NUMERICAL INVESTIGATION AND PREDICTIVE MODELING OF HEAT TRANSFER IN PULSATING NANOFLUID JETS USING CONVOLUTIONAL NEURAL NETWORKS

Regression analysis of the enhanced heat transfer performance of a pulsating nanofluid jet impinging on a heated surface was conducted in this study using a convolutional neural network (CNN) model. The well-known multilinear regression (MLR) model was also applied for comparison. A comprehensive numerical analysis was performed by using ANSYS-FLUENT commercial software to evaluate heat transfer enhancement. Additionally, a basic experimental study was carried out to verify the numerical results. The variable parameters considered to be effective on heat transfer, and particularly the Nusselt number, included (i) wave type (sinusoidal, rectangular, and triangular), (ii) frequency (10, 20, and 30 Hz), (iii) amplitude (0.3, 0.4, and 0.5 m/s<sup>2</sup> ), (iv) Reynolds number (1000, 5000, 7500, 10,000, and 15,000), (v) dimensionless jet-to-plate distance (2, 4, 5, and 6), and (vi) nanoparticle concentration in the Al<sub>2</sub>O<sub>3</sub>-water mixture (0%, 1%, 2%, 4%, and 5%). According to the convolutional neural network (CNN) results, the mean squared error, mean absolute error, root mean squared error, and coefficient of determination were found to be 48.882, 4.303, 9.097, and 0.9609, respectively. The results of the CNN method were compared to those of the MLR method. It was concluded that the CNN method provides more accurate and reliable predictions regarding the effects of design parameters on heat transfer.