Using artificial neural networks to predict grain boundary energies

• A solution to the unsolved problem of predicting the energy of a grain boundary given its misorientation is proposed.
• An average error of 4% for the predicted grain boundary energies is obtained.
• The capabilities of artificial intelligence methods and their applicability in the materials science domain are shown.

Artificial Neural Networks (ANNs) have been used in a few domains of materials science (Prechelt, 1997) [1], but never for the prediction of Grain Boundary (GB) energies. In the present article, an ANN is used to generate – for the first time – a function for the GB energy in terms of its five macroscopic degrees of freedom. The proposed approach is verified for GBs of body centred cubic iron. Part of the database calculated by Kim et al. (2011) [2] is used as training data for the ANN. After the ANN has been trained (i.e. after it has learned to replicate and predict the function), the magnitude of the errors in predicted GB energies for the remaining part of the database is about 4%4%, which is lower than the error of 10%10% that is typical for experimental GB energy measurements (Rohrer et al., 2010) [3].