UDC 004.724
APPLICATION OF RECURRENT NEURAL NETWORKS IN THE CLASSIFICATION PROBLEM OF FAILURES IN THE COMPLEX TECHNICAL SYSTEMS WITHIN THE FRAMEWORK OF PROACTIVE MAINTENANCE
L. A. Demidova, Dr. Sc. (Tech.), full professor RSREU Ryazan, Russia;
orcid.org/0000-0003-4516-3746, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
D. V. Marchev, assistant RSREU, Ryazan, Russia;
orcid.org/0000-0001-6635-6657, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The task of classifying the problem of complex technical systems using the technology of recurrent neural networks is considered. The aim of the work is to develop a model of a recurrent neural network, which will allow effective and quick classification of a probable class of error in the operation of equipment in complex technical systems as part of proactive maintenance activities. Models of recurrent neural networks such as LSTM and GRU are developed, which in the experimental dataset containing information on the operation of aircraft engines and freely available at the NASA Ames Research Center use the algorithm for back-propagation of time error. The results of training networks, the average time spent on training, variance and mathematical expectation, as well as graphical dependences characterizing the quality of training are given. Based on the results of training, the most effective model of a neural network is highlighted, as well as recommendations for further improvement of its work to improve the quality of classification.
Key words: neural networks, RNN, LSTM, GRU, machine learning, classification, proactive maintenance.