A New Methodology Based on Artificial Intelligence for Estimating the Compressive Strength of Concrete from Surface Images

Abstract

This study used digital image processing and an artificial neural network (ANN) to determine the compressive strength of concrete in reinforced concrete buildings without coring. First, 32 concrete samples were produced in the laboratory, with different water-to-cement ratios, aggregate types, amounts of binder, compression values applied to fresh concrete, and amounts of additive. Next, the locations of 192 cores were visualized, and the compressive strengths of their corresponding core samples were matched with the surface images of the concrete, which were then digitized by image processing. The digitized images were the input layer, and the training and testing procedures were performed using the ANN as an output layer. After testing, the model was validated in existing reinforced concrete buildings. For the verification process, 20 cores taken from randomly selected concrete buildings were used. Although the results obtained from the samples produced in the laboratory were satisfactory, the success rate of the samples taken from the field was limited. Finally, the findings of this study are compared against the literature on this subject, especially from the last two decades.