Structural damage alarm method based on piezoelectric guided wave-adaptive network probability association

Abstract

The invention discloses a structure damage alarm method based on piezoelectric guided wave-adaptive network probability association, and belongs to the technical field of aerospace structure health monitoring, and the method comprises the steps: employing probability statistical modeling combined with a dynamic window slippage method to obtain a reference alarm feature and a damage alarm feature of each channel; fusing the reference alarm features of the channels by adopting a truncated mean network fusion method to obtain a fused reference alarm feature, and fusing the damage alarm features of the channels to obtain a fused damage alarm feature; based on the fused reference alarm features, adaptively constructing a multivariate statistical feature fusion threshold by using statistical parameters; and when the fused damage alarm features are greater than a multivariate statistical feature fusion threshold, carrying out structural damage alarm. The implementation process is simple and efficient, the damage alarm sensitivity under the influence of the time-varying environment can be improved, and the false alarm probability can be effectively reduced while the damage of the monitored structure is comprehensively monitored.

Description

Technical Field

[0001] This application relates to the field of aerospace structural health monitoring technology, and in particular to a structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation. Background Technology

[0002] Structural health monitoring, as a comprehensive technology integrating sensing, computing, and diagnostics, has become an important support for the full life-cycle management of engineering structures. By embedding or attaching sensor networks to critical structural components, structural health monitoring systems can capture the mechanical response and environmental parameters of the structure in real time during service. Then, using data-driven or physical model-driven algorithms, early damage can be identified and its spatial distribution and severity quantified. Ultimately, this achieves multiple goals: improved safety, reduced maintenance costs, and extended service life.

[0003] Among existing structural health monitoring methods, waveguide-based technology is widely used due to its unique advantages. Waveguide technology exhibits high sensitivity to structural damage and can cover large monitoring areas. Furthermore, this method supports a flexible application mode combining online real-time monitoring and offline data analysis, demonstrating significant technical advantages in assessing the health status of complex structures. However, in practical engineering applications, time-varying environments pose significant challenges to the reliability and accuracy of structural health monitoring. The transient or slow evolution of temperature, humidity, external loads, and boundary constraints can cause drifts in material properties (such as elastic modulus, Poisson's ratio, density, and geometry) through thermo-mechanical coupling mechanisms, thereby altering the impedance distribution and energy dissipation path of the structure. Simultaneously, the propagation of guided waves within the structure is also affected by the time-varying environment, causing group / phase velocities, attenuation coefficients, and amplitudes to vary with environmental conditions. When signal disturbances introduced by the time-varying environment are superimposed on the changes caused by actual damage in the time-frequency domain, traditional damage identification frameworks based on feature residuals struggle to decouple the two. Especially when the rate of change of temperature gradient or load amplitude is similar to the time scale of damage evolution, environmental noise may drown out or spoof damage-sensitive features, causing false alarms or missed alarms, thus significantly reducing the robustness of the monitoring algorithm.

[0004] In the field of structural damage alarm, conventional damage alarm methods mainly suffer from two key problems. Firstly, due to the complexity of actual service environments, the damage alarm features extracted by conventional methods often exhibit significant fluctuations, making them difficult to directly apply to structural damage alarms. Secondly, pre-set fixed thresholds lack adaptability and cannot adapt to changes in actual monitoring conditions. Furthermore, most existing methods only consider damage monitoring of a single channel, failing to fully utilize the advantages of guided wave networks to achieve comprehensive monitoring of structural damage. Therefore, guided wave structure damage alarm methods under the influence of time-varying factors still need improvement. Summary of the Invention

[0005] This application aims to provide a structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation. The implementation process is simple and efficient, which can improve the damage alarm sensitivity under the influence of time-varying environment, provide comprehensive monitoring of the damage of the monitored structure, and effectively reduce the false alarm probability. This solves the engineering application problem of the reliability of damage monitoring under the influence of time-varying factors of aircraft structures.

[0006] A structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation includes, The baseline feature sample set and monitoring feature sample set of the monitored structure are obtained by using probabilistic statistical modeling combined with dynamic window sliding method, and the baseline alarm features and damage alarm features of each channel are obtained. The truncated mean network fusion method is used to fuse the baseline alarm features of each channel to obtain the fused baseline alarm features, and the damage alarm features of each channel are fused to obtain the fused damage alarm features. Based on the fused baseline alarm features, an adaptive multivariate statistical feature fusion threshold is constructed. During the damage monitoring phase of the monitored structure, when the fused damage alarm features are greater than the multivariate statistical feature fusion threshold, the monitored structure is determined to have suffered damage and a structural damage alarm is triggered.

[0007] Optional truncated mean network fusion methods include: Data to be merged from each channel Z= ( z 1, z 2,…, z n Sort in ascending order to get Z s = ( z (1) , z (2) ,…, z (n) ),in z (1) ≤ z (2) …≤ z (n) Before truncation m The minimum value and the subsequent m The maximum value; the data to be fused includes: baseline alarm features and damage alarm features; Calculate the arithmetic mean of the data to be fused from each channel to obtain the fusion result. Z’ ; Fusion results Z’ Includes: the fused baseline alarm features and the fused damage alarm features; fusion results. Z’ It is expressed as follows: in, Indicates the first l Data to be merged from each channel; n Indicates the number of channels; m This indicates the number of interruptions, which is set according to the actual alarm situation.

[0008] Optionally, the steps for constructing a multivariate statistical feature fusion threshold include: Calculate the mean and standard deviation of the fused baseline alarm characteristics; Calculate the kurtosis and slope of the fused baseline alarm features; Determine the coefficients for the mean value of the fused baseline alarm characteristics; Determine the coefficients of the standard deviation of the fused baseline alarm characteristics; The multivariate statistical feature fusion threshold is calculated based on the mean, standard deviation, and corresponding coefficients of the fused baseline alarm features.

[0009] Optionally, the kurtosis and slope of the fused baseline alarm features are calculated and represented as follows: in, Indicates the slope of the fused baseline alarm feature; Indicates the fused baseline alarm characteristics HDI' Peak value; Indicates the fused baseline alarm characteristics HDI' The first in the sequence i One value; Indicates the fused baseline alarm characteristics HDI' The mean value; N represents the number of guided wave signals collected when the monitored structure is in a healthy state.

[0010] Optionally, the coefficients of the mean of the fused baseline alarm features are expressed as follows: in, The coefficient representing the mean of the fused baseline alarm characteristics.

[0011] Optionally, the coefficient of the standard deviation of the fused baseline alarm features is expressed as follows: in, The coefficient representing the standard deviation of the fused baseline alarm characteristics.

[0012] Optional, the multivariate statistical feature fusion threshold (UCL) is expressed as follows: Wherein, UCL represents the threshold for fusion of multivariate statistical features; Indicates the fused baseline alarm characteristics HDI' The mean; Indicates the fused baseline alarm characteristics HDI' The standard deviation.

[0013] Optionally, a baseline feature sample set and a monitoring feature sample set of the monitored structure are obtained. A probabilistic statistical model combining dynamic window sliding is used to obtain the baseline alarm features and damage alarm features for each channel, including: Under the healthy condition of the monitored structure, guided wave signals from any set of piezoelectric sensor networks are collected, with one set serving as the reference signal. For each channel, all guided wave signals are adaptively truncated using a dynamic window sliding method to extract multiple damage factors, and any set of guided wave signals forms a reference feature sample set. After entering the damage monitoring stage, each channel again collects any set of guided wave signals, and all guided wave signals are adaptively truncated using a dynamic window sliding method to extract multiple damage factors, and any set of guided wave signals forms a monitoring feature sample set. The reference feature sample set and the monitoring feature sample set are used as inputs for probabilistic statistical modeling to obtain the reference residual sequence and the monitoring residual sequence. The absolute value operation is performed on the reference residual sequence and the monitoring residual sequence to obtain the reference alarm feature and damage alarm feature of the corresponding channel. Probabilistic statistical modeling combined with the dynamic window sliding method specifically includes: The theoretical flight time of the guided wave is calculated and converted into discrete sampling points. The starting position of the direct wave of each group of guided wave signals is used as the left boundary of the window sliding. in, d Indicates the spacing between piezoelectric sensors in the corresponding channel; v This represents the theoretical wave velocity of the guided wave. Indicates the sampling rate of the signal; Indicates the pre-trigger length of the signal acquisition system; Indicates the left boundary of the window's sliding action; Using the center frequency of the signal as a reference, the number of sampling points contained in the direct wave of the guided signal is calculated, and the theoretical cutoff sampling point of the direct wave of the guided signal is obtained as the right boundary of the window sliding. in, Indicates the center frequency of the signal; F Indicates the number of peaks; Indicates the right boundary of the window's sliding action; Determine the window size; in, Indicates the window size; Using 100 sampling points as the sliding step size, the window slides gradually from the left boundary to the right boundary along the time axis, and the damage factor of the direct wave band of the intercepted guided wave signal is calculated window by window. Based on the benchmark feature sample set H N×M =[ x 1,…, x i ,…, x N ], x i =( This 1 , This 2,…, This M ),Establish x i The vector autoregressive model, where, x i Indicates the first i Group of guided wave signals, This M Indicates the first M The damage factor, the first i Group of guided wave signals x i The vector autoregression model is represented as follows: Among them, Π p Indicates the first p An M×M parameter matrix; μ i Represents a white noise vector; c Represents an M×1 constant vector; p This indicates the lag order of the vector autoregressive model; Subtract both sides of the vector autoregression model equation. x i-1 To obtain the result of subtracting both sides simultaneously x i-1 The vector autoregression model ∆ x i It is expressed as follows: in, , indicating the first q The coefficient matrix is ​​defined by the parameter matrix of a vector autoregressive model. , representing the identity matrix The coefficient matrix defined by the parameter matrix; Will The decomposition is represented as follows: Where A and B both represent the... The coefficient matrix obtained by full-rank decomposition; the superscript T indicates the transpose sign; The coefficient matrix B is obtained using the maximum likelihood estimation method, and both sides are subtracted simultaneously. x i-1 The vector autoregression model ∆ x i The probability density function is f (∆ x i ), the probability density function f (∆ x i Substitute the white noise vector into the vector autoregressive model. μ i For probability density function f (∆ x i Take the logarithm on both sides to establish the maximum likelihood function. L , means as follows: ; The coefficient matrix B is obtained through the maximum likelihood function. L The first column vector of the coefficient matrix B is estimated to be... β= [ β 1, β 2,…, β M ] T For probabilistic statistical association features, the probabilistic statistical association features are... β Compared with the benchmark feature sample set H N×M Multiplication yields the first digit in a single channel. i The benchmark residuals of each benchmark feature sample , means as follows: ; e = [ e 1, e 2,…, e i ,…, e N [The reference residual sequence is shown.] When each set of guided wave signals is used to construct a monitoring feature sample set through the dynamic window sliding method S L×M and the monitoring feature sample set S L×MWhen the monitoring residual sequence is obtained by multiplying it with the probabilistic statistical correlation features as input for probabilistic statistical modeling, the method and steps used are the same as those used to obtain the baseline residual sequence.

[0014] A structural damage alarm system based on piezoelectric guided wave-adaptive network probabilistic correlation includes: one or more processors; and a storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described above, and perform structural damage alarm based on piezoelectric guided wave-adaptive network probabilistic correlation according to the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation.

[0015] A computer-readable storage medium includes a computer program that, when executed by a processor, implements the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described above.

[0016] The structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation proposed in this application eliminates extreme interference from multiple channels through a truncated mean network fusion method, and fully utilizes the advantages of piezoelectric sensor networks to achieve comprehensive monitoring of damage to the monitored structure. By constructing a multivariate statistical feature fusion threshold to eliminate the dependence on a pre-set fixed threshold, the reliability and adaptability of the alarm under time-varying conditions are improved. The algorithm is simple, accurate, and efficient, and performs large-area, high-coverage, and full-domain collaborative guided wave health monitoring of the monitored structure. It achieves accurate identification and reliable alarm of crack damage in complex structures under the influence of time-varying factors, and the alarm reliability is improved to 95%.

[0017] To make the above-mentioned features and advantages of the application more apparent and understandable, specific embodiments are provided below, and detailed descriptions are given in conjunction with the accompanying drawings. Attached Figure Description

[0018] Figure 1 This is a flowchart of the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation proposed in this application.

[0019] Figure 2 This is a schematic diagram of the monitored structure 1 and the piezoelectric sensor network arrangement in a specific embodiment of this application.

[0020] Figure 3 This is a schematic diagram of the damage arrangement of the monitored structure 1 according to a specific embodiment of this application.

[0021] Figure 4 This is a schematic diagram of the reference alarm features, damage alarm features, and multivariate statistical feature fusion threshold of the guided wave signal collected in a specific embodiment of this application.

[0022] In the accompanying drawings, similar reference numerals refer to the same elements. Detailed Implementation

[0023] To make the objectives and technical solutions of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the described embodiments of this application without creative effort are within the scope of protection of this application.

[0024] In one embodiment of this application, please refer to Figure 1 , Figure 1 This is a flowchart of the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation proposed in this application. The structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation proposed in this application includes steps S1 to S4.

[0025] Step S1: Obtain the baseline feature sample set of the monitored structure using probabilistic statistical modeling combined with dynamic window sliding method. H N×M and monitoring feature sample set S L×M The baseline alarm characteristics of each channel are obtained. HDI and damage alarm characteristics DDI .

[0026] Step S2: Employ the truncated mean network fusion method to analyze the baseline alarm characteristics of each channel. HDI The fusion process yields the fused baseline alarm characteristics. HDI' Damage alarm characteristics for each channel DDI The fusion process yields the fused damage alarm features. DDI' .

[0027] Step S3: Based on the fused baseline alarm characteristics HDI' Adaptively construct a multivariate statistical feature fusion threshold (UCL).

[0028] Step S4: During the damage monitoring phase of the monitored structure, when the fused damage alarm features... DDI' When the value exceeds the multivariate statistical feature fusion threshold (UCL), the monitored structure is determined to have suffered damage, and a structural damage alarm is triggered.

[0029] The structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation proposed in this application eliminates extreme interference from multiple channels through a truncated mean network fusion method, and fully utilizes the advantages of piezoelectric sensor networks to achieve comprehensive monitoring of damage to the monitored structure. By constructing a multivariate statistical feature fusion threshold to eliminate the dependence on a pre-set fixed threshold, the reliability and adaptability of the alarm under time-varying conditions are improved. The algorithm is simple, accurate, and efficient, and performs large-area, high-coverage, and full-domain collaborative guided wave health monitoring of the monitored structure. It achieves accurate identification and reliable alarm of crack damage in complex structures under the influence of time-varying factors, and the alarm reliability is improved to 95%.

[0030] In step S1, please refer to Figure 1 In step S1, a baseline feature sample set of the monitored structure is obtained by using probabilistic statistical modeling combined with dynamic window sliding method. H N×M and monitoring feature sample set S L×M The baseline alarm characteristics of each channel are obtained. HDI and damage alarm characteristics DDI .

[0031] As an example, obtain a baseline feature sample set of the monitored structure. H N×M and monitoring feature sample set S L×M Previously, this also included: deploying [a system] on the surface of the monitored structure. a An array of piezoelectric sensors, in which a The integer is greater than 2; by selecting one piezoelectric sensor as the excitation source and another piezoelectric sensor as the receiving end, a guided wave excitation-sensing channel can be formed. Multiple excitation-receiving combinations can then be used to form a... n One channel, n The channels together form a piezoelectric sensor network for structural state sensing, the number of channels... .

[0032] As an example, a baseline feature sample set of the monitored structure is obtained by using probabilistic statistical modeling combined with dynamic window sliding. H N×M and monitoring feature sample set S L×M The baseline alarm characteristics of each channel are obtained. HDI and damage alarm characteristics DDI This includes steps S11 to S13.

[0033] Step S11: Under the healthy condition of the monitored structure, collect data from the piezoelectric sensor network. N+1 group of guided wave signals, using one of these groups as the reference signal. For each channel, each group of guided wave signals is adaptively truncated using a dynamic window sliding method to extract... M Damage factors This , N A set of guided wave signals forms a reference feature sample set. H N×M =[ x 1,…, x i ,…, x N ],in, x i Indicates the first i Group of guided wave signals, x i =( This 1 , This 2,…, This M ), This M Indicates the first M Each damage factor.

[0034] Step S12, set the benchmark feature sample set H N×M As input for probabilistic and statistical modeling, calculate probabilistic and statistical correlation features. β and benchmark residual sequence e ; for the baseline residual sequence e Perform absolute value calculations to obtain the baseline alarm characteristics for the corresponding channel. HDI .

[0035] Step S13: Enter the damage monitoring stage of the monitored structure. L sets of guided wave signals are collected again in each channel. Multiple damage factors are obtained from each set of guided wave signals through the dynamic window sliding method. Any set of guided wave signals constitutes the monitoring feature sample set. S L×M =[ s 1,…, s j ,…, s L ],in, s j Indicates the first j Group of guided wave signals, s j =( This 1 , This 2,…, This M ). Monitor feature sample set S L×MAs input for probabilistic and statistical modeling, features associated with probabilistic and statistical data. β Multiply to obtain the monitoring residual sequence e s ; For monitoring residual sequences e s Perform absolute value calculations to obtain the damage alarm features for the corresponding channel. DDI .

[0036] Specifically, in step S11, the dynamic window sliding method includes steps S111 to S114.

[0037] Step S111: Calculate the theoretical time of flight of the guided wave and convert it into discrete sampling points to obtain the starting position of the direct wave of each group of guided wave signals as the left boundary of the window sliding. .

[0038] As an example, the left boundary of the window sliding. It is expressed as follows: in, d Indicates the spacing between piezoelectric sensors in the corresponding channel; v This represents the theoretical wave velocity of the guided wave. Indicates the sampling rate of the signal; This indicates the pre-trigger length of the signal acquisition system.

[0039] Step S112: Using the signal center frequency as a reference, calculate the number of sampling points contained in the direct wave of the guided signal, and obtain the theoretical cutoff sampling point of the direct wave of the guided signal as the right boundary of the window sliding. .

[0040] As an example, the right boundary of the window sliding. It is expressed as follows: in, Indicates the center frequency of the signal; F Indicates the number of peaks.

[0041] Specifically, when the trailing edge of the window touches the theoretical cutoff sampling point of the direct wave of the guided wave signal, it is determined to be the right boundary of the window sliding, and the window sequence immediately terminates the sliding.

[0042] Step S113: Determine window size .

[0043] As an example, window size It is expressed as follows: Step S114: Using 100 sampling points as the sliding step, slide along the time axis from the left boundary of the window. Slide to the right edge of the window By sliding the slider step by step, the damage factor of the direct waveband of the intercepted guided wave signal is calculated window by window.

[0044] As an example, the damage factor of the direct-arrival waveband of the Mth intercepted guided wave signal. It is expressed as follows: in, t 1 and t 2 represents the start and end times of the direct waveband of the intercepted guided wave signal, respectively; Indicates a reference signal; This represents the d-th segment of the guided wave signal intercepted during time period t.

[0045] As an example, the sliding step size can be dynamically adjusted based on the measured characteristics of the guided wave signal.

[0046] Specifically, in step S12, the probability and statistical modeling includes steps S121 to S124.

[0047] Step S121: Based on the benchmark feature sample set H N×M Establish the first i Group of guided wave signals x i The vector autoregressive model.

[0048] As an example, the first i Group of guided wave signals x i The vector autoregressive (VAR) model is represented as follows: Among them, Π p Indicates the first p An M×M parameter matrix; μ i Represents a white noise vector; c Represents an M×1 constant vector; p This represents the lag order of the vector autoregressive model.

[0049] Step S122: Subtract both sides of the vector autoregression model equation simultaneously. x i-1 To obtain the result of subtracting both sides simultaneously x i-1 The vector autoregression model ∆ x i It is expressed as follows: in, , indicating the first qThe coefficient matrix is ​​defined by the parameter matrix of a vector autoregressive model. , representing the identity matrix The coefficient matrix defined by the parameter matrix.

[0050] Furthermore, The decomposition is represented as follows: Where A and B both represent the... The coefficient matrix obtained by full-rank decomposition; the superscript T indicates the transpose symbol.

[0051] Step S123: Obtain the coefficient matrix B using the maximum likelihood estimation method, and subtract from both sides simultaneously. x i-1 The vector autoregression model ∆ x i The probability density function is f (∆ x i ), the probability density function f (∆ x i Substitute the white noise vector into the vector autoregressive model. μ i For probability density function f (∆ x i Take the logarithm on both sides to establish the maximum likelihood function. L , means as follows: Step S124: The coefficient matrix B is passed through the maximum likelihood function. L The first column vector of the coefficient matrix B is estimated to be... β= [ β 1, β 2,…, β M ] T For probabilistic statistical association features, the probabilistic statistical association features are... β Compared with the benchmark feature sample set H N×M Multiplication yields the first digit in a single channel. i The benchmark residuals of each benchmark feature sample , means as follows: Furthermore, e = [ e 1, e 2,…, e i,…, e N [ ] represents the baseline residual sequence, indicating the baseline feature sample set in a single channel. H N×M All baseline residuals.

[0052] Furthermore, for the baseline residual sequence e Perform absolute value calculations to obtain the baseline alarm characteristics for the corresponding channel. HDI Baseline alarm characteristics HDI It is a sequence.

[0053] Specifically, in step S13, each group of guided wave signals constructs a monitoring feature sample set using the dynamic window sliding method. S L×M and the monitoring feature sample set S L×M As input for probabilistic and statistical modeling, features associated with probabilistic and statistical data. β Multiply to obtain the monitoring residual sequence e s The principle, process, and steps of obtaining the benchmark residual sequence are as follows: S12 e The same applies, so I won't repeat it here; damage alarm characteristics DDI It is also a set of sequences.

[0054] Specifically, in the healthy state of the monitored structure and during the damage monitoring phase, the damage factor of each group of guided wave signals is adaptively truncated using the dynamic window sliding method. This The quantities are the same.

[0055] In step S2, please refer to Figure 1 In step S2, a truncated mean network fusion method is used to analyze the baseline alarm characteristics of each channel. HDI The fusion process yields the fused baseline alarm characteristics. HDI' Damage alarm characteristics for each channel DDI The fusion process yields the fused damage alarm features. DDI' .

[0056] As an example, the truncated mean network fusion method includes steps S21 to S22.

[0057] Step S21: Combine the data to be fused from each channel. Z= ( z 1, z 2,…, z n Sort the data in ascending order to obtain the data to be fused for each channel. Z s = ( z (1) ,z (2) ,…, z (n) ), before truncation m The minimum value and the subsequent m There are 1 maximum value. z n This represents the data to be fused in the nth channel; z (n) This represents the data to be fused in the nth channel after arrangement. z (1) ≤ z (2) …≤ z (n) .

[0058] Step S22: Calculate the arithmetic mean of the data to be fused in each channel to obtain the fusion result. Z’ .

[0059] Specifically, in step S21, the data to be fused includes: baseline alarm features. HDI and damage alarm characteristics DDI .

[0060] Specifically, settings are dynamically configured based on actual alarm conditions. m .

[0061] In one embodiment of this application, when the number of channels n When ≤10, the cutoff number m= 1; When the number of channels is 10≤ n When ≤30, the cutoff number m= 2; When the number of channels n When >30, the cutoff number m=floor (0.05× n This ensures that the remaining available channels are greater than 95%.

[0062] Specifically, in step S22, the fusion result Z’ Includes: fused baseline alarm features HDI' and the damage alarm characteristics after fusion DDI' Fusion results Z’ It is expressed as follows: in, Indicates the first l Data to be merged from each channel.

[0063] Specifically, the data to be fused is used as the damage alarm feature. DDI For example, the fused damage alarm features are obtained. DDI' , means as follows: in, Indicates the first l Damage alarm features of each channel DDI .

[0064] Specifically, the fused damage alarm features DDI' It is a sequence.

[0065] Furthermore, the fused baseline alarm characteristics are obtained. HDI' The principle and process of the fusion and the resulting damage alarm features DDI' The same applies, so I won't repeat it here; the fused baseline alarm characteristics HDI' It is also a set of sequences.

[0066] In step S3, please refer to Figure 1 In step S3, based on the fused baseline alarm characteristics... HDI' Adaptively construct a multivariate statistical feature fusion threshold (UCL).

[0067] As an example, the steps for constructing a multivariate statistical feature fusion threshold (UCL) include: steps S31 to S35.

[0068] Step S31: Calculate the fused baseline alarm features HDI' mean and standard deviation .

[0069] Step S32: Calculate the fused baseline alarm features HDI' peak and slope .

[0070] Step S33: Determine the fused baseline alarm characteristics HDI' mean coefficient .

[0071] Step S34: Determine the fused baseline alarm characteristics HDI' Standard deviation coefficient .

[0072] Step S35: Based on the fused baseline alarm characteristics HDI' mean Standard deviation The corresponding coefficients are used to calculate the multivariate statistical feature fusion threshold (UCL).

[0073] Specifically, in step S32, the fused baseline alarm features are calculated. HDI' peak and slope Used to measure the baseline alarm characteristics after fusion. HDI' The deviation from the normal distribution is represented as follows: in, Indicates the fused baseline alarm characteristics HDI' The first in the sequence i One value; Indicates the fused baseline alarm characteristics HDI' The mean.

[0074] Specifically, in step S33, the fused baseline alarm features HDI' mean coefficient , means as follows: .

[0075] Specifically, in step S34, the fused baseline alarm features HDI' Standard deviation coefficient , means as follows: .

[0076] Specifically, in step S35, the multivariate statistical feature fusion threshold (UCL) is expressed as follows: .

[0077] In step S4, please refer to Figure 1 In step S4 of the monitoring process, during the damage monitoring phase of the monitored structure, when the fused damage alarm features... DDI' When the value exceeds the multivariate statistical feature fusion threshold (UCL), the monitored structure is determined to have suffered damage, and a structural damage alarm is triggered.

[0078] Specifically, due to the fused damage alarm characteristics DDI' As a set of sequences, the fused damage alarm features DDI' If any feature in the sequence is greater than the multivariate statistical feature fusion threshold (UCL), then the monitored structure is determined to be damaged and a structural damage alarm is triggered. At the same time, any feature in the sequence that is greater than the multivariate statistical feature fusion threshold (UCL) indicates that the structure is in a damaged state.

[0079] In a specific embodiment of this application, a metal wing skin is used as the monitored structure. Considering the influence of temperature change, a typical time-varying factor, and combined with guided wave structure health monitoring technology, the implementation process of the method of the present invention is specifically illustrated using metal wing skin damage alarm as an example in a time-varying environment.

[0080] Please see Figure 2 , Figure 2This is a schematic diagram of the monitored structure 1 and the piezoelectric sensor network arrangement according to a specific embodiment of this application. The geometric dimensions of the monitored structure 1 and the spatial arrangement of the piezoelectric sensor network are shown below. Figure 2 As shown, a total of 32 piezoelectric sensors are arranged on the surface of the monitored structure 1, and are numbered according to a predetermined coordinate sequence from piezoelectric sensor PZT1 to piezoelectric sensor PZT32; the surface of the monitored structure 1 also includes structural ribs 11; any piezoelectric sensor can be selected as a guided wave excitation source, and any other piezoelectric sensor can be used as a response receiving unit. The two constitute a reversible excitation-sensing channel, and all channels form a large-area piezoelectric sensor network covering the monitored structure 1.

[0081] Furthermore, under the healthy condition of the monitored structure 1, guided wave signals propagating on the surface of the monitored structure 1 are collected. Guided wave signals of the monitored structure 1 under healthy condition are collected during a temperature change process from -40℃ to 0℃ as reference samples. The sampling rate in the guided wave signal acquisition parameters is... The signal center frequency is 10MHz. At 80kHz, number of peaks F Let the value be 3. The signal acquired in this stage is denoted as... , i =1, 2, …, 15, representing the i-th group of guided wave signals acquired, with the first acquired signal serving as the reference signal. Then, the damage monitoring stage of the monitored structure begins. Under healthy conditions, the temperature change process is repeated to acquire 16 groups of healthy monitoring signals as health monitoring samples. Finally, as shown... Figure 3 As shown, a 5mm crack was applied to the structural rib 11. After the crack was applied, the temperature change process was repeated. Nineteen sets of guided wave signals were collected from the damaged structure 1 as damage monitoring samples. A total of 35 sets of guided wave signals were collected during the damage monitoring stage of the monitored structure 1. , j =1, 2, …,35.

[0082] Furthermore, the window size is calculated in the dynamic window sliding parameters based on the signal acquisition parameters. Approximately 150, with a sliding step of 100 sampling points, the pre-trigger length of the signal acquisition system. The value is 500. After the calculation window is slid 8 times, the damage factor of the direct wave band of the intercepted guided wave signal is calculated window by window.

[0083] Furthermore, a baseline feature sample set for each channel is constructed. H 15×8 =[ x 1, x 2, …, x i , …, x 15 ],x i =( This 1, This 2,…, This 8) Construct a monitoring feature sample set for a channel. S 35×8 =[ s 1, s 2, …, s j , …, s 35 ], s j =( This 1, This 2,…, This 8). Based on the benchmark feature sample set H 15×8 Perform probabilistic statistical modeling to obtain a baseline feature sample set. H 15×8 Probability and statistical correlation features β= [ β 1, β 2,…, β 8] T and benchmark residual sequence e= [ e 1, e 2,…, e 15 [] For the baseline residual sequence e Perform absolute value operation element by element to obtain the reference alarm characteristics of one channel. HDI The baseline alarm characteristics of all channels were calculated using the same method. HDI The truncated mean network fusion method is used to combine the baseline alarm features of all channels. HDI The fused baseline alarm features HDI' .

[0084] Furthermore, the monitoring feature sample set S 35×8 Features associated with probability and statistics β Multiplication to obtain the monitoring feature sample set S 35×8 Monitoring residual sequence e s For monitoring residual sequences e s Perform absolute value calculations element by element to obtain the damage alarm characteristics of a channel. DDI Damage alarm characteristics for all channels were calculated using the same method. DDI The truncated mean network fusion method is used to combine the damage alarm features of all channels. DDI The damage alarm features are fused together. DDI' .

[0085] Furthermore, based on the fused baseline alarm characteristics HDI' Through kurtosis and slope Measuring the benchmark alarm characteristics HDI' To account for deviations from the normal distribution, a multivariate statistical feature fusion threshold (UCL) is adaptively constructed. When a fused damage alarm feature exists... DDI' When the value exceeds the multivariate statistical feature fusion threshold (UCL), the monitored structure 1 is determined to have suffered damage, and a structural damage alarm is triggered.

[0086] Please see Figure 4 , Figure 4 This diagram illustrates the baseline alarm features, damage alarm features, and multivariate statistical feature fusion threshold of the guided wave signal acquired in a specific embodiment of this application. When the damage alarm features of the damage monitoring sample exceed the damage alarm threshold line of the multivariate statistical feature fusion threshold, damage is determined to have occurred in the monitored structure 1, triggering a structural damage alarm.

[0087] This application also provides a structural damage alarm system based on piezoelectric guided wave-adaptive network probabilistic correlation, comprising: one or more processors; and a storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described above, and perform structural damage alarm based on piezoelectric guided wave-adaptive network probabilistic correlation according to the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation.

[0088] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described above.

[0089] The structural damage alarm method, system, and computer program product based on piezoelectric guided wave-adaptive network probabilistic correlation proposed in this application eliminate extreme interference from multiple channels through a truncated mean network fusion method, and fully utilize the advantages of piezoelectric sensor networks to achieve comprehensive monitoring of damage to the monitored structure. By constructing a multivariate statistical feature fusion threshold to eliminate the dependence on a pre-set fixed threshold, the reliability and adaptability of the alarm under time-varying conditions are improved. The algorithm is simple, accurate, and efficient, and performs large-area, high-coverage, and full-domain collaborative guided wave health monitoring of the monitored structure. It achieves accurate identification and reliable alarm of crack damage in complex structures under the influence of time-varying factors, and the alarm reliability is improved to 95%.

[0090] Although this application has been disclosed above with reference to embodiments, it is not intended to limit this application. Anyone skilled in the art may make some modifications and refinements without departing from the spirit and scope of this application. Therefore, the scope of protection of this application shall be determined by the appended claims.

Claims

1. A structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation, characterized in that, include, The baseline feature sample set and monitoring feature sample set of the monitored structure are obtained by using probabilistic statistical modeling combined with dynamic window sliding method, and the baseline alarm features and damage alarm features of each channel are obtained. The truncated mean network fusion method is used to fuse the baseline alarm features of each channel to obtain the fused baseline alarm features, and the damage alarm features of each channel are fused to obtain the fused damage alarm features. Based on the fused baseline alarm features, an adaptive multivariate statistical feature fusion threshold is constructed. During the damage monitoring phase of the monitored structure, when the fused damage alarm features are greater than the multivariate statistical feature fusion threshold, the monitored structure is determined to have suffered damage and a structural damage alarm is triggered.

2. The structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described in claim 1, characterized in that, Truncated mean network fusion methods include: Data to be merged from each channel Z= ( z 1, z 2,…, z n Sort in ascending order to get Z s = ( z (1) , z (2) ,…, z (n) ),in z (1) ≤ z (2) …≤ z (n) Before truncation m The minimum value and the subsequent m The maximum value; the data to be fused includes: baseline alarm features and damage alarm features; Calculate the arithmetic mean of the data to be fused from each channel to obtain the fusion result. Z’ ; Fusion results Z’ Includes: the fused baseline alarm features and the fused damage alarm features; fusion results. Z’ It is expressed as follows: in, Indicates the first l Data to be merged from each channel; n Indicates the number of channels; m This indicates the number of interruptions, which is set according to the actual alarm situation.

3. The structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described in claim 2, characterized in that, The steps for constructing a multivariate statistical feature fusion threshold include: Calculate the mean and standard deviation of the fused baseline alarm characteristics; Calculate the kurtosis and slope of the fused baseline alarm features; Determine the coefficients for the mean value of the fused baseline alarm characteristics; Determine the coefficients of the standard deviation of the fused baseline alarm characteristics; The multivariate statistical feature fusion threshold is calculated based on the mean, standard deviation, and corresponding coefficients of the fused baseline alarm features.

4. The structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described in claim 3, characterized in that, The kurtosis and slope of the fused baseline alarm features are calculated and represented as follows: in, Indicates the slope of the fused baseline alarm feature; Indicates the fused baseline alarm characteristics HDI Peak value; Indicates the fused baseline alarm characteristics HDI The first in the sequence i One value; Indicates the fused baseline alarm characteristics HDI The mean value; N represents the number of guided wave signals collected when the monitored structure is in a healthy state.

5. The structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described in claim 3, characterized in that, The coefficients of the mean of the fused baseline alarm features are expressed as follows: in, The coefficient representing the mean of the fused baseline alarm characteristics.

6. The structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described in claim 2, characterized in that, The coefficient of the standard deviation of the fused baseline alarm characteristics is expressed as follows: in, The coefficient representing the standard deviation of the fused baseline alarm characteristics.

7. The structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described in claim 3, characterized in that, The multivariate statistical feature fusion threshold (UCL) is expressed as follows: Wherein, UCL represents the threshold for fusion of multivariate statistical features; Indicates the fused baseline alarm characteristics HDI The mean; Indicates the fused baseline alarm characteristics HDI The standard deviation.

8. The structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described in claim 1, characterized in that, A baseline feature sample set and a monitoring feature sample set of the monitored structure are obtained. Probabilistic statistical modeling combining dynamic window sliding method is used to obtain the baseline alarm features and damage alarm features for each channel, including: Under the healthy condition of the monitored structure, guided wave signals from any set of piezoelectric sensor networks are collected, with one set serving as the reference signal. For each channel, all guided wave signals are adaptively truncated using a dynamic window sliding method to extract multiple damage factors, and any set of guided wave signals forms a reference feature sample set. After entering the damage monitoring stage, each channel again collects any set of guided wave signals, and all guided wave signals are adaptively truncated using a dynamic window sliding method to extract multiple damage factors, and any set of guided wave signals forms a monitoring feature sample set. The reference feature sample set and the monitoring feature sample set are used as inputs for probabilistic statistical modeling to obtain the reference alarm features and damage alarm features. Probabilistic statistical modeling combined with the dynamic window sliding method specifically includes: The theoretical flight time of the guided wave is calculated and converted into discrete sampling points. The starting position of the direct wave of each group of guided wave signals is used as the left boundary of the window sliding. in, d Indicates the spacing between piezoelectric sensors in the corresponding channel; v This represents the theoretical wave velocity of the guided wave. Indicates the sampling rate of the signal; Indicates the pre-trigger length of the signal acquisition system; Indicates the left boundary of the window's sliding action; Using the center frequency of the signal as a reference, the number of sampling points contained in the direct wave of the guided signal is calculated, and the theoretical cutoff sampling point of the direct wave of the guided signal is obtained as the right boundary of the window sliding. in, Indicates the center frequency of the signal; F Indicates the number of peaks; Indicates the right boundary of the window's sliding action; Determine the window size; in, Indicates the window size; Using 100 sampling points as the sliding step size, the window slides gradually from the left boundary to the right boundary along the time axis, and the damage factor of the direct wave band of the intercepted guided wave signal is calculated window by window. Based on the benchmark feature sample set H N×M =[ x 1,…, x i ,…, x N ], x i =( DI 1 , DI 2,…, DI M ),Establish x i The vector autoregressive model, where, x i Indicates the first i Group of guided wave signals, DI M Indicates the first M The damage factor, the first i Group of guided wave signals x i The vector autoregression model is represented as follows: Among them, Π p Indicates the first p An M×M parameter matrix; μ i Represents a white noise vector; c Represents an M×1 constant vector; p This indicates the lag order of the vector autoregressive model; Subtract both sides of the vector autoregression model equation. x i-1 To obtain the result of subtracting both sides simultaneously x i-1 The vector autoregression model ∆ x i It is expressed as follows: in, , indicating the first q The coefficient matrix is ​​defined by the parameter matrix of a vector autoregressive model. , representing the identity matrix The coefficient matrix defined by the parameter matrix; Will The decomposition is represented as follows: Where A and B both represent the... The coefficient matrix obtained by full-rank decomposition; the superscript T indicates the transpose sign; The coefficient matrix B is obtained using the maximum likelihood estimation method, and both sides are subtracted simultaneously. x i-1 The vector autoregression model ∆ x i The probability density function is f (∆ x i ), the probability density function f (∆ x i Substitute the white noise vector into the vector autoregressive model. μ i For probability density function f (∆ x i Take the logarithm on both sides to establish the maximum likelihood function. L , means as follows: ； The coefficient matrix B is obtained through the maximum likelihood function. L The first column vector of the coefficient matrix B is estimated to be... β= [ β 1, β 2,…, β M ] T For probabilistic statistical association features, the probabilistic statistical association features are... β Compared with the benchmark feature sample set H N×M Multiplication yields the first digit in a single channel. i The benchmark residuals of each benchmark feature sample , means as follows: ； e = [ e 1, e 2,…, e i ,…, e N [The reference residual sequence is shown.] The baseline alarm characteristics of the corresponding channel are obtained by performing a stepwise absolute value operation on the baseline residual sequence; When each set of guided wave signals is used to construct a monitoring feature sample set through the dynamic window sliding method S L×M and the monitoring feature sample set S L×M As input for probabilistic statistical modeling, when multiplying with probabilistic statistical correlation features to obtain the monitoring residual sequence, the method and steps used are the same as those used to obtain the baseline residual sequence; the damage alarm features of the corresponding channel are obtained by performing a stepwise absolute value operation on the monitoring residual sequence.

9. A structural damage alarm system based on piezoelectric guided wave-adaptive network probabilistic correlation, characterized in that, include: One or more processors; A storage device is configured to store one or more programs that, when executed by one or more processors, cause the one or more processors to implement the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described in any one of claims 1-8, and to perform structural damage alarm based on piezoelectric guided wave-adaptive network probabilistic correlation according to the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation.

10. A computer-readable storage medium comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the structural damage alarm method based on piezoelectric guided wave-adaptive network probabilistic correlation as described in any one of claims 1 to 8.

Images