Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
ARTIFICIAL NEURAL NETWORKS FOR PREDICTION OF THERMOSYPHON PERFORMANCE
Submission Author:
Thiago Antonini Alves , PR , Brazil
Co-Authors:
Thomas Siqueira Pereira, Yara de Souza Tadano, Hugo Siqueira, Thiago Antonini Alves
Presenter: Thiago Antonini Alves
doi://10.26678/ABCM.COBEM2023.COB2023-1505
Abstract
Thermosyphons are heat exchangers known for being versatile, easy to construct, and highly efficient for small temperature gradients. These components have highly complex equations with high error percentages. Because of this, achieving the necessary results is often complicated and highly time-consuming. Artificial Intelligence methods like the Artificial Neural Networks (ANN) are an excellent option to overcome this issue. ANN are computer algorithms based on the animal neural system that allows solving complex problems using only simple mathematical operations such as additions and multiplications. This algorithm uses known data of the proposed approach or similar cases to "learn" the system's behavior. For this experimental investigation, data was collected from different systems using thermosyphons and used to evaluate the capacity of proposed ANN to predict the thermal performance of thermosyphons. The experiment used thermosyphons made of copper tubes filled with distilled water as the working fluid. The heat source was simulated by a metal electric ribbon wrapped in the evaporator and heated by the Joule’s Effect, and the cooling was made by air forced convection in the condenser. Three ANN algorithms were used for the evaluation of the proposed systems: the Multilayer Perceptron (MLP), the Radial Basis Function (RBF), and the Extreme Learning Machine (ELM). The considered inputs were: the slop, the filling ratio, and the dissipated power, and as outputs the thermal resistance of each thermosyphon. The results showed that all ANN successfully predicted the experimental values with less than 15% error and that the ELM has better results with less computing time and no more than 10% error.
Keywords
Thermal performance, Extreme learning machine, phase change, artificial neural network
