Application of Artificial Neural Networks in predicting the thermal performance of grooved heat pipes

Resumo

Heat pipes are versatile, relatively easy to construct, and capable of exchanging large amounts of heat between small temperature differences, even without external pumping. On the other hand, these devices have complex equations, which usually complicates their development, generating more extended periods of research and expenses. Methods that use computational intelligence, such as Artificial Neural Networks (ANN), have the ideal characteristics for use in problems of this type. ANN are algorithms that can solve complex problems using only experimental data, even without knowledge about the physics of the problem, limited only by the quality of the data used and the available computational power. In many cases, the results found using ANN have lower error percentages than those obtained using conventional methods. The database used was generated from an experimental investigation of the thermal behavior of heat pipes with a wicked structure of axial grooves and using water as the working fluid. The results were used to train two different Artificial Neural Networks. The Neural Networks used were the Multi-Layer Perceptron (MLP) and the Extreme Learning Machine (ELM). Filling ratio, slope, and dissipated power were used as inputs to the networks, and as output, we have the expected thermal resistance of the heat pipe. The results show that both ANN were able to generalize the problem, presenting errors of less than 25%. It is also possible to note that the MLP presents better results, with an error of about 18%. These values show that ANN are viable as a tool to improve the development of grooved heat pipes.