Identification of the Damage Types for Reinforced Concrete Using CNN Models

Monitoring reinforced concrete structures makes it possible to prevent their further destruction, thereby significantly reducing the cost of their maintenance. Machine learning excels in precisely identifying and classifying various types of damage. The convolutional neural network, constructed based on deep learning principles, could precisely identify and categorize many occurrences of deterioration in pictures of reinforced concrete. The evaluation of the convolutional neural network’s performance in this study included assessing parameters such as accuracy, efficiency, and loss, and validating the enhanced architecture. Using a database that included 3200 images, it was found that a CNN containing 3 convolutional layers, 3 pooling layers, 1 rectified linear unit, filter 32/64/128, kernel size 4×4, and dropout 0.5 shows the best values in Accuracy (71.67–84.38%), Efficiency (14.4–2370 s), and Loss (0.99). The best results this CNN shows in the estimation of the combination of damages (crack + rebar exposure): Accuracy – 98.81%; Efficiency – 3945 s; Loss – 0.57. So, the use of the convolutional neural network model enables the identification of various forms of damage and the automation of the assessment of surface conditions in reinforced-concrete structures.