AI-based framework for concrete durability assessment using generative adversarial networks and bayesian neural networks

The durability and longevity of concrete structures are critical concerns in civil engineering, impacting safety, maintenance costs, and sustainability worldwide. A major challenge is the ingress of aggressive ions, such as chlorides and sulfates, which can lead to significant deterioration. Despite numerous studies, predicting and mitigating these attacks remains complex due to environmental and material variability. This study presents a novel framework to model and predict the ingress of chlorides and sulfates into concrete considering uncertainty using advanced Machine Learning (ML) algorithms. The primary objective is to enhance prediction accuracy by leveraging augmented datasets and innovative modelling techniques. A dataset comprising 403 chloride and 172 sulfate data points was collected from laboratory experiments, and from the literature, and significantly expanded using a Generative Adversarial Network (GAN). To reliably predict the ingress of aggressive ions, the augmented dataset was used to train a Bayesian Neural Network (BNN), which quantified the uncertainty associated with its predictions, providing a more reliable assessment of concrete degradation. R 2 values of 0.95 were achieved by this framework in predicting chloride and sulfate ingress. This framework significantly improves prediction capabilities, offering a valuable tool for degradation assessment, service life estimation, and health monitoring of concrete structures. • A novel ML framework developed to assess chloride and sulfate ingress in concrete. • Generative AI, Bayesian Neural Networks and Shapley additive explanations are used. • The proposed model showcases suitable performance. • The framework can be utilized for automation and classification of ingress levels.