A method for wing strain data representation and anomaly identification based on staged multivariate statistical process control

Abstract

This invention discloses a method for wing strain data characterization and anomaly identification based on staged multivariate statistical process control. The method collects strain signals from the wing structure and divides the flight process into takeoff, level flight, and landing phases. Six time-domain feature parameters—peak value, root mean square, standard deviation, peak-to-peak value, kurtosis, and skewness—are extracted from the strain signals of each flight phase to construct a six-dimensional feature vector. Based on the six-dimensional feature vectors from several normal flight sorties, Hotelling's T is established in stages. 2 A multivariate statistical control model was developed and the upper control limit was determined; the T value of the flight to be tested was calculated. 2 The statistical value is compared with the upper control limit to determine whether the strain status is abnormal; for abnormal flights, T is... 2 The statistical value is decomposed into the contribution values ​​of various time-domain characteristic parameters to determine the dominant characteristic parameters. This invention achieves multi-dimensional compact characterization of wing strain data and automatic identification of abnormal flights, and can trace the cause of anomalies, making it suitable for wing structural health monitoring.

Description

Technical Field

[0001] This invention relates to a method for characterizing and identifying anomalies in wing strain data based on staged multivariate statistical process control. Background Technology

[0002] Structural health monitoring (SHM) acquires response data through a sensor network integrated into critical aircraft structures. Signal processing enables structural condition assessment and damage identification, making it a crucial means of ensuring aircraft safety throughout its entire lifecycle. As the primary aerodynamic surface and load-bearing structure of an aircraft, the wing experiences significantly different loads during takeoff, level flight, and landing. Its strain response directly reflects the structural stress characteristics. Accurately characterizing the wing's strain characteristics at different flight stages is a prerequisite for assessing the wing's structural condition.

[0003] Currently, time-domain statistical parameters are widely used for feature extraction and state characterization of structural strain signals. Dimensional parameters such as peak value, root mean square (RMS), and standard deviation characterize signal properties from dimensions such as amplitude limits, energy levels, and dispersion, while dimensionless parameters such as kurtosis and skewness are highly sensitive to the impact characteristics and asymmetry in the waveform. However, existing methods for analyzing wing strain data still have the following shortcomings:

[0004] First, as the number of flights accumulates, the amount of strain data continues to grow. The strain curve of a single flight is insufficient to reveal the state evolution of the structure in multiple flights, and it is difficult to efficiently identify potential abnormal flights from massive strain data by relying on manual comparison.

[0005] Second, existing methods mostly employ univariate analysis strategies, which analyze each statistical parameter separately. This makes it difficult to capture the synergistic relationship between multiple characteristic parameters and easily misses abnormal situations where individual characteristic parameters have not exceeded their limits but the relationship between the characteristics has deviated from the normal state.

[0006] Third, existing methods can only determine whether the strain state is abnormal, lack a mechanism for tracing the cause of the abnormality, and cannot further locate the dominant characteristic parameters that cause the abnormality after the abnormality is detected, making it difficult to meet the needs of rapid diagnosis and response in engineering.

[0007] Fourth, the load patterns of the wing differ significantly during takeoff, level flight, and landing. Existing methods lack phased modeling strategies for different flight phases and use a single model to uniformly analyze each flight phase, which makes it difficult to adapt to flight states with vastly different load patterns and affects the accuracy of anomaly identification.

[0008] Therefore, there is an urgent need to establish a method that can automatically extract state information from massive amounts of wing strain data, accurately identify abnormal flights, and trace the causes of abnormalities. Summary of the Invention

[0009] The present invention provides a method for characterizing and identifying wing strain data based on staged multivariate statistical process control in order to solve the problems existing in the prior art.

[0010] The technical solutions adopted in this invention are as follows:

[0011] A method for wing strain data characterization and anomaly identification based on staged multivariate statistical process control includes the following steps:

[0012] S1: Collect strain signals of the wing structure during flight and divide the flight process into takeoff, level flight and landing phases;

[0013] S2: For each flight phase, extract the six time-domain feature parameters of the strain signal for that flight phase, namely peak value, root mean square, standard deviation, peak-to-peak value, kurtosis and skewness, and construct a six-dimensional feature vector corresponding to that flight phase.

[0014] S3: For each flight phase, construct a normal sample feature matrix from the six-dimensional feature vectors of several normal flight operations for that flight phase. Calculate the sample mean vector and sample covariance matrix of the normal sample feature matrix to establish Hotelling's T for that flight phase. 2 A multivariate statistical control model was developed, and the upper limit of control for this flight phase at a specified significance level was determined.

[0015] S4: For each flight to be tested, extract the six-dimensional feature vector of the flight to be tested in each flight phase. Based on the sample mean vector and sample covariance matrix of the corresponding flight phase, calculate the T of the flight to be tested in each flight phase. 2 Statistic; the T 2 Compare the statistic with the corresponding control limit for the flight phase:

[0016] If the T 2 If the statistic is less than or equal to the control limit of the corresponding flight phase, the strain status of the flight to be tested in that flight phase is determined to be normal.

[0017] If the T 2 If the statistical value is greater than the control limit for the corresponding flight phase, the strain status of the flight to be tested in that flight phase is determined to be abnormal.

[0018] S5: For flight sorties whose strain condition is determined to be abnormal, the T value of the corresponding flight phase will be adjusted accordingly. 2 The statistic is decomposed into the contribution values ​​of each of the six time-domain feature parameters, and the one or more time-domain feature parameters with the largest contribution values ​​are identified as the dominant feature parameters causing the anomaly.

[0019] Furthermore, in S2, the time-domain sequence of the strain signal is denoted as... Let i = 1, 2, ..., N, where N is the number of sampling points. The formulas for calculating the six time-domain characteristic parameters are as follows:

[0020] Peak value: ,

[0021] Root mean square: ,

[0022] Standard deviation: ,

[0023] Peak-to-peak value: ,

[0024] kurtosis: ,

[0025] Skewness: ,

[0026] In the formula: For the i-th sampling point of the time-domain sequence, This represents the average value of the time-domain sequence.

[0027] Furthermore, in S3, the formulas for calculating the sample mean vector and the sample covariance matrix are as follows:

[0028] ,

[0029] ,

[0030] In the formula: phase refers to any one of the flight phases, including takeoff, level flight, or landing. The sample mean vector for this flight phase The m-th element, , The six-dimensional feature vector of the j-th normal flight during this flight phase. The m-th element, Let be the sample covariance matrix for this flight phase, k be the number of normal flight sorties, and p be the number of time-domain feature parameters.

[0031] Furthermore, in S3, the upper control limit is determined according to the following formula:

[0032] ,

[0033] In the formula: This represents the upper control limit for the flight phase, where p is the number of time-domain characteristic parameters and k is the number of normal flight sorties. Let α be the right-tail quantile of the F-distribution with degrees of freedom p and k−p, and let α be the specified significance level.

[0034] Furthermore, in S4, T2 The statistic is calculated using the following formula:

[0035] ,

[0036] In the formula: For the test flight, the T value during the flight phase is... 2 Statistic, This represents the six-dimensional feature vector of the flight sequence to be tested during this flight phase. This is the sample mean vector for this flight phase. This is the sample covariance matrix for this flight phase.

[0037] Furthermore, in S5, the contribution value of the m-th time-domain characteristic parameter is calculated according to the following formula:

[0038] ,

[0039] In the formula: Let m be the contribution value of the m-th time-domain feature parameter. Let m be the m-th element of the six-dimensional eigenvector of the flight segment to be tested during the flight phase. Let m be the m-th element of the sample mean vector for this flight phase. is the element in the m-th row and n-th column of the inverse of the sample covariance matrix for this flight phase, and p is the number of time-domain feature parameters.

[0040] Furthermore, in S1, for each flight, the strain signals for 15 minutes of takeoff, 15 minutes of level flight, and 15 minutes of landing are extracted respectively.

[0041] Furthermore, in S1, the strain signal is acquired by a fiber optic strain sensor located at the rear spar edge of the wing structure.

[0042] The present invention has the following beneficial effects:

[0043] (1) This invention extracts six time-domain feature parameters, namely peak value, root mean square, standard deviation, peak-to-peak value, kurtosis and skewness, and compresses the massive strain sampling data of each flight stage into a six-dimensional feature vector. It quantifies the strain characteristics of the wing structure from six dimensions, namely amplitude limit, energy level, dispersion, alternating span, impact intensity and load symmetry, and realizes multi-dimensional characterization of wing strain data. It overcomes the shortcomings of manual comparison in efficiently identifying anomalies from massive data.

[0044] (2) The present invention establishes Hotelling's T based on the six-dimensional feature vector of normal flight sorties. 2The multivariate statistical control model incorporates six time-domain characteristic parameters and their interrelationships into a unified statistical measure for overall judgment. It can capture the coordinated changes among multiple characteristic parameters, avoid the omission of abnormal combination relationships in univariate analysis, and improve the accuracy and automation of abnormal flight identification.

[0045] (3) This invention establishes independent multivariate statistical control benchmarks for the three flight stages of takeoff, level flight and landing, fully considers the differences in load patterns in each flight stage, overcomes the shortcomings of a single model that is difficult to adapt to the load characteristics of multiple stages, and ensures the reliability of anomaly identification in each flight stage.

[0046] (4) Based on the multivariate statistical control model, this invention introduces a contribution map mechanism to decompose the T² statistic of abnormal flights into the contribution values ​​of various time-domain characteristic parameters. This enables further tracing and locating of the dominant characteristic parameters that cause the abnormality after the abnormality is detected, providing a basis for engineers to make rapid diagnosis and effective response, and overcoming the shortcomings of existing methods that can only detect abnormalities but cannot locate the cause. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the wing spars structure;

[0048] Figure 2 This is a flowchart for diagnosing abnormal strain states of airfoil structures based on staged multivariate statistical process control;

[0049] Figure 3 T refers to the takeoff, level flight, and landing phases of a normal flight sortie. 2 Control charts;

[0050] Figure 4 These are T² control charts for each flight phase of the six flights to be tested;

[0051] Figure 5 This is a contribution graph of 3 abnormal flights. Detailed Implementation

[0052] The invention will now be further described with reference to the accompanying drawings.

[0053] The wing is the most important aerodynamic surface and main load-bearing structure of an aircraft, generating lift to enable flight. The wing structure mainly consists of components such as skin, spars, ribs, and stringers. For example... Figure 1 As shown, the wing spars, as the most important longitudinal load-bearing component of the wing, consist of a web and a flange, and bear all or most of the bending moment of the wing. In this embodiment, fiber optic strain sensors are deployed at the critical load-bearing parts of the wing—the flanges—to collect strain response data of the wing structure under different flight conditions.

[0054] like Figure 2 As shown, the wing strain data characterization and anomaly identification method based on staged multivariate statistical process control provided by this invention takes feature vector construction, strain data characterization, control limit establishment, anomaly identification and judgment, and cause diagnosis as its core links, forming a complete wing structure strain data characterization and anomaly diagnosis mechanism, specifically including the following steps:

[0055] Step S1: Flight phase division and strain data acquisition.

[0056] Strain signals from the wing structure were collected during flight, and the flight process was divided into takeoff, level flight, and landing phases. For each flight, strain signals were extracted for 15 minutes during takeoff, 15 minutes during level flight, and 15 minutes during landing for subsequent analysis. This extraction strategy, while ensuring coverage of the three key flight phases, reduced the amount of data analyzed per flight to approximately 30% of the complete flight record, effectively improving analysis efficiency.

[0057] Step S2: Construction of six-dimensional feature vectors.

[0058] For each flight phase, six time-domain feature parameters of the strain signal for that flight phase are extracted: peak value, root mean square, standard deviation, peak-to-peak value, kurtosis, and skewness. A six-dimensional feature vector corresponding to that flight phase is constructed to achieve multi-dimensional characterization of the wing structure strain data.

[0059] Let the time-domain sequence of the strain signal be . Let i = 1, 2, ..., N, where N is the number of sampling points. The formulas for calculating the six time-domain characteristic parameters are as follows:

[0060] Peak value: ,

[0061] Root mean square: ,

[0062] Standard deviation: ,

[0063] Peak-to-peak value: ,

[0064] kurtosis: ,

[0065] Skewness: ,

[0066] In the formula: For the i-th sampling point of the time-domain sequence, The average value of the time-domain sequence; among the six time-domain characteristic parameters, the peak value... Directly reflects the maximum instantaneous strain amplitude of the wing; root mean square The average energy level of the signal is reflected; the standard deviation σ reflects the severity of load fluctuations; the peak-to-peak value... The full amplitude reflecting the alternation of positive load; kurtosis skewness measures the impact intensity in a signal. It reflects the asymmetry of load distribution.

[0067] Step S3: Establish a phased multivariate statistical control model.

[0068] For each flight phase, a normal sample feature matrix is ​​constructed by using the six-dimensional feature vectors of several normal flight operations for that phase. The sample mean vector and sample covariance matrix of the normal sample feature matrix are then calculated to establish the Hotelling T for that flight phase. 2 Multivariate statistical control model (Hotelling's T) 2 This is an existing technique in the field of multivariate statistical process control, and the control upper limit for this flight phase at a specified significance level is determined. The specific process is as follows:

[0069] There are k normal flight sorties. For each sortie, p time-domain feature parameters are extracted at a certain flight phase. The six-dimensional feature vectors of all normal flight sorties at that flight phase are used to construct the normal sample feature matrix. :

[0070]

[0071] In the formula: The m-th element of the six-dimensional feature vector of the j-th normal flight is represented, which is the value of the m-th time-domain feature parameter of the j-th normal flight in this flight phase. , and These represent the takeoff phase, level flight phase, and landing phase, respectively.

[0072] Calculate the p-dimensional sample mean vector That is, for the characteristic matrix Calculate the mean of each column, where the m-th element is:

[0073] ,

[0074] Calculate the sample covariance matrix :

[0075]

[0076] In the formula, Let represent the p-dimensional feature vector of the j-th normal flight at this stage.

[0077] Expanding the above equation into a p×p matrix

[0078] ,

[0079] In the formula, its diagonal elements The off-diagonal elements represent the degree of dispersion of the m-th time-domain feature parameter from the mean. Let be the covariance between the m-th time-domain feature parameter and the n-th time-domain feature parameter.

[0080] After establishing the model, T for each normal flight can be calculated. 2 The statistics are used to draw control charts and validate the model. For the j-th normal flight, its T during that flight phase... 2 The statistic is:

[0081]

[0082] According to multivariate statistical process control theory, when the process is under control and the sample follows a multivariate normal distribution, T 2 Statistics follow Distribution, based on which the upper limit of control is determined. :

[0083]

[0084] In the formula: Let α be the right-tail quantile of the F-distribution with degrees of freedom p and k−p, and let α be the specified significance level.

[0085] Step S4: Anomaly identification of the flight to be tested.

[0086] For each flight to be tested, a six-dimensional feature vector is extracted for each flight phase. Based on the sample mean vector and sample covariance matrix of the corresponding flight phase, the T value of the flight to be tested for each flight phase is calculated using the following formula. 2 Statistic:

[0087] ,

[0088] Expanding the above equation, it can be equivalently expressed as:

[0089]

[0090] In the formula: For the test flight, the T value during the flight phase is... 2 Statistic; This represents the six-dimensional feature vector of the flight sequence to be tested during this flight phase. It is the value of its m-th element, that is, the m-th time-domain characteristic parameter of the flight to be tested in this flight phase; This is the sample mean vector for this flight phase. It is its m-th element; This is the sample covariance matrix for this flight phase. The inverse matrix of the sample covariance matrix for this flight phase The element in the m-th row and n-th column.

[0091] T 2 Compare the statistic with the corresponding control limit for the flight phase: If That is, T 2 If the statistic is less than or equal to the control upper limit of the corresponding flight phase, the strain status of the flight under test in that flight phase is determined to be normal; if That is, T 2 If the statistical value is greater than the control limit for the corresponding flight phase, the strain status of the flight to be tested in that flight phase is determined to be abnormal.

[0092] Step S5: Trace the cause of the anomaly.

[0093] For flight sorties whose strain state is determined to be abnormal, a contribution plot (a current technique in the field of multivariate statistical process control) is introduced to visualize the flight sortie's T-segment in the corresponding flight phase. 2 The statistic is decomposed into the contribution values ​​of six time-domain feature parameters. By comparing the contribution values ​​of each time-domain feature parameter, the one or more time-domain feature parameters with the largest contribution values ​​are identified as the dominant feature parameters causing the anomaly.

[0094] The m-th time-domain feature parameter on T 2 The contribution value of a statistic is defined as:

[0095] ,

[0096] The meanings of the symbols in the formula are the same as in step S4.

[0097] This embodiment takes the rear sparsity edge of a certain type of aircraft wing as the research object and provides a detailed description of the method proposed in this invention.

[0098] Step 1: Divide the flight process of the wing structure into the takeoff phase, level flight phase and landing phase. For each flight of the wing rear spars, extract the strain signals for 15 minutes of the takeoff phase, 15 minutes of the level flight phase and 15 minutes of the landing phase. Establish independent multivariate statistical control benchmarks for each of the three flight phases.

[0099] Step 2: For each flight phase, extract six time-domain feature parameters of the strain signal for each flight phase: peak value, root mean square, standard deviation, peak-to-peak value, kurtosis, and skewness. Based on the six-dimensional time-domain feature vectors of 30 normal flight sorties in each flight phase, the strain of the wing structure in different flight phases is characterized.

[0100] Step 3: Create Hotelling's T 2 The process of establishing a multivariate statistical control model and determining the upper control limit is as follows:

[0101] Step 3.1: Construct a normal sample feature matrix Xphase (30×6 dimensions) from the six-dimensional feature vectors of 30 normal flight sorties of the wing structure in a certain flight phase, i.e., k=30, p=6;

[0102]

[0103] in, This represents the value of the j-th normal flight on the m-th feature. , and These represent the takeoff, level flight, and landing phases, respectively.

[0104] Step 3.2: Calculate the 6-dimensional sample mean vector ,in m = 1, 2, ..., 6;

[0105] Step 3.3: Calculate the sample covariance matrix:

[0106] .

[0107] Step 3.4: Determine the upper limit of control. In this embodiment, the upper limit of control is determined at a 99% confidence level (i.e., a significance level of α = 0.01):

[0108] .

[0109] like Figure 3 As shown, the T² values ​​for each normal flight of the wing during takeoff, level flight, and landing are all lower than the upper control limit (UCL=26.58) at the 99% confidence level, indicating that the constructed multivariate statistical control model can effectively characterize the strain fluctuation range under normal conditions.

[0110] Step 4: Based on the above normal state benchmark, anomaly diagnosis is performed on the six flight sorties to be tested (including three normal sorties and three abnormal sorties). The six-dimensional time-domain feature vectors of each sortie during takeoff, level flight, and landing are extracted. Based on the normal sample mean vector and covariance matrix derived in Step 3, the T0 values ​​for each sortie during each flight stage are calculated. 2 Statistics, T 2 Compare the statistic with the corresponding control upper limit for the flight phase: If T 2 If the statistic is less than or equal to the upper control limit, the response status of that sortie in that flight phase is considered normal; if T 2If the statistical value exceeds the upper limit of control, it is considered abnormal. The detailed steps are as follows:

[0111] Step 4.1: Extract the six-dimensional temporal feature vectors for each flight during takeoff, level flight, and landing phases:

[0112]

[0113] In the formula, This represents the m-th characteristic value of the test flight during this flight phase.

[0114] Step 4.2: Calculate T for each test flight at each stage. 2 Statistic:

[0115]

[0116] like If the wing's strain condition during that flight phase is normal, then the wing's strain condition during that flight is determined to be normal; if If so, it is judged as abnormal.

[0117] like Figure 4 As shown, the T values ​​for each flight phase of sorties 1, 4, and 5 are... 2 All values ​​were below the upper limit of control at the 99% confidence level (UCL=26.58), and were therefore considered normal; T values ​​during the level flight phase of flights 2, 3, and 6 were... 2 The values ​​all exceeded the control limit and were identified as abnormal, indicating that the load conditions of the above flights deviated from the normal baseline during the level flight phase.

[0118] Step 5: Based on the flight sorties identified as abnormal in Step 4, a contribution map is introduced to decompose the T² statistic into the contribution values ​​of each original feature parameter. The dominant feature parameter causing the abnormality is traced according to the magnitude of the contribution value.

[0119] Figure 5 The contribution plots for characteristic parameters of the three abnormal flights are shown. For flight 2, the contribution plot during the level flight phase shows that the main cause of the anomaly is the peak-to-peak value and standard deviation, indicating that the alternating load range of this flight was expanded and the fluctuations were more severe than normal. For flight 3, the contribution plot during the level flight phase shows that the main cause of the anomaly is the standard deviation, indicating that the load dispersion during the level flight phase of this flight was abnormally increased. For flight 6, the contribution plot during the level flight phase shows that the main cause of the anomaly is the standard deviation and peak-to-peak value, indicating that this flight exhibited both increased load fluctuations and expanded alternating range.

[0120] In summary, the anomalies in all three flights were concentrated during the level flight phase, but their dominant characteristic parameters differed, reflecting different types or degrees of unexpected load disturbances. Contribution plot analysis allowed for the accurate identification of the dominant characteristic parameters causing the anomalies, providing a basis for engineers to quickly diagnose and effectively respond.

[0121] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A method for characterizing and identifying anomalies in wing strain data based on staged multivariate statistical process control, characterized in that: Includes the following steps: S1: Collect strain signals of the wing structure during flight and divide the flight process into takeoff, level flight and landing phases; S2: For each flight phase, extract the six time-domain feature parameters of the strain signal for that flight phase, namely peak value, root mean square, standard deviation, peak-to-peak value, kurtosis and skewness, and construct a six-dimensional feature vector corresponding to that flight phase. S3: For each flight phase, construct a normal sample feature matrix from the six-dimensional feature vectors of several normal flight operations for that flight phase. Calculate the sample mean vector and sample covariance matrix of the normal sample feature matrix to establish Hotelling's T for that flight phase. 2 A multivariate statistical control model was developed, and the upper limit of control for this flight phase at a specified significance level was determined. S4: For each flight to be tested, extract the six-dimensional feature vector of the flight to be tested in each flight phase. Based on the sample mean vector and sample covariance matrix of the corresponding flight phase, calculate the T of the flight to be tested in each flight phase. 2 Statistic; the T 2 Compare the statistic with the corresponding control limit for the flight phase: If the T 2 If the statistic is less than or equal to the control limit of the corresponding flight phase, the strain status of the flight to be tested in that flight phase is determined to be normal. If the T 2 If the statistical value is greater than the control limit for the corresponding flight phase, the strain status of the flight to be tested in that flight phase is determined to be abnormal. S5: For flight sorties whose strain condition is determined to be abnormal, the T value of the corresponding flight phase will be adjusted accordingly. 2 The statistic is decomposed into the contribution values ​​of each of the six time-domain feature parameters, and the one or more time-domain feature parameters with the largest contribution values ​​are identified as the dominant feature parameters causing the anomaly.

2. The method for wing strain data characterization and anomaly identification based on staged multivariate statistical process control as described in claim 1, characterized in that: In S2, the time-domain sequence of the strain signal is denoted as... Let i = 1, 2, ..., N, where N is the number of sampling points. The formulas for calculating the six time-domain characteristic parameters are as follows: Peak value: , Root mean square: , Standard deviation: , Peak-to-peak value: , kurtosis: , Skewness: , In the formula: For the i-th sampling point of the time-domain sequence, This represents the average value of the time-domain sequence.

3. The method for wing strain data characterization and anomaly identification based on staged multivariate statistical process control as described in claim 1, characterized in that: In S3, the formulas for calculating the sample mean vector and the sample covariance matrix are as follows: ， ， In the formula: phase refers to any one of the flight phases, including takeoff, level flight, or landing. The sample mean vector for this flight phase The m-th element, , The six-dimensional feature vector of the j-th normal flight during this flight phase. The m-th element, Let be the sample covariance matrix for this flight phase, k be the number of normal flight sorties, and p be the number of time-domain feature parameters.

4. The method for wing strain data characterization and anomaly identification based on staged multivariate statistical process control as described in claim 1, characterized in that: In S3, the upper limit of control is determined according to the following formula: ， In the formula: This represents the upper control limit for the flight phase, where p is the number of time-domain characteristic parameters and k is the number of normal flight sorties. Let α be the right-tail quantile of the F-distribution with degrees of freedom p and k−p, and let α be the specified significance level.

5. The method for wing strain data characterization and anomaly identification based on staged multivariate statistical process control as described in claim 1, characterized in that: In S4, T 2 The statistic is calculated using the following formula: ， In the formula: For the test flight, the T value during the flight phase is... 2 Statistic, This represents the six-dimensional feature vector of the flight sequence to be tested during this flight phase. This is the sample mean vector for this flight phase. This is the sample covariance matrix for this flight phase.

6. The method for wing strain data characterization and anomaly identification based on staged multivariate statistical process control as described in claim 1, characterized in that: In S5, the contribution value of the m-th time-domain characteristic parameter is calculated according to the following formula: ， In the formula: Let m be the contribution value of the m-th time-domain feature parameter. Let m be the m-th element of the six-dimensional eigenvector of the flight segment to be tested during the flight phase. Let m be the m-th element of the sample mean vector for this flight phase. is the element in the m-th row and n-th column of the inverse of the sample covariance matrix for this flight phase, and p is the number of time-domain feature parameters.

7. The method for wing strain data characterization and anomaly identification based on staged multivariate statistical process control as described in claim 1, characterized in that: In S1, for each flight, the strain signals for 15 minutes during takeoff, 15 minutes during level flight, and 15 minutes during landing are extracted respectively.

8. The method for wing strain data characterization and anomaly identification based on staged multivariate statistical process control as described in claim 1, characterized in that: In S1, the strain signal is collected by a fiber optic strain sensor located on the rear spar edge of the wing structure.

Images