Utilizing artificial neural networks to predict the thermal performance of conical tubes with pulsating flow

A novel model of an artificial neural network is presented to predict friction factor and Nusselt number for pulsating flow in conically coiled tubes. The tubes are manufactured with coil torsions ranging between 0.02 and 0.052 and are subjected to a uniform heat flux. Experiments using pulsating flow with a pulsating frequency of 4 to 10Hz and a Reynolds number between 4836 and 12,562 were expanded. In terms of overall heat transfer improvement, the pulsing flow with a frequency of 4 Hz performs the best within the investigated range. Results demonstrated that the Nusselt number and thermal performance index increased when the Reynolds number and coil torsion decreased. The results predicted by a Feed-forward neural network were compared to the actual experimental data. A comparison showed that a feed-forward neural network could predict the Nusselt number and friction factor. The average root mean square error for the predicted Nusselt number was 4.6025. There is a good agreement between the predicted values and the experimental values. At the same Reynolds number and with a frequency of 4 Hz, reducing coil torsion from 0.052 to 0.02 improves Nusselt number by 23 % and 10 %, experimentally and in a feed-forward neural network, respectively.

Automated summary

This article presents a novel model of an artificial neural network that predicts friction factor and Nusselt number for pulsating flow in conically coiled tubes. Experimental results demonstrate that the best heat transfer performance is achieved at a pulsating frequency of 4Hz. Comparison with experimental data shows that a feed-forward neural network can accurately predict the Nusselt number and friction factor.