Thermal performance of a fan-driven airflow domain with AI analysis using LightGBM classifier and regression

Abstract

This research explores the thermal behavior of a fan-induced airflow system within a heated, grooved channel cavity, employing a hybrid approach that combines numerical simulation along with machine learning. The external airflow stream enters into a bottom heated grooved channel cavity, where a fan-shaped rotating body is placed at the center of the grooved cavity. The grooved cavity is fitted with horizontal inlet/ outlet ports. The Light Gradient Boosting Machine (LightGBM) algorithm is used to carry out classification and regression tasks for predicting temperature profiles and identifying fan positions under various rotational speeds (S2 = 2, 4, 6) and directions (clockwise and counter-clockwise). The governing equations for mass, momentum, and energy are solved using a finite volume method via ANSYS Fluent. Temperature data are collected at three vertical fan placements (H* = 0.25, 0.50, and 0.75) to evaluate the influence of fan speed and position on thermal performance. The results show that the average Nusselt number (Nu), increased by 44.71 % at S2 = 6, H* = 0.25 (CCW), compared to the stationary fan case. At H = 0.50*, Nu was 18.18 % higher with CCW rotation compared to CW at the same fan speed. Energy output enhancement with fan rotation relative to the non-rotating case at H* = 0.5 energy output increased by 117.14 % (CW) and 165.71 % (CCW). This shows that counter-clockwise rotation consistently enhanced energy output more effectively. Placing the fan at mid-height (H* = 0.50) resulted in the most efficient thermal mixing, regardless of rotation direction. The LightGBM classification model achieved an accuracy of 98.99 %, with F1-scores reaching up to 99.54 %, while regression analysis yielded R2 values as high as 99.97 %, confirming the model's strong predictive performance. The study demonstrates how integrating data-driven machine learning techniques with conventional computational methods can offer valuable insights into optimizing thermal systems. These findings have direct applications in areas such as HVAC design, energy storage systems, and industrial cooling, where precise thermal control is essential.