Experimental study and neural networks prediction on thermal performance assessment of grooved channel air heater

Abstract

This research presents a study on convective heat transfer and friction factor of turbulent airflow in a channel air heater fitted with isosceles triangular grooves only and isosceles triangular grooves combined with transverse baffles. Experiments were conducted within a rectangular channel with aspect ratio, AR=15 and height, H = 20 mm with three different groove depths (d) of 4, 6 and 8 mm, two different groove inclination angles (α) of 30° and 45° with two different triangular directions of down-stream (DG) and up-steam (UG) of air flow. The airflow rate was presented in terms of Reynolds numbers (Re) based on the inlet hydraulic diameter of the channel and was in the range of 6700 to 17,000. The experimental results showed a significant effect on the heat transfer rate and friction factor in the presence of the isosceles triangular grooves only compared to the smooth-wall channel, by about 1.23–1.91 and 1.12–3.76 times, respectively. While, the case study of isosceles triangular grooves (DG) combined with transverse baffles (TB) provided higher heat transfer rate and friction factor as compared to the smooth-wall channel, by about 1.32–2.00 and 1.25–3.83 times, respectively. The maximum thermal performance enhancement of 1.49 was achieved with the use of 30° down-stream isosceles triangular grooves with the depth of 8 mm combined with transverse baffles at Re ≈6700. Correlations for Nusselt number (Nu) and friction factor (f) for this work were also proposed. The neural networks models were designed for predicting the thermal enhancement factor (η) of a channel air heater with combined turbulators by using the Rapid-Miner Studio 9.5 software. The architecture of the model was 4–20–20–1 with Tanh–Tanh activation function and the use of 5–fold cross validation for the training and testing data segmentation together with the linear_sampling type at a number of the epoch of 1700 provided the best performance for the η prediction which had the coefficient of determination (R2) and mean squared error (MSE) of 0.998864 and 1.2 × 10−5, respectively.