Utilisation de la méthode des réseaux de neurones artificiels pour la prédiction du potentiel de corrosion

Abstract

Preventing corrosion in reinforced concrete structures is a major challenge for the construction industry. The use of non-destructive methods such as Canin+ has marked a significant advancement in assessing structural integrity. However, accurately predicting corrosion potential remains a crucial goal to ensure the durability of infrastructures. Although non-destructive methods have made progress, predicting corrosion remains challenging due to its complexity. Traditional mathematical models do not always capture the intricate interactions involved. Therefore, more sophisticated approaches are needed for precise prediction of corrosion potential in reinforced concrete structures. This research, conducted at the Civil Engineering and Public Works Laboratory of Aïn Témouchent University, developed several test devices to evaluate the risk of corrosion, notably through non-destructive methods like Canin+ on hardened concrete samples. Experimental data on corrosion potential were subjected to statistical analysis, including multiple linear regression, to assess their reliability. Subsequently, artificial neural networks (ANNs) were utilized to develop a model aimed at predicting the corrosion potential of reinforcements in concrete samples. The model based on artificial neural networks successfully predicts corrosion potential with a satisfactory level of accuracy