Real-time data driven structure sudden damage detection method and system and medium

Abstract

The invention provides a real-time data-driven structure sudden damage detection method and system and a medium, and belongs to the technical field of structure early warning, and the method comprises the steps: obtaining a response signal of an acceleration sensor of a to-be-detected structure; performing recursive singular spectrum analysis on the response signal to obtain a subspace coefficient curve and a reconstructed acceleration signal; performing Kalman filtering estimation on the reconstructed acceleration signal to obtain a time-varying autoregressive model coefficient, determining a difference value, and when the difference value suddenly changes, determining a damage generation moment; sub-space coefficient curves at different positions at the moment when damage is generated are counted, and variance and kurtosis around the abrupt change moment of the reconstructed acceleration signal are used as damage sensitive features; and determining the damage position in the to-be-detected complex structure according to the acceleration sensor mounting position corresponding to the sudden change generated by the damage sensitive feature. The method does not depend on a numerical or physical model, provides a technology for carrying out damage monitoring on the complex structure by adopting monitoring data, and provides reference for performance evaluation of the complex structure.

Description

Technical Field

[0001] The present invention relates to the technical field of structural health monitoring and safety early warning, and in particular to a method, system and medium for detecting sudden structural damage based on real-time data drive. Background Art

[0002] As time goes by, the demand for structural health monitoring is growing rapidly to ensure that complex structures can operate safely and stably. Structural health monitoring is a process that uses sensor arrays to obtain the response of the structure, extracts damage-sensitive indicators from the response, and determines the health of the structure through statistical analysis of the damage-sensitive indicators. At present, structural health monitoring systems use sparse and dense sensor arrangements. In this context, the development of a baseline-free, real-time damage detection solution remains a challenge.

[0003] Current methods of structural health monitoring based on damage identification use advanced signal processing techniques to collect vibration measurements from dense sensor arrays to monitor the condition of the structure. Commonly used damage methods currently use changes in modal properties to detect damage caused by various environmental conditions, artificial excitations or natural events. Through the substitution-assisted stochastic optimization inversion algorithm, static and dynamic parameters of the structure can be inverted to identify damage based on parameter changes; dynamic information of modal frequencies can be obtained through Hilbert transform combined with principal component analysis, and the instantaneous frequency is used as the damage index for damage detection; based on the mean square ratio between the two responses, only the modal parameter characteristics of the structure are relied on to detect damage changes. These damage detection methods can be broadly classified as model methods, which rely on high-precision FEM, so their computational cost is high. As the computational cost increases, the modeling errors and modeling parameter selection limit their application in real-time detection. Summary of the invention

[0004] To solve the above problems, the present invention provides a real-time data-driven structural sudden damage detection method, system and medium. The method does not rely on numerical or physical models, provides a technology for damage monitoring of complex structures using monitoring data, and provides a reference for performance evaluation of complex structures.

[0005] To achieve the above object, the present invention provides the following technical solutions.

[0006] A real-time data-driven structural sudden damage detection method, system and medium, comprising the following steps:

[0007] Obtaining response signals of acceleration sensors at different positions of the structure to be tested under environmental loads;

[0008] Perform recursive singular spectrum analysis on the response signal to obtain the subspace coefficient curve and reconstruct the acceleration signal;

[0009] The reconstructed acceleration signal is estimated by Kalman filtering to obtain the time-varying autoregressive model coefficients, and the difference between the coefficients of the time-varying autoregressive model coefficients at each time before and after is calculated. When the difference suddenly changes, the time when the damage occurs is determined;

[0010] The subspace coefficient curves of different acceleration sensor installation positions at the moment of damage and the variance and kurtosis around the moment of reconstructed acceleration signal mutation are statistically taken as damage sensitive features; the damage position of the structure to be detected is determined according to the acceleration sensor installation position corresponding to the mutation of the damage sensitive feature.

[0011] Preferably, the acceleration sensors are respectively arranged on the structure to be detected, and a plurality of acceleration sensors form a detection group. The response signal on the acceleration sensor at each position is collected by the acquisition system, and the acceleration response database at different positions is established using the response signal.

[0012] Preferably, the performing recursive singular spectrum analysis on the response signal comprises the following steps:

[0013] Collect the response signals of multiple acceleration sensors, select the L-dimensional sample vector at the current moment as the initial signal length, which contains l lag data in the one-dimensional time domain signal:

[0014] X k =[x k-l+1 ,x k-l+2 ,x k-l+3 ,...,x k ] T

[0015] The covariance estimate at the current k moment is expressed by the eigenvalue and eigenvector at the current moment, and the formula is as follows:

[0016]

[0017] In the formula, Q k is the Hankel covariance matrix at time k, X k is the sample vector at time k, Q k-1 is the Hankel covariance matrix at time k-1;

[0018] make Substituting into the above formula, we get:

[0019]

[0020] Where U k is the k-time eigenvector, U k-1 is the eigenvector at time k-1, and the superscript T represents the transposed matrix; kis the eigenvalue at time k;

[0021] According to Gershgorin's theorem, the above formula can be transformed into:

[0022]

[0023] The recursive update function of the feature subspace at time k-1 is obtained by taking the feature subspace at time k as the independent variable. The formula is as follows:

[0024] U k =U k-1 J k

[0025]

[0026] In the formula, Υ k is the eigenvalue at time k, U k is the k-time eigenvector, U k-1 is the eigenvector at time k-1;

[0027] According to the current moment feature space vector, the spatial principal component at the current moment is obtained, and the formula is as follows:

[0028]

[0029] SC=φ L

[0030] Where SC is the minimum eigenvector of the eigenvalue projected to the data at time k; φ i (k) is all the spatial components at time k;

[0031] From the above formula, we can get the i-th th principal component values, in the reconstruction step, the i-th th The principal components are projected back to their original subspace to obtain the last column of the corresponding elementary matrix:

[0032] R i (k)=(U i (k)) L×1 ×(φ i (k)) 1×1

[0033]

[0034] In the formula, r i (k) is R i (k) is the last item of the vector, n is the number of eigenvalues ​​that can explain more than 90%, and x(k) is the reconstructed signal value obtained recursively at each moment.

[0035] Preferably, the reconstructing the acceleration signal through Kalman filtering estimation comprises the following steps:

[0036] Construct the state equation and observation equation for reconstructing the acceleration signal:

[0037]

[0038] In the formula, is the state vector, a p (k) is the weight, A(k-1) is the unit matrix; Z(k) is the observation vector; W(k) and V(k) are Gaussian white noises, which are independent and uncorrelated; H(k) is the observation matrix H(k) = [X(k-1), X(k-2), ..., X(kp)];

[0039] Construct the one-step prediction equation for the state:

[0040]

[0041] In the formula, is the estimated value at time k-1, is the predicted value of the estimated value at time k;

[0042] Construct the one-step prediction equation for the mean square error:

[0043]

[0044] Where P k-1 / k-1 (k-1|k-1) is the estimated value of the mean square error matrix at time k-1, I is the unit matrix, is the variance of the noise W(k), P k / k-1 (k|k-1) is the predicted value of the estimated value of the mean square error at time k;

[0045] Construct the one-step prediction equation for the observation vector:

[0046]

[0047] Where Z(k|k-1) is the predicted value of the estimated value of the observation vector at time k; H(k) is the observation matrix; is the predicted value of the estimated value at time k;

[0048] Construct the one-step prediction equation for mean square error variance:

[0049]

[0050] In the formula, is the predicted value of the mean square error variance estimate at time k, is the variance of V(k); H(k) is the observation matrix at time k; H(k) T is the transposed matrix of the observation matrix; P k-1 / k-1 (k-1|k-1) is the estimated value of the mean square error matrix at time k-1;

[0051] Construct the filter gain equation:

[0052]

[0053] Where Kg(k) is the Kalman gain matrix; for The inverse matrix of T is the transposed matrix of the observation matrix;

[0054] Construct a state vector estimate:

[0055]

[0056] In the formula, is the estimated value at time k, is the predicted value of the estimated value at time k; Kg(k) is the Kalman gain matrix; Z(k) is the observed value at time k, and Z(k|k-1) is the predicted value of the estimated value of the observation vector at time k;

[0057] Construct the mean square error update matrix:

[0058] P k / k (k|k)=[I-Kg(k)H(k)]P k / k-1 (k|k-1)

[0059] Where P k / k (k|k) is the estimated value of the mean square error matrix at time k, P k / k-1 (k|k-1) is the predicted value of the estimated value of the mean square error at time k; I is the identity matrix; Kg(k) is the Kalman gain matrix; H(k) is the observation matrix at time k;

[0060] Compute differences for a time-varying autoregressive model:

[0061] X(k)=a1(k)X(k-1)+a2(k)X(k-2)+V(k)

[0062] ErTVAR=a1-a2

[0063] Where X(k) is the moment of capturing damage mutation. The ErTVAR coefficient changes as a whole at the moment of damage occurrence. a1(k) and a2(k) are the weights of the moment before and the two moments before the damage mutation.

[0064] Preferably, the subspace coefficient curves of different acceleration sensor installation positions and the variance and kurtosis around the moment of sudden change of the reconstructed acceleration signal are solved as shown in the following formula:

[0065]

[0066] Where μ is the average value within the range L at the time of damage, n is the number of data points within the range L, and X i is the data of each point within the range of L.

[0067] A real-time data-driven structural sudden damage detection method, system and medium, the system comprising:

[0068] processor;

[0069] a memory having stored thereon a computer program executable on the processor;

[0070] Wherein, when the computer program is executed by the processor, the method, system and medium for detecting sudden structural damage based on real-time data drive are implemented.

[0071] A computer-readable storage medium stores a data processing program, and when the data processing program is executed by a processor, the method, system and medium for detecting sudden structural damage based on real-time data drive are implemented.

[0072] Beneficial effects of the present invention:

[0073] The present invention proposes a method, system and medium for detecting sudden structural damage based on real-time data drive. The method collects the translational acceleration and rotational acceleration of the structure in the vibration direction under environmental load; uses multi-channel recursive singular spectrum to obtain the subspace coefficient curve and the reconstructed acceleration signal in the decomposition process; and combines the reconstructed signal with the time-varying autoregressive model to obtain the regression model coefficient difference. When a structural component suffers from sudden stiffness damage, the time when the damage occurs can be captured by the coefficient difference of the regression model, and the location of the damage can be determined based on the variance and kurtosis of the reconstructed signal and the subspace coefficient. The method does not rely on numerical or physical models; it can detect damage in real time without the need for baseline health data; it can quickly and easily analyze the massive data of each sensor, extract the time when sudden damage occurs, and identify the location of sudden damage. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 is a schematic diagram of a specific flow chart of a method according to an embodiment of the present invention;

[0075] Figure 2 It is a curve of the change of the translational acceleration of 12 sensors collected under the S1 working condition of an embodiment of the present invention over time;

[0076] Figure 3 is a curve of acceleration over time reconstructed by the sensor at the S1 working condition 12 of an embodiment of the present invention;

[0077] Figure 4 is a subspace coefficient curve obtained by sensor decomposition at S1 condition 12 of an embodiment of the present invention;

[0078] Figure 5 is a schematic diagram of damage in the S1 working condition of an embodiment of the present invention;

[0079] Figure 6 (a) is a schematic diagram of capturing the damage moment of the S1 working condition according to an embodiment of the present invention;

[0080] Figure 6 (b) is a schematic diagram of the damage position determination result of the S1 working condition according to an embodiment of the present invention;

[0081] Figure 7 is a schematic diagram of damage in the S3 working condition of an embodiment of the present invention;

[0082] Figure 8 (a) is a schematic diagram of capturing the damage moment of the S3 working condition according to an embodiment of the present invention;

[0083] Figure 8 (b) is a schematic diagram of the damage position determination result of the S3 working condition of an embodiment of the present invention;

[0084] Fig. 9 Schematic diagram of upper layer damage in the S2+S3+S4 working condition of an embodiment of the present invention;

[0085] Fig.10 (a) is a schematic diagram of capturing the damage moment of the upper sensor in the S2+S3+S4 working condition according to an embodiment of the present invention;

[0086] Fig.10 (b) is a schematic diagram of the result of determining the damaged position of the upper sensor in the S2+S3+S4 working condition according to an embodiment of the present invention;

[0087] Fig.11 Schematic diagram of the middle layer (front) damage in the S2+S3+S4 working condition of an embodiment of the present invention;

[0088] Fig.12 (a) is a schematic diagram of capturing the damage moment of the middle (front) sensor in the S2+S3+S4 working condition according to an embodiment of the present invention;

[0089] Fig.12 (b) is a schematic diagram of the result of determining the damage position of the middle (front) sensor in the S2+S3+S4 working condition according to an embodiment of the present invention;

[0090] Fig.13 Schematic diagram of the middle layer (front) damage in the S2+S3+S4 working condition of an embodiment of the present invention;

[0091] Fig.14 (a) is a schematic diagram of capturing the damage moment of the middle (front) sensor in the S2+S3+S4 working condition according to an embodiment of the present invention;

[0092] Fig.14 (b) is a schematic diagram of the damage position determination result of the middle (front) sensor in the S2+S3+S4 working condition of an embodiment of the present invention. DETAILED DESCRIPTION

[0093] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0094] Example 1

[0095] This embodiment proposes a method, system and medium for detecting sudden structural damage based on real-time data drive, such as Figure 1 As shown, Figure 1 This is a flow chart of a method, system and medium for detecting sudden structural damage based on real-time data drive. In the embodiment, a bridge is used as an object, and specifically includes the following steps:

[0096] Step 1: multiple sensors are arranged on the bridge. Multiple acceleration sensors form a detection group. The response signal on the acceleration sensor at each position is collected by the acquisition system, and the acceleration response database at different positions is established using the response signal.

[0097] Step 2: Import the acceleration response data collected in step 1, and perform recursive singular spectrum analysis on the time domain data at different positions. The specific operation is divided into the following four steps: sample selection, construction of recursive covariance matrix, recursive singular value decomposition, and reconstruction of time series. The subspace coefficient curve and reconstructed acceleration signal are obtained through recursive singular spectrum analysis. The specific steps are as follows:

[0098] STEP 1: Sample selection

[0099] The sample selects the L-dimensional sample vector (X k =[x k-l+1 ,x k-l+2 ,x k-l+3 ,...,x k ] T ) as the initial signal length, which contains L lag data in the one-dimensional time domain signal. L can be optional.

[0100] STEP2: Construct a recursive covariance matrix

[0101] The covariance estimate at the current k moment can be expressed by the eigenvalue and eigenvector at the current moment, and the formula is as follows:

[0102]

[0103]

[0104] Q k represents the Hankel covariance matrix at time k, X k represents the sample vector at time k, Q k-1 Represents the Hankel covariance matrix at time k-1

[0105] make Substituting into equation (2), we can get:

[0106]

[0107] U k represents the k-time eigenvector, U k-1 represents the eigenvector at time k-1, and the superscript T represents the transposed matrix.

[0108] STEP3: Recursive singular value decomposition

[0109] For a finite sample size (k) and a low-damping estimation system, It usually exhibits diagonally dominated behavior. Gershgorin's theorem can be applied to diagonally dominant matrices, and perturbation techniques can be used to provide recursive feature space updates at each time point. Application of Gershgorin's theorem ensures that the singular value decomposition of this term is The form of, where J is a standard orthogonal matrix, Ω k For a diagonal matrix, formula (3) can be transformed into the following formula:

[0110]

[0111] From formula (4), the feature subspace at time k-1 is used as the independent variable to obtain the recursive update function of the feature subspace at time k, which is as follows:

[0112]

[0113] Υ k is the eigenvalue at time k, U k represents the k-time eigenvector, U k-1 Represents the eigenvector at time k-1;

[0114] According to the current moment feature space vector, the spatial principal component at the current moment can be obtained, and the formula is as follows:

[0115]

[0116] SC=φ L (7)

[0117] SC represents the projection of the k-time data to the eigenvector with the minimum eigenvalue. i (k) represents all spatial components at time k.

[0118] STEP4: Reconstruct time series

[0119] From formula (7), we can get the i-th th In the reconstruction step, the i-th principal component value PC is th The principal components are projected back to their original subspace to obtain the last column of the corresponding elementary matrix.

[0120]

[0121] r i (k) represents R i The last item of the vector (k), n represents the number of eigenvalues ​​that can explain more than 90%, and x(k) represents the reconstructed signal value obtained recursively at each moment.

[0122] Step 3: Import the acceleration signal reconstructed in step 2, and solve the time-varying autoregressive time model coefficients through the Kalman filter method. The specific operation method is to construct the state equation and observation equation, time update matrix, and measurement update matrix. The time-varying autoregressive model coefficients are estimated through Kalman filtering, and the difference between the coefficients of the time-varying autoregressive model coefficients before and after is used as the damage sensitive feature to indicate the moment of structural mutation damage. The specific steps are as follows:

[0123] Step 1: Construct state equation and observation equation

[0124]

[0125] State Vector It is composed of the following components A(k-1) represents the identity matrix. Z(k) represents the observation vector. W(k), V(k) represent Gaussian white noise, which are independent and uncorrelated. H(k) represents the observation matrix H(k) = [X(k-1), X(k-2), ..., X(kp)].

[0126] Step 2: Time update matrix

[0127] i) One-step prediction equation of state:

[0128]

[0129] is the estimated value at time k-1, is the predicted value of the estimated value at time k.

[0130] ii) One-step prediction equation of mean square error:

[0131]

[0132] P k-1 / k-1 (k-1|k-1) is the estimated value of the mean square error matrix at time k-1, I is the identity matrix, is the variance of the noise W(k), P k / k-1 (k|k-1) is the predicted value of the estimated value of the mean square error at time k.

[0133] iii) One-step prediction equation for the observation vector:

[0134]

[0135] Z(k|k-1) is the predicted value of the estimated value of the observation vector at time k. H(k) is the observation matrix. The predicted value of the estimated value at time k.

[0136] iv) Mean square error variance one-step prediction equation:

[0137]

[0138] is the predicted value of the mean square error variance estimate at time k, is the variance of V(k). H(k) is the observation matrix at time k. T is the transposed matrix of the observation matrix. k-1 / k-1 (kk|k-1) is the estimated value of the mean square error matrix at time k-1.

[0139] Step 3: Measure and update the matrix

[0140] i): Filter gain equation:

[0141]

[0142] Kg(k) is the Kalman gain matrix. for The inverse matrix of H(k) T is the transposed matrix of the observation matrix

[0143] ii): State vector estimate:

[0144]

[0145] is the estimated value at time k, is the predicted value of the estimated value at time k. Kg(k) is the Kalman gain matrix. Z(k) represents the observed value at time k, and Z(k|k-1) is the predicted value of the estimated value of the observation vector at time k.

[0146] iii): Mean square error update matrix:

[0147] P k / k (k|k)=[I-Kg(k)H(k)]P k / k-1 (k|k-1) (18)

[0148] P k / k (k|k) is the estimated value of the mean square error matrix at time k, P k / k-1 (k|k-1) is the predicted value of the estimated value of the mean square error at time k. I is the identity matrix. Kg(k) is the Kalman gain matrix. H(k) is the observation matrix at time k.

[0149] Step 4: Difference of time-varying autoregressive model

[0150] X(k)=a1(k)X(k-1)+a2(k)X(k-2)+V(k) (19)

[0151] ErTVAR=a1-a2 (20)

[0152] X(k) represents the moment of capturing the damage mutation. The ErTVAR coefficient changes as a whole at the moment of damage occurrence. a1(k) and a2(k) are the weights of the moment before and two moments before the damage mutation.

[0153] The difference coefficients of the autoregressive model obtained in step 4 and step 3 can capture the moment of sudden damage, and the variance and kurtosis statistics are used as damage sensitive features. The variance and kurtosis around the sudden change moment of the subspace coefficient curve and the reconstructed acceleration signal are statistically analyzed. When the bridge is suddenly damaged, the reconstructed time curve and the damage sensitive features of the subspace coefficient curve reflect the same damage location. The damage sensitive features of the subspace coefficient curve are used as the reconstructed acceleration damage location verification.

[0154]

[0155] μ represents the average value within the range L at the time of damage, n represents the number of data points within the range L, and X i Represents the data of each point within the range of L.

[0156] In this embodiment, a numerical simulation is used to apply seismic loads to a steel truss bridge.

[0157] 1. Use ANSYS to construct a steel truss bridge. The entire bridge is about 108 meters long, 10 meters wide, and 16 meters high. The material used is structural steel with a density of 7800kg / m3 and an elastic modulus of 210GPa.

[0158] 2. The damage setting adopts the elastic modulus reduction method, with the reduction levels of 5%, 15%, and 20%.

[0159] 3. The damage is set to three conditions, two single damage conditions and one multiple damage condition.

[0160] The damage degree of damage condition S1 is 20%, and the damage is set to occur at 20s. Figure 5 .

[0161] The damage degree of single damage condition S3 is 15%, and the damage is set to occur at 20s. Figure 7 .

[0162] The multiple damage condition is S2+S3+S4, the damage is 5% each, and the damage occurs at 20s. Fig. 9 , 11 、13.

[0163] 4. Install sensors at each member of the steel truss bridge.

[0164] 5. Apply seismic loads to bridges.

[0165] 6. Collect the translational acceleration and rotational acceleration at different positions of the bridge.

[0166] 7. Import the acceleration response data collected in step 6 and perform recursive singular spectrum analysis on the time domain data at different positions. The specific operation is divided into the following four steps: sample selection, construction of recursive covariance matrix, recursive singular value decomposition, and reconstruction of time series. The subspace coefficient curve and reconstructed acceleration signal are obtained through recursive singular spectrum analysis.

[0167] 8. Import the acceleration signal reconstructed in step 7, and solve the time-varying autoregressive time model coefficients through the Kalman filter method. The specific operation method is to construct the state equation and observation equation, time update matrix, and measurement update matrix. The time-varying autoregressive model coefficients are estimated through Kalman filtering, and the difference between the coefficients before and after the time-varying autoregressive model coefficients is used as the damage sensitive feature to indicate the moment of structural mutation damage.

[0168] 9. The difference coefficient of the autoregressive model obtained in step 8 can capture the moment of sudden damage, and the variance and kurtosis statistics are used as damage sensitive features. The variance and kurtosis around the statistical subspace coefficient curve and the reconstructed acceleration signal mutation moment are obtained.

[0169] 10. Single Damage S1:

[0170] from Figure 5 It can be seen that the damage location of S1 is in the upper truss area of ​​the steel truss bridge. The translational acceleration of the 12 members of the upper truss along the load direction is collected, and zero-mean Gaussian white noise is added. It can be seen Figure 2 .from Figure 2 It is found that it is impossible to directly obtain the information of mutation damage in the original time domain signal. The regressed time domain signal and subspace coefficients are obtained by recursive singular spectrum analysis. For details, see Figure 3 , Figure 4 . The coefficient difference changes are obtained through the time-varying autoregressive model, see Figure 6 (a) From Figure 6 The ErTVAR diagram of (a) shows that the mutation occurred at 20 s; Figure 3 and Figure 4 The time information of the damage occurrence is confirmed in the regressed time domain signal and subspace coefficient. Finally, the pre-set damage sensitive feature variance shows that damage has occurred at position 11. Figure 6 (b) It is also found that rods 10 and 12 also experience similar damage fluctuations. This is because these rods are located around rod 11. The sudden damage of rod 11 will cause overall damage to the surrounding area. Figure 6 (b) verifies that the regressed time domain curve and subspace coefficients are consistent with the damage.

[0171] Single Damage S3:

[0172] from Figure 7 The S3 damage location is seen in the middle truss (front) area of ​​the steel truss bridge. Figure 8 (a) shows that the mutation damage occurs at 20s. Figure 8 In (b), the damaged area is found between positions 8 and 10, although the damage occurred at position 9, which shows that the entire nearby area is damaged.

[0173] Multiple injuries S2+S3+S4:

[0174] For the upper area, Fig.10 (a) It was found that the mutation damage occurred at 20s. Fig.10 In (b), the damaged area is found between position 3 and position 5, covering the damaged area from top to bottom. Fig.12 (a) It was found that the mutation damage occurred at 20s. Fig.12 (b) shows that the kurtosis diagram and subspace coefficient diagram of the regression time domain show that the damage area is divided into two blocks, one of which is between positions 4 and 6, and the other is between positions 9 and 11. Combined with the damage coverage area of ​​the upper truss, the location of the damage positions S4 and S3 is basically confirmed. For the middle truss (rear), from Fig.14(a) It was found that the mutation damage occurred at 20s. Fig.14 The kurtosis diagram and subspace coefficient diagram of the regression time domain in (b) show that the damaged area is between positions 4 and 6. Combined with the damage coverage area of ​​the upper truss, the location of the damaged position S2 is basically confirmed.

[0175] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A real-time data-driven structural sudden damage detection method, system and medium, characterized in that: The following steps are involved: Obtaining response signals of acceleration sensors at different positions of the structure to be tested under environmental loads; Perform recursive singular spectrum analysis on the response signal to obtain the subspace coefficient curve and reconstruct the acceleration signal; The reconstructed acceleration signal is subjected to Kalman filter estimation to obtain the time-varying autoregressive model coefficients, and the difference between the coefficients of the time-varying autoregressive model coefficients at each previous and next moment is determined. When the difference suddenly changes, the moment when the damage occurs is determined. The subspace coefficient curves of different acceleration sensor installation positions at the moment of damage and the variance and kurtosis around the moment of reconstructed acceleration signal mutation are statistically analyzed as damage sensitive features. The damage position of the structure to be detected is determined according to the installation position of the acceleration sensor corresponding to the sudden change of the damage sensitive characteristic.

2. The method, system and medium for detecting sudden structural damage based on real-time data drive according to claim 1, characterized in that: The acceleration sensors are arranged on the structure to be detected respectively, and a plurality of acceleration sensors form a detection group. The response signal on the acceleration sensor at each position is collected by the acquisition system, and the acceleration response database at different positions is established by using the response signal.

3. The method, system and medium for detecting sudden structural damage based on real-time data drive according to claim 1, characterized in that: The recursive singular spectrum analysis of the response signal comprises the following steps: Collect the response signals of multiple acceleration sensors, select the L-dimensional sample vector at the current moment as the initial signal length, which contains l lag data in the one-dimensional time domain signal: X k =[x k-l+1 ,x k-l+2 ,x k-l+3 ,...,x k ] T The covariance estimate at the current k moment is expressed by the eigenvalue and eigenvector at the current moment, and the formula is as follows: In the formula, Q k is the Hankel covariance matrix at time k, X k is the sample vector at time k, Q k-1 is the Hankel covariance matrix at time k-1; make Substituting into the above formula, we get: Where U k is the k-time eigenvector, U k-1 is the eigenvector at time k-1, and the superscript T represents the transposed matrix; γ k is the eigenvalue at time k; According to Gershgorin's theorem, the above formula can be transformed into: The recursive update function of the feature subspace at time k-1 is obtained by taking the feature subspace at time k as the independent variable. The formula is as follows: IN k =U k-1 J k In the formula, γ k is the eigenvalue at time k, U k is the k-time eigenvector, U k-1 is the eigenvector at time k-1; According to the current moment feature space vector, the spatial principal component at the current moment is obtained, and the formula is as follows: SC=φ L Where SC is the minimum eigenvector of the eigenvalue projected to the data at time k; φ i (k) is all the spatial components at time k; From the above formula, we can get the i-th th principal component values, in the reconstruction step, the i-th th The principal components are projected back to their original subspace to obtain the last column of the corresponding elementary matrix: R i (k)=(U i (k)) L×1 ×(φ i (k)) 1×1 In the formula, r i (k) is R i (k) is the last item of the vector, n is the number of eigenvalues ​​that can explain more than 90%, and x(k) is the reconstructed signal value obtained recursively at each moment.

4. The method, system and medium for detecting sudden structural damage based on real-time data drive according to claim 1, characterized in that: The reconstructed acceleration signal is subjected to Kalman filter estimation, comprising the following steps: Construct the state equation and observation equation for reconstructing the acceleration signal: In the formula, is the state vector, a p (k) is the weight, A(k-1) is the unit matrix; Z(k) is the observation vector; W(k) and V(k) are Gaussian white noises, which are independent and uncorrelated; H(k) is the observation matrix H(k) = [X(k-1), X(k-2), ..., X(kp)]; Construct the one-step prediction equation for the state: In the formula, is the estimated value at time k-1, is the predicted value of the estimated value at time k; Construct the one-step prediction equation for the mean square error: Where P k-1 / k-1 (k-1|k-1) is the estimated value of the mean square error matrix at time k-1, I is the unit matrix, is the variance of the noise W(k), P k / k-1 (k|k-1) is the predicted value of the estimated value of the mean square error at time k; Construct the one-step prediction equation for the observation vector: Where Z(k|k-1) is the predicted value of the estimated value of the observation vector at time k; H(k) is the observation matrix; is the predicted value of the estimated value at time k; Construct the one-step prediction equation for mean square error variance: In the formula, is the predicted value of the mean square error variance estimate at time k, is the variance of V(k); H(k) is the observation matrix at time k; H(k) T is the transposed matrix of the observation matrix; P k-1 / k-1 (k-1|k-1) is the estimated value of the mean square error matrix at time k-1; Construct the filter gain equation: Where Kg(k) is the Kalman gain matrix; for The inverse matrix of T is the transposed matrix of the observation matrix; Construct a state vector estimate: In the formula, is the estimated value at time k, is the predicted value of the estimated value at time k; Kg(k) is the Kalman gain matrix; Z(k) is the observed value at time k, and Z(k|k-1) is the predicted value of the estimated value of the observation vector at time k; Construct the mean square error update matrix: P.S k / k (k|k)[I-Kg(k)H(k)]P k / k-1 (k|k-1) Where P k / k (k|k) is the estimated value of the mean square error matrix at time k, P k / k-1 (k|k-1) is the predicted value of the estimated value of the mean square error at time k; I is the identity matrix; Kg(k) is the Kalman gain matrix; H(k) is the observation matrix at time k; Compute differences for a time-varying autoregressive model: X(k)=a1(k)X(k-1)+a2(k)X(k-2)+V(k) ErTVAR=a1-a2 Where X(k) is the moment of capturing damage mutation. The ErTVAR coefficient changes as a whole at the moment of damage occurrence. a1(k) and a2(k) are the weights of the moment before and the two moments before the damage mutation.

5. The method, system and medium for detecting sudden structural damage based on real-time data drive according to claim 2, characterized in that: The subspace coefficient curves of the different acceleration sensor installation positions and the variance and kurtosis around the moment of the reconstructed acceleration signal mutation are solved as shown in the following formula: Where μ is the average value within the range L at the time of damage, n is the number of data points within the range L, and X i is the data of each point within the range of L.

6. A method, system and medium for detecting sudden structural damage based on real-time data drive, characterized in that: The system comprises: processor; a memory having stored thereon a computer program executable on the processor; Wherein, when the computer program is executed by the processor, the steps of the method, system and medium for detecting sudden structural damage based on real-time data drive are implemented as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a data processing program, and when the data processing program is executed by the processor, the steps of the real-time data-driven structural sudden damage detection method, system and medium are implemented as described in any one of claims 1 to 5.

Images