Russian University Develops Neural Network to Detect Metal Defects
The system automatically identifies cracks, dents, and corrosion spots on steel surfaces using images from a standard camera.
Researchers at Novosibirsk State Technical University NETI have developed an intelligent quality-control system for industrial use. The system automatically detects cracks, dents, and corrosion spots on steel surfaces using photographs captured with a regular camera. The university’s press service told IT-Russia about the development.
Learning to “Understand” the Nature of a Defect
The system was developed by students from the university’s Faculty of Automation and Computer Engineering. At its core is a triplet neural network that does not require thousands of labeled images. Just a few examples are enough for it to accurately recognize a defect.
Existing systems typically require ideal images and massive datasets. What sets the university’s solution apart is its ability to accurately identify defect types while training on low-quality photos taken under poor lighting conditions and at oblique angles.
Quality Control and Predictive Maintenance
The system has already demonstrated its effectiveness. Detection accuracy exceeded 87 percent, significantly outperforming traditional methods. According to the developers, it is a ready-to-use solution for enterprises where collecting thousands of defect samples is either impossible or prohibitively expensive.
Looking ahead, the system could be adapted to monitor the condition of bridges, pipelines, and other structures where uninterrupted operation is critical.
Earlier, IT-Russia reported that a team from the Industrial Robotics Development Center at Innopolis University had developed the SPRUT ultrasonic tomography system. It examines metal structures and complex-shaped components, detecting internal defects 30 percent more effectively and increasing inspection sensitivity by 20 percent.
