Hybrid approach to predict the effective properties of heterogeneous materials using artificial neural networks and micromechanical models

In this article, an investigation was carried out to verify hybrid models capabilities to predict the effective properties of heterogeneous materials. A hybrid model ANN-phi is developed by combining artificial neural networks and micromechanical modeling. The homogenization approach used in this study is mainly based on Eshelby's inclusion problem. The ANN-phi model, once trained on an Eshelby's tensors database, showed an excellent predictive capabilities of the effective mechanical behavior and local stresses in heterogeneous materials. The obtained results with ANN-phi are compared to numerical estimations which are often costly in terms of computational time. The results presented in this work show that the developed hybrid model can provide a significant computational time saving by a factor up to 2000 for 104 phases while maintaining its accuracy and reliability.

Automated summary

This article investigates the capabilities of hybrid models in predicting the effective properties of heterogeneous materials. The developed ANN-phi model combines artificial neural networks and micromechanical modeling, showing excellent predictive capabilities once trained on an Eshelby's tensors database. The results indicate that the hybrid model can significantly reduce computational time while maintaining accuracy and reliability.