Structural reliability analysis of space vehicles using neural networks

Pomortsev Pavel Mikhailovich, Kireev Dmitry Gennadievich, Lesnichenko Roman Ivanovich, Lesnichenko Maxim Romanovich

Predicting the service life of spacecraft is an urgent problem due to the complexity of designs and inaccessibility. Before the launch of the mission, the spacecraft undergoes ground tests and effective training. However, many missions experience temporary or permanent failures resulting in failure. Although there is no perfect system that can prevent any failure, analysis of past failures shows that adequate testing, redundancy and flexibility are the keys to a reliable spacecraft failure recovery system. It was revealed that the causes of failures have a different physical nature, among which an important place is occupied by mechanical problems associated with the elasto-elastic properties of the materials of the supporting structures. To analyze the mechanical reliability of structural elements of space technology, it is proposed to use neural networks as a modern apparatus for analysis and modeling with a large amount of input data. Neural networks allow you to effectively analyze a large amount of data and obtain fairly accurate reliability estimates. The backpropagation neural network algorithm has been successfully used to obtain rough estimates of critical load factors and has shown significant reliability with respect to training set selection and network architecture in predicting failure probability.

Keywords:neural network, reliability, spacecraft, mechanical failure, error, modeling