А CLASSIFICATION APPROACH BASED ON A COMBINATION OF DEEP NEURAL NETWORKS FOR PREDICTING FAILURES OF COMPLEX MULTI-OBJECT SYSTEMS


UDC 004.02
DOI:10.26102/2310-6018/2020.29.2.037

K.A. Melekhina, P.P. Anan’ev, A.V. Plotnikova, S.A. Shestak

Scientific and technical progress has contributed to a rapid increase in the complexity of systems and their functions, which is especially characteristic of various fields of modern industry. Here, the cost of failure of equipment can be very high and sometimes lead to invaluable losses associated with the loss of life. Maintenance of such systems requires high material costs, but still does not exclude the possibility of failures. This indicates that the problem of ensuring the reliability of complex multi-object systems is still far from being solved. In this regard, the task of ensuring reliable operation of systems while minimizing the cost of their maintenance and maintenance is now in the first place. The solution of this problem is impossible without the development and implementation of intelligent systems that perform the functions of predictive analytics and predictive maintenance. This article proposes a hybrid neural network model for predicting failures of complex multi-object systems based on the classification approach, aimed at improving the operational reliability of equipment at minimal cost. The results of computational experiments confirming the high efficiency of the proposed solution are presented

Keywords:Forecasting failures, data-driven methods, deep neural networks, LSTM, CNN
Full text:
SaiShcherbakov_2_20_1.pdf