Deep learning method for predicting the strengths of microcracked brittle materials

The strengths of brittle or quasi-brittle materials strongly depend on the interaction of distributed microcracks. Traditional micromechanics methods are difficult to exactly predict the strengths of materials containing a large number of microcracks. In this paper, a micromechanics-based deep learning method is proposed to predict the strengths of two-dimensional microcracked brittle materials. Utilizing a numerical method based on Kachanov's theory of microcrack interaction, we generate a data set containing a large number of images of two-dimensional microcracked specimens and their load-bearing capacity under various in-plane loading. A deep neural network is formulated based on this data set to establish the implicit mapping between the load-bearing capacity of the specimens and the spatial distribution of microcracks. Numerical experiments demonstrate that the trained deep neural network can accurately and efficiently predict the loadbearing capacity of microcracked brittle materials.