CN-122018483-A - Spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity

CN122018483ACN 122018483 ACN122018483 ACN 122018483ACN-122018483-A

Abstract

A spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity comprises the steps of preprocessing telemetry parameters of all components of a spacecraft attitude and orbit control system, extracting a generalized two-dimensional fault pattern in a fixed format by combining stacked self-encoders, and finally realizing system-level high-precision fault diagnosis on the two-dimensional fault pattern through a deep convolutional neural network. According to the method, the fixed-format two-dimensional fault pattern is extracted from the high-dimensional coupling telemetry data through the stacked self-encoder, and the fault diagnosis is carried out by utilizing the deep convolutional neural network, so that the problems that the high-dimensional coupling telemetry data of all components of the attitude and orbit control system are difficult to consider and the universality among different system configurations is poor in the fault diagnosis in the existing method are solved.

Inventors

Liang hanyu
LIU CHENGRUI
LIU WENJING
ZHANG YAN
XING XIAOYU

Assignees

北京控制工程研究所

Dates

Publication Date: 20260512
Application Date: 20260120

Claims (8)

1. A spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity is characterized by comprising the following steps: (1) Collecting original telemetry sequences of all telemetry sites of the spacecraft in M fault modes, wherein the original telemetry sequences of all telemetry sites of the spacecraft in M fault modes are formed by arranging telemetry data collected by corresponding telemetry sites at time points 1,2, and N according to collection time from early to late; (2) Obtaining a fault pattern sequence of the spacecraft in each fault mode based on the stacked self-encoders, wherein the fault pattern sequence comprises fault patterns of the spacecraft at time points 1, 2 and N in the corresponding fault modes; (3) Inputting the fault pattern sequences of the spacecrafts in all fault modes into a deep convolutional neural network to finish training of the deep convolutional neural network; (4) And (3) after the telemetry data of the spacecraft to be subjected to fault detection are processed in the step (1) and the step (2), inputting the telemetry data into a trained deep convolution neural network, performing fault detection on the spacecraft by the deep convolution neural network, and outputting a fault mode of the spacecraft.
2. The spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity according to claim 1, wherein the specific steps of preprocessing the original telemetry sequence of all telemetry sites in each fault mode in the step (1) are as follows: performing interpolation compensation on telemetry sequences of all telemetry sites in each fault mode to obtain interpolation compensation telemetry sequences corresponding to each telemetry site; (1.2) performing outlier rejection on all interpolation compensation data in each fault mode based on a Laet criterion to obtain outlier rejection telemetry sequences corresponding to each telemetry site; (1.3) mapping the outlier rejection telemetry sequence of all telemetry sites in each fault mode to obtain a fault telemetry sequence of all telemetry sites, wherein the fault telemetry sequence comprises fault telemetry data of corresponding telemetry sites at time point 1, time point 2 and time point N.
3. The spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity according to claim 1, wherein mapping in the step (1.3) is to map all elements in a outlier rejection telemetry sequence to a preset normalized value range, and the preset normalized value range is a normalized value range of telemetry data measured when the corresponding telemetry site fails.
4. The method for diagnosing the high-dimensional coupling data fault of the spacecraft supporting the system configuration diversity according to claim 1, wherein the specific process of obtaining the fault pattern sequence of the spacecraft in each fault mode in the step (2) is as follows: (2.1) inputting fault telemetry data for all telemetry sites in the first fault mode at time point 1 into the stacked self-encoder; And (2.2) stacking the self-encoders to form a group of input sequences from fault telemetry data of all telemetry sites at a time point 1, and extracting features by using a deep encoder, after the features are extracted, stacking preset intermediate hidden layers in the encoders to convert the extracted features into fault patterns in a fixed format, thereby obtaining the fault patterns of the spacecraft at the time point 1 in a first fault mode, wherein the expression of the whole process is as follows: wherein P is a fault pattern, x is an input sequence, And Equivalent weights and equivalent offsets of the deep encoder respectively, Is an activation function; The method comprises the steps of (2.3) obtaining fault patterns of a spacecraft in a first fault mode at a time point 2, a time point 3, and a time point N, wherein the specific processes are the same as steps (2.1) - (2.2), and forming a fault pattern sequence of the spacecraft in the first fault mode by using the fault patterns of the spacecraft in the first fault mode at the time point 1, the time point 2, the time point N; And (2.4) acquiring a fault pattern sequence of the spacecraft in the second fault mode, the third fault mode and the M fault mode, wherein the specific process is the same as that of the steps (2.1) to (2.3).
5. The spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity according to claim 1, wherein the deep convolutional neural network used in the step (3) is a double-layer CNN network, and the double-layer CNN network comprises an input layer, a first convolutional layer, a first maximum pooling layer, a second convolutional layer, a second maximum pooling layer and a full connection layer.
6. The spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity is characterized in that when deep convolutional neural network training is conducted, an input layer conducts local area scanning on each input fault pattern of a spacecraft at each time point to form a group of high-dimensional input data, a first convolutional layer conducts local area scanning on each fault pattern in the high-dimensional input data through a first group of multiple shared weight matrixes, area feature extraction is achieved through convolution operation, and accordingly a first group of convolutional feature graphs are output, a first maximum value pooling layer conducts space feature extraction by keeping the maximum value of a preset first pooling area to reduce the size of the first group of convolutional feature graphs, and therefore the first group of space feature graphs are output, a second convolutional layer conducts local area feature extraction on the first group of space feature graphs through a second group of multiple shared weight matrixes, and therefore a second group of convolutional feature graphs are output, the second maximum value pooling layer conducts space feature extraction through the corresponding reduction of the maximum value of the preset second pooling area, and therefore the second group of spatial feature graphs are output, and the second maximum value pooling layer is used for generating a full-dimensional neural network fault vector fusion training of the space feature graphs.
7. The spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity according to claim 6, wherein the convolution layer outputs a feature map through convolution operation The calculation formula of (2) is as follows: Wherein, the For the input of the convolutional layer, For the feature map of the output of the convolutional layer, Is a matrix of weights for the convolution kernel, Offset for the convolutional layer; feature map of maximum value pooling layer based on convolution layer output The calculation formula for further outputting the space feature map is as follows: Wherein, the For the spatial feature map, k and j are respectively the output feature map of the convolution layer Row number and column number of (c).
8. A method for diagnosing high-dimensional coupling data of spacecraft with multiple supporting system configurations according to any one of claims 1-7, wherein the loss function used by deep convolutional neural network in training is cross entropy loss function The calculation formula is as follows: Wherein, the For switching value, when the class of the network output i is consistent with c, Taking 1, otherwise, Taking 0; and outputting the prediction probability of i belonging to the category c for the network.

Description

Spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity Technical Field The invention relates to a spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity, and belongs to the field of aerospace. Background With the rapid development of constellation of star groups in the field of aerospace, a limited ground station has difficulty in supporting timely diagnosis and data interpretation of faults of hundreds of spacecraft in the in-orbit running process, so that serious accidents such as service interruption or whole satellite failure are easily caused. In order to ensure the safe, reliable and stable operation of the large-scale spacecraft on orbit, the intelligent autonomous fault diagnosis capability is required to be provided for reducing the ground measurement and control pressure. As a typical high-reliability closed-loop control system of a spacecraft, the fault of the spacecraft has a closed-loop propagation characteristic, and the existing fault diagnosis method generally only analyzes a single component or a certain class of components, so that high-dimensional coupling telemetry data of all components of the spacecraft are difficult to consider. In addition, according to the on-orbit task demands of the spacecraft, different types of spacecraft have the problem of multiple system configurations, so that the system telemetry parameter dimension of different spacecraft is difficult to be defined, and unified means cannot be adopted to carry out generalized fault diagnosis on various types of spacecraft. Disclosure of Invention The technical problem of the invention is to overcome the defects of the prior art and provide a spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity. According to the method, the fixed-format two-dimensional fault pattern is extracted from the high-dimensional coupling telemetry data through the stacked self-encoder, and the fault diagnosis is carried out by utilizing the deep convolutional neural network, so that the problems that the high-dimensional coupling telemetry data of all components of the attitude and orbit control system are difficult to consider and the universality among different system configurations is poor in the fault diagnosis in the existing method are solved. The technical scheme of the invention is as follows: A spacecraft high-dimensional coupling data fault diagnosis method supporting system configuration diversity comprises the following steps: (1) Collecting original telemetry sequences of all telemetry sites of the spacecraft in M fault modes, wherein the original telemetry sequences of all telemetry sites of the spacecraft in M fault modes are formed by arranging telemetry data collected by corresponding telemetry sites at time points 1,2, and N according to collection time from early to late; (2) Obtaining a fault pattern sequence of the spacecraft in each fault mode based on the stacked self-encoders, wherein the fault pattern sequence comprises fault patterns of the spacecraft at time points 1, 2 and N in the corresponding fault modes; (3) Inputting the fault pattern sequences of the spacecrafts in all fault modes into a deep convolutional neural network to finish training of the deep convolutional neural network; (4) And (3) after the telemetry data of the spacecraft to be subjected to fault detection are processed in the step (1) and the step (2), inputting the telemetry data into a trained deep convolution neural network, performing fault detection on the spacecraft by the deep convolution neural network, and outputting a fault mode of the spacecraft. Further, the specific steps of preprocessing the original telemetry sequence of all telemetry sites in each failure mode in the step (1) are as follows: performing interpolation compensation on telemetry sequences of all telemetry sites in each fault mode to obtain interpolation compensation telemetry sequences corresponding to each telemetry site; (1.2) performing outlier rejection on all interpolation compensation data in each fault mode based on a Laet criterion to obtain outlier rejection telemetry sequences corresponding to each telemetry site; (1.3) mapping the outlier rejection telemetry sequence of all telemetry sites in each fault mode to obtain a fault telemetry sequence of all telemetry sites, wherein the fault telemetry sequence comprises fault telemetry data of corresponding telemetry sites at time point 1, time point 2 and time point N. Further, the mapping processing in the step (1.3) is to map all elements in the outlier rejection telemetry sequence to a preset normalized value range, wherein the preset normalized value range is the normalized value range of telemetry data measured when the corresponding telemetry site does not have a fault. Further, the specific process of obtaining the fault pattern sequence of the spacecraf