CN-122020442-A - Control moment gyro residual service life prediction method based on reconstruction error

CN122020442ACN 122020442 ACN122020442 ACN 122020442ACN-122020442-A

Abstract

The invention discloses a method for predicting the residual service life of a control moment gyro based on a reconstruction error, and belongs to the technical field of spacecraft health management. The method comprises the steps of collecting multidimensional sensor signals, carrying out wavelet denoising and normalization, carrying out singular value decomposition and feature extraction on the data, training a self-encoder model by using normal state features, calculating reconstruction errors to quantify abnormal degrees, determining thresholds based on the reconstruction errors, constructing health indexes, carrying out exponential moving average and self-adaptive Kalman filtering processing, calculating estimated degradation speed, and carrying out weighted fusion correction by combining theoretical design information to obtain the predicted residual service life. The method can fully excavate key information of the multidimensional sensor data, realize accurate identification of abnormal states and accurate prediction of residual service life, remarkably improve the accuracy and reliability of the control moment gyro health monitoring system and provide powerful guarantee for safe and stable operation of the spacecraft attitude control system.

Inventors

YU JINSONG
Liu baoding
LIU ZHIGANG
LIU HAO
TIAN LIMEI

Assignees

北京航空航天大学

Dates

Publication Date: 20260512
Application Date: 20251212

Claims (8)

1. A control moment gyro residual service life prediction method based on reconstruction errors is characterized in that the control moment gyro residual service life prediction method based on reconstruction errors extracts implicit characteristic information in original multidimensional data through Singular Value Decomposition (SVD), calculates health indexes and degradation trends based on differences of reconstruction errors of a self-encoder network relative to normal states, and carries out estimated service life correction by combining theoretical design information, so that accurate and stable health assessment and residual service life prediction of a Control Moment Gyro (CMG) are achieved, and the control moment gyro residual service life prediction method based on reconstruction errors comprises the steps of data acquisition and preprocessing, SVD feature extraction, self-encoder anomaly detection, health index calculation and correction and residual service life prediction.
2. The method for predicting the residual service life of the CMG according to claim 1, wherein the windowing process in the SVD feature extraction step of the method is characterized in that a fixed window size is adopted, the default window size is 64 samples, the step length is 64 samples, an adaptive parameter adjustment mechanism is built in, when the data quantity is small, the step length is automatically reduced to ensure that the window quantity is more than or equal to 512, the step length is adjusted to the power of the maximum 2 meeting the condition, the data matrix X is decomposed into the product of three matrices by Singular Value Decomposition (SVD), wherein X=U×S×V T , U is a left singular matrix, S is a singular value matrix, V is a right singular matrix, feature vectors are formed by extracting the first m singular values from the singular value matrix S, and m is the number of sensor parameter channels.
3. The method for predicting residual service life of a CMG according to claim 1, wherein a self-encoder model in the self-encoder anomaly detection step of the method adopts a fully connected network structure, the encoder comprises 3 hidden layers (the node numbers are 64, 32 and 16 in sequence), the decoder comprises 3 hidden layers (the node numbers are 16, 32 and 64 in sequence), the hidden layers adopt a ReLU activation function, the output layer adopts a Sigmoid activation function, and the model training aim is to minimize a Mean Square Error (MSE) between an input feature and a reconstruction feature, and the model training aim is to train by adopting an Adam optimizer.
4. The method for predicting the residual service life of the CMG according to claim 1, wherein the abnormal threshold value in the step of calculating the health index of the method is determined based on the statistical characteristics of the reconstruction errors of the normal data, and two methods are supported, namely a 3-sigma rule that the threshold value is the mean value +3 times standard deviation of the reconstruction errors of the normal data, and a MAD rule that the threshold value is the median +3 times absolute median difference of the reconstruction errors of the normal data.
5. The method for predicting remaining life of CMG according to claim 1, wherein the sliding window size in the step of calculating health index of the method is 1% of the total window number, and 10 windows minimum, and health index hi=1-anomaly ratio, wherein hi=1 indicates complete health, hi=0 indicates complete failure, and the CMG is judged to have failed when HI < 0.1.
6. The method for predicting residual life of CMG according to claim 1, wherein the Adaptive Kalman Filter (AKF) in the step of calculating health index of the method adapts to different degradation modes by adaptively adjusting filter gain, further smoothing health index curve and retaining degradation trend.
7. The method for predicting residual life of a CMG according to claim 1, wherein the estimated degradation rate in the step of predicting residual life of the CMG is calculated based on the estimated degradation rate of the current health index curve, and the calculation formula is: Wherein HI initial is the initial health index, the default value is 1, HI current is the current health index, T current is the current run time (unit: year), and the average degradation rate of the last 3 months is used as the estimated degradation rate in order to avoid short-term fluctuation effects.
8. The method according to claim 1, characterized in that: the weighted fusion method for correcting the residual service life by combining theoretical design information in the residual service life prediction step comprises the following steps: Wherein HI current is the current health index, HI failure is the failure threshold, the default value is 0.1, v theory is the theoretical degradation speed, v est is the estimated degradation speed, w is the weight of theoretical design information, the default value is 0.8, and the values can be dynamically adjusted according to the CMG operation stage, and the values of early value 0.9, middle value 0.8 and later value 0.7 are obtained.

Description

Control moment gyro residual service life prediction method based on reconstruction error Technical Field The invention belongs to the technical field of spacecraft health management, and particularly relates to a method for health state perception identification and residual service life (RUL) prediction of a Control Moment Gyro (CMG) on a spacecraft. Background The Control Moment Gyro (CMG) is a core executing mechanism of a spacecraft attitude control system, and provides accurate control moment for the spacecraft through angular change generated by high-speed rotation of a flywheel, so that the flexible adjustment and stable maintenance of the spacecraft attitude are realized. Considering the long-term nature and complexity of the on-orbit operation task of the spacecraft, the CMG generally has the design life requirement of more than 10 years, however, in the long-term on-orbit operation process, the CMG is subjected to comprehensive influence of multiple complex factors such as mechanical component abrasion, space extreme temperature alternation, orbit vibration interference, vacuum environment erosion and the like, the performance of the CMG can inevitably undergo progressive degradation, faults can be directly caused in severe cases, and fatal threats are formed to the attitude control precision of the spacecraft and even the smooth development of the whole on-orbit task. The reliability and stability of the attitude control system of the spacecraft are directly determined by the health state of the CMG, so that real-time and accurate health monitoring is realized, and scientific residual service life prediction is carried out based on monitoring data, and the method has important engineering significance for guaranteeing the safety of on-orbit operation of the spacecraft and improving the task completion quality. The CMG belongs to rotary machinery, at present, the existing health monitoring technology of the rotary machinery forms various technical paths, wherein a monitoring method based on vibration signals is a common type, fault identification and health state sensing are realized by analyzing the frequency spectrum characteristics of the vibration signals, but the CMG on a spacecraft is not provided with a vibration sensor, so that the method is difficult to apply, some researches focus on the temperature change rule of key parts of the CMG by using the monitoring method based on temperature signals, the health state is judged by temperature abnormal fluctuation, however, the change of the temperature signals is often delayed from the occurrence process of actual faults, and the delay of fault diagnosis is easy to cause. Besides the two methods, the monitoring method based on the current signal identifies faults by analyzing the change characteristics of the CMG driving current, but because the current signal has a strong coupling relation with the operation working condition of the CMG, the interference is eliminated by means of a complex working condition compensation algorithm, the technology implementation difficulty is greatly increased, the residual service life prediction scheme based on the statistical method generally assumes that the degradation process of the CMG is subject to specific statistical distribution (such as Weibull distribution), and the residual service life estimation is completed by fitting the degradation curve. In a comprehensive view, the existing CMG health monitoring and residual service life prediction technology still has a plurality of remarkable defects, and the actual requirements of high reliability guarantee of the spacecraft are difficult to meet. In the feature extraction link, the traditional time domain or frequency domain single-dimensional feature extraction method cannot fully mine the coupling information contained in the multidimensional sensor data, and a feature index system capable of comprehensively and accurately representing the health state of the CMG is difficult to construct. In the aspect of anomaly detection, the existing method mostly adopts a fixed threshold judgment mechanism, and the mechanism lacks adaptability to different types of CMG structural differences, performance parameters and different on-orbit running environments, so that anomaly judgment deviation is easy to occur. Notably, the performance degradation process of the CMG generally presents complex characteristics of strong nonlinearity and multi-factor coupling, and the degradation rule of the CMG is difficult to accurately describe by the traditional linear modeling method, so that the accuracy of the health state assessment is insufficient. In the aspect of residual service life prediction, the existing method relies on a single degradation model to carry out prediction, and cannot effectively fuse the difference of different models in a design stage and the complementary information of actual on-orbit degradation data, so that the prediction accuracy is difficult t