The Using of Artificial Neural Networks for Reinforced Concrete Structures Condition Monitoring

The paper considers the problem of defect detection on concrete products using convolutional neural networks. The analyzed dataset is considered and the structure of the YOLO model is described, and comparative metrics for different processes are obtained. Performance metrics of neural network algorithms are plotted and examples of test image processing are shown. By using data augmentation algorithms it was possible to increase the accuracy of defect recognition and localization up to 70%. At the same time the stability of the algorithm under different deformations is investigated. It should also be noted that a detailed analysis of the distribution of defects in the image images (it is the spatial distribution, not quantitative) has been performed. The developed models can significantly enhance the monitoring of concrete structures without the use of invasive methods.