Scientists from South Ural State University have developed a software package for detecting structural defects in transparent products. First experiments were conducted on glass insulators, which are used in high-voltage power lines. The technology can be applied at any enterprise where glass products of complex shapes are manufactured, which will allow to assess their quality without destructive external influence.
In manufacturing, there are methods for detecting defects in flat rolled glass (for example, in phone screens or display cases), where all defects are detected when light is refracted. These methods do not work on curved surfaces.
“It is quite hard to notice damage on complex transparent products,” says senior lecturer of the SUSU Department of Applied Mathematics and Programming Vladimir Surin. “Firstly, glares interfere if the light falls from the side. And when light falls from below, the contrast drops. In addition, complex objects have different patterns and textures that can be mistaken for a defect. The uniqueness of our development is that we have created an algorithm that increases contrast and based on neural networks searches for uncharacteristic inclusions in an object.”
At the first stage of the work, the researchers trained the neural network algorithm to detect visible defects several millimetres in size. Further development of the project implies using Fourier images for detecting defects of fractions of a millimetre in size and even flaws invisible to the eye.
“If we pass light rays through an object, they are somehow refracted by defects. The smaller the defect, the stronger the refraction,” explains postgraduate student of the Department of Optoinformatics Kirill Belov. “We will use this property to obtain Fourier images of transparent objects of complex shapes with defects. By studying the characteristics of such Fourier images, we want to prove that small defects in transparent objects can be detected in a similar way, regardless of the complexity of their geometry.”
SUSU scientists combined traditional methods with the mathematical ones applied at enterprises. Until now, no one in the world has used Fourier images for flaw detection in transparent macro-objects with complex geometry.
The unique project was created as a result of cooperation with AO UMEK, which is part of PO FORENERGO, and was then supported by the Ministry of Science and Higher Education of the Russian Federation. The scientists received a state assignment for a period of three years. The research team included staff members of the Department of Optoinformatics; the Department of Electric Power Generation Stations, Networks and Supply Systems; and the Department of Applied Mathematics and Programming. In addition to detecting defects, the scientists study the electrophysical properties of products with defects.
“First of all, we study the effect of defects on the electrical strength of transparent objects,” says Associate Professor of the Department of Electric Power Generation Stations, Networks and Supply Systems Petr Lonzinger. “The requirements for glass suspension insulators made on the basis of transparent dielectrics with complex geometry are regulated by GOST 6490-2017. We think that these requirements are incomplete. In particular, the standard does not regulate the location of permissible defects in a glass part with an accuracy of up to millimetres. By studying the electrophysical properties of transparent objects with defects, we want to provide scientifically based recommendations that clarify, in particular, the requirements for the parameters of permissible defects in glass insulators.”
The obtained results can potentially be of interest to any enterprise that is engaged in producing glass products. The uniqueness of the software package for computer vision for identifying defects in glass insulators is confirmed by a patent for an invention. The Chelyabinsk scientists are planning to introduce the new method into large-scale production by 2028.