Stochastic Modelling of Additively Manufactured Structures Using a Neural Network for Identification of Random Field Parameters

Abstract

Heterogeneous materials exhibit considerable spatial variations in properties, impacting structural performance and local stress and strain fields. Recent research has focused on considering material behaviour uncertainties and quantifying the impact of uncertainties on the structural response, requiring the definition of random fields for describing the material variability. In this paper, a stochastic modeling methodology for additively manufactured structures is implemented (forward model). The stochastic parameters are determined from experimental Digital Image Correlation (DIC) images (inverse problem) using a Convolutional Neural Network (CNN), and the CNN is trained using the forward model. Validation of the CNN estimates using previously unseen data shows adequate performance of the network, and consistent predictions are found when estimating the stochastic parameters from experimental results. The proposed methodology allows examination of the stochastic response and uncertainty quantification of additively manufactured structures, while requiring only minor experimental efforts to fully define the random fields. Once the CNN is trained, computational expense for predicting stochastic parameters is minimal.