Highlights

• •The machine-learning solutions of fracture toughness and path in the presence of micro-defects are presented.

• •The data set of fracture behaviors affected by defects is obtained by the phase field fracture and distributed dislocation methods.

• •Fracture toughness and path can be well predicted by neural-network models, and the square of correlation coefficient is more than 0.99.

Abstract

The effect of micro-defects on the fracture toughness and path is predicted by a machine learning method. The data set of fracture toughness is obtained based on the distributed-dislocation-technique solution, and the data set of fracture path is built based on the phase field fracture simulations. The neural network models are applied to approximate the nonlinear relationship between the micro-defect parameters (inputs) and the fracture parameters (outputs). The results show that the trained neural network models have a strong fitting ability, and the square of correlation coefficient is more than 0.99. Based on the trained models, the micro-crack toughening zones and the fracture path in the presence of a micro-void can be easily obtained, which is useful for toughening design and predicting fracture behaviors of brittle materials.

Link

https://doi.org/10.1016/j.engfracmech.2022.108900