AI System Detects Metal Defects in Real Time on Russian Production Lines
The technology is designed to replace unstable manual inspection and basic computer-vision algorithms.
Researchers at Moscow Polytechnic University have trained an artificial intelligence system to detect cracks, dents, and other defects in cast metal parts in real time directly on the production line.
So far, the quality of cast components in mechanical engineering has largely been checked by human inspectors. However, this outdated method cannot fully eliminate errors caused by the human factor, and the scale of modern manufacturing often makes thorough manual inspection difficult.
Simple computer-vision algorithms also struggle in complex situations, such as when metal surfaces are oxidized or the material itself is heterogeneous.
Diagnosing Defects Under Uncertain Conditions
The Moscow researchers combined a convolutional neural network that analyzes images of metal parts with a fuzzy-logic system designed to handle uncertainty. According to project author Sergey Kuzovov, the system does more than simply detect a crack – it evaluates the severity of the defect by accounting for a range of contextual factors.
The neural network will be trained on a labeled dataset of images showing defective parts. Researchers are collecting and annotating photographs, documenting the location, shape, and type of each defect. The fuzzy-logic module will handle ambiguous cases, producing a balanced assessment that accounts for uncertain conditions.
The final product is expected to be a fully integrated system ready for deployment in real manufacturing environments.
