CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
FAILURE PREDICTION OF COMPLEX MULTIPLE-COMPONENT SYSTEMS BASED ON A COMBINATION OF NEURAL NETWORKS: WAYS TO IMPROVE THE ACCURACY OF FORECASTING
 |
Read | Sai Van Cuong, Shcherbakov Maksim V. FAILURE PREDICTION OF COMPLEX MULTIPLE-COMPONENT SYSTEMS BASED ON A COMBINATION OF NEURAL NETWORKS: WAYS TO IMPROVE THE ACCURACY OF FORECASTING // Caspian journal : management and high technologies. — 2020. — №1. — pp. 49-60. |
/49-60.jpg)
Sai Van Cuong - Volgograd State Technical University, svcuonghvktqs@gmail.com
Shcherbakov Maksim V. - Volgograd State Technical University, maxim.shcherbakov@vstu.ru
The paper proposes a hybrid neural network model with two outputs based on convolutional neural networks (CNN) and long short - term memory networks (LSTM) for predicting failures of complex multi - component systems. CNN networks are used to extract spatial properties from multidimensional sensor data, and LSTM networks are used for temporal modeling of long - term dependencies. The first output of the proposed model is a classifier that allows you to predict whether the system may fail in the next n - steps of time in the future, in other words, it is an identifier of the stage of degradation of the equipment. The second output is a regressor that allows to predict the number of the remaining useful life (RUL) of the equipment at each time step. The results of computational experiments confirming the high efficiency of the proposed solution are presented.
Key words: предсказательное обслуживание, остаточный ресурс, глубокие нейронные сети, ансамблевые методы, гиперпараметрическая оптимизация, predictive maintenance, remaining useful life (RUL), deep neural network, ensemble method, hyperparameter optimization