The durability assessment of fly ash-based geopolymer high-performance concrete (HPC) under extreme environment treatment remains a subject of critical concern in structural material engineering sets. To enhance this performance, the present research develops a unique integrated framework based on physics informed and deep learning type analytical designs built to predict the durability and residual service life of geopolymer HPC subjected to freezing-and-thawing, sulfate, and thermal degradation. Firstly, it employs a Transformer Based Physics Informed Neural Network (TPINN) for modeling multi-degradation mechanisms with embedded physical constraints and self-attention mechanisms for accurate degradation mapping. Secondly, the Spatiotemporal Graph Neural Networks (ST-GNN) seize a localized pattern of crack propagation through time and space so as to step up the interpretability in damage evolutions prediction. Thirdly, the Deep Reinforcement Learning Based Mix Optimization (DRL MO) model seeks to maximize long-term retention of compressive strength while simultaneously searching for the best mix compositions free of fiber reinforcement constraints. Fourthly, Hypergraph Self Supervised Learning (HSSL) makes use of contrastive learning to model thermal deformability from micro to macro scales without requiring any labelled data samples. Finally, the Diffusion Probabilistic Model for Uncertainty Quantification (DPM-UQ) issues stochastic service-life predictions amounting to quantified failure risks in a risk-based decision-making process. This comprehensive framework considerably enhances predictive accuracy, interpretability, and decreases data need. Coupled with the insights produced from these models regarding service life extension and maintenance planning, this framework will hold considerable promise for suggesting a credible, scalable durability prediction of geopolymer HPC in extreme environments.
Vairagade et al. (Thu,) studied this question.