Optimizing heat convection in water heaters is crucial due to their industrial use. This research combines experiments and machine learning to enhance convection heat transfer using ultrasonic waves in a spiral heater within a cylindrical tank. Various parameters affect ultrasonic enhancement, including ultrasonic power (U.P.), transducers’ geometric configuration, heat source height (H.h.), and heater power (H.P.). Based on our experimental data, we obtained a correlation using ordinary least squares (OLS) and developed an averaging ensemble (AE) of classic machine learning (CML) models and a convolutional neural network (CNN) model. We aim to predict the temperature difference between the heater and the water bulk. The CNN model (R2 = 0.91) performed similarly to the AE of CML models (R2 = 0.90) and outperformed OLS (R2 = 0.8). Further, the applicability and sensitivity analyses revealed data validity (> 97%) and the most important parameters, respectively. Finally, we employ machine learning-based optimization. The optimal configuration, determined by the simplicial homology global optimizer, is 4 transducers at the corners of a square (configuration ID 1) with U.P. = 210 W, H.P. = 1000 W, and H.h. = 60 mm, resulting in Q/ΔT = 36.9 with a 1.11% error compared to experimental results.

Keywords