A method, system and terminal for evaluating a control moment gyroscope in orbit

Abstract

The present application belongs to the technical field of spacecraft attitude control, and relates to a control moment gyro (CMG) on-orbit evaluation method and system and a terminal. The evaluation method acquires on-board telemetry data covering CMG core parameters, including mechanical system, electrical system, environmental adaptation and task performance parameters. Based on the on-board telemetry data, the CMG index values in the reliability, on-orbit performance, stability and health state dimensions are calculated through a multi-source method of collaborative basic index calculation, neural network prediction and data feature analysis. Finally, the comprehensive evaluation result is generated in combination with the index values in each dimension. The present application realizes the all-around accurate characterization of the on-orbit state of the CMG by means of rich data collection, fusion of multi-source algorithms and construction of a multi-dimensional system, improves the evaluation accuracy and robustness, and guarantees the landing of the evaluation.

Description

Technical Field

[0001] This invention relates to the field of spacecraft attitude control technology, and in particular to an on-orbit evaluation method, system and terminal for a control moment gyroscope. Background Technology

[0002] The control moment gyroscope (CMG) is a core actuator of the spacecraft attitude control system. Its on-orbit operational status and reliability directly affect the successful implementation and even the success or failure of spacecraft missions. Conducting accurate and comprehensive on-orbit performance evaluation of CMGs has significant engineering application value and practical significance. Currently, CMG on-orbit evaluation technologies still suffer from prominent shortcomings such as an incomplete system and insufficient data support, making it difficult to meet the high-precision, multi-dimensional evaluation requirements of spacecraft on-orbit operations. The specific limitations of existing technologies are as follows: (i) The CMG evaluation system has structural defects. Currently, the mainstream CMG evaluation systems are mainly divided into three categories. All of these systems have obvious technical shortcomings and cannot achieve comprehensive and systematic on-orbit evaluation. 1. Systematic Error and Accuracy-Oriented Evaluation System This system focuses on CMG product-level and system-level error propagation analysis, studying parameters such as outer frame zero-position deviation, rotation error, installation configuration error, and control cycle error. Its core objective is to evaluate torque output accuracy. However, this type of system evaluates only the single dimension of control accuracy, completely ignoring key performance indicators such as CMG on-orbit reliability, lifetime decay patterns, and multi-space stress coupling adaptability. The evaluation results are highly biased and cannot provide comprehensive data support for CMG on-orbit fault early warning and remaining lifetime prediction, potentially leading to deviations in spacecraft attitude control decisions.

[0003] 2. Ground Integrated Stress Failure Boundary Evaluation System This system relies on ground-based simulations of comprehensive stress conditions, including vacuum, thermal environment, frame rotation speed, and spacecraft attitude rotation speed, to conduct CMG failure boundary tests and lifetime predictions. The evaluation data depends entirely on ground-based simulation experiments. However, because the ground environment cannot replicate the actual on-orbit microgravity, space radiation, and dynamic angular momentum exchange conditions, the differences between the Earth and space environments lead to significant deviations between lifetime predictions, performance evaluations, and the actual on-orbit conditions, making it difficult to meet the requirements for accurate on-orbit evaluation.

[0004] 3. Fragmented evaluation system based on single indicators This system targets specific operating conditions such as ultra-low speed operation and conducts special evaluations based on single performance indicators such as speed measurement accuracy and torque output stability. However, it lacks a systematic evaluation framework and cannot characterize the coupling relationship between parameters such as temperature, vibration, and load and core performance indicators. The evaluation results cannot fully reflect the overall on-orbit health status of the CMG and are difficult to support on-orbit operation and maintenance decisions in multiple scenarios.

[0005] In summary, existing evaluation systems generally suffer from common problems such as single evaluation dimensions, severe fragmentation, and low systematization. The evaluation results lack comprehensiveness, have limited reference value for engineering decisions, and do not yet possess the systematic evaluation capability to adapt to the real-world on-orbit environment.

[0006] (ii) Insufficient supporting data for CMG evaluation Currently, the basic data relied upon for CMG on-orbit evaluation mainly falls into two categories, neither of which can support high-precision, multi-dimensional evaluation work: The first type is ground-based simulation test data. This type of data is taken from ground-based thermal vacuum and accelerated life simulation tests. Due to the inherent differences between the space and ground environments, it cannot reflect the real operating conditions of CMG in orbit, such as microgravity, space radiation, and attitude dynamic coupling. The evaluation results based on this type of data deviate significantly from the actual state of CMG in orbit. The second type is traditional on-board telemetry data. Current technologies only collect a few key parameters such as frame rotation speed and output torque. The data dimensions are lacking and the analysis algorithms are simple, which cannot meet the needs of multi-dimensional and systematic evaluation, further exacerbating the one-sidedness of the evaluation results.

[0007] Therefore, it is urgent to break through the limitations of existing technologies, build a systematic evaluation framework, expand the dimensions of telemetry data acquisition, and rely on comprehensive on-orbit measurement data to achieve accurate and multi-dimensional on-orbit evaluation of CMG. Summary of the Invention

[0008] The purpose of this invention is to address the problems of limited dimensions, insufficient data support, and simplistic index algorithms in CMG on-orbit evaluation. It proposes an on-orbit evaluation method, system, and terminal for control moment gyroscopes. By enriching data acquisition, integrating multi-source algorithms, and constructing a multi-dimensional system, it achieves comprehensive and accurate characterization of the CMG's on-orbit status, improves evaluation accuracy and robustness, and ensures the effective implementation of the evaluation.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an on-orbit evaluation method for a control torque gyroscope, comprising the following steps: Acquire on-orbit telemetry data of the control torque gyroscope. The on-orbit telemetry data includes mechanical system parameters, electrical system parameters, environmental adaptation parameters, and mission performance parameters. Among them, the mechanical system parameters include at least rotor speed, frame rotation angle, and frame vibration data; the electrical system parameters include at least rotor motor current, high-speed motor current, and frame motor current; the environmental adaptation parameters include at least frame motor shell temperature; and the mission performance parameters include at least frame coarse and fine differences, torque output value, and angular momentum exchange state parameters. Based on on-board telemetry data, the index values ​​of the control torque gyroscope are calculated under multiple preset evaluation dimensions; the multiple preset evaluation dimensions include reliability dimension, on-orbit performance dimension, stability dimension and health status dimension; Based on the index values ​​under each evaluation dimension, the on-orbit comprehensive evaluation results of the control moment gyroscope are generated.

[0010] As one possible implementation, the reliability metric values ​​are obtained by fusing the reliability of the framework motor and the on-orbit reliability of a single unit: The reliability of the frame motor is calculated based on the logical comparison results of the real-time status of the frame motor shell temperature, frame motor current and frame coarse and fine difference with the corresponding preset thresholds and the normality rate. The reliability of single-unit on-orbit operation is obtained by first-order difference calculation, mutation detection, and entropy weighting summation based on time-series data of frame motor shell temperature, rotor motor current, high-speed motor current, and rotor speed.

[0011] As one possible implementation method, the reliability calculation method for the frame motor is as follows: For each sampling point, the frame motor housing temperature, frame motor current, and frame coarse and fine differences are compared with their respective preset safety threshold ranges to generate a status indicator indicating whether the status is normal or exceeds the limit. The normality rate of each parameter is calculated based on the ratio of the number of sampling points with normal status within a preset time period to the total number of valid sampling points. The average of the frame motor housing temperature normality rate, frame motor current normality rate, and frame coarse-precision difference normality rate is determined as the frame motor reliability; and / or, The method for calculating the on-orbit reliability of a single unit is as follows: First-order difference calculations were performed on the time-series data of frame motor housing temperature, rotor motor current, high-speed motor current and rotor speed to obtain the first-order difference sequence of each parameter; The difference in the absolute value of the first-order difference sequence of each parameter that exceeds the preset mutation threshold is determined as a sudden mutation, and the number of sudden mutations of each parameter within the preset period is counted. The single-parameter reliability of each parameter is calculated based on the ratio of the number of sudden mutations of each parameter to the total number of data points. The entropy weight method is used to assign weights to each parameter based on the degree of dispersion of the first-order difference sequence of each parameter; The on-orbit reliability of a single unit is obtained by weighted summation based on the individual parameter reliability and weight of each parameter.

[0012] As one possible approach, the on-orbit performance metric values ​​are obtained by fusing the overall on-orbit performance score, performance change trend score, and deterioration trend score. The overall on-orbit performance score is obtained by multi-dimensional scoring and entropy weighting based on the mean, range, and standard deviation of rotor motor current, high-speed motor current, frame motor current, rotor speed, frame coarse and fine difference, and frame motor shell temperature, as well as the corresponding engineering calibration parameters. The performance change trend score is calculated based on the change trend of the output anomaly probability value by inputting the time series data of rotor speed, frame motor housing temperature and high-speed motor current into the Long Short-Term Memory network LSTM. The deterioration trend score is calculated based on the time-series data of rotor motor current, high-speed motor current, frame motor current and rotor speed input into the denoising autoencoder (DAE) and the time-varying rate of change of its output reconstruction error.

[0013] As one possible implementation method, the on-orbit integrated performance score is calculated as follows: Calculate the mean, range, and standard deviation of each parameter within a preset period; For each parameter, a mean compliance score is determined based on the comparison between the mean and a preset rated range. The range compliance score is determined based on the comparison between the range and the preset maximum allowable range. The volatility stability score is determined based on the comparison between the standard deviation and the preset maximum allowable standard deviation. The mean compliance score, range compliance score, and fluctuation stability score are added together to obtain the single-parameter on-orbit performance score. The entropy weight method is used to determine the weight of each parameter, and the on-orbit performance scores of all individual parameters are weighted and summed to obtain the overall on-orbit performance score; and / or, The calculation method for the performance change trend score is as follows: Obtain the anomaly probability value of the LSTM network output at each time step; Perform first-order difference calculation on the sequence of abnormal probability values; The proportion of time steps with positive first-order differences is used as the proportion of improvement trend, and the proportion of time steps with negative first-order differences is used as the proportion of deterioration trend. Based on (Percentage of Improvement Trends - Percentage of Deterioration Trends) 10. Calculate the performance change trend score; and / or, The method for calculating the deteriorating trend score is as follows: Obtain the parameter reconstruction error output by the DAE network; Calculate the rate of change of the reconstruction error over time within the sliding window; Normalize the rate of change over time; Based on (1 - normalized rate of change over time) 10. Calculate the deterioration trend score.

[0014] As one possible implementation, the stability dimension index value is obtained by fusing the first stability score, the second stability score, the third stability score, and the fourth stability score; The first stability score is based on the time-series data of frame motor housing temperature, rotor motor current, high-speed motor current, frame motor current and rotor speed. The standard deviation and mean standard deviation of each parameter are calculated and determined in combination with the preset standard deviation stability threshold. The second stability score is determined by inputting time-series data of frame motor current and frame motor shell temperature into a decision tree model and performing statistical calculations based on the output state identification labels. The third stability score is determined by inputting time-series data of rotor speed and frame motor housing temperature into a support vector machine (SVM) model and performing statistical calculations based on the output state identification labels. The fourth stability score is based on the time-series data of frame motor housing temperature, rotor motor current, high-speed motor current and rotor speed. The mean value of the coefficient of variation (CV) of each parameter is calculated and determined in combination with the preset normal threshold value of the coefficient of variation.

[0015] As one possible implementation, the calculation process for the fourth stability score includes: For the time series data of each parameter, calculate the mean and standard deviation within a preset rolling window; Based on standard deviation / mean 100%, calculate the coefficient of variation (CV) for each window; The mean CV of each parameter is obtained by averaging the CV values ​​of all windows. Based on the comparison results between the mean CV and multiple preset threshold intervals, the stability score corresponding to each parameter is determined. The average stability score of each parameter is used as the fourth stability score.

[0016] As one possible implementation, the health status dimension index value is obtained by integrating the motor current health status and the multi-parameter comprehensive health status; The motor current health is calculated based on the deviation rates of the real-time current of the rotor motor, the real-time current of the high-speed motor, and the real-time current of the frame motor from their respective rated currents, combined with weights determined by the entropy weight method. Specifically, for each sampling point, the relative deviation rates of the currents of the rotor motor, the high-speed motor, and the frame motor are calculated; the average deviation rate of each motor is obtained by statistically analyzing the mean of the relative deviation rates of the currents of each motor within a preset period; the current weights of each motor are assigned based on the average deviation rates of each motor using the entropy weight method; and the motor current health is calculated and normalized according to Formula 1 - (rotor motor weight × rotor motor average deviation rate + high-speed motor weight × high-speed motor average deviation rate + frame motor weight × frame motor average deviation rate). The multi-parameter comprehensive health score is calculated by inputting the core operating parameters of the control moment gyroscope into a random forest model, an XGBoost model, a linear discriminant analysis (LDF) model, an isolated forest model, and a multilayer perceptron (MLP) model, respectively, based on the anomaly probability value or health status probability value output by each model. Specifically: the anomaly probability values ​​output by the random forest model, XGBoost model, and isolated forest model are obtained and converted into corresponding health scores according to the formula (1 - anomaly probability value) × 10; the LDF score output by the LDF model is obtained and normalized based on a pre-determined normal sample LDF score threshold, and converted into corresponding health scores; the health status probability value output by the MLP model is obtained and multiplied by 10 to obtain the corresponding health score; the minimum value among the health scores corresponding to the random forest model, XGBoost model, LDF model, isolated forest model, and MLP model is determined as the multi-parameter comprehensive health score.

[0017] Secondly, the present invention provides an on-orbit evaluation system for a control torque gyroscope, comprising: The data acquisition module is used to acquire on-board telemetry data of the control torque gyroscope during its on-orbit operation. The on-board telemetry data includes mechanical system parameters, electrical system parameters, environmental adaptation parameters, and mission performance parameters. The index calculation module is used to calculate the index values ​​of the control moment gyroscope under multiple preset evaluation dimensions based on on-board telemetry data. The multiple preset evaluation dimensions include reliability, on-orbit performance, stability and health status. The comprehensive evaluation module is used to generate on-orbit comprehensive evaluation results of the control moment gyroscope based on the index values ​​under each evaluation dimension.

[0018] Thirdly, the present invention provides an on-orbit evaluation terminal for a control moment gyroscope, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the on-orbit evaluation method for a control moment gyroscope provided by the present invention.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The present invention proposes an on-orbit evaluation method for control moment gyroscopes (CMGs). By expanding the telemetry data acquisition range and enriching the dimensions of on-board telemetry data acquisition, it comprehensively covers various key parameters such as CMG mechanical system, electrical system, environmental adaptation, and mission performance, providing sufficient and comprehensive data support for multi-dimensional evaluation. This addresses the technical bottleneck of insufficient evaluation data dimensions and inadequate support from the source, ensuring the effective implementation of the evaluation system and calculation method.

[0020] 2. The present invention proposes an on-orbit evaluation method for control moment gyroscopes (CMGs), which innovatively integrates multi-source evaluation calculation methods. For example, for the stability evaluation dimension, it adopts a collaborative calculation mode of basic index calculation, neural network prediction, and data feature analysis. It directly quantifies core parameters through telemetry data, mines hidden failure patterns based on neural networks, and extracts abnormal signals based on feature analysis. After multi-source result fusion processing, it significantly improves the accuracy and robustness of CMG on-orbit evaluation and solves the problems of existing algorithms being simple and lacking evaluation accuracy.

[0021] 3. The present invention proposes an on-orbit evaluation method for control moment gyroscopes (CMGs), which innovatively constructs a multi-dimensional evaluation system, breaks through the limitations of existing single-dimensional evaluations, and establishes a comprehensive evaluation framework covering stability, reliability, health status, and on-orbit performance, based on the core requirements of CMG on-orbit operation. This enables a comprehensive and accurate characterization of CMG on-orbit performance and effectively solves the technical problems of one-sided evaluation results and biased decision-making in existing systems. Attached Figure Description

[0022] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of the on-orbit evaluation method for the control torque gyroscope in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the process of obtaining the index values ​​for the on-orbit reliability evaluation of the control moment gyroscope in this embodiment of the invention. Figure 3 This is a flowchart illustrating the process of obtaining on-orbit performance dimension index values ​​for the control torque gyroscope in this embodiment of the invention. Figure 4 This is a flowchart illustrating the process of obtaining the index values ​​for the on-orbit stability dimension of the control moment gyroscope in this embodiment of the invention. Figure 5 This is a flowchart illustrating the process of obtaining index values ​​for the on-orbit evaluation of the health status dimension of the control torque gyroscope in this embodiment of the invention. Detailed Implementation

[0023] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, the first threshold and the second threshold are merely used to distinguish different thresholds and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0024] It should be noted that in this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0025] In this invention, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one of a, b, or c" can represent: a, b, c, a combination of a and b, a combination of a and c, a combination of b and c, or a, b, and c, where a, b, and c can be single or multiple.

[0026] In a first aspect, embodiments of the present invention provide an on-orbit evaluation method for a control torque gyroscope, comprising the following steps: See [link to relevant documentation] Figure 1 ; (a) Parameter Acquisition Acquire on-orbit telemetry data of the control torque gyroscope. The on-orbit telemetry data includes mechanical system parameters, electrical system parameters, environmental adaptation parameters, and mission performance parameters. Among them, the mechanical system parameters include at least rotor speed, frame rotation angle, and frame vibration data; the electrical system parameters include at least rotor motor current, high-speed motor current, and frame motor current; the environmental adaptation parameters include at least frame motor shell temperature; and the mission performance parameters include at least frame coarse and fine differences, torque output value, and angular momentum exchange state parameters. Collecting multi-dimensional telemetry data can fill gaps in parameter coverage, comprehensively capture the actual on-orbit operational characteristics and multi-stress coupling patterns of CMG, break through the limitations of single-parameter evaluation, and provide data support for building a comprehensive and systematic evaluation system. At the same time, it can support coupled correlation analysis and refined algorithm modeling, avoid the one-sidedness of evaluation, and provide accurate data support for CMG on-orbit fault early warning, life prediction, attitude control optimization, and full life cycle operation and maintenance, thereby improving the credibility of evaluation and the value of engineering decision-making.

[0027] Rotor speed is a fundamental parameter determining the gyroscope's angular momentum. Its stability directly affects the constancy of the output torque, and speed fluctuations can lead to attitude control deviations. Rotor speed is continuous numerical time-domain data. Under normal conditions, the value is close to the constant design value with minimal fluctuations, highly concentrated data distribution, and a significantly smaller standard deviation. Under abnormal conditions, the frequency and amplitude of fluctuations increase significantly, the standard deviation increases, and extreme values ​​are more likely to exceed the preset safety range, resulting in a substantial increase in dispersion.

[0028] The rotor motor current reflects the load condition driving the rotor. Abnormal current may indicate rotor dynamic imbalance or motor winding faults, threatening speed stability. The rotor motor current is a time-domain current signal. Under normal conditions, the value is stable without obvious abrupt changes, the waveform is regular and smooth, and the spectrum contains only components matching the normal operating frequency, without additional harmonics or noise. Under abnormal conditions, the current rises and falls sharply, the waveform is distorted, and harmonic peaks of specific frequencies appear in the frequency domain, disrupting the numerical stability.

[0029] Among them, the high-speed motor current is related to the driving state of high-speed rotating components. Its changes can reflect the load fluctuations and health status of the motor in a timely manner, avoiding the impact of high-speed motor failure on the overall gyroscope function. The high-speed motor current is also a time-domain current signal. Under normal conditions, it changes smoothly with the driving load demand, the value fluctuation matches the load intensity, the waveform is continuous without abrupt jumps, and the spectrum is clean without abnormal components. Under abnormal conditions, the load fluctuates frequently, the value changes suddenly, the waveform is irregular, and abnormal spectral peaks deviating from the normal range appear in the frequency domain.

[0030] Among them, the frame motor current corresponds to the driving force requirement for frame rotation. An excessively large current peak may indicate mechanical jamming of the frame or overload of attitude commands, affecting the response efficiency of attitude adjustment. The frame motor current is a continuous numerical time-domain data. Under normal conditions, it increases and decreases smoothly in sync with attitude commands, and the trend of numerical change is consistent with the command requirements, with the peak value controlled within the design upper limit. Under abnormal conditions, the peak value far exceeds the design threshold, the current suddenly increases, the trend of change is decoupled from the command, and irregular jumps or continuous high values ​​occur.

[0031] Among them, the frame coarse and fine difference is the core indicator of frame positioning accuracy. Exceeding the tolerance will cause the torque output direction to deviate and aggravate attitude drift. The frame fine and coarse difference is a numerical error data. Under normal circumstances, it fluctuates around the zero value in a small range, the data is concentrated within the accuracy threshold, the frequency of exceeding the tolerance is extremely low, and statistical indicators such as the 95th percentile are close to zero. When abnormal, the fluctuation range expands, the quantile exceeds the accuracy limit, the extreme value deviates significantly from the zero value, and the frequency of exceeding the tolerance increases significantly.

[0032] Among them, the frame motor housing temperature is crucial to the reliability of the motor. Excessive temperature will accelerate insulation aging and may even cause the motor to stall, which is essential for the long-term stable operation of the control torque gyroscope. The frame motor housing temperature is a continuous numerical time-domain data. Under normal conditions, it changes slowly within the safe temperature range, with a gentle trend of increase and decrease, and the distribution follows a normal law without sudden rises or falls. Under abnormal conditions, the temperature rises suddenly and sharply, the extreme value exceeds the safe upper limit, the trend of change is steep, the distribution deviates from normal, and local high temperature peaks appear.

[0033] (ii) Calculation of index values Based on on-board telemetry data, the index values ​​of the control torque gyroscope are calculated under multiple preset evaluation dimensions; the multiple preset evaluation dimensions include reliability dimension, on-orbit performance dimension, stability dimension and health status dimension; 1. Reliability dimension As one possible implementation, reliability metrics are obtained by integrating the reliability of the fusion-frame motor and the on-orbit reliability of a single unit; see [link to relevant documentation]. Figure 2 ; As an example, a weighted summation method is used to integrate the reliability of the frame motor and the on-orbit reliability of a single unit, as follows: The weights are assigned based on the coupling characteristics of the CMG mechanical-electric system. The on-orbit reliability of a single unit is assigned a weight of 0.4, and the reliability of the frame motor is assigned a weight of 0.6. The overall reliability is denoted as R, and the calculation formula is: Overall reliability R = Reliability of frame motor × 0.6 + Reliability of single unit on-orbit operation × 0.4, rounded to two decimal places; Overall reliability level classification: Excellent (98≤R≤100), Good (90≤R<98), Medium (80≤R<90), Poor (65≤R<80), Very Poor (R<65).

[0034] The reliability of the frame motor is calculated based on the logical comparison results of the real-time status of the frame motor shell temperature, frame motor current and frame coarse and fine difference with the corresponding preset thresholds and the normality rate. The reliability of single-unit on-orbit operation is obtained by first-order difference calculation, mutation detection, and entropy weighting summation based on time-series data of frame motor shell temperature, rotor motor current, high-speed motor current, and rotor speed.

[0035] (1) Reliability of frame motor As one possible implementation method, the reliability calculation method for the frame motor is as follows: For each sampling point, the frame motor housing temperature, frame motor current, and frame coarse and fine differences are compared with their respective preset safety threshold ranges to generate a status indicator indicating whether the status is normal or exceeds the limit. The safety threshold range is the boundary for determining whether a parameter is normal, and it needs to be calibrated based on motor design specifications and ground test data.

[0036] Among them, the frame motor housing temperature affects the insulation and bearing life of CMG. Abnormally excessive frame motor housing temperature will accelerate insulation aging and bearing wear, directly shortening the service life of CMG. For example, the preset safe threshold range for frame motor housing temperature is 20℃~60℃.

[0037] If the frame motor current is abnormal, it can directly reflect hidden faults such as motor load imbalance, winding short circuit or open circuit; for example, the preset safe threshold range of frame motor current is 1.2A~2.5A.

[0038] Poor frame quality can indicate excessive clearance, jamming, or reduced transmission accuracy in the mechanical transmission mechanism.

[0039] As an example, ensure that the percentage of valid sampling points is ≥98% within a day to avoid monitoring blind spots caused by missing data.

[0040] As an example, data with lost telemetry packets and abnormal jump values ​​without any apparent cause are excluded. For instance, data with a single jump amplitude exceeding three times the normal fluctuation without the influence of external factors such as attitude adjustment or load change.

[0041] As an example, the sampling frequency is synchronized with the telemetry downlink frequency, for example, 1~10Hz, to ensure that temperature changes are captured in real time and that a day is a complete monitoring period.

[0042] The normality rate of each parameter is calculated based on the ratio of the number of sampling points with normal status within a preset time period to the total number of valid sampling points. As an example, perform a two-branch logical operation on the real-time data of each sampling point: If the lower threshold ≤ parameter ≤ upper threshold, the status is determined to be "normal" and the status variable is recorded as 0. If the parameter is less than the lower threshold or greater than the upper threshold, it is determined to be "state out of limit", the state value is recorded as 1, and a unified identifier for exceeding the limit is used to focus on rapid early warning. In order to avoid misjudgment caused by instantaneous interference, anti-shake logic is added. That is, the parameter is finally confirmed to be valid only when three consecutive sampling points are determined to be 1. If the parameter exceeds the limit and then immediately returns to normal, it is determined to be instantaneous interference, and the state value is corrected to 0 to ensure the accuracy of the early warning.

[0043] As an example, to calculate the normality rate of each parameter, for each valid sampling point within a day, count the number of times "state variable = 0" and calculate the normality rate: The result should be rounded to two decimal places.

[0044] The average value of the frame motor housing temperature normality rate, frame motor current normality rate, and frame coarse and fine difference normality rate is determined as the frame motor reliability. As an example, in addition to outputting the reliability of the frame motor, it also outputs status indicators, daily statistics, and the normality rate of each parameter; The status indicators include: 0 = normal, 1 = exceeding the limit; if the limit is exceeded, the output includes the type of the limit exceedance and the core data of the limit exceedance, such as high temperature of the casing exceeding the limit, current overload exceeding the limit; current parameter value, the start time of the limit exceedance (accurate to the second), and duration; The daily statistics include: the total number of times each parameter exceeds the limit and the maximum deviation exceeding the limit; In summary, the method for calculating the reliability of the CMG frame motor provided by this invention enables quantitative monitoring of the multi-dimensional status of the CMG frame motor; single-parameter status determination quickly identifies anomalies, providing a basis for local fault location; and the normality rate and reliability indicators intuitively reflect the overall health level of the motor, supporting CMG on-orbit performance evaluation and maintenance decisions.

[0045] (2) On-orbit reliability of a single unit The method for calculating the on-orbit reliability of a single unit is as follows: First-order difference calculations were performed on the time-series data of frame motor housing temperature, rotor motor current, high-speed motor current and rotor speed to obtain the first-order difference sequence of each parameter; The first-order difference is used to capture sudden changes in data, calculated as the difference between two adjacent data points. By calculating the difference between adjacent data points, sudden changes in data are captured. Its core function is to monitor sudden changes in rotor motor current and assist in judging abnormal current frequency caused by faults such as winding short circuits.

[0046] As an example, the first difference = the next data point - the previous data point; calculate the difference between each adjacent data point to form a first difference sequence.

[0047] The difference in the absolute value of the first-order difference sequence of each parameter that exceeds the preset mutation threshold is determined as a sudden mutation, and the number of sudden mutations of each parameter within the preset period is counted. As an example, the preset cycle is 1 day.

[0048] The single-parameter reliability of each parameter is calculated based on the ratio of the number of sudden mutations of each parameter to the total number of data points. As an example, the single-parameter reliability is denoted as S, and its expression is: S = (Number of normal occurrences / Total number of occurrences) × 100% Wherein, normal number of times = total number of times - number of sudden mutations.

[0049] The entropy weight method is used to assign weights to each parameter based on the degree of dispersion of the first-order difference sequence of each parameter; The on-orbit reliability of a single unit is obtained by weighted summation based on the single-parameter reliability and weight of each parameter. As an example, the reliability of a single machine is denoted as... The expression is: in, This indicates the weight of each parameter.

[0050] As an example, a first-order difference sequence formed by the differences between consecutive adjacent data points can form a difference fluctuation curve; After assessment, the result of the sudden mutation was recorded as either "mutation present" or "no mutation present".

[0051] In summary, the method for calculating the on-orbit reliability of a single unit provided by this invention achieves multi-dimensional quantification and visualization of single-unit operational reliability. Single-parameter reliability can directly pinpoint the weak parameters of single-unit operation, assisting the ground team in optimizing operational parameters for these parameters. Single-unit reliability and comprehensive reliability are condensed indicators of the continuous operational capability of a single unit, providing quantitative basis for the scheduling and maintenance planning of CMG on-orbit missions.

[0052] 2. On-orbit performance dimension As one possible implementation, the on-orbit performance metric is obtained by fusing the overall on-orbit performance score, performance change trend score, and deterioration trend score; see [link to relevant documentation]. Figure 3 ; Among them, the overall on-orbit performance reflects the level of stability and compliance under the current operating conditions, the performance change trend predicts the long-term trend, and the deterioration trend assessment focuses on the rate of degradation. All three comprehensively cover performance characteristics from the current, long-term, and growth rate perspectives, and are equally important.

[0053] As an example, during the fusion process, the on-orbit comprehensive performance score, performance change trend score, and deterioration trend score are uniformly normalized to the range of 0-10. For example, when the performance change trend score is <0, it is counted as 0, and when the performance change trend score is >10, it is counted as 10; the deterioration trend score is counted in the same way.

[0054] As an example, the overall performance score P = (on-orbit overall performance score + performance change trend score + deterioration trend score) ÷ 3, with the result rounded to one decimal place.

[0055] As an example, the on-orbit performance rating rules are: Excellent 9~10 points, Good 7~8.9 points, Pass 6~6.9 points, and Unpass <6 points.

[0056] The overall on-orbit performance score is obtained by multi-dimensional scoring and entropy weighting based on the mean, range, and standard deviation of rotor motor current, high-speed motor current, frame motor current, rotor speed, frame coarse and fine difference, and frame motor shell temperature, as well as the corresponding engineering calibration parameters. Insufficient overall on-orbit performance will directly affect the output accuracy of CMG control torque, increasing the risk of spacecraft attitude control deviation.

[0057] The performance change trend score is calculated based on the change trend of the output anomaly probability value by inputting the time series data of rotor speed, frame motor housing temperature and high-speed motor current into the Long Short-Term Memory network LSTM. The deterioration trend score is calculated based on the time-series data of rotor motor current, high-speed motor current, frame motor current and rotor speed input into the denoising autoencoder (DAE) and the time-varying rate of change of its output reconstruction error.

[0058] (1) Overall performance score on orbit Real-time telemetry of rotor motor current, high-speed motor current, frame motor current, rotor speed, frame coarse and fine difference, and frame motor shell temperature. These parameters directly reflect the on-orbit operating status of each system of the single unit. The rated ranges, maximum allowable ranges, and maximum allowable standard deviations of each parameter are preset; the maximum allowable standard deviation is determined based on the CMG design specifications and ground tests; these parameters are the benchmark boundaries for performance scoring.

[0059] Data acquisition: The sampling frequency is synchronized with the telemetry downlink frequency. For example, the sampling frequency is 1~10Hz to ensure complete capture of parameter status. Data loss due to telemetry and abnormal jump values ​​without cause are removed. For example, a single jump amplitude exceeding the rated range by 20% without any change in external operating conditions is considered an abnormal jump value without cause and is removed. The proportion of valid sampling points within the data acquisition period is ≥95% to avoid performance evaluation bias caused by missing data.

[0060] As one possible implementation method, the on-orbit integrated performance score is calculated as follows: Calculate the mean, range, and standard deviation of each parameter within a preset period; Among them, the mean reflects the overall operating level of the parameters, the range reflects the extreme value fluctuation range, and the standard deviation represents the degree of discrete fluctuation. The combination of the three can comprehensively evaluate the on-orbit stability and compliance capability of a single unit. As an example, the extreme value fluctuation range of each parameter can be obtained by the difference between the maximum and minimum values; For each parameter, a mean compliance score is determined based on the comparison between the mean and a preset rated range. As an example, the average compliance score is set to a maximum of 4 points; The mean is completely within the specified range → 4 points; If the mean value exceeds the rated range but deviates from the rated center value by ≤5%, then 2 points are awarded. If the mean exceeds the rated range and deviates from the rated center value by more than 5%, the score will be 0.

[0061] The range compliance score is determined based on the comparison between the range and the preset maximum allowable range. As an example, the maximum score for poor compliance is set at 3 points; Range ≤ Maximum Allowable Range → 3 points; Maximum allowable range < range ≤ 1.2 × maximum allowable range → 1 point; Range > 1.2 × maximum allowable range → 0 points.

[0062] The volatility stability score is determined based on the comparison between the standard deviation and the preset maximum allowable standard deviation. As an example, the maximum score for fluctuation stability is set at 3 points; Standard deviation ≤ maximum allowable standard deviation → 3 points; Maximum allowable standard deviation < standard deviation ≤ 1.2 × maximum allowable standard deviation → 1 point; Standard deviation > 1.2 × maximum allowable standard deviation → 0 points.

[0063] The mean compliance score, range compliance score, and fluctuation stability score are added together to obtain the single-parameter on-orbit performance score. The entropy weight method is used to determine the weight of each parameter, and the on-orbit performance scores of all individual parameters are weighted and summed to obtain the overall on-orbit performance score.

[0064] As an example, the output includes a single-parameter on-orbit performance score, a single-unit on-orbit comprehensive performance score, and a performance level. In summary, the on-orbit comprehensive performance score calculation method provided by this invention achieves multi-dimensional and quantitative evaluation of on-orbit performance; the multi-dimensional scoring of mean, range, and standard deviation covers the full-state characteristics of overall parameter level, extreme value fluctuation, and discrete fluctuation, which is more comprehensive than single-index evaluation; the entropy weight method can reflect the differences in the impact of each parameter on the performance of a single unit, improving the rationality of the comprehensive score; the clear performance level classification (excellent / good / qualified / unqualified) facilitates the ground team to quickly determine the on-orbit status of a single unit, providing an intuitive basis for mission payload allocation and maintenance priority ranking.

[0065] (2) Performance change trend score This paper applies Long Short-Term Memory (LSTM) networks to calculate performance change trend scores. LSTM is an improved model of recurrent neural networks. Through gating units, including input gates, forget gates, and output gates, it solves the problem of long sequence dependencies and can capture long-term variation patterns in time-series data, making it suitable for device performance trend prediction tasks. LSTM can selectively retain or forget historical information, output the state probability at each time step, and predict the state in future periods, thus achieving trend analysis.

[0066] The time-series data of CMG rotor speed, frame motor housing temperature, and high-speed motor current directly reflect the performance fluctuations of CMG, providing a time series basis for capturing long-term trends. The time-series data of the above parameters are input into LSTM.

[0067] LSTM outputs the anomaly probability value at each time step, including the predicted values ​​for the current period and future periods. These data can reflect the temporal variation pattern of CMG state.

[0068] The calculation method for the performance change trend score is as follows: Obtain the anomaly probability value of the LSTM network output at each time step; Perform first-order difference calculation on the sequence of abnormal probability values; As an example, the statistical difference results are positive or negative; a positive difference indicates performance improvement, while a negative difference indicates performance degradation.

[0069] The proportion of time steps with positive first-order differences is used as the proportion of improvement trend, and the proportion of time steps with negative first-order differences is used as the proportion of deterioration trend. Based on (Percentage of Improvement Trends - Percentage of Deterioration Trends) 10. Calculation performance trend score; A higher performance trend score indicates that the performance is improving, while a lower score indicates that the performance is deteriorating. The performance trend score can be used to predict the performance trend of CMG in advance.

[0070] (3) Deterioration trend score A denoising autoencoder (DAE) is applied to calculate the degradation trend. A DAE is an unsupervised deep learning model that learns normal patterns in data and identifies abnormal states that deviate from these patterns, making it suitable for detecting device degradation trends. The DAE adds noise to the input data, trains the model to reconstruct the original data, and calculates the reconstruction error between the input and the reconstructed output. The larger the error, the more significant the deviation of the sample from the normal pattern. The degradation trend can be analyzed by combining the rate of change of the error.

[0071] The CMG's timing data for rotor motor current, high-speed motor current, frame motor current, and rotor speed exhibit clear normal pattern characteristics, providing a foundation for learning normal reconfiguration modes. The timing data of the above parameters are then input into the DAE.

[0072] The method for calculating the deteriorating trend score is as follows: Obtain the parameter reconstruction error output by the DAE network; Among them, the larger the parameter reconstruction error, the more significant the deviation between the current state and the normal mode, and the higher the degree of degradation.

[0073] Calculate the rate of change of the reconstruction error over time within the sliding window; Among them, the rate of change over time can reflect the speed of deterioration, and the distribution of the rate of change can identify the rate of increase in the deterioration trend.

[0074] Normalize the rate of change over time; Based on (1 - normalized rate of change over time) 10. Calculate the deterioration trend score.

[0075] Among them, the lower the deterioration trend score, the faster the performance deteriorates and the higher the risk. When the score is below 3, a higher priority on-orbit monitoring process needs to be triggered.

[0076] 3. Stability dimension As one possible implementation, the stability dimension index value is obtained by fusing the first stability score, the second stability score, the third stability score, and the fourth stability score; see [link to relevant documentation]. Figure 4 ; Among them, the first stability score, the second stability score, the third stability score and the fourth stability score evaluate stability from four perspectives: parameter fluctuation quantification, real-time state classification, real-time state classification (SVM) and normalized fluctuation comparison. They are all directly related to the consistency of CMG torque output and component wear risk, and have no priority difference. Therefore, the arithmetic mean fusion is adopted.

[0077] As an example, the stability scores for the first through fourth levels are all on a scale of 0 to 10, requiring no additional conversion. As an example, the overall stability score S = (standard deviation score + decision tree score + SVM score + CV score) ÷ 4, with the result rounded to two decimal places; As an example, stability levels are categorized as follows: 9≤S≤10 is "Excellent", 8≤S<9 is "Good", 6≤S<8 is "Medium", 3≤S<6 is "Acceptable", and 0≤S<3 is "Very Poor". This grading method is adapted to the early warning needs of engineering operation and maintenance.

[0078] (1) First stability score The first stability score is based on the time series data of frame motor housing temperature, rotor motor current, high-speed motor current, frame motor current and rotor speed. The standard deviation and mean standard deviation of each parameter are calculated and determined in combination with the preset standard deviation stability threshold. Among them, the fluctuation stability of the core parameters such as frame motor housing temperature, rotor motor current, high-speed motor current, frame motor current and rotor speed directly determines the attitude control accuracy of the control torque gyroscope, the reliability of components and the adaptability to on-orbit missions. Abnormal fluctuations in parameters will lead to a decrease in torque output consistency and increased wear of core components. Long-term deterioration may also cause equipment shutdown.

[0079] The standard deviation stability threshold serves as the benchmark for stability scoring.

[0080] As an example, time-series data of frame motor housing temperature, rotor motor current, high-speed motor current, frame motor current, and rotor speed are collected: sampling frequency 1Hz, 1 data point per second. This setting conforms to standard engineering configurations and requires no additional hardware upgrades; the default scrolling window is 30 seconds, containing 30 data points, which can be adjusted to 10 seconds or 60 seconds as needed. 10 seconds provides fast response, while 60 seconds offers better smoothness; extreme anomalies are removed, such as single temperature jumps >5℃, current exceeding the rated range by 20%, and speed deviation from the design value ±10%. After removal, the data from the previous moment is used to replace the missing data. Removing extreme anomalies avoids interference with standard deviation calculations; ensuring no data loss is achieved, if data is missing at a certain moment, the average value of the previous window is temporarily used to fill the gap, ensuring the continuity of scrolling calculations.

[0081] As an example, Single-window parameter standard deviation calculation: Take continuous data points from the time series, for example, n>30, and calculate the standard deviation of the data within that window using the following formula: Multi-window standard deviation mean calculation: Calculate the arithmetic mean of the standard deviations of all windows for each parameter. As an example, Single-parameter stability score calculation: The ratio r of the mean standard deviation of the parameter to the benchmark threshold is used as the core of the score. The smaller r is, the stronger the stability and the higher the score. Scoring rules: Score = 10 - 10 × r (r ≤ 1); Score = 0 (r > 1); When r > 1, the parameter fluctuations have exceeded the reasonable range, the stability is extremely poor, and a score of 0 is given directly.

[0082] As an example, the first stability score is calculated as follows: Based on the dispersion of the standard deviation of each parameter window, the entropy weight method is used to calculate the weight of each parameter. The higher the dispersion, the greater the weight, reflecting the more significant the impact of the parameter on the overall stability. The overall stability score of CMG is calculated by weighted summation, using the following formula: First stability score = . In summary, the first stability score is a refined quantification and stability assessment of the fluctuations in single-machine operation; the window standard deviation and mean can accurately capture the short-term fluctuation characteristics of parameters and help locate potential anomalies such as high-frequency fluctuations.

[0083] (2) Second stability score The second stability score is determined by inputting time-series data of frame motor current and frame motor shell temperature into a decision tree model and performing statistical calculations based on the output state identification labels. Among them, decision trees are supervised machine learning algorithms that are highly interpretable and computationally efficient, making them suitable for state classification tasks of industrial equipment. In CMG (Continuous Metal Gear) operation status determination, they can intuitively present the correlation logic between features and states. Decision trees select optimal features based on indicators such as information gain and Gini coefficient, recursively partition the feature space to construct a tree structure, with each internal node corresponding to the judgment condition of the feature, and leaf nodes corresponding to the classification result, until leaf node samples belong to the same category or meet the stopping condition, ultimately achieving a binary classification of normal / abnormal.

[0084] As an example, the two key telemetry parameters of CMG, namely frame motor current and frame motor case temperature, are input into the decision tree. Among them, frame motor current directly reflects the electrical load operating status of the motor, and frame motor case temperature reflects the thermal stability of the motor. The two are jointly related to the core operating conditions of CMG, providing a basis for feature division of the decision tree.

[0085] As an example, the decision tree outputs the status identification label of the CMG, clearly determining whether the current working condition is normal or abnormal. This classification method is highly interpretable and can intuitively present the feature conditions corresponding to the status determination; for example, if the frame motor shell temperature is >60℃, it is determined to be abnormal.

[0086] As an example, the second stability score calculation method: The sliding window statistics include the number of normal labels, the number of consecutive abnormal labels, and the total number of labels within a given period. Second stability score = max{0, (number of normal labels in the window / total number of labels) × 10 - (number of consecutive anomalies × 0.5)}; The higher the second stability score, the smaller the operational fluctuation and the stronger the stability, which can be directly used as a decision-making basis for temporarily adjusting the CMG operating parameters.

[0087] (3) Third stability score The third stability score is determined by inputting time-series data of rotor speed and frame motor housing temperature into a support vector machine (SVM) model and performing statistical calculations based on the output state identification labels. Support Vector Machine (SVM) is a classic supervised machine learning algorithm, suitable for binary classification tasks with small samples and high-dimensional features, especially in scenarios where the boundaries between normal and abnormal categories are unclear in industrial equipment condition monitoring. SVM is maximum margin classification, which searches for the optimal hyperplane in a high-dimensional feature space to maximize the margin between the hyperplane and the normal and abnormal state samples. For non-linearly separable data, kernel functions, such as radial basis function kernels, can be used to map low-dimensional data to a high-dimensional space to achieve linear separability. Finally, the classification boundary is determined by the support vector closest to the hyperplane, ensuring the model's generalization ability and avoiding overfitting.

[0088] As an example, two key telemetry parameters of the CMG (Central Molecular Generator) – rotor speed and frame motor housing temperature – are input into the support vector machine (SVM). The rotor speed determines the fundamental rate of momentum exchange in the CMG, directly impacting torque output stability. The frame motor housing temperature reflects the motor's thermal operating state; abnormal temperatures can lead to decreased insulation performance and bearing wear. These two parameters cover the core states of the CMG's momentum and thermal systems, providing data support for SVM to handle high-dimensional feature classification tasks, aligning with its application characteristics in equipment condition monitoring.

[0089] As an example, the SVM output is a CMG status identification label, which clearly determines whether the current operating condition is normal or abnormal, realizing binary distinction of operating conditions. The classification result has strong robustness and can effectively resist slight noise interference in telemetry data.

[0090] As an example, the third stability score calculation method: The sliding window counts the number of normal labels, the number of consecutive occurrences of abnormal labels, and the total number of labels within a given period. The third stability score = max{0, (number of normal labels in the window / total number of labels) × 10 - (number of consecutive anomalies × 0.5)}.

[0091] The third stability score quantifies the stability of CMG's real-time operation on a 10-point scale. The higher the score, the smaller the operational fluctuations and the stronger the stability, which can serve as a basis for the ground team to make quick decisions on temporarily adjusting CMG operating parameters.

[0092] (4) Fourth stability score The fourth stability score is based on the time-series data of frame motor housing temperature, rotor motor current, high-speed motor current and rotor speed. The mean value of the coefficient of variation (CV) of each parameter is calculated and determined in combination with the preset normal threshold value of the coefficient of variation.

[0093] Among them, the coefficient of variation (CV) of the frame motor housing temperature, motor current, and rotor speed of the control torque gyroscope (CMG) is the core indicator for quantifying its operational stability. The rolling trend can accurately reflect the dynamic changes in the performance of the CMG drive system, thermal control system, and the health status of core components.

[0094] Using the coefficient of variation (CV) as the core analytical indicator, the system performs multi-step processing on the time-series data of key CMG operating parameters to achieve dynamic evaluation of the stability of temperature, current, and speed fluctuations. Combined with threshold determination and visualization, this ultimately forms a quantifiable single-unit stability index, providing a standardized analytical process for CMG operating status monitoring and anomaly early warning.

[0095] The calculation process for the fourth stability score includes: For the time series data of each parameter, calculate the mean and standard deviation within a preset rolling window; As an example, when collecting time-series data of parameters, the sampling frequency is consistent with the telemetry downlink frequency; for example, the sampling frequency is 1~10Hz to ensure that the data density supports window calculation; the time window adopts a fixed rolling window, for example, 1 hour / 1 orbital mission cycle, and the proportion of effective data points in each window is ≥95%; physical anomalies are removed, such as T<-10℃ or>60℃, current exceeding the rated range by 20%, speed deviating from the design value by ±10%, and special operating condition data, such as equipment start-up and shutdown, attitude maneuvering switching periods.

[0096] As an example, the normal threshold for the coefficient of variation is denoted as CV0 and uniformly set to 10%. The normal threshold for the coefficient of variation needs to be set in combination with the actual engineering situation and serves as a benchmark for judging whether the parameter fluctuation is normal. If it exceeds the threshold, it is considered to be an abnormal stability.

[0097] As an example, the arithmetic mean formula is applied to calculate the mean μ of each parameter within each time window to reflect the overall level of the parameter during that period.

[0098] As an example, the standard deviation σ of the parameter within each window is calculated to quantify the fluctuation of the parameter from the mean. The larger the standard deviation, the worse the operational stability.

[0099] Based on standard deviation / mean 100%, calculate the coefficient of variation (CV) for each window; The CV calculation method is the standard deviation normalization method, which can eliminate the influence of the mean on the fluctuation assessment and realize cross-parameter and cross-operating condition horizontal comparison.

[0100] The mean CV of each parameter is obtained by averaging the CV values ​​of all windows. Among them, the mean CV can be used as the core basis for stability scoring and threshold determination.

[0101] Based on the comparison results between the mean CV and multiple preset threshold intervals, the stability score corresponding to each parameter is determined. As an example, set a CV threshold: The normal threshold (CV0) is 10%; the attention threshold (CV1=1.2×CV0); the warning threshold (CV2=1.5×CV0); and the emergency threshold (CV3=2.0×CV0).

[0102] The mean CV value is less than CV0, the score is 10, and the performance is stable. The mean CV value is between CV0 and CV1, with a score of 8, indicating a relatively stable performance. The mean CV value exceeds CV1, the score is 6, and the fluctuation is increasing; The mean CV value exceeds CV2, with a score of 4, indicating a regression. The mean CV value exceeds CV3, the score is 2, which is abnormal.

[0103] The average stability score of each parameter is used as the fourth stability score.

[0104] In summary, the fourth stability score, which uses the coefficient of variation to normalize and standardize the fluctuation assessment, breaks through the limitations of different parameter dimensions and mean differences on stability comparison, and can realize cross-type parameter stability comparisons such as shell temperature, current, and rotation speed.

[0105] The four-level threshold scoring rule constructs a full-link early warning system for stable, basically stable, increasing fluctuation, deterioration, and anomaly. The early warning levels are clear and the judgment criteria are explicit, which makes it easy for the ground team to quickly identify the stability degradation stage of CMG.

[0106] 4. Health Status Dimension As one possible implementation, the health status dimension index value is obtained by integrating the motor current health status and the multi-parameter comprehensive health status; As examples, multi-parameter comprehensive health includes: multi-parameter comprehensive health (random forest), multi-parameter comprehensive health (XGBoost), multi-parameter comprehensive health (LDF), multi-parameter comprehensive health (isolation forest), and multi-parameter comprehensive health (MLP).

[0107] As an example, both the motor current health score and the multi-parameter comprehensive health score are rated on a 0-10 scale, requiring no external conversion; As an example, the health scores of Random Forest, XGBoost, LDF, Isolation Forest, and MLP5 are extracted under a multi-parameter comprehensive health assessment, and the minimum value is denoted as H. min Taking the minimum value can focus on the weakest health risk points and avoid underestimating the potential for deterioration in a single dimension.

[0108] Health status dimension index value H=(H min +Motor current health status H I ÷ 2, and round the result to two decimal places; The indicators for the health status dimension are divided into levels: healthy 9.0≤H≤10 points, sub-healthy 8.0≤H<9.0 points, poor health 5.0≤H<8.0 points, and faulty 0≤H<0.5 points.

[0109] Among them, the CMG motor current is the core indicator reflecting the health status of its electrical performance. An excessive current deviation rate will directly reflect the increased loss of motor windings, abnormal load or fluctuations in the power supply system. Long-term deviation will lead to motor overheating, decreased insulation performance, and ultimately abnormal CMG output torque.

[0110] (1) Motor current health The motor current health status is calculated based on the deviation rate between the real-time current of the rotor motor, the real-time current of the high-speed motor, and the real-time current of the frame motor and their respective rated currents, combined with the weights determined by the entropy weight method. Among them, the real-time current Ir of the rotor motor, the real-time current Ih of the high-speed motor, and the real-time current If of the frame motor (unit: A) directly reflect the real-time electrical load status of each motor. As an example, the sampling frequency is synchronized with the telemetry downlink frequency. For example, the sampling frequency is 1~10Hz to ensure real-time capture of current changes; lost telemetry data and current jump values ​​without cause are eliminated, such as a single jump amplitude exceeding the rated current by 10% without external factors such as load adjustment; the proportion of valid sampling points within the data acquisition period is ≥98% to avoid evaluation bias caused by missing data.

[0111] Specifically, for each sampling point, the relative deviation rate of the current of the rotor motor, high-speed motor and frame motor is calculated; As an example, the relative deviation rate of current for each motor: in, This is the rated current.

[0112] The average deviation rate of each motor current is obtained by calculating the mean of the relative deviation rate of each motor within the preset period. As an example, the average deviation rate of each motor: M is the number of valid sampling points within the data acquisition period; The entropy weight method is used to assign current weights to each motor based on the average deviation rate of each motor. The motor current health is calculated and normalized according to Formula 1 - (rotor motor weight × rotor motor average deviation rate + high-speed motor weight × high-speed motor average deviation rate + frame motor weight × frame motor average deviation rate). As an example, motor current health status: Among them, H I The health score ranges from 0 to 10. A score closer to 10 indicates a healthier current system. ω r ω h ω f The current weights are for rotor motors, high-speed motors, and frame motors, respectively. The weights reflect the degree of influence of the current deviation of each motor on the overall health. The motor with the greater deviation fluctuation has a higher weight.

[0113] In summary, this technical solution quantifies the health level of each motor current through relative deviation rate calculation, entropy weight method weight allocation, and health normalization, providing a quantitative basis for on-orbit health management and fault early warning of the CMG electrical system. Specifically, the current deviation rate and average deviation rate accurately capture the degree of deviation of each motor current, directly locating motors with abnormal current fluctuations; the entropy weight method weight allocation reflects the priority of each motor current's impact, improving the rationality of the health assessment; the normalized health score (0-10 points) intuitively reflects the health level of the current system and can be linked with other health indicators (such as temperature and stability) to form a comprehensive health assessment system for the CMG, providing quantitative support for on-orbit fault early warning and maintenance decisions.

[0114] (2) Comprehensive health status based on multiple parameters The multi-parameter comprehensive health status is calculated by inputting the core operating parameters of the control moment gyroscope into the random forest model, XGBoost model, linear discriminant analysis LDF model, isolated forest model and multilayer perceptron MLP model respectively, and based on the abnormal probability value or health status probability value output by each model. Random forest is an ensemble learning algorithm composed of multiple decision trees. It has strong generalization ability and good resistance to noise interference, making it suitable for probabilistic assessment tasks of device health status. Through bootstrap sampling (randomly sampling the training set) and random subspace (randomly selecting features), multiple independent decision trees are constructed. In classification tasks, the results are output through majority voting, ultimately converting them into probability values ​​of whether a sample belongs to an abnormal category, reducing the risk of overfitting from a single tree.

[0115] XGBoost is an optimization algorithm for gradient boosting trees, characterized by high prediction accuracy and strong ability to handle nonlinear relationships, making it suitable for precise assessment of device health status. XGBoost iteratively trains weak classifiers (decision trees), fitting the residuals of the previous model each time, while introducing a regularization term to control model complexity. It optimizes the objective function based on gradient descent and ultimately outputs the anomaly probability value of the sample, capturing the complex correlation between parameters and health status.

[0116] LDF (Latent Distributed Dimensionality Reduction) is a supervised dimensionality reduction and classification algorithm. It is computationally efficient, suitable for feature discrimination with small samples, and widely used in feature difference analysis of equipment health status. LDF maximizes inter-class distance and minimizes intra-class distance, projecting high-dimensional features onto a low-dimensional discrimination space. The LDF scores of samples in this space are used to distinguish categories. The LDF scores of normal samples are concentrated in a specific interval, while those of abnormal samples deviate from this interval.

[0117] Isolation Forest is an unsupervised anomaly detection algorithm that requires no labeled data, is computationally efficient, and is suitable for anomaly detection tasks in industrial equipment. Isolation Forest constructs multiple isolated trees by randomly selecting features and split points. Anomaly samples, due to their features deviating from the normal distribution, are isolated to leaf nodes more quickly. The path length of a sample within the multiple trees is calculated; the shorter the path length, the more likely the sample is to be anomaly, which is then converted into an anomaly probability value.

[0118] Among them, MLP is a fundamental neural network model in deep learning. It has strong fitting ability and can learn complex nonlinear relationships between features, making it suitable for accurate probabilistic assessment of device health status. MLP consists of an input layer, a hidden layer, and an output layer. It introduces nonlinearity through activation functions, calculates the output through forward propagation, optimizes the weights through backpropagation, and finally outputs the probability value of a sample belonging to a healthy state.

[0119] Specifically: Obtain the anomaly probability values ​​output by the Random Forest model, XGBoost model, and Isolation Forest model, and convert them into corresponding health scores according to the formula (1 - anomaly probability value) × 10; Obtain the LDF score output by the LDF model, normalize it based on a pre-determined normal sample LDF score threshold, and convert it into the corresponding health score; Obtain the health status probability value output by the MLP model, multiply it by 10 to obtain the corresponding health score; Determine the minimum value among the health scores corresponding to the Random Forest model, XGBoost model, LDF model, Isolation Forest model, and MLP model as the multi-parameter comprehensive health score.

[0120] As an example, the core operating parameters of the control torque gyroscope include rotor speed, high-speed motor current, and frame motor housing temperature. The rotor speed reflects the basic operating state of the momentum system, the high-speed motor current reflects the load level of the electrical system, and the frame motor housing temperature reflects the thermal stability. These three types of parameters cover the characteristics of multiple CMG systems, providing sufficient feature dimensions for ensemble learning.

[0121] Input the core operating parameters of the control torque gyroscope into the random forest model. The output CMG anomaly probability value ranges from 0 to 1. The closer the probability value is to 1, the higher the probability that the sample belongs to an abnormal state; the closer it is to 0, the closer the running state is to normal. Health score = (1 - abnormal probability value) × 10; The CMG health score quantifies the overall health level of the CMG on a 10-point scale. A higher score indicates a lower degree of degradation in core modules such as motor components and thermal control systems, and a better health status.

[0122] As an example, the CMG's rotor motor current, frame motor current, high-speed motor current (electrical system), rotor speed (momentum system), and frame motor case temperature (thermal system) are input to XGBoost; XGBoost outputs the anomaly probability value of CMG, which is in the range of 0-1. Compared with traditional ensemble algorithms, its probability value has higher accuracy and can more accurately distinguish different health levels such as mild degradation and severe degradation.

[0123] Health score = (1 - abnormal probability value) × 10 A higher CMG health score indicates a better CMG health status. When the score is below 6, the potential deterioration risk of the component needs to be closely monitored.

[0124] As an example, the rotor motor current and rotor speed of the CMG are input into the LDF. The rotor motor current and rotor speed reflect the core states of the electrical system and momentum system, respectively. Their characteristic differences can effectively distinguish the normal / abnormal operating conditions of the CMG and provide a basis for the characteristic projection of the LDF.

[0125] The LDF output score reflects the position of a sample in the optimal discrimination space. The further the score deviates from the concentration range of normal samples, the higher the risk of anomaly.

[0126] The maximum LDF score of normal samples in the training set is used as the normal threshold. The LDF score of normal samples is normalized based on a pre-determined threshold and converted into a corresponding health score; specifically: Obtain the LDF score of the sample. If the LDF score is less than or equal to the normal threshold, the normalized value is 1 - (LDF score / normal threshold); if the LDF score is greater than the normal threshold, the normalized value is 0.5 - (LDF score - normal threshold) / (2 × normal threshold). Health score = normalized value × 10, CMG health score. The higher the score, the smaller the deviation of the sample from the normal state and the higher the health level.

[0127] As an example, the electrical system (rotor, high-speed motor current), thermal system (frame motor shell temperature), and momentum system (rotor speed) parameters of CMG are input into an isolated forest. The normal distribution characteristics of these parameters are obvious, making them suitable for unsupervised anomaly detection.

[0128] The anomaly probability value (0-1) of the CMG output by the isolated forest is closer to 1 for samples with shorter path lengths, reflecting the anomaly risk of the current state.

[0129] Health score = (1 - abnormal probability value) × 10. The higher the score, the closer the CMG's operating status is to the normal mode.

[0130] As an example, the core parameters of the CMG's electrical system (rotor motor current, high-speed motor current, frame motor current), thermal system (frame motor case temperature), and momentum system (rotor speed) are input into the MLP; The MLP outputs the health status probability of CMG (0-1). The closer the probability value is to 1, the higher the probability that CMG is in a healthy state. Health score = probability of being in good health × 10; A higher health score indicates a better overall health level for CMG.

[0131] (III) Overall Evaluation Results Based on the index values ​​under each evaluation dimension, the on-orbit comprehensive evaluation results of the control moment gyroscope are generated.

[0132] Secondly, embodiments of the present invention provide an on-orbit evaluation system for a control torque gyroscope, comprising: The data acquisition module is used to acquire on-board telemetry data of the control torque gyroscope during its on-orbit operation. The on-board telemetry data includes mechanical system parameters, electrical system parameters, environmental adaptation parameters, and mission performance parameters. The index calculation module is used to calculate the index values ​​of the control moment gyroscope under multiple preset evaluation dimensions based on on-board telemetry data. The multiple preset evaluation dimensions include reliability, on-orbit performance, stability and health status. The comprehensive evaluation module is used to generate on-orbit comprehensive evaluation results of the control moment gyroscope based on the index values ​​under each evaluation dimension.

[0133] Thirdly, embodiments of the present invention provide an on-orbit evaluation terminal for a control moment gyroscope, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the on-orbit evaluation method for a control moment gyroscope provided by the present invention.

[0134] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the description of the drawings, in carrying out the claimed invention. In this specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several of the functions listed in the specification. While certain measures are described in different embodiments, this does not mean that these measures cannot be combined to produce good results.

[0135] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the invention and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is also intended to include such modifications and modifications.

Claims

1. A method for on-orbit evaluation of a control torque gyroscope, characterized in that, Includes the following steps: Acquire on-orbit telemetry data of the control torque gyroscope. The on-orbit telemetry data includes mechanical system parameters, electrical system parameters, environmental adaptation parameters, and mission performance parameters. Among them, the mechanical system parameters include at least rotor speed, frame rotation angle, and frame vibration data; the electrical system parameters include at least rotor motor current, high-speed motor current, and frame motor current; the environmental adaptation parameters include at least frame motor shell temperature; and the mission performance parameters include at least frame coarse and fine differences, torque output value, and angular momentum exchange state parameters. Based on on-board telemetry data, the index values ​​of the control torque gyroscope are calculated under multiple preset evaluation dimensions; the multiple preset evaluation dimensions include reliability dimension, on-orbit performance dimension, stability dimension and health status dimension; Based on the index values ​​under each evaluation dimension, the on-orbit comprehensive evaluation results of the control moment gyroscope are generated.

2. The on-orbit evaluation method for control torque gyroscopes according to claim 1, characterized in that, The reliability metrics are obtained by combining the reliability of the integrated framework motor and the on-orbit reliability of a single unit: The reliability of the frame motor is calculated based on the logical comparison results of the real-time status of the frame motor shell temperature, frame motor current and frame coarse and fine difference with the corresponding preset thresholds and the normality rate. The reliability of single-unit on-orbit operation is obtained by first-order difference calculation, mutation detection, and entropy weighting summation based on time-series data of frame motor shell temperature, rotor motor current, high-speed motor current, and rotor speed.

3. The on-orbit evaluation method for control torque gyroscopes according to claim 2, characterized in that, The method for calculating the reliability of a frame motor is as follows: For each sampling point, the frame motor housing temperature, frame motor current, and frame coarse and fine differences are compared with their respective preset safety threshold ranges to generate a status indicator indicating whether the status is normal or exceeds the limit. The normality rate of each parameter is calculated based on the ratio of the number of sampling points with normal status within a preset time period to the total number of valid sampling points. The average of the frame motor housing temperature normality rate, frame motor current normality rate, and frame coarse-precision difference normality rate is determined as the frame motor reliability; and / or, The method for calculating the on-orbit reliability of a single unit is as follows: First-order difference calculations were performed on the time-series data of frame motor housing temperature, rotor motor current, high-speed motor current and rotor speed to obtain the first-order difference sequence of each parameter; The difference in the absolute value of the first-order difference sequence of each parameter that exceeds the preset mutation threshold is determined as a sudden mutation, and the number of sudden mutations of each parameter within the preset period is counted. The single-parameter reliability of each parameter is calculated based on the ratio of the number of sudden mutations of each parameter to the total number of data points. The entropy weight method is used to assign weights to each parameter based on the degree of dispersion of the first-order difference sequence of each parameter; The on-orbit reliability of a single unit is obtained by weighted summation based on the individual parameter reliability and weight of each parameter.

4. The on-orbit evaluation method for a control torque gyroscope according to claim 1, characterized in that, The on-orbit performance metrics are obtained by fusing the overall on-orbit performance score, performance change trend score, and deterioration trend score. The overall on-orbit performance score is obtained by multi-dimensional scoring and entropy weighting based on the mean, range, and standard deviation of rotor motor current, high-speed motor current, frame motor current, rotor speed, frame coarse and fine difference, and frame motor shell temperature, as well as the corresponding engineering calibration parameters. The performance change trend score is calculated based on the change trend of the output anomaly probability value by inputting the time series data of rotor speed, frame motor housing temperature and high-speed motor current into the Long Short-Term Memory network LSTM. The deterioration trend score is calculated based on the time-series data of rotor motor current, high-speed motor current, frame motor current and rotor speed input into the denoising autoencoder (DAE) and the time-varying rate of change of its output reconstruction error.

5. The on-orbit evaluation method for a control torque gyroscope according to claim 4, characterized in that, The method for calculating the overall on-orbit performance score is as follows: Calculate the mean, range, and standard deviation of each parameter within a preset period; For each parameter, a mean compliance score is determined based on the comparison between the mean and a preset rated range. The range compliance score is determined based on the comparison between the range and the preset maximum allowable range. The volatility stability score is determined based on the comparison between the standard deviation and the preset maximum allowable standard deviation. The mean compliance score, range compliance score, and fluctuation stability score are added together to obtain the single-parameter on-orbit performance score. The entropy weight method is used to determine the weight of each parameter, and the on-orbit performance scores of all individual parameters are weighted and summed to obtain the overall on-orbit performance score. And / or, The calculation method for the performance change trend score is as follows: Obtain the anomaly probability value of the LSTM network output at each time step; Perform first-order difference calculation on the sequence of abnormal probability values; The proportion of time steps with positive first-order differences is used as the proportion of improvement trend, and the proportion of time steps with negative first-order differences is used as the proportion of deterioration trend. Based on (Percentage of Improvement Trends - Percentage of Deterioration Trends) 10. Calculation performance trend score; And / or, The method for calculating the deteriorating trend score is as follows: Obtain the parameter reconstruction error output by the DAE network; Calculate the rate of change of the reconstruction error over time within the sliding window; Normalize the rate of change over time; Based on (1 - normalized rate of change over time) 10. Calculate the deterioration trend score.

6. The on-orbit evaluation method for a control torque gyroscope according to claim 1, characterized in that, The stability dimension index value is obtained by integrating the first stability score, the second stability score, the third stability score, and the fourth stability score; The first stability score is based on the time-series data of frame motor housing temperature, rotor motor current, high-speed motor current, frame motor current and rotor speed. The standard deviation and mean standard deviation of each parameter are calculated and determined in combination with the preset standard deviation stability threshold. The second stability score is determined by inputting time-series data of frame motor current and frame motor shell temperature into a decision tree model and performing statistical calculations based on the output state identification labels. The third stability score is determined by inputting time-series data of rotor speed and frame motor housing temperature into a support vector machine (SVM) model and performing statistical calculations based on the output state identification labels. The fourth stability score is based on the time-series data of frame motor housing temperature, rotor motor current, high-speed motor current and rotor speed. The mean value of the coefficient of variation (CV) of each parameter is calculated and determined in combination with the preset normal threshold value of the coefficient of variation.

7. The on-orbit evaluation method for a control torque gyroscope according to claim 6, characterized in that, The calculation process for the fourth stability score includes: For the time series data of each parameter, calculate the mean and standard deviation within a preset rolling window; Based on standard deviation / mean 100%, calculate the coefficient of variation (CV) for each window; The mean CV of each parameter is obtained by averaging the CV values ​​of all windows. Based on the comparison results between the mean CV and multiple preset threshold intervals, the stability score corresponding to each parameter is determined. The average stability score of each parameter is used as the fourth stability score.

8. The on-orbit evaluation method for a control torque gyroscope according to claim 1, characterized in that, The health status index values ​​are obtained by integrating the motor current health score and the multi-parameter comprehensive health score; The motor current health is calculated based on the deviation rates of the real-time current of the rotor motor, the real-time current of the high-speed motor, and the real-time current of the frame motor from their respective rated currents, combined with weights determined by the entropy weight method. Specifically, for each sampling point, the relative deviation rates of the currents of the rotor motor, the high-speed motor, and the frame motor are calculated; the average deviation rate of each motor is obtained by statistically analyzing the mean of the relative deviation rates of the currents of each motor within a preset period; the current weights of each motor are assigned based on the average deviation rates of each motor using the entropy weight method; and the motor current health is calculated and normalized according to Formula 1 - (rotor motor weight × rotor motor average deviation rate + high-speed motor weight × high-speed motor average deviation rate + frame motor weight × frame motor average deviation rate). The multi-parameter comprehensive health score is calculated by inputting the core operating parameters of the control moment gyroscope into a random forest model, an XGBoost model, a linear discriminant analysis (LDF) model, an isolated forest model, and a multilayer perceptron (MLP) model, respectively, based on the anomaly probability value or health status probability value output by each model. Specifically: the anomaly probability values ​​output by the random forest model, XGBoost model, and isolated forest model are obtained and converted into corresponding health scores according to the formula (1 - anomaly probability value) × 10; the LDF score output by the LDF model is obtained and normalized based on a pre-determined normal sample LDF score threshold, and converted into corresponding health scores; the health status probability value output by the MLP model is obtained and multiplied by 10 to obtain the corresponding health score; the minimum value among the health scores corresponding to the random forest model, XGBoost model, LDF model, isolated forest model, and MLP model is determined as the multi-parameter comprehensive health score.

9. An on-orbit evaluation system for a control torque gyroscope, characterized in that, include: The data acquisition module is used to acquire on-board telemetry data of the control torque gyroscope during its on-orbit operation. The on-board telemetry data includes mechanical system parameters, electrical system parameters, environmental adaptation parameters, and mission performance parameters. The index calculation module is used to calculate the index values ​​of the control torque gyroscope under multiple preset evaluation dimensions based on on-board telemetry data; the multiple preset evaluation dimensions include reliability dimension, on-orbit performance dimension, stability dimension and health status dimension; The comprehensive evaluation module is used to generate the on-orbit comprehensive evaluation results of the control moment gyroscope based on the index values ​​under each evaluation dimension.

10. An on-orbit evaluation terminal for a control torque gyroscope, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the on-orbit evaluation method for the control moment gyroscope as described in any one of claims 1 to 8.

Images