A team of scientists from the SUSU Centre of Vibration Testing and Monitoring of the State of Structures is developing a neural network that could instantly predict fatigue of metals and prevent production accidents.
As part of implementing a series of grants from the Russian Science Foundation (RSF), the research group, headed by Deputy Director of the Centre of Vibration Testing and Monitoring of the State of Structures, Associate Professor of the Department of Aircrafts, Candidate of Sciences (Engineering) Aleksey Erpalov, is actively working with its industrial partners in order to create a unique software product. The project's main goal is to provide engineers with a tool that in a fraction of a second could predict the fatigue life of metal structures.
In essence, hybrid neural networks are used to abandon the traditional and extremely time-consuming mechanical simulation. In the classical approach, engineers have to wait for the results of complex numerical computations, while the new system trained on verified data on the mechanics of destruction yields a prediction practically instantaneously.
"We do not improve the accuracy, we rather strive to speed up the process," explains Aleksey Erpalov. "Today, the pace of life and of industry development requires high speeds. Things need to be done very quickly, but with good quality at the same time. For example, a design engineer changes a mounting arm or enhances a structure's rigidity, and needs to immediately understand how it will influence the life of the structure. Our model will provide him with an answer, even if the loads are of random or fluctuating character, like, for instance, vibrations in a car on different types of road pavement."
Despite placing the stakes on the speed, the accuracy of predictions still remains quite high: around 95 – 99%, that is, comparable to classical methods, but achieved several-fold quicker. The specific feature of this approach is the focus on large structures made of isotropic materials (steel, aluminium alloys) without welded joints and operated in a multiple-cycle-fatigue mode for years on end.
This development has two strategically important applications for industry:
-
Designing: speeding-up the work of design engineers that helps quickly assess the impact of a part's geometry on the strength, without the need to attract domain specialists on complicated calculations.
-
Online monitoring: creating digital twins of equipment (for instance, of rolling mills) for the transition from a scheduled repair system to condition-monitored maintenance. The system will be able to analyse current anomalies, vibrations, and to predict the remaining life and prevent sudden breakdowns.
This project is being implemented with the support from the "Step into the Future" V.B. Khristenko's Grants Program and is of a clearly pronounced applied nature. The team is already actively collaborating with the Chelyabinsk Region's enterprises and big businesses, including Magnitogorsk Iron & Steel Works, Severstal company, and Automobile Plant URAL. According to Aleksey Erpalov, there is no publicly available ready-made analogue of such system either in Russia, or abroad.
The work will result in registering the software for computers (result of intellectual activity). In the first phase, this software can be used by design engineers to cut the price of and speed up the process of creating complex mechanical-engineering products. In the future, it is planned to expand the range of parameters taken into account by including the analysis of the impact of corrosion and microstructure of materials.
The project is being fulfilled as part of the "Step into the Future" V.B. Khristenko's Grants Program. This is a program of providing annual financial support for the university development under the Priority 2030 project. Grants are allocated for conducting advanced research, forming a personnel reserve, and implementing unique academic programmes (components) that determine the vector of development for SUSU and the Chelyabinsk Region.