (Bootstrap Aggregating), Categorical Boosting (CatBoost) and Random Forest (RF) are employed to optimize and model the performance of a radiator. Model development is performed by collecting experimental data by varying hot water flow rate (0.5-2.5l/min), cold air velocity (1-5m/s) and feeding temperature (50-60 °C). The best radiator effectiveness was figured out to be 71.9%. The model RSM was found to be capable of predicting effectiveness with a coefficient of determination (R^2) of 0.966 as compared to Bagging, CatBoost and RF techniques with an R^2 of 0.99, 0.99 and 0.98, respectively. The Mean Squared Error (MSE) of Bagging, CatBoost and RF techniques were found as 6.242, 2.876 and 8.837, respectively while the Root Mean Squared Error (RMSE) for Bagging, CatBoost and RF techniques were found as 2.498, 1.696 and 2.973, respectively. The CatBoost model emerged as the best model because it had the highest coefficient of determination, along with the lowest mean square error and root mean square error. SHapley Additive exPlanations (SHAP) analysis indicated that the ML models relied most heavily on feed temperature (mean |SHAP| of 10.23, 9.65 and 6.67 for Bagging, RF and CatBoost, respectively), whereas ANOVA of the RSM model identified air velocity and water flow rate as the statistically significant factors. The reasons for this difference are discussed. This research highlights the importance of a data-driven approach to optimizing and modeling radiators.