A CMG failure test design method based on Gaussian process regression
The CMG failure test design method based on Gaussian process regression solves the problems of high cost and long time in CMG failure testing, realizes low-cost and short-cycle test design, and improves CMG design and verification capabilities.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING INST OF SPACECRAFT ENVIRONMENT ENG
- Filing Date
- 2023-11-08
- Publication Date
- 2026-05-19
AI Technical Summary
Existing CMG failure tests are difficult to realistically simulate the on-orbit environment on the ground, resulting in high test costs and long test times, and a lack of low-cost, short-cycle test design methods.
A CMG failure test design method based on Gaussian process regression was adopted. By determining the test factors and indicators, establishing a prediction model, performing simulation calculations and cluster analysis, and selecting key working conditions for physical testing, simulation errors were reduced and the test design was optimized.
This approach enables low-cost, short-cycle CMG failure test design, improves test efficiency, reduces the number of physical tests, and saves resource consumption.
Smart Images
Figure CN117494314B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a CMG failure test design method based on Gaussian process regression, which is applicable to the multi-stress coupling environment failure test design of spacecraft CMG products. Background Technology
[0002] A control moment gyroscope (CMG) is an inertial actuator used for attitude control in spacecraft, widely applied in space missions requiring rapid maneuvering and high pointing accuracy, such as space stations and high-precision space monitoring. CMGs have complex structures with high-speed rotating shafts. Operating environments such as vacuum and alternating hot and cold temperatures can affect the service life and reliability of these shafts, potentially leading to on-orbit failures. Ground testing is crucial for understanding their operational characteristics. Common ground tests include thermal vacuum tests and life tests, but these struggle to accurately simulate on-orbit conditions, especially the CMG's motion. CMG failure testing simulates the environmental / operating stresses experienced by the CMG on the ground, elucidating its operational characteristics and determining whether it experiences operational stalls or startup difficulties under specific conditions, thus verifying its performance. Typical CMG failure tests require simulating environmental stresses such as vacuum and temperature, as well as operating stresses such as frame rotation speed and satellite rotation speed. These tests are complex, time-consuming, and expensive. To save testing time and costs, there is an urgent need to research low-cost, short-cycle test design methods to improve the design and verification capabilities of CMGs.
[0003] Simulation calculations and physical testing are commonly used design verification methods. Using computer simulation software, calculations can effectively simulate the structure and performance characteristics of complex products in actual use, allowing for the early identification of design problems and the calculation of product defects. However, it suffers from low accuracy. Physical testing, based on the actual usage principles of the product, involves fixing a physical model or prototype in an artificial environment, artificially creating realistic operating conditions to simulate various complex real-world states and obtain accurate test performance data. However, product manufacturing costs and actual testing are very high, and the human and material resources consumed in testing are enormous. Given that simulation, as a virtual numerical simulation, can provide a basis and preliminary data for physical test design, predicting the operating characteristics of CMGs through the integration of simulation calculations and physical testing will be one of the important trends in future technological development. However, a mature integration method is currently lacking. Summary of the Invention
[0004] To address the problems existing in the prior art, this invention provides a CMG failure test design method based on Gaussian process regression, achieving low-cost and short-cycle test design.
[0005] To achieve the above objectives, the present invention adopts the following solution:
[0006] This invention provides a CMG failure test design method based on Gaussian process regression, comprising the following steps:
[0007] S1: Determine the experimental factors and experimental indicators;
[0008] S2: Determine the range of variation of the experimental factors. Based on the design requirements and application scenarios, determine the range of variation of the experimental factors as the parameter setting boundary for simulation and experiment.
[0009] S3: Initial simulation calculation: Based on the number of test factors and the range of variation of the test factors, the initial simulation conditions are set by uniformly sampling each test factor.
[0010] S4: Establish a prediction model based on Gaussian process regression. Based on the simulation data D obtained in step S3, construct a prediction model based on Gaussian process regression.
[0011] S5: Establish the utility function;
[0012] S6: Select simulation conditions at the extreme points of the utility function;
[0013] S7: Obtain simulation results;
[0014] S8: Obtain the model prediction results;
[0015] S9: Determine whether the simulation and prediction results meet the error requirements;
[0016] S10: Perform cluster analysis on all simulation conditions;
[0017] S11: Select sample points for testing in each type of simulation working condition.
[0018] Further, in step S1, the test factor is the environmental / working stress that needs to be controlled during the test, including temperature, frame rotation speed, and satellite rotation speed; the test index is the test result that can determine whether the CMG working characteristics meet the design requirements, including the reduction in radial clearance of the frame bearing and whether there is operational jamming / starting difficulty.
[0019] Further, step S3 specifically involves: the x-th experimental factor within the range [F] x,min ,F x,max If the area is divided into m equal parts, then the initial number of simulation conditions is E0 = (m + 1). N The simulation data D0 = [X, y] is obtained, where the matrix X = [X1, X2, ..., Xy]. N ] represents the combination of experimental factors in group E0, and y represents the experimental index of group E0 experimental factors obtained based on simulation.
[0020] Further, in step S4, the prediction model is denoted as P(f) = GP(f, K). X,X ), where f represents the mean predicted by the Gaussian process, and K X,X The covariance matrix represents the degree of correlation between operating conditions.
[0021] Furthermore, step S5 specifically involves defining any point x in the N-dimensional space. p =(x 1,p ,x 2,p ,...,x N,p Utility function at ) Where var(xp) is the point x in N-dimensional space. p The variance at the location.
[0022] Further, step S6 includes the following steps:
[0023] S6-1: At two adjacent points x p With x p+1 Between, where x p x p+1 Let ∈X, and denote the utility function ε(x). p The point of maximum value is denoted as This leads to the maxima of all utility functions in the X space. The corresponding utility function value is
[0024] S6-2: The number of simulation conditions is set to C. Sort the data in descending order, and select the first C utility function values to form the first sampling condition, denoted as...
[0025] Furthermore, in step S7, let X be... C1 The simulation result is y C1 Step S8 specifically involves: placing X C1 Substitute the values into the prediction model established in step S4 to obtain the model prediction result f. C1 .
[0026] Furthermore, step S9 specifically includes:
[0027] If f C1 With y C1 If the average error is less than the given error, proceed to step S10; otherwise, [X] C1 ,y C1 As supplementary sample points, return to step S4, update the prediction model, and repeat steps S5 to S7 until the simulation and prediction results meet the given error requirements.
[0028] Furthermore, step S10 specifically includes:
[0029] Let the union of the initial simulation condition and each of the previous sampling conditions be denoted as . The DBSCAN clustering analysis method is used to perform clustering analysis on all sample points in D, and all sample points in D are clustered into c categories (c is greater than or equal to 2), so that the similarity between individuals in the same category is as high as possible, and the similarity between different categories is as low as possible.
[0030] Further, step S11 specifically involves: based on the clustering results in step S10, selecting one or more operating points in each cluster for physical testing.
[0031] The beneficial effects of this invention are:
[0032] The method of this invention enables low-cost, short-cycle experimental design.
[0033] In some embodiments, the present invention also has the following effects.
[0034] In step S4, a prediction model based on Gaussian process regression was established using simulation results, which can predict the changing trend of experimental indicators based on limited simulation data.
[0035] In step S5, a utility function that can characterize the uncertainty of sampling points is constructed by utilizing the predictable variance of Gaussian process regression.
[0036] In step S6, the next simulation condition is set at the maximum value of the utility function, which effectively identifies the condition that contributes the most to the predicted product test indicators and improves simulation efficiency.
[0037] In step S10, all simulation conditions are clustered according to the principle of maximizing the similarity of individuals within the same class and minimizing the similarity between different classes. This divides the relatively large number of conditions into a few classes, thereby maximizing the representativeness of the limited number of physical test conditions selected in step S10. These conditions can represent the test index characteristics within the entire test factor space, effectively reducing the number of physical tests. Attached Figure Description
[0038] Figure 1 A flowchart of a CMG failure test design method based on Gaussian process regression is provided for an embodiment of this application. Detailed Implementation
[0039] To make the technical solutions and advantages of the present invention clearer, the technical solutions of the embodiments of the present invention will be fully described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0040] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0041] To comprehensively evaluate the operational characteristics of a certain type of CMG product under different temperature, frame rotation speed, and satellite rotation speed conditions, a multi-stress environment test bench was built in a ground-based vacuum environment simulator. This bench can simulate environmental / operating stresses such as temperature, frame rotation speed, and satellite rotation speed of the CMG, thereby observing whether the CMG experiences operational jamming under a set stress combination environment. Simultaneously, a thermo-mechanical coupling simulation calculation model considering the effects of temperature, frame rotation speed, and satellite rotation speed was established. This model can calculate the reduction in radial clearance δ of the frame bearing under specific stress combinations. When δ ≥ 41 μm, the CMG may experience operational jamming; when δ < 41 μm, the CMG will not experience operational jamming.
[0042] The experimental design is now carried out according to the experimental design method described in this application.
[0043] S1. Determine the test factors and test indicators: The test factors are temperature (denoted as X1), frame rotation speed (denoted as X2), and satellite rotation speed (denoted as X3); the test indicator is the reduction in radial clearance of the frame bearing (denoted as y).
[0044] S2. Determine the range of variation of test factors: Based on the design requirements and application scenario, determine the temperature range as [-20℃, 60℃], the frame rotation speed range as [0° / s, 60° / s], and the satellite rotation speed as [1° / s, 5° / s].
[0045] S3. Initial Simulation Calculation: Set the initial simulation conditions according to the uniform sampling method for each experimental factor, and divide the variation range of each experimental factor into two equal parts. as follows:
[0046] Serial Number <![CDATA[x1]]> <![CDATA[x2]]> <![CDATA[x3]]> Serial Number <![CDATA[x1]]> <![CDATA[x2]]> <![CDATA[x3]]> Serial Number <![CDATA[x1]]> <![CDATA[x2]]> <![CDATA[x3]]> 1 -20 0 1 10 20 0 1 19 60 0 1 2 -20 0 3 11 20 0 3 20 60 0 3 3 -20 0 5 12 20 0 5 21 60 0 5 4 -20 30 1 13 20 30 1 22 60 30 1 5 -20 30 3 14 20 30 3 23 60 30 3 6 -20 30 5 15 20 30 5 24 60 30 5 7 -20 60 1 16 20 60 1 25 60 60 1 8 -20 60 3 17 20 60 3 26 60 60 3 9 -20 60 5 18 20 60 5 27 60 60 5
[0047] The initial simulation calculations were carried out according to the above initial simulation conditions to obtain the corresponding simulation result y0.
[0048] S4. Establish a prediction model based on Gaussian process regression: Based on the simulation data D0=[X0,y0] obtained in S3, construct a prediction model based on Gaussian process regression, denoted as P(f)=GP(f,K X,X ).
[0049] S5. Establish the utility function:
[0050] S6. Select simulation conditions at the extreme points of the utility function:
[0051] S6.1 Obtain the maximum points of all utility functions in the X0 space. The corresponding utility function value is
[0052] S6.2 The number of simulation conditions for each sampling is set to 10. Sort by value from largest to smallest, and take the points corresponding to the top 10 utility function values. This is the first sampling condition.
[0053] S7. Obtain simulation results: Let X be... C1 The simulation result is y C1 .
[0054] S8. Obtain the model prediction results: X C1 Substitute the values into the prediction model established in S4 to obtain the model prediction result f. C1 .
[0055] S9 determines whether the simulation and prediction results meet the error requirements: f C1 With y C1 The average error is greater than 5%, [X] C1 ,y C1 As supplementary sample points, return to S4, update the prediction model, and repeat S5 to S7 to obtain sample points [X] sequentially. C2 ,y C2 ]、[X C3 ,y C3 ]、……、[X C14 ,y C14 ], f C14 With y C14 The average error meets the requirement of being less than 5%.
[0056] S10 performs cluster analysis on all simulation conditions: the union of the initial simulation condition and each of the previous sampling conditions is denoted as... The DBSCAN clustering analysis method was used to perform cluster analysis on all sample points in D, and all sample points in D were clustered into 6 categories, so that the similarity between individuals in the same category is as high as possible, and the similarity between different categories is as low as possible.
[0057] S11 selects sample points in each type of simulated operating point for testing: Based on the 6 results obtained from clustering in S10, 1 operating point is selected in each type for physical testing to verify the operating characteristics of the CMG product.
[0058] The preferred sequential sampling method based on utility functions obtained a prediction model that met the error requirements using 167 sets of simulation data. Compared with the Latin hypercube sampling method (which requires 512 sets of simulation data), the number of simulation samples was reduced by 67%.
[0059] Based on this design method, only a minimum of 6 sets of experimental data are needed to verify the dynamic characteristics of CMG within the entire design space, effectively improving experimental efficiency.
[0060] In the description of this specification, references to terms such as "an embodiment" and "example" refer to specific features, structures, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms are not necessarily intended to refer to corresponding embodiments or examples in a suitable manner.
[0061] It must be pointed out that the above description of the embodiments is not intended to limit the invention but only to help understand the core idea of the invention. For those skilled in the art, any improvements to the invention and equivalent alternatives made to the invention without departing from the principle of the invention are also within the scope of protection of the claims of the invention.
Claims
1. A CMG failure test design method based on Gaussian process regression, characterized in that, Includes the following steps: S1: Determine the experimental factors and experimental indicators; S2: Determine the range of variation of the experimental factors. Based on the design requirements and application scenarios, determine the range of variation of the experimental factors as the boundary for setting parameters in simulation and experiment. S3: Initial simulation calculation: Based on the number of test factors and the range of variation of the test factors, the initial simulation conditions are set by uniformly sampling each test factor. S4: Establish a prediction model based on Gaussian process regression. Based on the simulation data D obtained in step S3, construct a prediction model based on Gaussian process regression. S5: Establish the utility function; S6: Select simulation conditions at the extreme points of the utility function; S7: Obtain simulation results; S8: Obtain the model prediction results; S9: Determine whether the simulation and prediction results meet the error requirements; S10: Perform cluster analysis on all simulation conditions; S11: Select sample points for testing in each type of simulation operating condition; In step S1, the test factors are the environmental stress and working stress that need to be controlled during the test, including temperature, frame rotation speed, and satellite rotation speed; the test indicators are the test results that can determine whether the CMG working characteristics meet the design requirements, including the reduction of the radial clearance of the frame bearing, whether there is operational jamming or starting difficulty. Step S5 specifically involves defining any point in N-dimensional space. utility function at point ,in, For a point in N-dimensional space variance at location; S6-1: At two adjacent points Between, among Denote the utility function The point of maximum value is denoted as This allows us to obtain the maximum points of all utility functions in the X space. The corresponding utility function value is ; S6-2: The number of simulation conditions is set to C. Sort the data in descending order, and select the first C utility function values to form the first sampling condition, denoted as... .
2. The method according to claim 1, characterized in that, Step S3 specifically refers to: x The test factors vary within the range [F] x,min ,F x,max If the area is divided into m equal parts, then the initial number of simulation conditions is: To obtain simulation data , where the matrix for The combination of group experimental factors, represent The experimental factors are based on experimental indicators obtained from simulation.
3. The method according to claim 1, characterized in that, In step S4, the prediction model is denoted as ,in, Characterizes the mean predicted by the Gaussian process. The covariance matrix represents the degree of correlation between operating conditions.
4. The method according to claim 1, characterized in that, In step S7, record The simulation results are Step S8 specifically involves: ... Substitute the values into the prediction model established in step S4 to obtain the model prediction results. .
5. The method according to claim 1, characterized in that, Step S9 is as follows: like and If the average error is less than the given error, proceed to step S10; otherwise, proceed to step S10. As supplementary sample points, return to step S4, update the prediction model, and repeat steps S5 to S7 until the simulation and prediction results meet the given error requirements.
6. The method according to claim 1, characterized in that, Step S10 is as follows: Let the union of the initial simulation condition and each of the previous sampling conditions be denoted as . The DBSCAN clustering analysis method is used to perform clustering analysis on all sample points in D, and all sample points in D are clustered into c categories, where c is greater than or equal to 2, so that samples in the same category are clustered and samples in different categories are dispersed.
7. The method according to claim 1, characterized in that, Step S11 specifically involves selecting one or more operating points in each cluster based on the clustering results in step S10 for physical testing.