The key idea is to deliver an automated, machine-learning (ML) based algorithm to combine the results from different non-destructive testing (NDT) measurements on a reinforced concrete wall in order to generally improve the reliability of the corrosion assessment. To this aim, a set of data was acquired with various NDT methods. The data include concrete cover measurements, half-cell potential mapping data, electrical resistance measurements (Wenner probe) in different grid sizes, and images from the concrete surface. After registration and preprocessing (interpolations and feature scaling), a ML algorithm was applied to cluster the data into groups. The aim was to examine the outcome of ML in comparison to traditional data analysis with manual cross-links between the individual measurements to locate corrosion damages such as cracks or cross-sectional due to chloride-induced corrosion. The surface images were analyzed by a convolutional neural network trained beforehand for the classification task (concrete crack/no concrete crack on the surface) to gain an additional numerical feature. The actual corrosion state of the reinforcement (ground truth) was examined at several locations. The preliminary outcomes of the ML are comparable to the traditional method.

Key words: Reinforced concrete, Inspection, Machine Learning, potential mapping, corrosion, NDT