Library Subscription: Guest

ISSN Online: 2377-424X

ISBN CD: 1-56700-226-9

ISBN Online: 1-56700-225-0

International Heat Transfer Conference 13
August, 13-18, 2006, Sydney, Australia

REPRESENTATION OF THE PERFORMANCE OF A FIN TUBE HEAT EXCHANGER BY ARTIFICIAL NEURAL NETWORKS

Get access (open in a dialog) DOI: 10.1615/IHTC13.p18.50
12 pages

Abstract

This paper employs experimental data to train artificial neural networks to represent the heat transfer characteristics of a compact, fin-tube heat exchanger with air and water/ethylene glycol anti-freeze mixtures as the working fluids. The experimental measurements were undertaken over a range of flow rates and inlet temperatures and with various ethylene glycol concentrations. In addition, the inlet air flow was distorted by obstructing part of the inlet duct near the front face of the exchanger. This simulated the type of flow maldistributions that can occur in practice due to, for example, the use of a damper or simple butterfly valve to control the air flow rate. A proportion of the measurements were used to train the neural networks and the remainder of the results was employed as unseen testing data to assess the accuracy of the network predictions. The fluid flow rates, inlet temperatures, mixture concentrations, and the percentage obstruction of the inlet duct were employed as inputs to the networks. These networks were able to predict the overall rate of heat transfer in the exchanger with a high degree of accuracy. Similar good agreement was obtained between the measured and predicted spatial variations in the exit air temperatures across the outlet face of the exchanger even under highly distorted flow conditions. The artificial neural networks were therefore able to predict both the overall and detailed thermal behaviour of the compact heat exchanger. The results indicate that, once trained, these networks can predict heat exchanger performance from knowledge of the composition of the fluid mixtures and a crude estimate of the inlet flow maldistribution. Detailed information on the transport properties of the fluids and the inlet velocity distributions are not necessary. The neural network models were then employed for “fault detection” by identifying and classifying the deterioration in exchanger performance associated with different levels of inlet obstruction.