Satellite component health state evaluation method based on empirical mode decomposition

Abstract

The application relates to a satellite component health state evaluation method based on empirical mode decomposition, which comprises the following steps: step one, establishing a statistical parameter data set based on a satellite component product characteristic parameter set: time domain statistical analysis is carried out by using remote sensing data of the product characteristic parameters, and each statistical parameter data set for evaluating the health state of the product is obtained; step two, extracting main components of a signal sequence based on empirical mode decomposition (EMD) decomposition: main component extraction and reconstruction are carried out on the statistical parameter data sample set; and step three, evaluating the health state based on the main components of the statistical parameters. The method realizes a satellite health state quantitative evaluation technology based on data driving, and under the premise that product composition and structural relationship are not needed, quantitative and qualitative evaluation results of the health state of a satellite component product can be given by constructing a health state scoring system based on historical remote sensing data. The health evaluation accuracy of the application can reach 90%.

Description

Technical Field

[0001] This invention belongs to the field of spacecraft testing technology, specifically relating to a method for assessing the health status of satellite components based on empirical mode decomposition. Background Technology

[0002] Satellite health status assessment is a crucial aspect of spacecraft on-orbit operation management. Satellite systems are complex, highly redundant, and exhibit characteristics such as reconfigurability and nonlinearity, making effective health status assessment challenging. Currently, with the increasing number of satellites of different platforms and series in my country, ground stations and operation control centers have accumulated a wealth of on-orbit measurement data, as well as information on satellite anomalies and malfunctions. However, most existing health status assessment methods are qualitative assessments at the system structure level, often requiring detailed system structure design data, and the assessment results are frequently inaccurate. Furthermore, research on health assessment at the component and product levels is limited. Moreover, system structure-level health assessment methods rely on structural relationships and fault propagation logic, resulting in very low utilization of telemetry data, especially for component-level products where hierarchical structures are difficult to define. Therefore, there is a need to develop a quantitative health status assessment method for satellite products that can fully utilize historical telemetry data and is data-driven, providing a rapid, accurate, and efficient health assessment technology for the stable and reliable on-orbit operation of large and complex spacecraft such as satellites and even space stations. Summary of the Invention

[0003] To address the aforementioned problems, this invention proposes a method for assessing the health status of satellite components based on empirical mode decomposition.

[0004] This invention provides a method for assessing the health status of satellite components based on empirical mode decomposition, comprising the following steps:

[0005] Step 1: Establish a statistical parameter dataset based on the original telemetry data parameter set of satellite component products: First, select parameters with actual physical significance from the original telemetry data of satellite component products to construct a feature parameter set. Then, perform data preprocessing and recombination on the feature parameter set to construct a satellite product feature parameter data sample set. Finally, use time-domain analysis methods to extract the statistical parameters of the feature parameter data sample set to obtain various statistical parameter datasets used for product health status assessment.

[0006] Step 2: Extract principal components of the statistical parameter sequence based on Empirical Mode Decomposition (EMD) and Multi-Scale Permutation Entropy (MPE): The signal is decomposed into several intrinsic mode functions (IMFs) using Empirical Mode Decomposition (EMD). Then, Multi-Scale Permutation Entropy (MPE) is used to measure the complexity of the IMFs. Finally, the principal components and minor components are distinguished according to the threshold of MPE, thus completing the reconstruction and extraction of the principal components of the signal sequence.

[0007] Step 3: Assess product health status based on principal components of statistical parameters: Establish threshold ranges for different health statuses using the principal components of statistical parameters and Euclidean distance method to achieve health status assessment of satellite component products.

[0008] In step one, the construction of the product feature parameter data sample set utilizes the original telemetry data of the product. First, parameters with actual physical significance are selected from the original telemetry data to construct the feature parameter set. Then, the telemetry data of each feature parameter of the product are preprocessed. Finally, they are reorganized and arranged in a certain order to construct the product feature parameter data sample set.

[0009] In step one, for each preprocessed feature parameter, samples are taken sequentially in chronological order with N consecutive data points, resulting in M ​​samples. Each sample is then arranged column-wise according to chronological order, thus obtaining a new data sample set. That is, the original one-dimensional dataset is rearranged into an N×M two-dimensional matrix. In this matrix, each column represents an independent sample.

[0010] In step one, the method for constructing the statistical parameter set is as follows: Time-domain statistical analysis is performed using telemetry data of product characteristic parameters, for the sequence X = {x} i Let i = 1, 2, ..., N, and extract K time-domain statistical parameters. Based on the obtained two-dimensional sample dataset, for each sample, K time-domain statistical parameters can be obtained using time-domain statistical analysis. Then, the K time-domain statistical parameters of each sample are arranged in rows to obtain the statistical parameter set of that sample. That is, each sample, originally a sequence containing the original data, becomes a new sequence containing K statistical parameters after time-domain statistical analysis. Each statistical parameter sequence is then arranged in columns to obtain a statistical parameter data sample set C of size K×M based on the feature parameter telemetry data sample set.

[0011] In step two, the Empirical Mode Decomposition (EMD) process is as follows:

[0012] (1) Finding extreme points: First, obtain all the maxima and minima of the signal sequence;

[0013] (2) Fitting the envelope curve: By using the maximum and minimum value groups of the signal sequence, two smooth peak / valley fitting curves are obtained through cubic spline interpolation, namely the upper envelope and lower envelope of the signal.

[0014] (3) Mean envelope: The mean envelope is obtained by averaging the two extreme curves;

[0015] (4) Intermediate signal: Subtract the mean envelope from the original signal to obtain the intermediate signal;

[0016] (5) Determine the intrinsic mode function (IMF): The IMF needs to meet two conditions: 1) In the entire data segment, the number of extreme points and the number of zero crossings must be equal or differ by no more than one; 2) At any time, the average value of the upper envelope formed by the local maximum point and the lower envelope formed by the local minimum point is zero, that is, the upper and lower envelopes are locally symmetrical with respect to the time axis.

[0017] (6) If the new data obtained by subtracting the envelope average from the original data still has negative local maxima and positive local minima, it means that this is not an intrinsic mode function and the process (1)-(5) needs to be repeated; the first intermediate signal that satisfies the IMF condition is the first intrinsic mode function component IMF1 of the original signal.

[0018] (7) After obtaining the first IMF using the above method, subtract IMF1 from the original signal to obtain the new original signal, and repeat the process (1)-(6) to obtain IMF2. This process is repeated to complete the EMD decomposition.

[0019] For a signal sequence, after EMD decomposition, several intrinsic mode functions are obtained.

[0020] In step two, the method for distinguishing between the principal and minor components of the multi-scale permutation entropy (MPE) and determining the MPE threshold is as follows: the principal and minor components of the feature index are distinguished based on the MPE magnitude of the intrinsic mode function (IMF) of the feature index.

[0021] The process of calculating the multi-scale permutation entropy of a signal sequence is as follows:

[0022] (1) Coarsening the time series: For a time series of length M, P = {p i The sequence}, i = 1, 2, ..., M is coarse-grained to obtain a coarse-grained sequence. Its expression is:

[0023]

[0024] Where s is the scaling factor, s≤M; [M / s] represents rounding down M / s, and when s=1, the coarse-grained sequence is the original sequence;

[0025] (2) Reconstruct the phase space for each coarse-grained sequence;

[0026] (3) The reconstructed components are arranged in ascending order, and the probability of each symbol sequence is calculated.

[0027] (4) According to The permutation entropy of each coarse-grained sequence is calculated and normalized to obtain the multi-scale permutation entropy value.

[0028] The process for determining the MPE threshold that distinguishes between major and minor components is as follows:

[0029] (1) Construct several trend sequences and random noise sequences to represent different principal components and detail components, and form new time series respectively;

[0030] (2) Perform EMD decomposition on the new time series composed in (1) to obtain several intrinsic mode functions (IMFs). Reconstruct the IMFs smaller than the MPE value into principal components according to different MPE values.

[0031] (3) Calculate the Euclidean distance between the main components reconstructed with different MPE values ​​in (2) and the original trend sequence. Determine the MPE threshold with the goal of minimizing the Euclidean distance and making the reconstructed components contain less fluctuation.

[0032] For time series, after decomposing the time series EMD, the main components of the time series are extracted after reconstructing the specific IMF based on the MPE value.

[0033] In step three, the methods for extracting the main components of the statistical parameters and constructing the sample set are as follows:

[0034] Each row in the statistical parameter set C is a statistical parameter sequence of length M. EMD decomposition is performed on the K statistical parameter sequences: for the k-th statistical parameter sequence... When performing EMD decomposition, m consecutive data points are used as the objects, i.e. A total of M-m+1 EMD decompositions are performed to obtain M-m+1 different sets of intrinsic mode functions (IMFs). Then, the multi-scale permutation entropy (MPE) value of each signal in the IMF set is calculated. Finally, the principal components are extracted based on the MPE threshold. That is, for the k-th statistical parameter sequence... Finally, we can obtain M-m+1 new principal component sequence samples, each containing m data points.

[0035] In step three, the method for determining the thresholds for different health states based on Euclidean distance is as follows:

[0036] The sample points contain n variables, and the first sample point A = (x1, x2, ..., xn). nFrom the second sample point B = (y1, y2, ..., y...) n The distance between them can be expressed as:

[0037]

[0038] If we take M samples, all of which are data from a healthy state, then the process of determining the threshold range of the k-th statistical parameter for a healthy state using this sample set is as follows:

[0039] (1) Extract the M-m+1 principal component sequences of the k-th statistical parameter of the M samples, and calculate the mean of the M-m+1 samples. Then, a mean vector is constructed using this mean.

[0040] (2) Calculate the Euclidean distance between each of the M-m+1 samples and the mean vector μ1, and obtain the Euclidean distance set {d1(i)}, i=1,2,…,M-m+1;

[0041] (3) Calculate the mean of the set of Euclidean distances. with standard deviation Based on the Wright criterion, for those elements in this set that exceed... Replace the distance value of the range with the mean value. This results in a new set of distances.

[0042] (4) Calculate the new distance set mean with standard deviation Therefore, the threshold for health status is set with a lower limit of 0 and an upper limit of [value missing]. That is, the threshold range

[0043] The threshold range for sub-health status needs to be determined using the mean vector of the healthy state. Using this as a baseline, the calculation process is similar to the threshold determination process for healthy states. For a sub-healthy state with R samples, the specific process for determining the threshold is as follows:

[0044] (1) Extract the R-m+1 principal component sequences of the kth statistical parameter of the R samples, calculate the Euclidean distance between each sample in the R-m+1 samples and the mean vector μ1 of the health status, and obtain the Euclidean distance set {d2(i)}, i=1,2,…,R-m+1;

[0045] (2) Calculate the mean of the set of Euclidean distances. with standard deviation Based on the Wright criterion, for those elements in this set that exceed... The distance value of the range is taken as the average value. This results in a new set of distances.

[0046] (3) Calculate the new distance set mean with standard deviation Since this health state is the next level below the health state, for this state, the lower limit is the upper limit of the health state threshold, and the upper limit is... That is, the distance threshold range

[0047] For the failure state, the threshold range is:

[0048] In step three, the health status quantification method based on threshold mapping is as follows:

[0049] By setting a threshold range for health status and its quantified score, then for the distance value d... s The corresponding quantitative scoring results are as follows:

[0050] 1) If 0 ≤ d s ≤a:

[0051]

[0052] 2) If a≤d s ≤b:

[0053]

[0054] 3) If d s >b:

[0055]

[0056] in, This represents the upper limit of the health status threshold. This represents the upper limit of the sub-healthy state threshold.

[0057] In step three, the health status assessment method based on comprehensive evaluation of statistical parameters is as follows:

[0058] Health status is assessed using a test sample with a sample size greater than m. The assessment result represents the health status of the first sample. The test sample is then tested, and each statistical parameter corresponds to a quantified score.

[0059] {score(C k )}, k=1,2,…,K

[0060] Based on this result, the final quantitative score of the test sample can be obtained:

[0061]

[0062] Beneficial effects

[0063] This invention's method, based on temporal statistical parameters from satellite component telemetry data, utilizes Empirical Mode Decomposition (EMD) and Multiscale Permutation Entropy (MPE) theory to extract the principal components of these statistical parameters. Then, based on these principal components, the Euclidean Distance (ED) method is used to calculate threshold ranges for different health states, establishing a health status assessment system to achieve health assessment of satellite component products. This method realizes a data-driven quantitative assessment technology for satellite health status. Without requiring information on product composition or structural relationships, it constructs a health status scoring system based on historical telemetry data, providing both quantitative and qualitative assessment results for the health status of satellite component products. Extensive health assessment examples have verified that the proposed health assessment accuracy can reach 90%. Attached Figure Description

[0064] Figure 1 This is an example diagram illustrating the construction of a two-dimensional data sample set using preprocessed data sequences according to the present invention.

[0065] Figure 2 This is an example diagram illustrating the construction of a statistical parameter set using a two-dimensional data sample set in this invention.

[0066] Figure 3 This is a schematic diagram illustrating the reconstruction and extraction of major and minor components based on EMD decomposition according to the present invention.

[0067] Figure 4 This is a schematic diagram illustrating the extraction of the main components of the statistical parameters and the construction of the sample set in this invention. Detailed Implementation

[0068] The following combination Figures 1 to 4 This implementation method will be described in detail.

[0069] This invention provides a satellite component health status assessment method based on empirical mode decomposition, comprising the following steps:

[0070] Step 1: Establish a statistical parameter dataset based on the original telemetry data parameter set of satellite component products.

[0071] First, parameters with practical physical significance are selected from the raw telemetry data of satellite components to construct a feature parameter set. Then, the feature parameter set undergoes data preprocessing and rearrangement to construct a satellite product feature parameter data sample set. Finally, time-domain analysis methods are used to extract statistical parameters from the feature parameter data sample set, resulting in various statistical parameter datasets used for product health status assessment. The details are as follows:

[0072] Step 1-1: Construction of Product Feature Parameter Data Sample Set

[0073] Using the product's original telemetry data, parameters with actual physical significance are first selected from the original telemetry data to construct a set of characteristic parameters. Then, the telemetry data of each characteristic parameter of the product are preprocessed. Finally, they are reorganized and arranged in a certain order to construct a sample set of product characteristic parameter data.

[0074] Step 1-1-1: Preprocessing of telemetry data

[0075] For raw telemetry data, the data preprocessing methods include outlier removal and missing value handling.

[0076] (1) Outlier removal

[0077] Because the acquisition process of telemetry data is subject to interference from sensors, converters, and wireless transmission, the received data often contains abnormal jump points, which are incorrect points caused by the measuring equipment and the transmission process. These data points that deviate from the variation pattern of the measured signal are usually called outliers. Outliers in telemetry data provide erroneous information and affect the processing and analysis results of telemetry signals. Therefore, outlier removal is an important step in telemetry data preprocessing. By eliminating randomly occurring measurement values ​​with large errors, the authenticity of telemetry data can be guaranteed to a certain extent, and the reliability of data analysis can be improved.

[0078] The Wright criterion method assumes that the distribution of measurement data approximates a normal distribution. Based on this assumption, using a given confidence probability of 99.7% as the standard and three times the standard deviation of the measurement as the criterion, any measurement value exceeding this limit is considered an outlier. For a given telemetry measurement sequence x = {x...} i}, i = 1, ..., N, where N is the number of sampling points. The basic operation flow of the Wright criterion is as follows:

[0079] 1): Calculate the mean of the measurement value series:

[0080]

[0081] 2): Calculate the standard deviation of the measurement series:

[0082]

[0083] 3): Remove outliers:

[0084]

[0085] (2) Handling missing values

[0086] After removing outliers, missing values ​​need to be processed. Methods used include mean imputation and regression imputation. Mean imputation uses the mean of the known data as a substitute for the missing values. Let the data sequence used to impute the missing values ​​be Y = {y...} i}, i = 1, 2, ..., N, where N is the number of data points, and the missing value to be filled is y. * ,but:

[0087]

[0088] Regression imputation: This method involves constructing a regression equation for the known data and using this equation to predict missing points, thus achieving data imputation. Let the known data sequence of the variable to be imputed be Y = {y...} i}, i = 1, 2, ..., N, where the missing value of the variable to be filled is y. * The sequence of auxiliary variables used to construct the regression equation and the known sequence of the variable to be filled is X = {x}. j}, j=1,2,…,N, where the auxiliary variable value corresponding to the missing value of the variable to be filled is x. * And x * Given x and y, construct a regression equation:

[0089] y i =f(x) i )

[0090] Using the above formula and x * Predictable y * for:

[0091] y * =f( x * )

[0092] Step 1-1-2: Rearrangement of feature parameters and construction of data sample set

[0093] Based on step 1-1-1, for each preprocessed feature parameter, samples are taken sequentially in chronological order with N consecutive data points, yielding M samples. Each sample is then arranged column-wise according to chronological order, resulting in a new data sample set. That is, the original one-dimensional dataset is rearranged into an N×M two-dimensional matrix. In this matrix, each column represents an independent sample (see Appendix). Figure 1 .

[0094] Steps 1-2: Construction of the statistical parameter set

[0095] Based on step 1-1, time-domain statistical analysis is performed using telemetry data of product characteristic parameters. For a sequence X = {x} containing N data points... i}, i = 1, 2, ..., N, extract K time-domain statistical parameters, taking the following 8 commonly used time-domain statistical parameters as examples:

[0096] (1) Maximum value: X max =max{X}

[0097] (2) Minimum value: X min =min{X}

[0098] (3) Peak-to-peak value: X ppv =X max -X min

[0099] (4) Mean:

[0100] (5) Absolute average amplitude:

[0101] (6) Root amplitude:

[0102] (7) Standard deviation:

[0103] (8) Effective value (root mean square value):

[0104] Based on the two-dimensional sample dataset obtained in step 1-1-2, for each sample, K time-domain statistical parameters can be obtained using time-domain statistical analysis. These K time-domain statistical parameters for each sample are then arranged row-wise to obtain the statistical parameter set for that sample. In other words, each sample, originally a sequence containing the original data, becomes a new sequence containing K statistical parameters after time-domain statistical analysis. Each statistical parameter sequence is then arranged column-wise to obtain a statistical parameter data sample set C of size K×M based on the feature parameter telemetry data sample set (see Appendix). Figure 2 .

[0105] Step 2: Extract principal components of statistical parameter sequences based on Empirical Mode Decomposition (EMD) and Multiscale Permutation Entropy (MPE).

[0106] To better utilize trend changes in component product evaluation, it is necessary to extract and reconstruct the principal components of the statistical parameter data sample set. The principal components, i.e., the low-frequency part of the statistical parameter signal, mainly reflect the trend characteristics of the signal; the secondary components, i.e., the high-frequency part of the signal, mainly reflect some irregular fluctuations. Principal components of the statistical parameter sequence are extracted based on Empirical Mode Decomposition (EMD) and Multi-Scale Permutation Entropy (MPE): EMD decomposes the signal into several Intrinsic Mode Functions (IMFs), then MPE is used to measure the complexity of the IMFs. Finally, the principal and secondary components are distinguished based on the MPE threshold, completing the reconstruction and extraction of the principal components of the signal sequence. The details are as follows:

[0107] Step 2-1: Obtaining the intrinsic mode functions based on Empirical Mode Decomposition (EMD)

[0108] The process of empirical mode decomposition is as follows:

[0109] (1) Finding extreme points: First, obtain all the maxima and minima of the signal sequence;

[0110] (2) Fitting the envelope curve: By using the maximum and minimum value groups of the signal sequence, two smooth peak / valley fitting curves are obtained through cubic spline interpolation, namely the upper envelope and lower envelope of the signal.

[0111] (3) Mean envelope: The mean envelope is obtained by averaging the two extreme curves;

[0112] (4) Intermediate signal: Subtract the mean envelope from the original signal to obtain the intermediate signal;

[0113] (5) Determine the intrinsic mode function (IMF): The IMF needs to meet two conditions: 1) The number of extreme points and the number of zero crossings must be equal or differ by no more than one in the entire data segment; 2) At any time, the average value of the upper envelope formed by the local maximum point and the lower envelope formed by the local minimum point is zero, that is, the upper and lower envelopes are locally symmetrical with respect to the time axis.

[0114] (6) If the new data obtained by subtracting the envelope average from the original data still contains negative local maxima and positive local minima, it indicates that this is not an intrinsic mode function and the process (1)-(5) needs to be repeated. The first intermediate signal that satisfies the IMF condition is the first intrinsic mode function component IMF1 of the original signal;

[0115] (7) After obtaining the first IMF using the above method, subtract IMF1 from the original signal to obtain the new original signal, and repeat the process (1)-(6) to obtain IMF2. This process is repeated to complete the EMD decomposition.

[0116] For a signal sequence, after EMD decomposition, several intrinsic mode functions are obtained.

[0117] Step 2-2: Method for distinguishing between the principal and minor components of multi-scale permutation entropy (MPE) and determining the MPE threshold.

[0118] Multiscale Permutation Entropy (MPE) is an effective metric for measuring the complexity of a signal. The primary and secondary components of a feature index are distinguished by the magnitude of its intrinsic modulus function (IMF).

[0119] The process of calculating the multi-scale permutation entropy of a signal sequence is as follows:

[0120] (1) Coarse-graining the time series: For a sequence of length M, P = {p i The sequence}, i = 1, 2, ..., M is coarse-grained to obtain a coarse-grained sequence. Its expression is:

[0121]

[0122] Where s is the scaling factor, s≤M. [M / s] represents rounding down M / s. When s=1, the coarse-grained sequence is the original sequence.

[0123] (2) Reconstruct the phase space for each coarse-grained sequence;

[0124] (3) The reconstructed components are arranged in ascending order, and the probability of each symbol sequence is calculated.

[0125] (4) According to The permutation entropy of each coarse-grained sequence is calculated and normalized to obtain the multi-scale permutation entropy value.

[0126] The primary component mainly reflects the trend characteristics of the signal, while the secondary component mainly reflects some irregular fluctuations in the signal. The process of determining the MPE threshold that distinguishes between the primary and secondary components is as follows:

[0127] (1) Construct several trend sequences and random noise sequences to represent different principal components and detail components, and form new time series respectively;

[0128] (2) Perform EMD decomposition on the new time series composed in (1) to obtain several intrinsic mode functions (IMFs). Reconstruct the IMFs smaller than the MPE value into principal components according to different MPE values.

[0129] (3) Calculate the Euclidean distance between the main components reconstructed with different MPE values ​​in (2) and the original trend sequence. Determine the MPE threshold with the goal of minimizing the Euclidean distance and making the reconstructed components contain less fluctuation.

[0130] Therefore, for a time series, based on step 2-1 and the above method: after decomposing the time series EMD, and then reconstructing the specific IMF according to the MPE value, the main components of the time series can be extracted. (See Appendix) Figure 3 .

[0131] Step 3: Assess the product's health status based on statistical parameters and key components.

[0132] Based on the statistical parameter dataset established in step one and the principal component extraction method in step two, threshold ranges for different health states are established using the principal components of the statistical parameters and the Euclidean distance method, thereby enabling the assessment of the health status of satellite component products. Specifically:

[0133] Step 3-1: Methods for extracting the main components of statistical parameters and constructing the sample set.

[0134] Each row in the statistical parameter set C is a statistical parameter sequence of length M. EMD decomposition is performed on the K statistical parameter sequences: for the k-th statistical parameter sequence... When performing EMD decomposition, m consecutive data points are used as the objects, i.e. A total of M-m+1 EMD decompositions are performed to obtain M-m+1 different sets of intrinsic mode functions (IMFs). Then, the multi-scale permutation entropy (MPE) value of each signal in the IMF set is calculated. Finally, the principal components are extracted based on the MPE threshold. That is, for the k-th statistical parameter sequence... Ultimately, we can obtain M-m+1 new principal component sequence samples, each containing m data points, as shown in the appendix. Figure 4 .

[0135] Step 3-2: Method for determining different health status thresholds based on Euclidean distance

[0136] Euclidean distance is a commonly used definition of distance, referring to the true distance between two points in space, or the natural length of a vector (i.e., the distance from that point to the origin). If the sample points contain n variables, then the first sample point A = (x1, x2, ..., x...). n From the second sample point B = (y1, y2, ..., y...) n The distance between them can be expressed as:

[0137]

[0138] In step 3-1, M samples are taken, all of which are data samples from the healthy state. The process of determining the threshold range of the k-th statistical parameter for the healthy state using this sample set is as follows:

[0139] (1) Based on step 3-1, extract the M-m+1 principal component sequences of the k-th statistical parameter of the M samples, and calculate the mean of the M-m+1 samples. Then, a mean vector is constructed using this mean.

[0140]

[0141] (2) Calculate the Euclidean distance between each of the M-m+1 samples and the mean vector μ1, and obtain the Euclidean distance set {d1(i)}, i=1,2,…,M-m+1;

[0142] (3) Calculate the mean of the set of Euclidean distances. with standard deviation Based on the Wright criterion, for those elements in this set that exceed... Replace the distance value of the range with the mean value. This results in a new set of distances.

[0143] (4) Calculate the new distance set mean with standard deviation Therefore, the threshold for health status is set with a lower limit of 0 and an upper limit of [value missing]. That is, the threshold range

[0144] The threshold range for sub-health status needs to be determined using the mean vector of the healthy state. Using this as a baseline, the calculation process is similar to the threshold determination process for healthy states. For a sub-healthy state with R samples, the specific process for determining the threshold is as follows:

[0145] (1) Extract the R-m+1 principal component sequences of the kth statistical parameter of the R samples, calculate the Euclidean distance between each sample in the R-m+1 samples and the mean vector μ1 of the health status, and obtain the Euclidean distance set {d2(i)}, i=1,2,…,R-m+1;

[0146] (2) Calculate the mean of the set of Euclidean distances. with standard deviation Based on the Wright criterion, for those elements in this set that exceed... The distance value of the range is taken as the average value. This results in a new set of distances.

[0147] (3) Calculate the new distance set mean with standard deviation Since this health state is the next level below the health state, for this state, the lower limit is the upper limit of the health state threshold, and the upper limit is... That is, the distance threshold range

[0148] For the failure state, the threshold range is:

[0149] Step 3-3: Health Status Quantification Method Based on Threshold Mapping

[0150] Based on the threshold range established in step 3-2, health status: Sub-health state: Failure status: set up The threshold ranges and quantified scores for each health state can be assumed to be as shown in the table below:

[0151] healthy Sub-health Failure Threshold range [0,a] (a,b) >b Quantitative scoring [100,80] (80,60] <60

[0152] Then for the distance value d s The corresponding quantitative scoring results are as follows:

[0153] 1) If 0 ≤ d s ≤a:

[0154]

[0155] 2) If a≤d s ≤b:

[0156]

[0157] 3) If d s >b:

[0158]

[0159] Steps 3-4: Health Status Assessment Method Based on Comprehensive Evaluation of Statistical Parameters

[0160] If a test sample is to be used to assess the health status over a certain period of time, the number of samples must be greater than m; otherwise, EMD decomposition cannot be performed. The result of the assessment represents the health status of the first sample.

[0161] Based on steps 3-1 to 3-3, an assessment system for evaluating health status using the k-th statistical parameter was established. Similarly, for other statistical parameters, steps 3-1 to 3-3 were repeated to establish an assessment system for each parameter. Then, this assessment system was used to test the sample, yielding a quantitative score for each statistical parameter.

[0162] {score(C k )}, k=1,2,…,K

[0163] Based on this result, the final quantitative score of the test sample can be obtained:

[0164]

[0165] The above description of the present invention is only a preferred embodiment of the present invention and is not intended to limit the implementation of the present invention. Those skilled in the art can easily make corresponding modifications or alterations based on the main concept and spirit of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection claimed in the claims.

Claims

1. A method for evaluating health condition of a satellite component based on empirical mode decomposition, characterized in that, Comprising the following steps: Step one, establishing a statistical parameter data set based on the satellite component product original telemetry data parameter set: first, constructing a characteristic parameter set from the satellite component product original telemetry data with actual physical significance, then performing data preprocessing and reordering on the characteristic parameter set to construct a satellite product characteristic parameter data sample set, and finally extracting statistical parameters of the characteristic parameter data sample set using time domain analysis method to obtain respective statistical parameter data sets for product health state evaluation; Step two, extracting main components of the statistical parameter sequence based on empirical mode decomposition (EMD) and multi-scale permutation entropy (MPE): decomposing the signal into a plurality of intrinsic mode functions (IMFs) using EMD, then measuring the complexity of the IMFs using MPE, and finally distinguishing the main components from the secondary components according to the MPE threshold value to complete the reconstruction and extraction of the main components of the signal sequence; The method for distinguishing the main components from the secondary components based on MPE and determining the MPE threshold value is as follows: distinguishing the main components from the secondary components of the characteristic index according to the MPE size of the characteristic index IMF; The process of calculating the multi-scale permutation entropy (MPE) of the signal sequence is as follows: (1) Coarsening the time series: Coarsen the time series of length N to obtain a coarsened series whose expression is: ； Where s is the scale factor, ; Indicates to Rounding down, when s=1, the coarse-grained sequence is the original sequence; (2) Space reconstruction is performed on each coarse-grained sequence; (3) The reconstructed components are arranged in ascending order, and the probability of each symbol sequence appearing is calculated; (4) According to The permutation entropy of each coarse-grained sequence is calculated and normalized, and then the multi-scale permutation entropy value is obtained. The process of determining the MPE threshold value for distinguishing the main components from the secondary components is as follows: (1) A plurality of trend sequences and random noise sequences representing different main components and detail components are constructed to form new time sequences; (2) The new time sequences formed in (1) are decomposed using EMD to obtain a plurality of intrinsic mode functions (IMFs), and the IMFs less than the MPE value are reconstructed as main components according to different MPE values; (3) The Euclidean distance between the main components reconstructed with different MPE values in (2) and the original trend sequence is calculated, and the MPE threshold value is determined by taking the minimum Euclidean distance and the MPE value with fewer fluctuations as the target; For a time sequence, the main components of the time sequence are extracted by reconstructing a specific IMF after EMD decomposition and according to the MPE value; Step three, evaluating the product health state based on the main components of the statistical parameters: establishing the threshold range of different health states using the main components of the statistical parameters and the Euclidean distance method to realize the health state evaluation of the satellite component product.

2. The empirical mode decomposition-based satellite component health assessment method according to claim 1, wherein, In step one, the product characteristic parameter data sample set is constructed using the product original telemetry data, first, the characteristic parameter set is constructed from the original telemetry data with actual physical significance, then the telemetry data of each product characteristic parameter is preprocessed, and finally it is reordered in a certain order to construct the product characteristic parameter data sample set.

3. The empirical mode decomposition-based satellite component health assessment method according to claim 2, wherein, In step one, after the original telemetry data is preprocessed, for each characteristic parameter, a time sequence is formed with N seconds as a monitoring period, As a sample, each sample is arranged in turn according to the sequence, so that a new data sample set is obtained, that is, for the original one-dimensional data set, after recombination and arrangement, it becomes a two-dimensional matrix; for the matrix, each column represents different samples, the time interval between each two adjacent samples is equal, and each row represents telemetry data arranged in time in a sample.

4. The empirical mode decomposition-based satellite component health assessment method according to claim 3, wherein, In step one, the construction method of the statistical parameter set is as follows: The time domain statistical analysis is performed on the telemetry data of the product characteristic parameters, and k time domain statistical parameters are extracted Based on the obtained two-dimensional sample data set, for each sample, the method of time domain statistical analysis can respectively obtain k time domain statistical parameters, and then the k time domain statistical parameters of each sample are arranged in rows to obtain a statistical parameter set of the sample. That is, each sample is originally a sequence containing original data, and after the time domain statistical analysis, it becomes a new sequence containing k statistical parameters of the sequence. Each sample is arranged in turn in columns to obtain a statistical parameter data sample set based on the characteristic parameter telemetry data sample set.

5. The empirical mode decomposition-based satellite component health assessment method according to claim 1, wherein, In step two, the process of empirical mode decomposition (EMD) is as follows: (1) Finding extreme points: first, obtain all the maximum and minimum values of the signal sequence; (2) Fitting envelope curve: two smooth peak / trough fitting curves, i.e. upper envelope curve and lower envelope curve of the signal, are obtained by using cubic spline interpolation method through the maximum and minimum value groups of the signal sequence; (3) Mean envelope curve: the mean envelope curve is obtained by averaging the two extreme value curves; (4) Intermediate signal: the intermediate signal is obtained by subtracting the mean envelope curve from the original signal; (5) Judgment of intrinsic mode function IMF: IMF needs to meet two conditions: 1) the number of extreme points and the number of zero-crossing points must be equal or differ by at most one in the entire data segment; 2) at any time, the average of the upper envelope curve formed by the local maximum points and the lower envelope curve formed by the local minimum points is zero, i.e. the upper and lower envelope curves are locally symmetric with respect to the time axis; (6) The new data obtained by subtracting the envelope average from the original data, if there are still negative local maximum values and positive local minimum values, it means that it is not an intrinsic mode function, and the process (1)-(5) needs to be repeated; the first intermediate signal that meets the IMF condition is the first intrinsic mode function component IMF1 of the original signal; (7) After obtaining the first IMF using the above method, subtract IMF1 from the original signal as the new original signal, and repeat the process (1)-(6), IMF2 can be obtained, and so on, to complete the EMD decomposition; For a signal sequence, EMD decomposition is performed to obtain several intrinsic mode functions.

6. The empirical mode decomposition-based satellite component health assessment method according to claim 1, wherein, In step three, the extraction of main components of statistical parameters and the construction method of sample set are as follows: In the EMD decomposition, the peak-peak value statistical parameters of m samples are sequentially taken as objects, i.e. A total of n-m+1 times of EMD decomposition are performed to obtain n-m+1 groups of different intrinsic mode function sets, then the multi-scale permutation entropy values MPE of each signal in the intrinsic mode function set are calculated, and finally the extraction of principal components is completed according to the MPE threshold value; that is, for the peak-peak value statistical parameters of the n samples, finally n-m+1 new principal component sequence samples each containing m data can be obtained.

7. The empirical mode decomposition-based satellite component health assessment method according to claim 6, wherein, In step three, the determination method of different health state threshold values based on Euclidean distance is as follows: The sample points contain n variables, the first sample point to the second sample point The distance between them can be expressed as: ； Taking n sample data samples in healthy state, the process of determining the peak-to-peak value statistical parameter threshold range of the health state is as follows: (1) extracting n-m+1 principal component sequences of peak-to-peak value statistical parameters of the n samples, calculating the mean value of the n-m+1 samples , and then constructing a mean vector using the mean value ; (2) calculate the Euclidean distance of each sample in the n-m+1 samples from the mean vector , and obtain a Euclidean distance set ; (3) calculating the mean of the set of Euclidean distances and the standard deviation ; Based on the Wright criterion, for the distances in the set that exceed the range [a, b], replace them with the mean value of the set, so that the new set of distances is [a, b] ; (4) Calculate the new distance set mean with standard deviation Therefore, the threshold for health status is set with a lower limit of 0 and an upper limit of [value missing]. That is, the threshold range For the threshold range of other health states, the calculation process is similar to the threshold determination process for the health state. Taking the sub-health state, the next level of health, as an example, assuming there are x samples in total, the specific process is as follows: (1) extracting x-m+1 principal component sequences of the peak-to-peak value statistical parameters of the x samples, calculating the Euclidean distance between each sample in the x-m+1 samples and the health state mean vector , and obtaining a Euclidean distance set ; (2) calculating the mean of the set of Euclidean distances and the standard deviation ; Based on the Wite criterion, for the distance values in the set that exceed the range [0, 1], take the mean value from the set, so that the new distance set is [0, 1] ; (3) Calculate the new distance set mean with standard deviation Since this health state is the next level below the health state, for this state, the lower limit is the upper limit of the health state threshold, and the upper limit is... That is, the distance threshold range ; For the failure state, the threshold range is .

8. The empirical mode decomposition-based satellite component health assessment method according to claim 7, wherein, In step three, the health state quantification method based on threshold mapping is as follows: Setting the threshold range corresponding to the health state and its quantitative score, then for the distance value The corresponding quantitative score result is: 1) If : ； 2) if : ； 3) if : ； wherein, is a health status upper threshold, is a sub-health status upper threshold.

9. The empirical mode decomposition-based satellite component health assessment method according to claim 8, wherein, In step three, the health state evaluation method based on statistical parameter comprehensive evaluation is: Using test samples to evaluate the health state, the sample number is greater than m, and the evaluation result represents the health state of the first sample; test the test samples, and each parameter corresponds to a quantitative score result: ； Wherein, n represents different time domain statistical parameters; based on the result, the final quantitative score of the test sample can be obtained: 。

Images