Scientists of Perm Poly have created a neural network model that will allow recognizing the properties of metals and alloys from digital images. Artificial intelligence and process automation will help businesses quickly and inexpensively create new materials that are unique in their properties.
Industrialists, for example, have a constant interest in developing new materials. Therefore, "New materials and substances" is one of the areas of the Perm REC.
According to scientists of Perm Poly, neural network models will be able to optimize the work of engineers: reduce time and financial costs for research.
To select the optimal metals and alloys, engineers need to conduct a series of experiments to study the microstructure and measure the qualities of materials. Scientists have developed a technology for selecting such components, in which the neural network itself "recognizes" promising views from digital images of samples (micro-glyphs) without the need for experiments and assigns each of them to a specific hardness class.
Perm researchers used the VGG deep neural network for training and found out that it classifies microstructures of steels with high accuracy by hardness. They processed the initial information using the ResNet deep neural network and compared the results with experimental data. Scientists have proved that the neural network can be used as the core of an intelligent system for complex evaluation of materials.
— Unlike our analogs, we used deeper neural networks based on real data, not synthesized data. According to various estimates, we managed to achieve an accuracy of 66.2% to 92.1% in the model's operation. In addition, we conducted a unique study on the stability of the neural network and found out how many incorrectly marked images can distort the result — " says the author of the project, Candidate of Physical and Mathematical Sciences, Associate Professor of the Department of Computational Mathematics, Mechanics and Biomechanics at Perm Poly Andrey Klyuev.
According to the researchers, the development will be of interest to enterprises in the real sector of the economy. For example, in the aircraft industry, functional materials can be used to reduce the weight of aircraft and engines. This will increase the ship's competitiveness and reduce production costs. In addition, the development can be used in engineering and construction. In the future, the neural network model will become an "intelligent assistant" for the engineer at the enterprise, who will automatically select the method of manufacturing structural elements, determine the chemical composition of alloys and the program for their thermomechanical processing.
In the future, scientists plan to expand the range of properties by which the neural network will help developers select suitable materials for new industrial products.
The developers implemented the project as part of a grant from the federal target program aimed at research and development in priority areas of development of the Russian scientific and technical complex.
