Time-Dependent Reliability Assessment of Mass Concrete Structures: Leveraging an LSTM-Based Surrogate Model for Early-Age Thermal Cracking under Parametric Uncertainty

Mass concrete structures encounter a significant challenge with heat dissipation during hydration, often leading to thermal cracks due to nonuniform temperature distribution. An accurate reliability estimate concerning thermal cracking is essential to mitigate the risks of failure. Since concrete gains strength over time through hydration, this reliability assessment must inherently be time-dependent. In this context, a case study of mass concrete lift during dam construction was undertaken to evaluate the probability of structural failure over time. A finite element (FE) model, parametrized with critical factors such as lift height, pouring temperature, time-dependent material properties (thermal and mechanical), and heat generation due to hydration under thermal and mechanical constraints, was developed. These parameters were treated probabilistically, following appropriate distributions considering the material uncertainty to assess the risks of thermal cracking and its progression. To avoid the complexities associated with simulating a high-fidelity finite element model for the thermomechanical analysis of a mass concrete continuum, a data-driven surrogate model utilizing a long-short-term memory (LSTM) framework was established. This model aimed to evaluate the failure probability related to the mass concrete lift while maintaining an acceptable level of precision. The assessment of failure probabilities was based on outcrossing failure events, which highlighted significant risks inherent in the approach. In response to these risks, the implementation of a reduction in placement temperature was worked out as a risk mitigation strategy that effectively helped minimize the occurrence of cracking. The findings of this study contribute valuable insights for developing robust risk mitigation strategies aimed at preventing thermal cracking at an early age.