Reduced order surrogate modeling of opening along the interface due to debonding in non-metallic reinforced cementitious composites

Abstract

The key model component required for a realistic characterization of cementitious composites with non-metallic reinforcement is the crack bridge model, which can capture the triggering mechanism of splitting cracks in diverse design configurations and for different types of bond-slip laws. A crack-bridge model with a 3D representation of the stress state and inelastic material behavior is computationally intensive. Therefore, a reduced-order surrogate model that is able to map a variety of bond-slip laws to desired output can be effectively used, especially where such a mapping function needs to be called numerous times for parameter identification or design optimization. We considered a simplified crack bridge model in the form of an initial boundary value problem in which the cross-section of yarn is idealized circularly and pulled out by displacement-controlled loading from both sides while embedded in a cylindrical concrete specimen in the axisymmetry configuration. The geometry, boundary condition, concrete, and reinforcement material model parameters are assumed to be fixed and, consequently, are not considered in the mapping domain. A sequence of finite element simulations is performed over the pre-defined domain of the selected parameters of the interface material model. Under a monotonically increasing displacement load, the relative displacement vector, maximum hoop stress evolution in the matrix, concrete reaction, and pullout force are monitored. As an initial step in a research plan aiming to conduct a crack bridge surrogate model, in this paper we present a reduced-order surrogate model based on proper orthogonal decomposition (POD) to predict the opening field along the cohesive interface between the reinforcement and concrete. For different desired tolerance levels, the low-ranked surrogate model of the opening field is created, and the accuracy is evaluated for parameter sets newly inserted into the domain by comparing the prediction with the FEM using interpolation on POD coefficients. Finally, the results, potential methods for accuracy improvements, and future research plans are discussed.