Uncertainty Quantification in Material Properties of Additively Manufactured Materials for Application in Topology Optimization

Abstract

This research presents an approach to measuring the inherent randomness in properties of materials fabricated by the fused filament fabrication (FFF) method. Defects associated with the layer-by-layer process introduce significant variability in the elastic modulus field of materials printed. To describe the random distribution in Young’s modulus fields, statistical properties of mean, variance, and correlation length must be estimated for bulk regions (the printed filaments) and fusion regions (the thin regions connecting printed filaments). The goal is to estimate the random properties from the surface strain fields calculated by digital image correlation (DIC) analysis. A machine learning algorithm is developed that can estimate the spatial variations in the elastic modulus. The model is trained on a dataset of simulated two-dimensional strain fields with known random distributions in the corresponding elastic modulus fields generated by finite element (FE) simulations. On the test data, we achieved the R2 score of 0.93 and 0.95 for the mean in the bulk and fusion Young’s modulus fields, respectively. Also, for the variance in bulk and fusion areas, the R2 score of 0.74 and 0.83 are achieved, respectively. The results demonstrate the feasibility of the proposed approach in measuring the randomness in material properties of FFF-based printed materials.