• A fast surrogate material model for architected AM weaves is presented. • The training data is created using an automated labeling process. • The labeling finds the weaves’ internal forces using full-field displacement tests. • The results show a mean accuracy of above 90% for both normalized load and shape. Additively Manufacturing (AM) textiles enables the fabrication of architected structures that have tuned local mechanical properties to customize their design, e.g. , for personalization to the human body. For the inverse design of architected AM textiles, creating a computationally efficient mechanical simulation is still a major challenge. Continuum mechanics approaches lose their inherent simulation speed for the high element count that is required to simulate spatially varying designs and yarn-level behavior. Truss-based textile simulations typically have a low simulation time and high adaptivity, but the complexity of textile mechanics is challenging to model using simple truss elements. In this paper, we model weave structures as planar trusses with truss elements that have variable stiffness depending on strain, textile shear and textile design. We present an automated framework to evaluate the correlation between textile design, textile deformation and textile mechanics on a truss member level. This correlation is modeled using a discrete constitutive manifold, whose datapoints represent an empirical surrogate for traditional constitutive models. A design-sensitive surrogate material model formulated as an Artificial Neural Network (ANN) is trained on the constitutive manifold and is validated by evaluating the simulation accuracy of textile samples that vary their weave direction. The results show the capability of the ANN to simulate AM weaves with spatially varying designs on the order of seconds. This paves the way towards an efficient inverse design of architected AM textiles.
Wirth et al. (2026) studied this question.