Approach to the generalized parameters formation of the complex technical systems technical condition using neural network structures

Abstract

A variant of the generalized parameters process forming of the complex technical systems technical condition is considered. This approach is relevant for modern robotic systems equipped with built-in telemetry tools. The proposed approach to the generalized parameters formation is based on heterogeneous telemetry parameters weighted summation using information about structural and functional relationships in a complex technical system, followed by digital lowfrequency filtering of the weighted summation results. This solution makes it possible to increase the complex technical systems technical condition reliability assessment by the values of generalized parameters in the external control loop. The form of generalized parameter representation in the technical condition gradations, which corresponds to the normal functioning, emergency situations and complex technical system partially operational state, is chosen. A multilevel hierarchical model of the generalized parameters formation of the complex technical systems technical condition on the basis of telemeasurements based on a variety of neural network structures allowing to take into account the nonlinear nature of the parameters being telemetered and the mutual influence between them has been developed. The model uses a variety of digital low-frequency filters that reduce the level of disturbances in the generalized parameters time series. The occurrence of disturbances is associated with the uncertainty of changes in the values of the telemetered and generalized parameters near the tolerance limits set by experts with expanded gradations according to technical condition. Information about the limits of tolerances characterizes not only the situations of regular and non-standard functioning, but also the complex technical systems partially operational state. The results of the generalized parameters formation of the spacecraft onboard system technical condition using multilayer neural networks, Kolmogorov–Gabor polynomials and digital filtering methods are presented. The advantages of using multilayer neural networks and median filters in the developed model are shown. The use of generalized parameters will significantly reduce the information load on the transmission channels of telemetry information as well as the means of its processing and analysis in the external control loop. The proposed solutions based on the basic operations of weighted summation and nonlinear transformation can be effectively implemented on promising vector-matrix and tensor processors that support their execution at the hardware level.

Keywords

• technical system
• technical condition
• telemetric parameters
• generalized parameter
• neural networks
• digital filtering