PNIPU: Program for predicting the characteristics of alloys

2024: Introduction of a program for predicting the characteristics of alloys

Perm Polytechnic scientists have developed a program to predict the characteristics of alloys. The university announced this on September 20, 2024.

Titanium alloys are used in the aerospace, medical and automotive industries due to their high strength, low weight and corrosion resistance. However, the lack of experimental data creates difficulties in predicting their characteristics, which slows down and worsens production. Scientists at the Perm Polytechnic University have developed a program for neural networks that accurately predicts the roughness of the alloy surface. The wear of the part when rubbing with other mechanisms or surfaces, as well as corrosion resistance, depends on it.

Certificate No. 2024668654 has been issued for development.

In recent years, machine learning methods have been used in various industries. One of the key features is the need for a large amount of data to train models. But in real industrial conditions, data collection is difficult or becomes financially costly. This is especially true for high-precision and complex processes, such as cutting titanium alloys.

So, for example, in the production of engines, alloy is used for the manufacture of parts of the air collector, housing, blades and compressor disks. To get a high-quality surface, you need to optimize cutting modes. This requires information about the effect of different parameters on the quality of processing in order to predict the final result. Neural networks are engaged in predicting roughness indicators. To expand the training sample and reduce the cost of conducting experiments, the method of artificially increasing the amount of data is used - augmentation.

Perm Polytechnic scientists investigated the regression model of augmentation and based on it created a program that solves the problem. Such models are built on a limited set of information. These were used to generate additional data, thereby creating an extended database of 2,000 examples. The results were used to train neural networks that predict the surface roughness of the VT6 alloy (common in aviation and rocket science).

The results showed that neural networks trained on augmented data achieved high prediction accuracy on the MAPE metric of 3.97%. This suggests that the error was only 3.97% of the actual values. In other words, this method is effective in conditions of a limited amount of initial data, - explained Vadim Danelyan, graduate student of the Department of Computational Mathematics, Mechanics and Biomechanics, head of the group of the Youth Design Technological Bureau of the Advanced Engineering School "Higher School of Aviation Engine Building" PNIPU.

We have created a program that not only selects the necessary cutting modes for titanium alloys VT6 to achieve the desired level of roughness, but can also be used in the processing of steels and other alloys, "said Andrey Klyuev, Candidate of Physical and Mathematical Sciences, Associate Professor of the Department of Computational Mathematics, Mechanics and Biomechanics at PNIPU.

The application of the program of scientists of the Perm Polytechnic allows you to expand the sample, implement the correct process of training artificial neural networks and achieve high accuracy in predicting the future characteristics of the alloy, including the roughness indicator, on which the wear period of parts depends. The proposed approach can be adapted for other material processing processes in various areas of industrial production.