ICT2014 Abstracts

Prediction of performance of thermoelectric cooling material (ZT) with artificial neural network was carried out using simulation results obtained from finite element method as a training data. A total of 357 input vectors for ZT obtained from the 4 TEC modules modelled using the FEM analysis [1] were available in the training set to a back propagation feed forward neural network. The training data was obtained for dimensions ranging from 1 cm to 1 µm of the length and width from 100 cm² to 100 µm² area of ceramic plate. In the neural network model developed the input parameters were cold side temperature, length and width of thermoelectric leg and square area of sandwich material, current passing through them and voltage across the terminals of the module, while the output parameter was the performance of thermoelectric material (ZT) with temperature. The number of hidden layers and neurons in each layer were determined through trial and error to be optimal including with different transfer functions as hyperbolic tangent, sigmoid and hybrid. After the network was trained, better results were obtained from the network formed by the sigmoid transfer function in the 1^st and 3^rd hidden layers and the tangent hyperbolic transfer function in the 2^nd and 4^th hidden layers and output layer and predicted the ZT. The maximum contribution from input nodes was width of thermoelectric leg while minimum one was the voltage across the terminals of the module [Table 1]. An average correlation and maximum prediction error were found to be 99.99% and 0.12%, respectively for the ZT trained. The standard deviation of values also was 0.019%. A set of test data, different from the training data was used to investigate the network performance. The average correlation and maximum prediction error were found to be 99.99% and 0.07%, respectively for the TEC module tested.

Table 1. Percent contribution of input nodes to estimate the performance of thermoelectric material.

[1] F. Guldiken, Modelling of effect of physical dimensions of thermoelectric modules on thermoelectric cooling capacity using finite element method, M. Sc. Thesis, Uludag University, Bursa Turkey, 2011.