A Fault Detection Method for Spaceborne Turntable Based on Multi-Mode Threshold Optimization

Abstract

The present invention relates to a fault detection method for a spaceborne turntable based on multi-mode threshold optimization. The method includes: constructing a multi-mode threshold optimization framework based on evidence reasoning; determining the initial interval reference values of each key parameter under the multi-mode threshold optimization framework; performing multi-mode threshold iterative optimization based on interval reference value evidence reasoning according to the initial interval reference values of each key parameter to generate optimized thresholds; and performing online fault detection of the spaceborne turntable based on the optimized thresholds. The present invention innovatively uses the evidence reasoning algorithm to fuse multi-mode thresholds and monitoring data, and optimizes the multi-mode thresholds, so as to realize the on-orbit fault detection of the spaceborne turntable, and solves the problems such as low fault detection rate caused by unreasonable setting of multi-mode initial thresholds before orbit injection and large difference between the initial thresholds and the on-orbit operation states. The method of the present invention can fully fuse expert knowledge, improve the on-orbit fault detection ability of the spaceborne turntable, and has good engineering application value.

Description

Technical Field

[0001] The present invention relates to the field of measurement technology, and in particular to a method for detecting on-orbit faults of a spaceborne turntable based on multi-mode threshold optimization. Background Art

[0002] Currently, with the rapid development of the level of science and technology, countries have accelerated the pace of exploring outer space, and large spacecraft such as launch vehicles, space shuttles, manned spacecraft, and artificial satellites have emerged continuously. Among them, remote sensing and various space exploration satellites, as an important means for humans to explore space and the earth, account for an increasing proportion of the artificial satellites launched by various countries. As an important payload on the satellite, the spaceborne turntable plays a crucial role in the satellite's on-orbit detection and space communication tasks. However, during the on-orbit operation of the spaceborne turntable, due to the changes in the space environment and the degradation of the performance of the rotating components, the rotational resistance of the turntable often increases due to the increase in the friction of the turntable shaft system, and even the turntable gets stuck instantaneously in orbit, resulting in the instability of the satellite attitude, which not only affects the execution efficiency of the satellite's on-orbit tasks but also endangers the safety of the on-orbit satellite itself. Therefore, it is crucial to detect the on-orbit faults of the spaceborne turntable in a timely and rapid manner to ensure the quality of the on-orbit satellite tasks and the safe operation.

[0003] Currently, the on-orbit fault detection of each component of the satellite system mainly adopts the telemetry parameter judgment method. The spacecraft telemetry data is transmitted to the ground at a certain frame rate. The management personnel focus on monitoring the key parameters of each component (such as motor current, etc.), and set corresponding over-limit alarm thresholds in the monitoring system in combination with its on-orbit working mode. If a certain parameter is detected to exceed the threshold range, an alarm message is given to complete the fault detection. This fault detection method is simple. As long as the over-limit alarm threshold is set reasonably, the on-orbit anomalies of the components can be detected in a timely manner. Obviously, the setting of the alarm threshold is the premise for quickly and accurately detecting faults. However, in the actual operation process, there are still three problems with this method: (1) The initial alarm threshold is generally determined by the satellite research and development department according to the design indexes of the components. The set initial threshold range is relatively conservative, which is not conducive to discovering the trend slow-varying faults within the threshold range, and it is easy to have the problem of missed fault reports, which is likely to cause the further deterioration of the faults and develop into serious faults; (2) There is often a certain difference between the set threshold range and the actual on-orbit operation state of the components, so that after the performance of the components degrades, the adaptability of the initial threshold becomes poor. (3) When the values of some satellite parameters change due to reasons such as working mode switching and operating environment changes, the alarm threshold cannot be accurately set, and using a certain fixed threshold cannot cover all the changes in working conditions, which is likely to cause a large false alarm rate or missed alarm rate of faults. Therefore, it is necessary to optimize the initial threshold according to the actual on-orbit working mode and operating state of the components under the initial threshold conditions to obtain a more reasonable fault alarm threshold, which can not only meet the detection requirements of slow-varying faults under different working modes but also adapt to the influence of the state changes brought about by the long-term operation of the components.

[0004] At present, the optimization methods for fault alarm thresholds mainly include model-based, knowledge-based, and data-based methods. Among them, model-based methods require an accurate understanding of the system operation mechanism model. For complex systems such as on-orbit satellite turntables, it is difficult to accurately establish the mechanism model of the system. Data-driven threshold optimization methods mainly achieve threshold updates through the best utilization of historical data, but they require an accurate understanding of the prior distribution of system parameters, and most of these methods belong to the "black box" model, showing deficiencies in model interpretability. In addition, most of the previous threshold optimization methods generally optimize from the perspective of the single mode of the system and do not consider the multi-mode working characteristics of the system. Therefore, their scope of application is limited. When the above existing threshold optimization methods are applied to satellite turntable fault detection, due to the unreasonable setting of multi-mode initial thresholds before entering orbit, problems such as low fault detection rate caused by the large difference between the initial threshold and the on-orbit operation state often occur.

[0005] In contrast, the multi-source information fusion method that combines expert knowledge and historical data can effectively utilize domain expert knowledge and combine historical observation data to establish a relatively accurate system model, which has been widely used in multi-attribute decision-making and achieved good results. As a typical MADM method, the Evidential Reasoning (ER) method combines multiple independent evidences to achieve multi-source information fusion and can effectively handle various uncertainties. To further enhance the modeling ability of the ER method for interval uncertainty, Wang Yingming et al. extended the ER method to the ER approach with interval belief degrees (ER-IBD). The parameter threshold, as an effective means to monitor the on-orbit operation state of satellites, is a typical domain expert knowledge and also has a certain degree of interval uncertainty. The ER-IBD method can be used to handle this type of interval uncertainty. Therefore, it is an urgent problem to propose a multi-mode threshold optimization method that can not only effectively fuse expert knowledge and historical observation data but also take into account different working modes of the system. Summary of the Invention

[0006] The purpose of the present invention is to provide a satellite turntable fault detection method based on multi-mode threshold optimization to achieve threshold optimization in multiple modes of the satellite turntable and improve the on-orbit fault detection ability of large rotating components of the satellite under multi-mode conditions.

[0007] To achieve the above purpose, the present invention provides the following solutions:

[0008] A satellite turntable fault detection method based on multi-mode threshold optimization, including:

[0009] Construct a multi-mode threshold optimization framework based on evidence reasoning;

[0010] Determine the initial interval reference values of each key parameter under the multi-mode threshold optimization framework;

[0011] Perform multi-mode threshold iterative optimization based on interval reference value evidence reasoning according to the initial interval reference values of each key parameter to generate optimized thresholds;

[0012] Perform on-orbit fault detection of the spaceborne turntable based on the optimized thresholds.

[0013] Optionally, the construction of the multi-mode threshold optimization framework based on evidence reasoning specifically includes:

[0014] Obtain the on-orbit steady-state working modes of the spaceborne turntable; the steady-state working modes include low-speed steady-state operation mode and high-speed steady-state operation mode;

[0015] Extract the monitoring thresholds of the key parameters of the spaceborne turntable in each of the steady-state working modes; the key parameters include turntable motor current, power supply monitoring voltage, and turntable speed;

[0016] Take the steady-state working modes as the evaluation levels of evidence reasoning, and use the monitoring thresholds in different steady-state working modes as the interval reference values for each evaluation level to construct a multi-mode threshold optimization framework for each key parameter based on interval reference value evidence reasoning.

[0017] Optionally, the determination of the initial interval reference values of each key parameter under the multi-mode threshold optimization framework specifically includes:

[0018] Determine the initial working threshold interval values of each key parameter in different steady-state working modes according to the design indicators of the spaceborne turntable components and the ground test data;

[0019] Under the constructed multi-mode threshold optimization framework, assign the initial working threshold interval values of each key parameter to the interval reference values of each key parameter at each evaluation level.

[0020] Optionally, the performance of multi-mode threshold iterative optimization based on interval reference value evidence reasoning according to the initial interval reference values of each key parameter to generate optimized thresholds specifically includes:

[0021] Perform multi-mode threshold iterative optimization based on interval reference value evidence reasoning in the case of intersection of interval reference values of adjacent two evaluation levels to generate optimized thresholds; or

[0022] Perform multi-mode threshold iterative optimization based on interval reference value evidence reasoning in the case of independence of interval reference values of adjacent two evaluation levels to generate optimized thresholds.

[0023] Optionally, when the reference values of adjacent two evaluation level intervals cross, perform multi-mode threshold iterative optimization based on interval reference value evidence reasoning to generate an optimized threshold, specifically including:

[0024] Obtain a training data set and a test data set; the training data set includes normal historical observation data of the key parameters in the steady-state operating mode; the test data set includes historical observation data of the key parameters at two evaluation levels; the historical observation data includes normal data and abnormal data;

[0025] Use a utility-based information conversion method to calculate the interval confidence degrees of adjacent two training samples in the training data set at different evaluation levels;

[0026] According to the interval confidence degrees of the adjacent two training samples at different evaluation levels, calculate the interval probability masses of the adjacent two training samples at different evaluation levels; the interval probability masses and the interval reference values together form the parameter to be optimized;

[0027] Use a projection covariance adaptive evolution algorithm to optimize the parameter to be optimized to generate an optimized parameter;

[0028] Perform false alarm rate and miss rate tests on the test samples in the test data set according to the optimized parameter, and determine whether the optimization cut-off condition is satisfied;

[0029] If the optimization cut-off condition is satisfied, use the optimized parameter as the optimized threshold;

[0030] If the optimization cut-off condition is not satisfied, update the interval reference value and the iteration number, and return to the step of using the utility-based information conversion method to calculate the interval confidence degrees of adjacent two training samples in the training data set at different evaluation levels.

[0031] Optionally, when the reference values of adjacent two evaluation level intervals are independent of each other, perform multi-mode threshold iterative optimization based on interval reference value evidence reasoning to generate an optimized threshold, specifically including:

[0032] Obtain a training data set and a test data set; the training data set includes normal historical observation data of the key parameters in the steady-state operating mode; the test data set includes historical observation data of the key parameters in the steady-state operating mode; the historical observation data includes normal data and abnormal data;

[0033] Use a utility-based information conversion method to calculate the interval confidence degrees of adjacent two training samples in the training data set at different evaluation levels;

[0034] Calculate the interval probability mass of two adjacent training samples at different evaluation levels according to the interval confidence levels of the two adjacent training samples at different evaluation levels; the interval probability mass and the interval reference value together form the parameter to be optimized;

[0035] Optimize the parameter to be optimized by using the projection covariance adaptive evolution algorithm to generate the optimized parameter;

[0036] Perform false alarm rate and missed alarm rate tests on the test samples in the test dataset according to the optimized parameter, and determine whether the optimization cut-off condition is satisfied;

[0037] If the optimization cut-off condition is satisfied, use the optimized parameter as the optimization threshold;

[0038] If the optimization cut-off condition is not satisfied, update the interval reference value and the iteration number, and return to the step of calculating the interval confidence levels of two adjacent training samples in the training dataset at different evaluation levels by using the utility-based information conversion method.

[0039] Optionally, the on-orbit fault detection of the spaceborne turntable based on the optimization threshold specifically includes:

[0040] Perform online fault detection based on a single-variable threshold on multiple key parameters of the spaceborne turntable in different steady-state working modes according to the optimization threshold to generate a single-variable threshold detection result;

[0041] Perform fault alarm according to the single-variable threshold detection result.

[0042] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention:

[0043] The present invention provides a spaceborne turntable fault detection method based on multi-mode threshold optimization. The method includes: constructing a multi-mode threshold optimization framework based on evidence reasoning; determining the initial interval reference value of each key parameter under the multi-mode threshold optimization framework; performing multi-mode threshold iterative optimization based on interval reference value evidence reasoning according to the initial interval reference value of each key parameter to generate an optimization threshold; performing on-orbit fault detection of the spaceborne turntable based on the optimization threshold. The present invention innovatively uses the evidence reasoning algorithm to fuse multi-mode thresholds and monitoring data, and optimizes the multi-mode thresholds to achieve on-orbit fault detection of the spaceborne turntable, and solves problems such as low fault detection rate caused by unreasonable setting of multi-mode initial thresholds before orbit injection and large differences between the initial thresholds and the on-orbit operating states. The method of the present invention can fully fuse expert knowledge, improve the on-orbit fault detection ability of the spaceborne turntable, and has good engineering application value. Description of the Drawings

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0045] Figure 1 It is a flowchart of a satellite-borne turntable fault detection method based on multi-mode threshold optimization provided by the present invention;

[0046] Figure 2 It is a schematic diagram of the reference value crossover of adjacent evaluation level intervals provided by the present invention (Case 1);

[0047] Figure 3 It is a schematic diagram of the independence of the reference values of adjacent evaluation level intervals provided by the present invention (Case 2);

[0048] Figure 4 It is a multi-mode threshold optimization flowchart provided by the present invention (Case 1);

[0049] Figure 5 It is a single-mode threshold optimization flowchart provided by the present invention (Case 2);

[0050] Figure 6 It is a schematic diagram of the optimization result of the turntable angular velocity threshold in the high-speed steady state mode provided by the embodiment of the present invention;

[0051] Figure 7 It is a schematic diagram of the optimization result of the turntable motor current threshold in the high-speed steady state mode provided by the embodiment of the present invention;

[0052] Figure 8 It is a schematic diagram of the optimization result of the turntable power supply monitoring voltage threshold in the high-speed steady state mode provided by the embodiment of the present invention;

[0053] Figure 9 It is a schematic diagram of the fault detection result when the turntable motor current increases abnormally provided by the embodiment of the present invention. Detailed implementation manners

[0054] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0055] The object of the present invention is to provide a spaceborne turntable fault detection method based on multi-mode threshold optimization, so as to realize the threshold optimization of the spaceborne turntable under multiple modes and improve the on-orbit fault detection ability of large rotating components of the spaceborne under multi-mode conditions.

[0056] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] Figure 1 The flowchart of a spaceborne turntable fault detection method based on multi-mode threshold optimization provided by an embodiment of the present invention is as follows. Figure 1 As shown, a spaceborne turntable fault detection method based on multi-mode threshold optimization of the present invention includes:

[0058] Step 101: Construct a multi-mode threshold optimization framework based on evidential reasoning.

[0059] The method of the present invention first constructs a multi-mode threshold optimization framework based on evidential reasoning with interval reference values. The process of constructing the multi-mode threshold optimization framework based on evidential reasoning with interval reference values is as follows:

[0060] Process 1.1: Analyze and determine the main on-orbit working modes of the spaceborne turntable, such as low-speed steady-state operation mode, high-speed steady-state operation mode, etc.;

[0061] Process 1.2: Extract the working thresholds of parameters such as the current of the turntable motor, the monitored power supply voltage, and the turntable speed in each working mode;

[0062] Process 1.3: Take the working mode of the turntable as the evaluation level of evidential reasoning, and use the working thresholds in different modes as the interval reference values for each evaluation level to construct a corresponding evidential reasoning with interval reference values (ER-IRV) framework.

[0063] Therefore, the specific steps of constructing the multi-mode threshold optimization framework based on evidential reasoning in step 101 include:

[0064] Step 1.1: Obtain the on-orbit steady-state working modes of the spaceborne turntable; the steady-state working modes include low-speed steady-state operation mode and high-speed steady-state operation mode.

[0065] Analyze and determine the main on-orbit steady-state working modes (referred to as working modes for short) of the spaceborne turntable, such as low-speed steady-state operation mode, high-speed steady-state operation mode, etc. For example: a multi-mode system identifies N steady-state working modes, which are respectively defined as H1, H2,..., H N .

[0066] Step 1.2: Extract the monitoring thresholds of the key parameters of the spaceborne turntable in each of the steady-state operating modes; the key parameters include the turntable motor current, the power supply monitoring voltage, and the turntable rotation speed.

[0067] Extract the monitoring thresholds (also known as operating thresholds) of the key parameters of the spaceborne turntable (such as parameters like the turntable motor current, the power supply monitoring voltage, and the turntable rotation speed, etc.) in each operating mode. For example: A total of L key variables (i.e., key parameters) are extracted, denoted as x1, x2, …, x L . For the i-th variable, its monitoring threshold in the n-th steady-state operating mode is denoted as

[0068] Step 1.3: Take the steady-state operating mode as the evaluation level for evidence reasoning, and use the monitoring thresholds in different steady-state operating modes as the interval reference values for each evaluation level to construct a multi-mode threshold optimization framework for each key parameter based on evidence reasoning with interval reference values.

[0069] Take the turntable operating mode as the evaluation level for evidence reasoning, and use the monitoring thresholds in different modes as the interval reference values for each evaluation level to construct a multi-mode recognition framework for each parameter based on evidence reasoning with interval reference values (ER-IRV). Under this framework, each steady-state mode of the system serves as each evaluation level in the framework, and all steady-state modes constitute the identification framework of ER-IRV (i.e., the multi-mode threshold optimization framework), denoted as Θ = {H1, H2, …, H N}}. In addition, the reference value for each evaluation level in the identification framework is represented as an interval reference value. For example: For the evaluation level H n , the upper bound of its interval reference value is the upper bound of the monitoring threshold of the i-th variable in the n-th steady-state operating mode and its lower bound is the lower bound of the monitoring threshold of the i-th variable in the n-th steady-state operating mode

[0070] Step 102: Determine the initial interval reference values of each key parameter under the multi-mode threshold optimization framework.

[0071] Step 102 determines the initial interval reference values of each key parameter. The process of determining the initial interval reference values includes:

[0072] Process 2.1: Given the initial operating threshold interval values of the spaceborne turntable components in different operating modes by experts according to the design specifications and ground test data of the spaceborne turntable components

[0073] Process 2.2: Under the constructed interval reference value evidential reasoning (ER-IRV) framework, assign the initial threshold to the interval reference values of each evaluation level. Among them, assign to the lower bound of the interval reference value of the i-th parameter at the n-th evaluation level H n , and assign to the upper bound of the interval reference value of the i-th parameter at the n-th evaluation level H n .

[0074] Therefore, the specific steps for step 102 to determine the initial interval reference values of each key parameter under the multi-mode threshold optimization framework include:

[0075] Step 2.1: According to the design indicators of the spaceborne turntable component and the ground test data, determine the initial working threshold interval values of each key parameter under different steady-state working modes;

[0076] Given by experts according to the design indicators of the spaceborne turntable component and the ground test data, the initial monitoring thresholds of the selected key variables under different steady-state working modes

[0077] Step 2.2: Under the constructed multi-mode threshold optimization framework, assign the initial working threshold interval values of each key parameter to the interval reference values of each key parameter at each evaluation level.

[0078] Under the constructed interval reference value evidential reasoning (ER-IRV) framework, assign the initial threshold to the interval reference values of each evaluation level. Among them, assign to the lower bound of the interval reference value of the i-th variable at the n-th evaluation level H n , and assign to the upper bound of the interval reference value of the i-th variable at the n-th evaluation level H n .

[0079] Step 103: Perform multi-mode threshold iterative optimization based on interval reference value evidential reasoning according to the initial interval reference values of each key parameter to generate optimized thresholds.

[0080] The step 103 performs multi-mode threshold iterative optimization based on interval reference value evidential reasoning, and the multi-mode threshold iterative optimization of interval reference value evidential reasoning is mainly divided into two cases:

[0081] Case 1: The interval reference values of two adjacent evaluation levels (e.g., evaluation levels n and n + 1) cross, and the thresholds under the two evaluation levels can be optimized. The threshold optimization process is as follows:

[0082] Process 3.1: Acquisition of training dataset and test dataset. Import the historical observation data of a certain variable in mode n as the training dataset; in addition, import the historical observation data (including normal data and abnormal data) of this variable in modes n and n+1 as the corresponding test datasets in the two modes respectively.

[0083] Process 3.2: Conversion of training data to interval confidence. According to the obtained training dataset, determine the positional relationship between the training data x i (t j ) and the interval reference values of the i-th parameter, and locate the reference value interval into which the training sample falls; convert the training sample into the corresponding interval confidence in different modes according to the utility-based information conversion technology and form the corresponding interval confidence structure.

[0084] Step Process 3.3: Solving of interval probability mass. In each round of iteration, extract two adjacent training data sample points of the system in mode n in chronological order, and obtain the corresponding interval probability mass according to the interval confidence of each training sample in different modes and where n = 1,..., N.

[0085] Process 3.4: Evidence fusion and threshold optimization. Construct a parameter optimization model based on ER-IRV, and the objective function of the parameter optimization model is The constraint conditions are

[0086]

[0087] and where e j represents the evidence formed by two training samples, j = 1, 2; k represents the number of iterations. The parameters to be optimized are:

[0088]

[0089] The optimization method uses the projection covariance adaptive evolution algorithm (P-CMA-ES) to optimize the parameters, and obtains the optimized parameters in the k-th round, including and

[0090] Process 3.5: Judgment of optimization termination condition. Considering the optimal threshold characteristics of the Receiver operating characteristic (ROC) curve, according to the reference values obtained after each round of optimization and Test the false alarm rate (FAR) and missed alarm rate (MAR) of the test data for the \(i\)-th parameter in mode \(n\) and mode \(n + 1\). The defined optimization cut-off condition is: \(|FAR n,i - MAR n,i | \leq \varepsilon\) and \(|FAR n+1,i - MAR n+1,i | \leq \varepsilon\). If the cut-off condition is met, the corresponding interval reference value is taken as the optimal threshold for the \(i\)-th parameter in mode \(n\) and mode \(n + 1\), that is and The optimization ends. Otherwise, enter process 3.6;

[0091] Process 3.6: Update the interval reference value. Update the interval reference value optimized in this round at a certain update speed. Subsequently, update the iteration count such that \(k = k + 1\), import the next two training sample points in the training data, return to process 3.2, and repeat processes 3.2 to 3.6 until the optimization cut-off condition is met.

[0092] Case 2: The interval reference values of two adjacent evaluation levels are independent. In this case, it is only used to optimize the threshold for one evaluation level. The threshold optimization process is as follows:

[0093] Process 3.1): Obtain the training data set and test data set. Import the historical observation data of a certain variable in mode \(n\) as the training data set; in addition, import the historical observation data (including normal data and abnormal data) of this variable in mode \(n\) as the corresponding test data set for this mode.

[0094] Process 3.2) Convert the training data to interval confidence. Based on the obtained training data set, determine the positional relationship between the training data \(x i (t j ) and each interval reference value of the \(i\)-th parameter, and locate the reference value interval into which the training sample falls; convert the training sample to the corresponding interval confidence in different modes according to the utility-based information conversion technology and form the corresponding interval confidence structure.

[0095] Process 3.3): Solve the interval probability mass. In each round of iteration, extract two adjacent training data sample points of the system in mode \(n\) in sequence according to the time series. According to the interval confidence of each training sample in different modes, calculate the corresponding interval probability mass and where \(n = 1, \ldots, N\).

[0096] Process 3.4): Evidence fusion and threshold optimization. Construct a parameter optimization model based on ER-IRV. The objective function of the parameter optimization model is The constraint conditions are where j = 1, 2, representing the evidences formed by two training samples; k represents the number of iterations. The parameters to be optimized are:

[0097]

[0098] The optimization method uses the projected covariance adaptive evolution algorithm (P-CMA-ES) to optimize the parameters and obtain the optimized parameters in the k-th round, including and

[0099] Process 3.5): Optimization termination condition judgment. Considering the optimal threshold characteristics of the Receiver operating characteristic (ROC) curve, according to the reference values and obtained after each round of optimization, test the false alarm rate (FAR) and missed alarm rate (MAR) of the test data of the i-th parameter in mode n. The defined optimization termination condition is: |FAR n,i - MAR n,i | ≤ ε. If the termination condition is met, the corresponding interval reference value is used as the optimal threshold of the i-th parameter in mode n, that is The optimization ends. Otherwise, go to Process 3.6);

[0100] Process 3.6): Update the interval reference value. Update the interval reference value after this round of optimization at a certain update speed. Then update the number of iterations so that k = k + 1, import the next two training sample points in the training data, return to Process 3.2), and repeat Processes 3.2) to 3.6) until the optimization termination condition is met.

[0101] Therefore, the step 103 performs multi-mode threshold iterative optimization based on interval reference value evidence reasoning according to the initial interval reference values of each key parameter, and generates the optimized threshold, specifically including:

[0102] Step 3.1: Perform multi-mode threshold iterative optimization based on interval reference value evidence reasoning in the case of intersection of adjacent two evaluation grade interval reference values to generate the optimized threshold.

[0103] Step 103 is implemented by distinguishing two cases: Case 1, where the interval reference values of two adjacent evaluation levels (e.g., evaluation levels n and n+1) overlap, as shown in Appendix Figure 2 ; Case 2, where the interval reference values of two adjacent evaluation levels (e.g., evaluation levels n and n+1) are independent, as shown in Appendix Figure 3 .

[0104] Figure 4 This is the multi-mode threshold optimization flowchart (Case 1) provided by the present invention. Refer to Figure 4 , and step 3.1 specifically includes:

[0105] Step 3.1.1: Obtain a training data set and a test data set; the training data set includes normal historical observation data of the key parameters in the steady-state operating mode; the test data set includes historical observation data of the key parameters at two evaluation levels; the historical observation data includes normal data and abnormal data.

[0106] Import the training data and the test data. Collect and obtain the normal historical observation data of the i-th variable in the steady-state operating mode n as the training data set, denoted as {x n,i (t-T n,i ), x n,i (t-T n,i +h), …, x n,i (t)}, where h is the sampling interval of the training data, t is the sampling moment, and T n,i is the total sampling duration of the training data set of the i-th variable in the steady-state operating mode n. In addition, obtain the historical observation data (including normal data and abnormal data) of the i-th variable in the steady-state operating modes n and n+1 as two sets of test data sets, denoted as {x n,i (t1-T), x n,i (t1-T+1), …, x n,i (t1)} and {x n+1,i (t2-T), x n+1,i (t2-T+1), …, x n+1,i (t2)}. Where T is the total sampling duration of the test data set of the i-th variable in the steady-state operating modes n and n+1.

[0107] Step 3.1.2: Calculate the interval confidence levels of two adjacent training samples in the training data set at different evaluation levels by using a utility-based information conversion method.

[0108] This step is to obtain the interval confidence levels. In the k-th iteration (k = 1 in the first iteration), select two adjacent data points (training samples) in the training data set, that is, x n,i (t-T n,i+(k - 1)·h) and x n,i (t - T n,i + k·h). Assign the above two training data points to x n,i (e1) and x n,i (e2) respectively, and determine the positional relationship between the two training data and the reference values of each interval of the i-th parameter. Convert the training samples into the corresponding interval confidence degrees in different modes according to the utility-based information conversion technology and form the corresponding interval confidence structure. The corresponding conversion method is as follows: As shown in the appendix Figure 2 The training data x n , i (e j ) may fall into one of the intervals ①, ②, or ③. Based on this, three states can be distinguished for conversion

[0109] State 1: When the training data point falls into interval ①, that is According to the utility-based information conversion method, the training data is evaluated as grade H n and H n+1 The calculation of the interval confidence degrees is as follows

[0110] and

[0111] and

[0112] State 2: When the training data point falls into interval ②, that is The training data is evaluated as grade H n and H n+1 The calculation of the interval confidence degrees is as follows

[0113] and

[0114] and

[0115] State 3: When the training data point falls into interval ③, that is The training data is evaluated as grade H n and H n+1 The calculation of the interval confidence degrees is as follows

[0116] and

[0117] and

[0118] For an incomplete assessment, there is global ignorance, and the interval confidence assigned to the entire set \(H\) of the identification framework can be obtained by the following formula:

[0119]

[0120]

[0121] From formulas (1)-(8), the interval confidence structure after the conversion of each training data point is as follows:

[0122]

[0123] Step 3.1.3: Calculate the interval probability mass of two adjacent training samples at different evaluation levels according to the interval confidence of the two adjacent training samples at different evaluation levels; the interval probability mass and the interval reference value together constitute the parameter to be optimized.

[0124] This step is the calculation of the interval probability mass. According to the interval confidence of each training sample in different modes obtained, the corresponding interval probability mass is obtained according to the following formula and where \(n = 1,\ldots,N\).

[0125]

[0126]

[0127]

[0128] where, \(w\) j is the relative weight of the evidence \(e\) j . Since the two training data samples are equally important relative to each other, therefore, \(w_1 = w_2 = 0.5\) is set. In addition, \(m\) n,i \((e\) j ) is the probability mass assigned by the evidence \(e\) j to the evaluation level \(H\) n . and are the upper and lower bound values corresponding to the probability mass \(m\) n,i \((e\) j ). and are the two parts of the probability mass assigned by the evidence \(e\) j to the entire set \(H\) of the identification framework. and are the upper and lower bound values of the probability mass respectively. The sum of all probability masses satisfies the following constraint conditions:

[0129]

[0130] So far, all the interval probability masses and interval reference values are used as the parameters to be optimized, and a corresponding parameter vector P is formed. k , which is expressed as follows:

[0131]

[0132] Step 3.1.4: Use the projected covariance matrix adaptation evolution algorithm to optimize the parameters to be optimized and generate the optimized parameters.

[0133] This step is for evidence fusion and threshold optimization. A parameter optimization model based on ER-IRV is constructed. The optimization method uses the projected covariance matrix adaptation evolution algorithm (P-CMA-ES) to optimize the parameter P k for optimization. The number of update generations in the optimization algorithm is set to 100 generations. Thus, the optimized parameters in the k-th round can be obtained, including and The parameter optimization model is as follows:

[0134]

[0135]

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142]

[0143]

[0144] Step 3.1.5: According to the optimized parameters, perform false alarm rate and missed alarm rate tests on the test samples in the test dataset to determine whether the optimization cut-off condition is satisfied.

[0145] This step is for judging the optimization cut-off condition. According to the reference values obtained after each round of optimization (i.e., the optimized parameters) and for the test data of the i-th parameter in mode n and mode n + 1 (i.e., {x n,i (t1 - T), xn,i (t1 - T + 1), …, x n,i}(t1)} and {x n+1,i (t2 - T), x n+1,i (t2 - T + 1), …, x n+1,i (t2)}) are tested for false alarm rate (FAR) and missed alarm rate (MAR). The defined optimization cut-off condition is: |FAR n,i - MAR n,i | ≤ ε and |FAR n+1,i - MAR n+1,i | ≤ ε. Where ε is a very small constant value satisfying 0 ≤ ε << 1, and |·| is the absolute value operation. FAR n,i and MAR n,i respectively represent the false alarm rate and missed alarm rate of the i-th key parameter in the steady-state operating mode n; FAR n+1,i and MAR n+1,i respectively represent the false alarm rate and missed alarm rate of the i-th key parameter in the steady-state operating mode n + 1.

[0146] Step 3.1.6: If the optimization cut-off condition is satisfied, take the optimized parameter as the optimization threshold.

[0147] If the cut-off condition is satisfied, take the corresponding interval reference value as the optimal threshold (i.e., the optimization threshold) of the i-th parameter in modes n and n + 1, and the optimization ends.

[0148] The optimization threshold is expressed by formula (25):

[0149]

[0150] Step 3.1.7: If the optimization cut-off condition is not satisfied, update the interval reference value and the iteration count, and return to the step of calculating the interval confidence of adjacent two training samples in the training dataset at different evaluation levels using the utility-based information conversion method.

[0151] This step updates the interval reference value. Update the interval reference value optimized in this round at a certain update speed. Specifically, update the interval reference value optimized in this round according to the following formula:

[0152]

[0153]

[0154]

[0155]

[0156] Among them, c1 to c4 are called update speed factors and satisfy 0 < c l < 1; l = 1, …, 4. The update speed factor can be considered to be adjusted to control the threshold update speed. r1 to r4 are 4 unit random numbers, and their values are between 0 and 1.

[0157] Subsequently, update the iteration number so that k = k + 1, import the next two training sample points in the training data, return to step 3.1.2, and repeat steps 3.1.2 to 3.1.7 until the optimization cut-off condition is met.

[0158] Step 3.2: Perform multi-mode threshold iterative optimization based on interval reference value evidence reasoning under the condition that the reference values of adjacent two evaluation grade intervals are independent, and generate optimized thresholds.

[0159] Figure 5 This is the multi-mode threshold optimization flowchart (case two) provided by the present invention. Refer to Figure 5 and the specific steps of step 3.2 include:

[0160] Step 3.2.1: Obtain a training data set and a test data set; the training data set includes normal historical observation data of the key parameters in the steady-state working mode; the test data set includes historical observation data of the key parameters in the steady-state working mode; the historical observation data includes normal data and abnormal data.

[0161] Import the training data and the test data. Collect and obtain the normal historical observation data of the i-th variable in the steady-state working mode n as the training data set, denoted as {x n,i (t - T n,i ), x n,i (t - T n,i + h), …, x n,i (t)}, where h is the sampling interval of the training data. Obtain the historical observation data (including normal data and abnormal data) of the i-th variable in the steady-state working mode n as the test data set, denoted as {x n,i (t1 - T), x n,i (t1 - T + 1), …, x n,i (t1)}.

[0162] Step 3.2.2: Calculate the interval confidence degrees of adjacent two training samples in the training data set at different evaluation grades by using a utility-based information conversion method.

[0163] This step is for obtaining the interval confidence degrees. In the k-th iteration (k = 1 in the first iteration), select two adjacent data points (training samples) in the training data set, that is, x n,i(t - T n,i +(k - 1)·h) and x n,i (t - T n,i +k·h). Assign the above two training data points to x n,i (e1) and x n,i (e2) respectively, and judge the positional relationship between the two training data and the reference values of each interval of the i-th parameter. Convert the training samples into the corresponding interval confidence degrees in different modes according to the utility-based information conversion technology and form the corresponding interval confidence structure. The corresponding conversion method is as follows: As shown in the appendix Figure 3 , the training data x n,i (e j ); j = 1, 2 may fall into one of the intervals ④, ⑤ or ⑥. Based on this, three states (State) can be distinguished for conversion.

[0164] State 1: When the training data point falls into interval ④, that is it means that the training data completely belongs to the evaluation level H n , that is, the confidence degree for level H n is 1, and the confidence degrees for other evaluation levels are all 0. Then the corresponding interval confidence structure is as follows:

[0165] M(x i (e j )) = {(H1, [0, 0]), …, (H n , [1, 1]), …, (H N , [0, 0]), (H, [0, 0]); n = 1, …, N}(30)

[0166] State 2: When the training data point falls into interval ⑤, that is the training data does not belong to the evaluation levels H n and H n+1 . The corresponding interval confidence structure is as follows:

[0167] M(x i (e j )) = {(H1, [0, 0]), …, (H n , [0, 0]), (H n+1 , [0, 0]), …, (H N , [0, 0]), (H, [1, 1]); n = 1, …, N} (31)

[0168] State 3: When the training data point falls into interval ⑥, that is the training data completely belongs to the evaluation level H n+1 , that is, the confidence degree for level H n+1If the confidence level for one is 1 and the confidence levels for other evaluation levels are all 0, the obtained interval confidence structure is as follows:

[0169] M(x i (e j ))={(H1,[0,0]),…,(H n+1 ,[1,1]),…,(H N ,[0,0]),(H,[0,0]);n=1,…,N} (32)

[0170] For an incomplete evaluation, there is global ignorance, and the interval confidence level assigned to the entire set H of the identification framework and can be obtained by the following formula:

[0171]

[0172]

[0173] From formulas (30)-(34), the interval confidence structure after the conversion of each training data point is as follows:

[0174]

[0175] Step 3.2.3: Calculate the interval probability mass of two adjacent training samples at different evaluation levels according to the interval confidence levels of the two adjacent training samples at different evaluation levels; the interval probability mass and the interval reference value together constitute the parameters to be optimized.

[0176] This step is for calculating the interval probability mass. According to the interval confidence levels of each training sample obtained in different modes, the corresponding interval probability mass is obtained according to the following formula and where n = 1,…,N.

[0177]

[0178]

[0179]

[0180] Among them, w j is the relative weight of the evidence e j . Since the two training data samples are equally important relative to each other, therefore, w1 = w2 = 0.5 is set. In addition, m n,i (e j ) is the probability mass assigned by the evidence e j to the evaluation level H n , and is the probability mass m n,i (e j ) corresponding upper and lower bound values. and are the two parts of the probability mass assigned by the evidence e j to the entire set H of the identification framework. and are respectively the upper and lower bound values of the probability mass . The sum of all probability masses satisfies the following constraint:

[0181]

[0182] So far, taking all the interval probability masses and interval reference values as the parameters to be optimized, a corresponding parameter vector P k is formed, which is expressed as follows:

[0183]

[0184] Step 3.2.4: Use the projected covariance adaptive evolutionary algorithm to optimize the parameters to be optimized, and generate the optimized parameters.

[0185] This step is for evidence fusion and threshold optimization. A parameter optimization model based on ER-IRV is constructed. The optimization method uses the projected covariance matrix adaptive evolutionary algorithm (P-CMA-ES) to optimize the parameters to be optimized P k . The number of update generations in the optimization algorithm is set to 100 generations. Thus, the optimized parameters in the k-th round can be obtained, including and The described parameter optimization model is as follows:

[0186]

[0187]

[0188]

[0189]

[0190]

[0191]

[0192] and

[0193]

[0194] Step 3.2.5: Perform false alarm rate and missed alarm rate tests on the test samples in the test dataset according to the optimized parameters, and determine whether the optimization cut-off condition is satisfied.

[0195] This step is for judging the optimization cut-off condition. According to the reference values obtained after each round of optimization (i.e., the optimized parameters) and perform false alarm rate (FAR) and missed alarm rate (MAR) tests on the test data of the i-th variable in mode n (i.e., {x n,i (t1 - T), x n,i (t1 - T + 1), …, x n,i (t1)}). The defined optimization cut-off condition is: |FAR n,i - MAR n,i | ≤ ε. Wherein, the definition of ε is as in Case 1.

[0196] Step 3.2.6: If the optimization cut-off condition is satisfied, take the optimized parameters as the optimization threshold.

[0197] If the cut-off condition is satisfied, take the corresponding interval reference value as the optimal threshold of the i-th parameter in mode n (i.e., the optimization threshold), and the optimization ends.

[0198] The optimization threshold is expressed as formula (50):

[0199]

[0200] Step 3.2.7: If the optimization cut-off condition is not satisfied, update the interval reference value and the iteration number, and return to the step of calculating the interval confidence degrees of two adjacent training samples in the training dataset at different evaluation levels by using the utility-based information conversion method.

[0201] This step updates the interval reference value. Update the interval reference value optimized in this round at a certain update speed. Specifically, update the interval reference value optimized in this round according to the following formula:

[0202]

[0203]

[0204] Among them, c1 to c2 are called update speed factors and satisfy 0 < c l < 1; l = 1, 2. The update speed factors can be considered to be adjusted to control the update speed. r1 to r2 are two unit random numbers, and their values are between 0 and 1.

[0205] Subsequently, update the iteration count such that k = k + 1, import the next two training sample points in the training data, return to step 3.2.2, and repeat steps 3.2.2 to 3.2.7 until the optimization cutoff condition is met.

[0206] By sequentially optimizing the initial thresholds in various steady-state modes for L different variables x1, x2, …, x L the optimal thresholds of the corresponding variables in different steady-state modes can be obtained, that is

[0207] Step 104: Perform on-orbit turntable online fault detection based on the optimized thresholds.

[0208] This step is the online fault detection based on the optimized thresholds. The fault detection process based on the optimized thresholds is as follows:

[0209] Process 4.1: Based on the optimized thresholds, perform online fault detection based on single-variable thresholds for multiple attribute values of the turntable in different modes;

[0210] Process 4.2: According to the detection results, if multiple consecutive points exceed the optimized thresholds, then issue a fault alarm.

[0211] Therefore, step 104 performs on-orbit turntable online fault detection based on the optimized thresholds, specifically including:

[0212] Step 4.1: According to the optimized thresholds, perform online fault detection based on single-variable thresholds for multiple key parameters of the on-orbit turntable in different steady-state operating modes, and generate single-variable threshold detection results.

[0213] According to the variable x in step 103 i the optimized threshold in mode n perform single-variable threshold detection on the selected key variables x1, x2, …, x L respectively.

[0214] Step 4.2: Perform a fault alarm according to the single-variable threshold detection results.

[0215] According to the detection results, if multiple consecutive points exceed the optimized thresholds, then issue a fault alarm.

[0216] Assume that the real-time online data of L variables collected at time t are respectively: x1(t), x2(t), …, x L (t). For each real-time online data x i (t), if the online data at 3 consecutive sampling times all exceed the optimized thresholds in N steady-state modes (i.e., If \(n = 1,\ldots,N\); \(i = 1,\ldots,L\)), then the variable is considered abnormal, that is, a fault alarm is issued. The fault judgment criterion is defined as follows:

[0217]

[0218] where \(M\) s (x i (t)) represents the result of single-variable threshold detection for three consecutive real-time data \(x\) i (t).

[0219] The advantages of the method of the present invention are that it realizes the optimization of monitoring thresholds and on-orbit fault detection under various working modes of the spaceborne turntable. The method of the present invention can use the evidence reasoning algorithm with interval reference values to fuse the system working thresholds and historical observation data under multiple modes, optimize the multi-mode thresholds offline, and realize the on-orbit fault detection of the spaceborne turntable. The method of the present invention is suitable for the threshold optimization and fault detection of components with multiple working modes, and can be applied to fault detection in the case of large parameter changes brought about by multi-mode switching of the system. At the same time, the threshold optimization process based on interval reference value evidence reasoning can realize the transparency of the fusion process, and use the false alarm rate and missed alarm rate under different thresholds as the optimization cut-off conditions, ensuring the interpretability of the optimization results.

[0220] A specific embodiment is provided below to further illustrate the effectiveness of the method of the present invention.

[0221] Taking the on-orbit telemetry data of a certain spaceborne turntable as an example to illustrate the implementation process. The spaceborne turntable mainly has two steady-state working modes, namely the low-speed steady-state mode and the high-speed steady-state mode. The working mode of the turntable mainly depends on the satellite mission requirements. Usually, there are mainly two mission modes for the satellite in orbit. Under mission mode 1, the turntable needs to rotate at a low speed and uniformly; under mission mode 2, the turntable needs to be driven to rotate at a high speed and uniformly. Under different steady-state working modes of the turntable, there are significant differences in the working thresholds corresponding to its key variables (key parameters). Three key variables, namely the turntable angular velocity \(x1\), the turntable motor current \(x2\), and the power supply monitoring voltage \(x3\), generated during the on-orbit operation of the spaceborne turntable are selected for its fault detection. Among them, the turntable angular velocity \(x1\) is mainly used to reflect the index of the actual operating speed of the turntable, and can reflect the speed holding ability of the turntable in the corresponding mode; the turntable motor current \(x2\) mainly reflects the monitoring index of the driving torque that the turntable motor can provide, and can reflect the magnitude of the turntable friction torque; the power supply monitoring voltage \(x3\) is mainly used to reflect the index of the performance of the turntable power supply system, and can reflect the performance status of the turntable drive power supply system.

[0222] Based on the above analysis, an ER-IRV multi-mode threshold optimization framework for the spaceborne turntable is constructed. The identification framework of this framework consists of two evaluation levels, denoted as Θ={H1, H2}, where H1 represents the low-speed steady-state mode and H2 represents the high-speed steady-state mode. The reference value of each evaluation level is represented as an interval reference value, and each interval reference value can be set according to the initial thresholds of the key variables in Table 1. For example, for the turntable motor current, the evaluation level H 1,2 of its interval reference value is set to

[0223] Table 1. Initial Thresholds of Key Monitoring Variables of the Turntable System

[0224] Variable name Low-speed steady-state initial threshold High-speed steady-state initial threshold Unit <![CDATA[Rotary table angular velocity x1]]> [2.0,4.0] [59.0,61.0] ° / s <![CDATA[Rotary table motor current x2]]> [0,0.4] [0.9,1.7] A <![CDATA[Power supply monitoring voltage x3]]> [5.0,7.0] [4.0,5.5] V

[0225] After the framework is constructed, according to the process of step 103, the thresholds of the three key variables in different modes can be optimized by distinguishing two cases. According to Table 1, the initial threshold intervals of the two variables of the turntable angular velocity x1 and the turntable motor current x2 are independent of each other in the two steady-state modes, and the threshold optimization can be carried out according to the process of case 2 in step 103; while the initial threshold intervals of the power supply monitoring voltage x3 in the two steady-state modes overlap, and the threshold optimization needs to be carried out according to the process of case 1 in step 103. Through threshold optimization, the threshold optimization results of each variable in the high-speed steady-state mode are as Figure 6 , 7 and shown in Figure 8. The threshold optimization results of each variable in different steady-state modes are shown in Table 2:

[0226] Table 2. Threshold Optimization Results of Key Monitoring Variables of the Turntable System

[0227] Variable name Low-speed steady-state optimization threshold High-speed steady-state optimization threshold Unit <![CDATA[Rotary table angular velocity x1]]> [2.9555,3.0508] [59.4940,60.4761] ° / s <![CDATA[Rotary table motor current x2]]> [0.1252,0.2748] [1.1372,1.4748] A <![CDATA[Power supply monitoring voltage x3]]> [5.5744,5.7994] [4.5989,4.8529] V

[0228] According to the optimized thresholds obtained in Table 2, univariate threshold detection is performed on the three key variables respectively. Taking the anomaly that the turntable motor current increases due to the increase of the bearing friction torque during the on-orbit operation of the spaceborne turntable as an example, the corresponding fault detection results are as Figure 9 shown. As can be seen from Figure 9 , the fault detection rate using the initial threshold is 30.6%, while the improvement value of the fault detection rate using the optimized threshold is 97.6%. It can be seen that using the optimized threshold for turntable fault detection can effectively improve the fault detection rate, verifying the effectiveness of the method of the present invention.

[0229] The method of the present invention innovatively uses the evidential reasoning algorithm to fuse multi-mode thresholds and monitoring data, and optimizes the multi-mode thresholds to achieve on-orbit fault detection of the spaceborne turntable, solving the problems of low fault detection rate caused by unreasonable setting of multi-mode initial thresholds before orbit injection and large differences between the initial thresholds and the on-orbit operating states. The present invention can fully integrate expert knowledge, improve the on-orbit fault detection ability of the spaceborne turntable, and has good engineering application value.

[0230] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts between the embodiments, reference can be made to each other.

[0231] Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A satellite turntable fault detection method based on multi-mode threshold optimization, characterized in that, Including: Construct a multi-mode threshold optimization framework based on evidence reasoning; The construction of the multi-mode threshold optimization framework based on evidence reasoning specifically includes: Obtain the on-orbit steady-state working modes of the spaceborne turntable; the steady-state working modes include a low-speed steady-state operation mode and a high-speed steady-state operation mode; a multi-mode system identifies N steady-state working modes, which are respectively defined as H1, H2, …, H N ; Extract the monitoring thresholds of the key parameters of the spaceborne turntable in each of the steady-state operating modes; the key parameters include the turntable motor current, the power supply monitoring voltage, and the turntable speed; a total of L key parameters are extracted, which are respectively represented as x1, x2, …, x L ; for the i-th variable, its monitoring threshold in the n-th steady-state operating mode is expressed as The evaluation level is inferred based on the evidence of the steady-state operating mode, and the monitoring thresholds under different steady-state operating modes are used as the interval reference values for each evaluation level to construct a multi-mode threshold optimization framework for each key parameter based on interval reference value evidence reasoning; for the evaluation level H n For it, the upper bound of the interval reference value is the upper bound of the monitoring threshold of the i-th variable in the n-th steady-state operating mode The lower bound is the lower bound of the monitoring threshold of the i-th variable in the n-th steady-state operating mode Determine the initial interval reference value of each key parameter under the multi-mode threshold optimization framework; The determination of the initial interval reference value of each key parameter under the multi-mode threshold optimization framework specifically includes: Based on the design specifications of the spaceborne turntable component and the ground test data, determine the initial working threshold interval values of each key parameter under different steady-state working modes Under the constructed multi-mode threshold optimization framework, the initial working threshold interval value of each key parameter is given the interval reference value of each key parameter at each evaluation level; among them, is given the lower bound of the interval reference value of the i-th variable at the n-th evaluation level H n , and is given the upper bound of the interval reference value of the i-th variable at the n-th evaluation level H n . Perform multi-mode threshold iterative optimization based on interval reference value evidence reasoning according to the initial interval reference value of each key parameter to generate an optimized threshold; The performing of multi-mode threshold iterative optimization based on interval reference value evidence reasoning according to the initial interval reference value of each key parameter to generate an optimized threshold specifically includes: Perform multi-mode threshold iterative optimization based on interval reference value evidence reasoning to generate an optimized threshold when the interval reference values of two adjacent evaluation levels cross; or Perform multi-mode threshold iterative optimization based on interval reference value evidence reasoning to generate an optimized threshold when the interval reference values of two adjacent evaluation levels are independent of each other; The performing of multi-mode threshold iterative optimization based on interval reference value evidence reasoning to generate an optimized threshold when the interval reference values of two adjacent evaluation levels cross specifically includes: Obtain a training data set and a test data set; the training data set includes normal historical observation data of the key parameter in the steady-state working mode; the test data set includes historical observation data of the key parameter at two evaluation levels; the historical observation data includes normal data and abnormal data; Calculate the interval confidence of two adjacent training samples in the training data set at different evaluation levels by using a utility-based information conversion method; Calculate the interval probability mass of two adjacent training samples at different evaluation levels according to the interval confidence of two adjacent training samples at different evaluation levels; the interval probability mass and the interval reference value jointly form the parameter to be optimized; Optimize the parameter to be optimized by using a projection covariance adaptive evolution algorithm to generate an optimized parameter; Perform false alarm rate and miss rate tests on the test samples in the test dataset according to the optimized parameters, and determine whether the optimization cut-off condition is satisfied; the defined optimization cut-off condition is: |FAR n,i -MAR n,i | ≤ ε and |FAR n+1,i -MAR n+1,i | ≤ ε; where ε is a very small constant value, satisfying 0 ≤ ε << 1, |·| is the absolute value operation; FAR n,i and MAR n,i respectively represent the false alarm rate and miss rate of the i-th key parameter in the steady-state operating mode n; FAR n+1,i and MAR n+1,i respectively represent the false alarm rate and miss rate of the i-th key parameter in the steady-state operating mode n + 1; If the optimization cut-off condition is satisfied, use the optimized parameter as the optimized threshold; If the optimization cut-off condition is not satisfied, update the interval reference value and the iteration number, and return to the step of calculating the interval confidence of two adjacent training samples in the training data set at different evaluation levels by using the utility-based information conversion method; Perform on-orbit turntable online fault detection based on the optimized threshold.

2. The method according to claim 1, characterized in that, The performing of multi-mode threshold iterative optimization based on interval reference value evidence reasoning to generate an optimized threshold when the interval reference values of two adjacent evaluation levels are independent of each other specifically includes: Obtain a training data set and a test data set; the training data set includes normal historical observation data of the key parameter in the steady-state working mode; the test data set includes historical observation data of the key parameter in the steady-state working mode; the historical observation data includes normal data and abnormal data; Calculate the interval confidence of two adjacent training samples in the training data set at different evaluation levels by using a utility-based information conversion method; Calculate the interval probability mass of two adjacent training samples at different evaluation levels according to the interval confidence of the two adjacent training samples at different evaluation levels; the interval probability mass and the interval reference value together constitute the parameter to be optimized. Optimize the parameter to be optimized by using the projection covariance adaptive evolution algorithm to generate the optimized parameter. Perform false alarm rate and miss alarm rate tests on the test samples in the test dataset according to the optimized parameter, and determine whether the optimization cut-off condition is satisfied. If the optimization cut-off condition is satisfied, use the optimized parameter as the optimization threshold. If the optimization cut-off condition is not satisfied, update the interval reference value and the iteration number, and return to the step of calculating the interval confidence of two adjacent training samples in the training dataset at different evaluation levels by using the utility-based information conversion method.

3. The method according to claim 1, characterized in that, The on-line fault detection of the spaceborne turntable based on the optimization threshold specifically includes: Perform on-line fault detection based on a single variable threshold on multiple key parameters of the spaceborne turntable in different steady-state working modes according to the optimization threshold to generate a single variable threshold detection result; perform fault alarm according to the single variable threshold detection result.

Images