Estimation of forced heat convection in a rectangular channel with curved-winglet vortex generator: A machine learning approach

There are a limited number of studies in the literature on the effect of curved fins on heat transfer. In this study, the effect of novel fin geometry and angle of attack of winglet (alpha = 30 degrees, 60 degrees, and 90 degrees) on heat convection is estimated using a machine learning method. Airflow in the rectangular channel is investigated under constant heat flux (q=100 W/m2) and turbulence regime (5683 <= Re <= 17049) by experimental studies. Improvements in heat transfer are observed at different temperature values (T = 30 degrees C, 50 degrees C and 70 degrees C) of the plate on which the blades were attached. In order to investigate the effect of input parameters on the prediction accuracy, an artificial neural network structure consisting of curved fin angle, Reynolds number and heater plate temperature parameters is preferred. Heat transfer is estimated by feedforward backpropagation (FFBP) and multi-layer perceptron (MLP) neural network algorithm using a data set containing 45 empirically obtained values in the forced convection condition. The training network has been calculated through Scaled Conjugate Gradient with ten neurons in the hidden layer. The results obtained from the experimental data assessed in consequence of the comparison with the target value within the +/- 4 % diversion interval for all Nusselt numbers. Optimum mean square error (MSE), average relative deviation (ARD%) and correlation coefficient (R2) is obtained as 1.4x10-2, 0.1,771 and 0.9879 respectively in architecture of average Nusselt number datasets.

Automated summary

In this study, the effect of curved fin geometry and winglet angle of attack on heat transfer is estimated using a machine learning method. Experimental studies are conducted to investigate airflow in a rectangular channel under specific conditions, and an artificial neural network is used to predict heat transfer. The results show a high accuracy in predicting heat transfer compared to experimental data.