Application of neural networks for detecting defects and damage in metal structures

The rapid development of neural networks has led to the integration of these technologies into various industrial sectors. At the same time, improving the accuracy and efficiency of detecting defects and damages, including in real-time, remains a critical task. By combining neural networks with the Internet of Things (IoT) and technologies for data collection, storage and protection, it is possible to create a comprehensive and effective information-measurement system for surface defect detection. In this context, the present work highlights recent advances in the application of artificial intelligence for quality control, as well as the detection of defects and damages in structures. The focus is on the development and training of neural networks capable of effectively identifying and classifying various types of defects. The study demonstrates how these technologies significantly improve the speed and accuracy of diagnostics compared to traditional visual and instrumental inspection methods. The results of model testing on real industrial data confirm the high efficiency of the proposed approach. Additionally, the authors have developed an algorithm and implemented software for the automatic annotation of images in a format suitable for modern architectures such as YOLO. This approach enables the effective application of the model for detecting damages on the surfaces of structures and systems using widely available types of datasets.