Recurrent Neural Networks’ Configurations in the Predictive Maintenance Problems
Liliya Demidova
Abstract
Abstract The possibilities of various configurations of the recurrent neural networks in solving the problems of the maintenance performance based on the multidimensional time series have been investigated. The typical examples of the maintenance performance’ problems from technical and medical diagnostics have been considered. The configurations’ examples of the one- and two-layer recurrent neural networks with the RNN, LSTM, and GRU neurons for the aircraft engine maintenance problems have been given, the graphical dependencies of the development’ results of the neural network models, the estimates of the development time, and the estimates of the accuracy indicator have been presented. The conclusions about the advantages of the recurrent neural networks with the LSTM and GRU neurons have been made.