Abstract: An accuracy self-adaptive model for fin-in-tube heat exchangers is established, in which two artificial neural networks are combined with a simplified traditional mathematical model. One of the neural networks is used to compensate the difference between the distributed-parameter model and the simplified one, the other is used to improve the model accuracy by adaptively learning from experimental data. The model is used for predicting fin-in-tube heat exchangers and compared with experimental results. For condensers, it is shown that the average and maximum deviations of heat flow rate are 0.63% and 1.72% respectively, while the average and maximum deviations of subcooling are 0.9 ℃ and 3.2 ℃ respectively. For evaporators, the average and maximum deviations of heat flow rate are 1.56% and 11.0% respectively, while the average and maximum deviations of superheat is 1.5 ℃ and 9.8 ℃ respectively. For condensers and evaporators, the computational speed with the new model is about two orders of magnitude faster than that with the distributed-parameter model.

Key words: Artificial neural network, Heat exchanger, Model