Prediction of absolute Seebeck coefficients at ITS-90 temperatures using an artificial neural network

A three layer feed-forward artificial neural network (ANN) was utilised to process the complex dependence of the absolute Seebeck coefficients (ASC’s) of pure palladium and platinum on their thermodynamic properties. The latter were computed using molecular dynamics (MD) simulations, which, together with experimental ASC’s data from the literature formed the training data for a neural network. A further test set was predicted at an rms of 0.3, enabling the interpolation of ASC’s at sixteen ITS-90 temperatures to be predicted. These ASC’s can be used to extend the response range of thermocouples.