A device anomaly detection method for wind-induced deformation measurement of large deployable structures

Abstract

The present application relates to the field of structural health detection, and particularly relates to a device abnormality detection method for wind-induced deformation measurement of large deployable structures, comprising the following steps: obtaining time series response data of a healthy structure under wind load excitation to form baseline reference data; and normalizing the data; combining empirical mode decomposition and Hurst index analysis to carry out signal denoising; establishing a Hankel matrix of each channel data based on the denoised data; calculating the orthogonal projection vector of the matrix; obtaining a state matrix and an output matrix of the large deployable structure through matrix decomposition; and calculating a reference mode matrix of the structure in a healthy state; in the service process of the structure, data acquisition, preprocessing and signal denoising are carried out in the same way as the above steps to obtain a current mode matrix corresponding to the service state; the distance between the current mode matrix and the reference mode matrix is calculated; the damage position of the deployable structure is determined by comparing the distance changes of each order; and the abnormality detection of the structure is realized.

Description

Technical Field

[0001] This invention relates to the field of structural health testing, and more specifically to a method for detecting equipment anomalies in the measurement of wind-induced deformation of large deployable structures. Background Technology

[0002] Large-scale engineering structures are widely used in aerospace, heavy machinery, civil engineering, and other fields. Alternating vibrations, sudden impacts, and fatigue loads generated in their working environments pose significant challenges to the safe and healthy service life of these structures. To ensure structural service life, formulate operation and maintenance strategies, and prevent major safety accidents, it is necessary to research corresponding structural anomaly detection methods (health monitoring methods) to determine the structural condition in real time using sensor data.

[0003] Structural anomaly detection methods can be divided into two categories: model-driven and data-driven. Model-driven methods establish a descriptive model of the target object using structural dynamics knowledge and judge the structural health status based on changes in model parameters. However, due to the difficulty in accurately modeling complex structures in the real physical world, model-driven methods have significant limitations. Data-driven methods preprocess, extract features, and perform pattern recognition on response signals obtained from sensor networks to construct a mapping relationship between recognition results and structural states, and are currently the mainstream approach. Data-driven structural health detection methods face two main problems. First, the accuracy of state assessment is greatly affected by factors such as data quality, sample size, and feature extraction methods; measurement errors exist between sensor measurements and actual physical quantities of the structure during actual field detection. Furthermore, data on structural damage in real-world scenarios is often difficult to obtain, making methods like deep learning, which require complete data samples for training, prone to failure.

[0004] This invention aims to propose a data-driven structural detection method that addresses any of the aforementioned problems by requiring only structural health operation data and eliminating the need for prior acquisition of samples from different damage states. The goal is to achieve unsupervised damage identification of large deployable structures under wind loads.

[0005] Specifically, this invention provides a method for detecting equipment anomalies in measuring wind-induced deformation of large deployable structures, characterized by the following steps:

[0006] S1: Obtain the time-series response data of a healthy large deployable structure under wind load excitation to form a baseline reference dataset;

[0007] S2: Normalize the baseline reference data to form a normalized baseline reference dataset;

[0008] S3: Denoising detection data by combining empirical mode decomposition and Hearst exponent analysis;

[0009] S4: Establish the Hankel matrix for each channel based on the noise-reduced data. and , and Let be the number of rows and columns of the matrix. and These represent historical data and future data, respectively, with a specific point in time as the dividing line. Where m is the number of sensing points, n is the number of rows in the matrix, and m is the number of columns in the matrix.

[0010] S5: Calculate the matrix exist The orthogonal projection vectors on the matrix are used to obtain the state matrix and output matrix of the large developable structure through matrix decomposition, and the reference mode matrix of the structure in the healthy state is calculated.

[0011] S6: During the service of the large deployable structure, data acquisition, preprocessing, and signal noise reduction are performed in the same manner as steps S1 to S5 to obtain the current mode shape matrix corresponding to the service state;

[0012] S7: Calculate the distance between the current mode shape matrix and the reference mode shape matrix, and determine the location of damage in the developable structure by comparing the changes in distance at each order, thereby realizing structural anomaly detection.

[0013] In one implementation, the baseline reference dataset in step S1 contains signals from N channels, i.e.:

[0014]

[0015] In one implementation, the normalization process of the baseline reference data in step S2 can be expressed as follows:

[0016] ,

[0017] In the formula For the first Data from each channel, and The maximum and minimum values ​​among all channel data. For the first Normalized samples for each channel The normalization factor is the normalized baseline reference dataset. It can be represented as:

[0018] ,

[0019] In one implementation, the empirical mode decomposition in step S3 includes targeting For each channel signal, calculate the Intrinsic Mode Function (IMF) of different orders for each channel;

[0020] In one implementation, the Hearst exponent analysis includes obtaining the Hearst exponents of intrinsic mode functions (IMFs) of different orders through detrended volatility analysis (DFA).

[0021] In one implementation, the noise reduction of detection data by combining empirical mode decomposition and Hurst exponent analysis includes extracting... The local extreme points of each channel signal are obtained through interpolation. upper and lower envelopes and ; Calculate the envelope mean curve Then, the residual signal is solved. After determining that the residuals satisfy the convergence criterion, the iteration is recorded. The subsequent residuals The first intrinsic mode function After stepwise decomposition, we obtain The intrinsic mode matrix composed of the first-order intrinsic mode functions The Hurst exponent is used to describe the intrinsic mode functions (IMFs) of different orders. After filtering out the IMFs corresponding to noise terms, the remaining IMFs are summed to obtain the denoised signal. , can be represented as:

[0022] ,

[0023] In one implementation, in step S4, the first The Hankel matrix corresponding to the noise-reduced signal of each channel and It can be represented as:

[0024] ,

[0025] ,

[0026] In the formula For the first The sensor data of the first channel is in the... The value of the moment. and These represent the number of rows and columns of the matrix, respectively.

[0027] In one implementation, in step S5, the first Orthogonal projection vectors corresponding to the detection data of each channel It can be represented as:

[0028] ,

[0029] against use Decomposition and Singular Value Decomposition The equation can be obtained as follows:

[0030] ;

[0031] in and These are the system observation matrix and the output matrix, respectively. Based on matrix analysis, we can obtain:

[0032] ,

[0033] ,

[0034] The mode shape under the reference state can be obtained by decomposing the eigenvalues ​​of the observation matrix. :

[0035] ,

[0036] ,

[0037] In the formula, The system is a discrete-time diagonal matrix. Let be the complex eigenvector matrix of the system.

[0038] In one implementation scheme, the location of structural damage anomalies is determined by comparing the difference between the current mode matrix and the reference mode matrix using the following formula;

[0039] ,

[0040] in The current mode shape matrix, This is the reference matrix.

[0041] In one implementation, the calculation of the normalized baseline reference dataset also takes into account transfer accuracy.

[0042] The solution of the present invention has the following effects.

[0043] This invention requires only data from structurally healthy states to construct a baseline dataset. Structural anomaly detection can be achieved by correlation matching with the modal matrix of measured data. A transfer model between strain measured by fiber optic sensors and structural strain is considered to compensate for errors in the detection signal (wherein, the strain signal acquired by the fiber optic grating sensor is followed by the structural deformation calculated based on the structural modulus parameters). This method is an unsupervised health detection method that requires no training and has broad algorithmic applicability. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the composition of the large deployable structure of the present invention;

[0045] Figure 2 This is a flowchart of the identification method of the present invention;

[0046] Figure 3 This is a structural diagram of the surface-mount fiber optic sensor of the present invention;

[0047] Figure 4 This is the finite element simulation model of the large deployable structure of this invention;

[0048] Figure 5 This is a schematic diagram of the measuring point arrangement for the large deployable structure of the present invention;

[0049] Figure 6 is a structural diagram of the column skeleton beam with crack damage according to the present invention;

[0050] Figure 7 Figure 6 shows a cross-sectional view of the crack and the surrounding structure.

[0051] Figure 8 shows the strain transmissivity of the bonded fiber Bragg grating sensor with different core 2022 dimensions according to the present invention: (a) Core 2022 inner diameter 62.5 (b) Fiber core 2022, inner diameter 12.5 mm (c) Fiber core 2022 inner diameter 25 (d) Core 2022, inner diameter 37.5 mm (e) Core 2022 inner diameter 50 (f) Core 2022 inner diameter 75 ;

[0052] Figure 9 shows the detection data of measurement point 1 after adding noise according to the present invention, where (a) is the original signal and (b) is the noise-added signal;

[0053] Figure 10 The noise-added signal at measurement point 1 of this invention is of each order. picture;

[0054] Figure 11 This is a comparison diagram of the original signal and the denoised signal of this invention;

[0055] Figure 12 shows the abnormal detection results of health indicators under different column skeleton beam structure damage states of the present invention, where A is a diagram of beam No. 1; B is a diagram of beam No. 2; C is a diagram of beam No. 3; D is a diagram of beam No. 4; E is a diagram of beam No. 5; and F is a diagram of beam No. 6.

[0056] In the picture: Detailed Implementation

[0057] To make the technical solutions and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Some technical terms and expressions used herein have the same meaning as understood by those skilled in the art to which this application pertains.

[0058] like Figure 1 As shown, the large folding and unfolding mechanism of the present invention includes a scissor mechanism 100, a frame 200, and multi-stage telescopic arms. The multi-stage telescopic arms are driven by multi-stage cylinders and integrated in a multi-stage nested manner. The scissor mechanism 100 is connected to each stage of the telescopic arms via equally spaced pins. When the multi-stage cylinders extend, they drive the scissor mechanism 100 to unfold horizontally. The ends of the multi-stage cylinders are connected to a large back frame structure, which can be simplified as a fixed support end.

[0059] like Figure 2 The present invention provides a method for detecting equipment anomalies in measuring wind-induced deformation of large deployable structures, comprising the following steps:

[0060] S1: Obtain the time-series response data of the healthy structure under wind load excitation to form baseline reference data;

[0061] S2: Normalize the baseline reference data to form a normalized baseline reference dataset;

[0062] S3: Denoising of detection data is carried out by combining Empirical Mode Decomposition (EMD) and Hurst exponent analysis;

[0063] S4: Establish the Hankel matrix for each channel based on the noise-reduced data. and , and Let be the number of rows and columns of the matrix. m is the number of sensing points, m is the number of rows in the matrix, and n is the number of columns in the matrix;

[0064] S5: Calculate the matrix exist orthogonal projection vector on The state matrix of the large expandable structure is obtained through matrix decomposition. and output matrix The modal matrix of the structure under the reference state is calculated. ;

[0065] S6: During the structural service process, data acquisition, preprocessing, and signal noise reduction are performed in the same manner as steps S1 to S5 to obtain the mode shape matrix corresponding to the current state. ;

[0066] S7: Calculate the current mode shape matrix and reference formation matrix By comparing the distances between different orders, the location of damage to the expandable structure can be determined, thereby enabling structural anomaly detection.

[0067] The proposed solution acquires multi-channel strain signals of large deployable structures under wind loads (the strain signals obtained through fiber optic grating sensors and the structural deformation can be calculated based on structural modulus parameters), reducing measurement noise introduced by environmental factors and minimizing measurement errors caused by sensor arrangement and packaging configuration. This enables accurate acquisition of overall structural strain response information and precise solution of physical parameters, allowing for the identification of abnormal states in large deployable structures. This method can achieve structural anomaly detection based solely on response signals, without requiring detection data under damaged conditions or pre-training and optimization processes; it is an unsupervised method. It can provide a theoretical basis for structural optimization design, service status assessment, and remaining life prediction. This invention is verified through a specific implementation scheme using finite element simulation examples. Compared with other common data-driven methods, this method has advantages such as fast computation, strong robustness, and high accuracy.

[0068] In one implementation, the baseline reference dataset in step S1 contains signals from multiple channels, namely:

[0069]

[0070] Specifically, sensors are deployed at multiple locations on the large deployable structure under initial health conditions and connected to a multi-channel signal acquisition system to acquire real-time temporal response data of different locations on the deployable structure under wind load excitation, forming baseline reference data. ;

[0071] In one implementation, the normalization process of the baseline reference data in step S2 can be expressed as: The signals from each channel are preprocessed and normalized. Samples from each channel It can be represented as:

[0072]

[0073] In the formula For the first Data from each channel, and The maximum and minimum values ​​among all channel data. For the first Normalized samples for each channel This is the normalization factor. The normalized baseline reference dataset. It can be represented as:

[0074]

[0075] When measuring strain signals during structural service using fiber Bragg grating sensors, measurement errors can occur due to factors such as installation method and packaging configuration. Therefore, a strain transfer model needs to be considered to compensate for the sensor signal.

[0076] In one implementation, the empirical mode decomposition in step S3 includes targeting For each channel signal, calculate the Intrinsic Mode Function (IMF) of different orders for each channel.

[0077] Specifically, first extract Local extreme point sequence of signals in each channel and The extreme points are processed by cubic spline interpolation to obtain... upper and lower envelopes and .

[0078]

[0079] In the above formula, , , and These are coefficient terms of different orders. After determining the computational step size, a system of equations is solved using the constraint that the first and second derivatives of the polynomial are the same at different nodes. , , and It can be represented as:

[0080]

[0081] In the above formula Given a tridiagonal matrix, the nodal values ​​and endpoint conditions can be substituted into the matrix equation, and Gaussian elimination can be used to obtain the mean envelope curve. , can be represented as:

[0082]

[0083] The difference between the reference data and the envelope mean As shown below:

[0084]

[0085] Update via iterative loop After the convergence criterion is satisfied, let the updated residual be... This is the first-tier IMF. minus The subsequent loop performs calculations to obtain IMFs of different orders and establishes the intrinsic mode matrix. ,in It refers to the order.

[0086] In one implementation, the Hurst index analysis includes obtaining the Hurst index for different orders of IMFs through detrended volatility analysis (DFA).

[0087] Specifically, with the first First-order intrinsic mode function For example, calculate its cumulative summation deviation. :

[0088]

[0089] Will Divided into A length of non-overlapping sequences, where Subsequently, regarding the first The fitted values ​​of each segmented sequence were calculated using the least squares method. , length is The difference between the original signal and the fitted value As shown below:

[0090]

[0091] This is a detrended time series. Then, by calculating the... The squared mean of all data points in a segmented sequence can be expressed as:

[0092]

[0093] Calculate the average and square root of all segmented samples to obtain the DFA fluctuation function. :

[0094]

[0095] By changing the length of the segmented intervals , obtain Follow The logarithmic coordinate curve of the curve can be fitted to a function that can be expressed as:

[0096]

[0097] slope in the above formula This refers to the Hurst exponent. Since the Hurst exponent of the wind load spectrum is generally between 0.7 and 1.0, the IMF corresponding to a Hurst exponent less than 0.5 is considered a noise term. Let the order sequence corresponding to the noise term be... ,and The noise-reduced detection data is shown below:

[0098]

[0099] By performing the same operation on all channels of data, the noise-reduced dataset can be obtained. .

[0100] In one implementation, the discretized state equation used in step S4 can be written as:

[0101]

[0102] In the formula The discrete state space matrix, For discrete input matrices, For discrete output matrices, This is a direct-feed matrix. Assume a noise term. and To obtain white noise, simplifying the above equation yields:

[0103]

[0104] In the formula, Characterizes the superposition of model error, external random excitation, and process noise. This is the sum of observation noise and external random excitation. The Hankel matrix corresponding to the noise reduction signal of each channel and It can be represented as:

[0105]

[0106]

[0107] In the formula For the first The sensor data of the first channel is in the... The value of the moment. and These represent the number of rows and columns of the matrix, respectively.

[0108] In one implementation, step S5 is specifically performed as follows: calculating the matrix. exist orthogonal projection vector on As shown in the following formula:

[0109]

[0110] According to Kalman filter estimation, state-space equations, and matrix analysis, the orthogonal projection vector can be expressed as the product of the system's state matrix and output matrix. For... use Decomposition and Singular Value Decomposition The equation can be obtained as follows:

[0111]

[0112] in and These are the system observation matrix and the output matrix, respectively. Based on matrix analysis, we can obtain:

[0113]

[0114]

[0115] The mode shape under the reference state can be obtained by decomposing the eigenvalues ​​of the observation matrix. :

[0116]

[0117]

[0118] In the formula, Let be the system's discrete-time diagonal matrix. Let be the complex eigenvector matrix of the system.

[0119] Furthermore, the mode shape matrix corresponding to the current state is calculated in the same manner. In step 6, by comparison and The difference between the two can be used to determine the location of structural damage anomalies.

[0120]

[0121] In one implementation, the sensor is a surface-mount fiber optic sensor. Such sensors are more sensitive and accurate, and are easy to install. Specifically, such as... Figure 3 As shown, the surface-mount fiber optic sensor includes a substrate 201, an optical fiber 202, and an adhesive 203. The optical fiber 202 includes a core 2022 and a protective layer 2021. The protective layer 2021 surrounds the outer periphery of the core 2022 and is connected to the substrate 201 by the adhesive 203. A grating region 2023 is disposed on the optical fiber 202. The length of the grating strain measurement point region of the optical fiber is 2L, and the radii of the core 2022 and the protective layer 2021 are respectively... and The fiber optic sensor is constantly subjected to strain along the axial direction.

[0122] In one implementation, in addition to considering the selection of sensor type to improve detection accuracy as described above, the detection method also considers transmission accuracy.

[0123] Specifically

[0124] Along the fiber optic sensor axis Establish force balance equations based on direction:

[0125]

[0126] Transforming formula (26), we obtain the following expression:

[0127]

[0128] Simplifying the above equation, we get:

[0129]

[0130] Because the length of the fiber optic sensor is much larger than the diameter of the 2022 fiber core, that is Therefore, the first term in equation (28) can be ignored, and the protective layer 2021 and the fiber core 2022 are at any point. The shear stress at the point can be expressed as:

[0131]

[0132] The shear stress value can also be determined by the compatibility condition of the axial deformation of the optical fiber, as shown in the following formula:

[0133]

[0134] In the formula, , and These represent the deformations of the substrate material, the fiber core 2022, and the protective coating, respectively. The above equation shows that the displacement of the substrate 201 is the sum of the shear deformation of the protective coating and the axial displacement of the fiber. Since all materials are within the linear elastic range, Hooke's theorem can be used to obtain:

[0135]

[0136] in , The shear strain and shear modulus of the protective layer 2021.

[0137] According to the principles of elasticity, we know that... ,in and These represent the Young's modulus and Poisson's ratio of the protective layer 2021 material, respectively. Under the assumption of small deformation, the shear strain of the protective layer 2021 can be obtained. Combining formula (29), the deformation of the protective layer 2021 can be expressed as follows:

[0138]

[0139] The axial stresses of the matrix 201 material and the fiber core 2022 are as follows, where... , , and , , The values ​​represent the stress, elastic modulus, and strain of the matrix material 201 and the fiber core 2022, respectively.

[0140]

[0141]

[0142] The axial deformations of the matrix 201 and the fiber core 2022 can be obtained by integrating the strains of both along the axial direction:

[0143]

[0144]

[0145] In the formula The axial tensile force on fiber core 2022 can be obtained by the following formula:

[0146]

[0147] Substituting the expressions on the right side of equations (32), (35), and (36) into the deformation compatibility equation (30) and simplifying it, we obtain the following integral equation:

[0148]

[0149] Differentiating the above equation, we get:

[0150]

[0151] According to the strain compatibility condition, it can be known that... Therefore, the leftmost and last terms in the above equation can be canceled out, thus yielding:

[0152]

[0153] Differentiating the above equation yields the following equation:

[0154]

[0155] in:

[0156]

[0157] The above equation is a typical second-order homogeneous linear differential equation with constant coefficients. According to the solution method for this type of differential equation, its general solution form is as follows:

[0158]

[0159] Since the strain-sensitive section of the fiber Bragg grating sensor is concentrated in the middle of grating region 2023, the boundary condition of the fiber axial tension can be obtained as follows:

[0160]

[0161]

[0162] The above boundary conditions indicate that the strain of the fiber grating core 2022 at its axis of symmetry (and the middle part of the sensor) is the same as the strain of the substrate material, and gradually decreases to zero at both ends. Substituting equations (44) and (45) into equation (43) and solving the system of equations, the integral constant can be obtained. and for:

[0163]

[0164] obtained and Substituting the general solution (18) yields the shear stress distribution function describing the interface between fiber core 2022 and fiber optic protective layer 2021:

[0165]

[0166] Substituting the above formula into the axial tensile force formula, we can obtain the expression for the axial tensile force of fiber core 2022:

[0167]

[0168] Therefore, the axial tensile force per unit area is:

[0169]

[0170] Dividing both sides of the formula by the elastic modulus of fiber core 2022, we can obtain the strain transfer rate formula between fiber core 2022 along the axial direction and matrix 201:

[0171]

[0172] set up The range of values ​​is Then the average transfer rate of the strain-sensitive region of the fiber grating can be obtained. :

[0173] .

[0174] In summary, the final strain transfer modulus of the fiber Bragg grating sensor can be represented by the following set of equations.

[0175]

[0176] The actual surface strain of the structure can be expressed as:

[0177]

[0178] In the formula, the common parameters of fiber Bragg gratings are: , , , , ,but The average strain transmissibility is .

[0179] but, .

[0180] Then, using replace Continue with the identification steps described above.

[0181] The following describes the working process and verification of the unsupervised damage identification and detection method for large deployable structures, using specific embodiments.

[0182] Figure 1 The large deployable antenna structure of the present invention shown is illustrated in the finite element model of the large deployable structure established using ANSYS, as follows: Figure 4 As shown, the length, width, and thickness of the 3D truss scissor mechanism in this simulation example are 8540mm and 6000mm, respectively. Structural 45 steel is selected as the structural material. Specific parameters are shown in Table 1. The entire structure is meshed using SOLID185 solid elements, with a mesh size of 54021.

[0183] Table 1 Material Properties

[0184]

[0185] II. Selection of Candidate Measurement Points and Dynamic Response of Structures under Wind Load

[0186] The first to sixth beams were selected as the research objects, and two measuring points were evenly set on each beam. The arrangement of the measuring points is as follows: Figure 5 As shown in Figure 6, a side view of the large deployable structure along its extension direction is presented. The entire deployable structure consists of a multi-stage telescopic cylinder, a scissor mechanism, connecting pins, and a column frame. The multi-stage cylinder drives the scissor mechanism for free extension and retraction in this direction. The column frame is 6000 mm long and 40 mm high. Structural damage is simulated by introducing cracks into the column frame. The length and depth of the cracks are 50 mm and 20 mm, respectively. Figure 7 As shown. Wind loads were applied perpendicular to the plane of the deployable structure, and dynamic response analysis was performed. First, response data were acquired at 12 measuring points in a healthy, undamaged state. Then, the first to sixth column frames were alternately replaced with beams containing cracks. The large deployable structure with healthy and damaged column frames is shown below. Figure 5 As shown, multi-channel damage data at different locations were obtained through sequential dynamic numerical calculations.

[0187] 3. Determine the strain transmissibility and perform error compensation based on the fiber optic sensor configuration.

[0188] The strain measurement region length, fiber core 2022 radius, protective layer 2021 radius, and grating region 2023 length of the fiber optic sensor all affect the actual strain transmissibility. Figure 8 shows the transmissibility for different fiber core 2022 sizes and strain measurement region lengths.

[0189] As shown in the figure, the sensor's measurement accuracy is correlated with its length. A larger detection area results in a flatter strain transmissibility near the axis of symmetry, allowing more structural strain to be transferred to the fiber core 2022, which is more beneficial for reducing measurement errors. Changing the inner diameter of the fiber core 2022 reveals that a smaller cross-sectional area of ​​the fiber grating increases its sensitivity to strain changes, leading to a higher strain transmissibility. Conversely, a larger diameter of the fiber optic protective layer 2021 significantly reduces the transmissibility at different locations. Therefore, while the protective layer 2021 improves the survival rate of the fiber optic sensor, it should not be too thick to avoid affecting measurement accuracy. In this example, the inner diameter of the fiber core 2022 is 62.5 mm. The total length of the optical fiber 2L is 60mm, and the length of the grating region 2023 is 20mm. After determining the transmission rate, error compensation is achieved according to formula (53) to obtain the actual strain of the structure.

[0190] III. Multi-channel data preprocessing and filtering / noise reduction

[0191] against The samples in the data are normalized according to equation (2). The structural dynamic response obtained by ANSYS numerical simulation does not contain any noise, which is inconsistent with the actual detection scenario. In order to verify the proposed data denoising method, the data is processed by ANSYS numerical simulation. Introducing periodic fluctuations and Gaussian noise, as shown below:

[0192]

[0193] In the formula, It is Gaussian noise. and These are the amplitude coefficients of periodic disturbance and Gaussian noise, respectively. The data samples are after adding noise. After normalizing the samples according to equation (2), the EMD algorithm is used to obtain the intrinsic mode functions of each order. For this example, the order is decomposed. Therefore, the intrinsic mode functions obtained from the decomposition of data in each channel can be represented as Taking the detection data of measurement point 1 after adding noise as an example, its original data and noisy data are shown in Figure 9. The decomposed IMFs of each order are... Figure 10 The information is provided in the text.

[0194] The Hurst exponents obtained according to equations (9) to (13) are shown in Table 2. From the table, it can be seen that... ~ The Hurst exponent is less than 0.5, indicating noise interference. Using equation (14)... ~ The noise-reduced signal is obtained by summing the results. A comparison of the original signal and the noise-reduced signal is shown below. Figure 11 As shown, the two have a good degree of agreement.

[0195] Table 2 Hurst index for IMFs of different orders

[0196]

[0197] IV. Calculate local vibration modes to achieve structural damage detection.

[0198] Following the same steps, All data are processed, and mode shape information is obtained using the parameter identification method of equations (15)-(24). To identify structural anomaly areas, the mode shape of each beam is calculated using samples from two measuring points on each beam. The mode shape matrix of all beams under both the reference and damaged states can then be expressed as follows:

[0199]

[0200]

[0201] In the above formula, In this example, six operating states were obtained through numerical simulation, corresponding to the damage of beams 1 through 6. For the vibration modes obtained on each beam, the mean of the reference vibration mode and the vibration mode obtained from the actual detection data were subtracted to determine the vibration mode under different states. The ratio of the vibration modes on the root beam to the modal differences is defined as the first modal difference ratio. The damage index of the root beam can be expressed as:

[0202]

[0203] In the formula The mean mode shape was used to detect the vibration data. The damage index obtained under the damage conditions of each beam is shown in Table 3.

[0204] Table 3 Damage Index Results

[0205]

[0206] By calculating the median of each measuring point under different conditions and comparing the deviations sequentially, abnormal column frames can be detected under each condition, as shown in Figure 12. As can be seen from the table and Figure 12, damaged column frames at different locations correspond one-to-one with the abnormal vibration modes of the corresponding orders, enabling accurate identification of abnormal column frames.

[0207] Large deployable structures, characterized by their large scale, multi-element nature, and rigid-flexible coupling, are prone to structural fatigue damage under long-term wind loads, necessitating real-time monitoring and in-service diagnostics via distributed multi-channel measurement systems. The coupling effect between the structural system's own excitation and wind load, along with the sensor arrangement and packaging configuration, introduces disturbance noise and measurement errors into the sensing signals, significantly impacting the accuracy of condition assessment. This invention employs empirical modal analysis and wind load characteristics to filter out the system's periodic natural noise from the signal. A strain transfer model is established based on the fiber optic grating sensor packaging configuration to compensate for detection errors. Based on the form of large deployable structures, this invention proposes using the differences and ratios of mode shapes in different column skeleton structures to detect abnormal structures. This invention is validated through a specific implementation scheme using finite element simulation examples. Compared with other common data-driven methods, this method offers advantages such as fast computation, strong robustness, and high accuracy.

[0208] It should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting equipment anomalies in the measurement of wind-induced deformation of large deployable antenna structures, characterized in that: Includes the following steps: S1: Obtain the timing response data of a healthy large deployable antenna structure under wind load excitation to form a baseline reference dataset; S2: The baseline reference data is normalized to form a normalized baseline reference dataset. The calculation of the normalized baseline reference dataset also needs to consider the transmission accuracy. S3: Denoising detection data by combining empirical mode decomposition and Hearst exponent analysis; S4: Establish the Hankel matrix for each channel based on the noise-reduced data. and , and Let be the number of rows and columns of the matrix. and These represent historical data and future data, respectively, with a specific point in time as the dividing line. This represents the number of sensing points; S5: Calculate the matrix exist The orthogonal projection vectors on the surface are used to obtain the state matrix and output matrix of the large deployable antenna structure through matrix decomposition, and the reference mode matrix of the structure in the healthy state is calculated. S6: During the service life of the large deployable antenna structure, data acquisition, preprocessing, and signal noise reduction are performed in the same manner as steps S1 to S5 to obtain the current mode shape matrix corresponding to the service status; S7: Calculate the distance between the current mode shape matrix and the reference mode shape matrix, and determine the location of damage in the developable structure by comparing the changes in distance at each order, thereby realizing structural anomaly detection; The large deployable antenna structure includes a scissor mechanism, a frame, and multi-stage telescopic arms. The multi-stage telescopic arms are driven by multi-stage cylinders and integrated in a multi-stage nested manner. The scissor mechanism is connected to each stage of the telescopic arms by equally spaced pins. When the multi-stage cylinders extend, they drive the scissor mechanism to unfold horizontally. In step S1, the baseline reference dataset contains signals from N channels, namely: ； ； The normalization process for the baseline reference data in step S2 can be expressed as follows: In the formula For the first Data from each channel, and The maximum and minimum values ​​among all channel data. For the first Normalized samples for each channel The normalization factor is the normalized baseline reference dataset. It can be represented as: The method of combining empirical mode decomposition and Hurst exponent analysis for noise reduction of detection data includes extracting... The local extreme points of each channel signal are obtained through interpolation. upper and lower envelopes and ; Calculate the envelope mean curve Then, the residual signal is solved. After determining that the residuals satisfy the convergence criterion, the iteration is recorded. The subsequent residuals The first intrinsic mode function After stepwise decomposition, we obtain The intrinsic mode matrix composed of the first-order intrinsic mode functions The intrinsic mode functions (IMFs) of different orders are described using the Hearst exponent. After filtering out the IMFs corresponding to noise terms, the remaining IMFs are summed to obtain the denoised signal. , can be represented as: 。 2. The method for detecting equipment anomalies in the measurement of wind-induced deformation of large deployable antenna structures according to claim 1, characterized in that: Step S3, empirical mode decomposition includes targeting For each channel signal, calculate the Intrinsic Mode Function (IMF) of different orders for each channel.

3. The method for detecting equipment anomalies in the measurement of wind-induced deformation of large deployable antenna structures according to claim 2, characterized in that: The Hearst exponent analysis includes obtaining the Hearst exponents of Intrinsic Mode Functions (IMFs) of different orders through detrended volatility analysis (DFA).

4. The equipment anomaly detection method for measuring wind-induced deformation of large deployable antenna structures according to claim 3, characterized in that: In step S4, the first The Hankel matrix corresponding to the noise reduction signal of each channel and It can be represented as: In the formula For the first The sensor data of the first channel is in the... The value of the moment. and These represent the number of rows and columns of the matrix, respectively.

5. The equipment anomaly detection method for measuring wind-induced deformation of large deployable antenna structures according to claim 4, characterized in that: In step S5, the first Orthogonal projection vectors corresponding to the detection data of each channel It can be represented as: against use Decomposition and Singular Value Decomposition The equation can be obtained as follows: in and These are the system observation matrix and the output matrix, respectively. Based on matrix analysis, we can obtain: The mode shape under the reference state can be obtained by decomposing the eigenvalues ​​of the observation matrix. : In the formula, Let be the system's discrete-time diagonal matrix. Let be the complex eigenvector matrix of the system.

6. The method for detecting equipment anomalies in the measurement of wind-induced deformation of large deployable antenna structures according to claim 5, characterized in that: The location of structural damage anomalies can be determined by comparing the difference between the current mode shape matrix and the reference mode shape matrix using the following formula. in The current mode shape matrix, This is the reference mode shape matrix.

Images