Estimation of macroscopic failure strength of heterogeneous geomaterials containing inclusion and pore with artificial neural network approach

Abstract

This work is devoted to estimation of macroscopic failure strength of heterogeneous rock-like and cement-based materials. Three representative microstructures are considered, respectively with a random distribution of pores, stiff inclusions, and both pores and inclusions in a pressure-sensitive plastic solid matrix. In the first part, a series of direct numerical simulations are performed by using a nonlinear fast Fourier transform (FFT) method. Different values of porosity and inclusion volume fraction are considered. The respective influences of pores, inclusions and their interactions on the macroscopic failure stresses are investigated for a large range of mean stress. The obtained results provides a new insight on the effect of interaction between pores and inclusion at the same scale. For this case, it is very difficult to obtain analytical solutions. In the second part, a specific model based on artificial neural network (ANN) is constructed for the prediction of macroscopic failure strength by using porosity and inclusion volume fraction as input parameters. This model is trained by using a dataset based on the results obtained from the numerical simulations. The accuracy of the ANN-based model is verified through different statistic indicators. The good performance of this model is finally shown through the comparisons between its predictions and the references solutions from the direct numerical simulations for three groups of heterogeneous materials.