On dislocation-based artificial neural network modeling of flow stress

Computational design of materials processes has received great interests during the past few decades. Successful designs require accurate assessment of material properties, which can be influenced by the internal microstructure of materials. This work aim to develop a novel computational model based on dislocation structures to predict the flow stress properties of metallic materials. To create sufficient training data for the model, the flow stress of a precipitation–hardening aluminum alloy was measured by characterizing the dislocation structure of specimens from interrupted mechanical tests using a high resolution electron backscatter diffraction technique. The density of geometrically necessary dislocations was calculated based on analysis of the local lattice curvature evolution in the crystalline lattice. For three essential features of dislocation microstructures – substructure cell size, cell wall thickness, and density of geometrically necessary dislocations – statistical parameters of their distributions were used as the input variables of the predictive model. An artificial neural network (ANN) model was used to back-calculate the in situ non-linear material parameters for different dislocation microstructures. The model was able to accurately predict the flow stress of aluminum alloy 6022 as a function of its dislocation structure content. In addition, a sensitivity analysis was performed to establish the relative contribution of individual dislocation parameters in predicting the flow stress. The success of this approach motivates further use of ANNs and related methods to calibrate and predict inelastic material properties that are often too cumbersome to model with rigorous dislocation-based plasticity models.