Predicting the thermal performance of a rectangular channel with asymmetric protrusion surfaces using machine learning algorithms

This study aims to simulate the thermal performance of asymmetric protrusion surfaces in a two-dimensional rectangular channel enhanced by machine learning analysis. The surfaces have varying heights and two distinct radii of curvature. We examined the impact of these factors on thermal performance, with the radius of curvature and height of protrusions ranging from 0.15 to 0.7 and 0.1 to 0.275 times the hydraulic diameter, respectively. We generated 108 data cases, covering Reynolds numbers from 4000 to 20,000, to train and validate the machine learning models. The results showed that the channel Nusselt number increased with the Reynolds number, while the Nusselt ratio decreased. Therefore, channels with protruded surfaces exhibited better thermal performance at low Reynolds numbers. Using machine learning across all Reynolds flow ranges, we identified the optimal dimensions for the best local performance. The analysis revealed that a surface with a pitch of 7.2 times the hydraulic diameter achieved the highest thermal performance of 2.154 at Re = 4000.