Nikolai Obidin, a student at the Faculty of Applied Mathematics and Computer Science at Novosibirsk State Technical University, created a crack detection system that uses artificial intelligence and machine learning to analyze images and video from surveillance cameras. The development will avoid the risk of accidents and economic losses.
"Cracks in concrete may be invisible to the naked eye, but they lead to serious damage. Their untimely detection increases the likelihood of structural collapse. Regular monitoring helps to avoid serious financial costs: repair of emergency buildings is more expensive than preventive maintenance. Modern technologies such as machine learning open up new possibilities for automated diagnostics. As part of the project, a large data set was collected, including images from various objects, and a basic neural network was created. I trained it to detect cracks in concrete based on images processed and segmented using the contour method (the contour is made on a special application, it is highlighted so that the neural network can recognize these cracks). The system analyzes video streams from cameras, then the data is processed: artificial intelligence detects cracks based on a trained model," said Nikolai Obidin.
The created model showed an accuracy of 95% crack detection. Currently, a prototype of the system has been developed, which includes a video processing module. The prototype has been successfully tested. During the pilot project, the system detected 15 cracks, of which 10 were missed during manual inspection.
The advantages of the development include timely identification of problems: the system allows you to detect cracks at an early stage, preventing the development of serious damage and emergencies; saving time and resources: automated monitoring reduces labor costs for manual control, as well as reduces repair and maintenance costs. The use of artificial intelligence and machine learning guarantees high crack detection accuracy, which is important for making informed decisions.
The target audience of the project is construction companies interested in diagnosing the condition of buildings to prevent accidents and save on repairs; municipalities and government agencies responsible for the safety of public buildings, bridge structures and infrastructure in general; gas and oil companies, engineering firms, owners of commercial real estate.
The plans include model optimization and scaling, user interface development, testing and refinement, and launching pilot projects. "In the future, I would like to see a trained model implemented in drones, as well as underwater vehicles that will monitor the condition of reinforced concrete structures and fix cracks both above and below water," adds Nikolai Obidin.
The presentation of the project took place within the framework of the NSTU-NETI REACTOR acceleration program, which is implemented within the framework of the federal project "Platform of University Technological Entrepreneurship" of the state program "Scientific and Technological Development of the Russian Federation".