Title: Prediction of Time-to-Corrosion Cracking of Reinforced Concrete using Deep Learning Approach
Author(s): Bakhta Boukhatem, Ablam Zidol and Arezki Tagnit-Hamou
Publication: Symposium Paper
Appears on pages(s): 629-647
Keywords: deep learning, fly ash concrete, genetic algorithm, time-to-corrosion cracking
This study presents an accurate corrosion prediction through an intelligent approach based on deep learning. The deep learning is used to predict the time-to-corrosion induced cover cracking in reinforced concrete elements exposed to chlorides ions. The key parameters taken into consideration include thickness, quality and condition of the concrete cover. The prediction performance of the deep learning model is compared against traditional machine learning approaches using neural network and genetic algorithms. Results show that the proposed approach provides better prediction with higher generalization ability. The efficiency of the method is validated by an accelerated corrosion test conducted on 91 and 182-day moist cured reinforced fly ash concrete samples with different water-to-binder ratios. The results are in agreement with the model predictions. They also show that using the proposed model for numerical investigations is very promising, particularly in extracting the effect of fly ash on reducing the extent of corrosion. Such an intelligent prediction will serve as an important input in order to assist in service life prediction of corroding reinforced concrete structures as well as repair evaluation.