Aplicação de redes neurais artificiais na predição do desempenho térmico de tubos de calor e termossifões

Abstract

With the increasing energy consumption generated by technological advances, efficiency in energy use becomes more important every day. The loss of energy in the form of heat is a common problem in almost all areas of industry, and heat exchangers play an essential role in increasing efficiency and reducing energy waste. Heat pipes and thermosyphons are heat exchangers known for their high versatility and high efficiency in systems. In contrast to their high utility, these devices have, in many cases, complex theoretical equations with high percentages of error that can limit their development and cause dependence on empirical methods that generate waste of time and material, resulting in significant expenses and reducing the viability of its use. Thus, techniques that use computational intelligence, such as Artificial Neural Networks (ANN), can be an excellent option to facilitate the construction and development of this type of equipment since they can generate good results without knowledge of the complex theory behind the problem. ANN are computational algorithms based on the nervous system of higher organisms, which can solve complex problems using experimental data about the problem in question. These algorithms use known data from the proposed system to “learn” its behavior. In many cases, an ANN can achieve results as good or even better than those obtained by analytical methods and are limited only by the quality of the database and the available computational power. This investigation uses experimental data from previous studies to evaluate the ability of ANN to predict the thermal performance of thermosyphons and heat pipes. A database of thermosyphons and heat pipes was organized and applied to neural networks to evaluate their ability to simulate the proposed systems. The neural networks used are the Multilayer Perceptron (MLP), the Neural Network with Radial Basis Function (RBF), and the Extreme Learning Machine (ELM). Design properties were used as inputs to the neural networks, and the thermal resistance of thermosyphons and heat pipes was expected as output. The results show that the ANN can successfully simulate the properties of thermosyphons and heat pipes. The best results for thermosyphons showed Mean Absolute Percentage Error (MAPE) of less than 6.00% using the ELM network. The lowest error for heat pipes was achieved with the MLP network, resulting in a MAPE of around 13.93%. Thus, using ANN in developing thermosyphons and heat pipes is an excellent option for accelerating and improving the project phase, reducing the loss of material, time and other resources.