Auto-Regressive U-Net for Full-Field Prediction of Shrinkage-Induced Damage in Concrete

TL;DR

The study uses an auto-regressive U-Net model to predict the evolution of the scalar damage field in a unit cell given microstructural geometry and evolution of an imposed shrinkage profile, and uses the predicted damage output as input for subsequent predictions to facilitate the continuous assessment of damage progression.

Abstract

This paper introduces a deep learning approach for predicting time-dependent full-field damage in concrete. The study uses an auto-regressive U-Net model to predict the evolution of the scalar damage field in a unit cell given microstructural geometry and evolution of an imposed shrinkage profile. By sequentially using the predicted damage output as input for subsequent predictions, the model facilitates the continuous assessment of damage progression. Complementarily, a convolutional neural network (CNN) utilises the damage estimations to forecast key mechanical properties, including observed shrinkage and residual stiffness. The proposed dual-network architecture demonstrates high computational efficiency and robust predictive performance on the synthesised datasets. The approach reduces the computational load traditionally associated with full-field damage evaluations and is used to gain insights into the relationship between aggregate properties, such as shape, size, and distribution, and the effective shrinkage and reduction in stiffness. Ultimately, this can help to optimize concrete mix designs, leading to improved durability and reduced internal damage.

Figures (29)

• Figure 1: Schematic representation of the modelling approach for the simulation of a concrete mesostructure subjected to uniform shrinkage of mortar restrained by elastic aggregates (Uniform shrinkage scenario, Section \ref{['sec:uniform-simul']}).
• Figure 2: Schematic representation of the modelling approach for the simulation of a representative section of an internally restrained concrete beam subjected to nonuniform shrinkage of mortar (Nonuniform shrinkage scenario, Section \ref{['sec:nonuniform-simul']}).
• Figure 3: Shrinkage profiles prescribed to the mortar phase within the concrete RVE for the non-uniform shrinkage scenario (Section \ref{['sec:nonuniform-simul']}). Solid colored lines represent scaled reference profiles exported from a hygro-mechanical FEM simulation HAVLASEK_creep_2025, while the black dashed lines illustrate the linear interpolation performed for two selected values of pseudo-time.
• Figure 4: Examples of the generated microstructures with the level-set method.
• Figure 5: Auto-regression scheme. Two networks, a U-Net (depicted in blue) and a CNN (depicted in orange) are used to predict time-dependent full-field damage $\omega(\bm{x},t)$, macroscopic shrinkage $\varepsilon_{o,sh}(t)$, and effective residual stiffness $k(t)$. The output at the pseudo-time step $t$ is used as an input at the pseudo-time step $t+1$.