Inspenet, August 19, 2023.
The results of this research have been published in the journal Energies.
Russian scientists have developed a failure prediction system for gas turbines and other power generation facilities. This invention is a neural network designed to detect irregularities and proves to be twice as accurate compared to similar solutions, promising a decrease in costs related to equipment maintenance and repair.
Researchers from the Volgograd State Technical University (VolgGTU) have highlighted the vital importance of ensuring the reliability of power plants and other components within the field of energy and fuels.
Therefore, one of the ways to optimize energy costs and improve its reliability is to develop flexible methods to monitor equipment wear. The implementation of such systems could guarantee a continuous operation in energy companies and an efficient repair of their facilities.
Gas turbine failure prediction system: a complete success
The innovative approach, based on deep learning neural network models, has been successfully tested in gas turbines . The creators of this technique claim that in the near future, it can be used in a wide variety of industrial engines and power plants.
“For every company, it is essential to know what the current state of the equipment is and what needs to be done to prolong its useful life. Our method allows us to carry out this evaluation and reduce by half the error in the prediction of equipment failure time in comparison with existing analogues,” said Maxim Scherbakov, a scientist at the VolgGTU Faculty of Electronic and Computer Engineering.
According to the report, this discovery has the ability to generate accurate forecasts with only a minimal amount of initial data on equipment characteristics. It is important to highlight in this context the use of neural network mechanisms that do not require a large amount of preprocessed data.
“The classic methods for the useful life of the equipment are based on failure statistics, but they have certain limitations. Equipment manufacturers do not always provide detailed statistics and, in general, the number of recorded failures of this equipment is not enough even to the most powerful machine learning algorithms, which are ‘trained’ on data prepared in advance,” Shcherbakov explained.
The decrease in the possibility of making errors gives specialists a greater margin to make decisions, since the engineer will be informed not only of the moment in which a possible failure could occur, but also of the modifications in the performance of the plant caused by the wear of its components.
“By having an accurate estimate of when an equipment failure will occur, it is possible to adjust the maintenance schedule or optimize the mode of operation of the equipment. This will allow moving to a qualitatively new level of maintenance,” he stressed.
Additionally, the authors emphasized that the results can be used by equipment manufacturers to establish supplementary services for operators, such as repair quality assessment.
This breakthrough is within the framework of the intellectual management platform of the technical condition of equipment for energy and fuel companies that is being developed at the Volgograd State Technical University (VolgGTU). This platform, in turn, is a strategic project of the university framed within the Priority-2030 program.