A prediction model of artificial neural networks in development of thermoelectric materials with innovative approaches

Abstract

The fact that the properties of thermoelectric materials are to be estimated with Artificial Neural Networks without production and measurement will help researchers in terms of time and cost. For this purpose, figure of merit, which is the performance value of thermoelectric materials, is estimated by Artificial Neural Networks without an experimental study. P-and n-type thermoelectric bulk samples were obtained in 19 different compositions by doping different elements into Ca2.7Ag0.3Co4O9- and Zn0.98Al0.02O-based oxide thermoelectric materials. The Seebeck coefficient, electrical resistivity and thermal diffusivity values of the bulk samples were measured from 200 °C to 800 °C with an increase rate of 100 °C, and figure of merit values were calculated. 7 different Artificial Neural Network models were created using 123 measured results of experimental data and the molar masses of the doping elements. In this system aiming to predict the electrical resistivity, thermal diffusivity and figure of merit values of thermoelectric materials, the average R value and accuracy rate of these values were estimated to be 94% and 80%, respectively.