Structural damage clustering monitoring method and system based on strain distribution similarity measure

Abstract

The application provides a structure damage clustering monitoring method based on strain distribution similarity measurement, and belongs to the technical field of aviation structure health monitoring. The method sets multiple monitoring paths around a damage monitoring area, and arranges strain sensors on the monitoring paths. Strain distribution data are collected in a structure health state and a damage monitoring state respectively under the same loading condition. The strain distribution Fréchet distance is calculated based on a strain distribution benchmark of each monitoring path, and the strain distribution Fréchet distance is used as a high-dimensional space point coordinate to perform DBSCAN density clustering. Whether the structure appears abnormal strain distribution caused by damage is determined through the clustering result, and damage monitoring is realized. The method is easy to implement, efficient, and accurate in structure damage identification, and can be used for aircraft structure damage monitoring.

Description

Technical Field

[0001] This invention belongs to the field of aerospace structural health monitoring technology, and in particular relates to a structural damage clustering monitoring method and system based on strain distribution similarity measurement. Background Technology

[0002] With the continuous development of aviation science and technology, aircraft structural design and manufacturing are constantly moving towards the goals of lightweight, high strength and long service life. In order to ensure the long-term safe use and condition-based maintenance of new aircraft structures during service, it is necessary to conduct long-term and effective monitoring of the health status of the structure to determine whether structural damage that affects flight safety has occurred. Therefore, the importance of aviation structural health monitoring technology is becoming increasingly prominent.

[0003] Currently, there are many types of technologies for monitoring the health of aerospace structures, such as piezoelectric active waveguide technology, smart coating technology, strain monitoring methods, acoustic emission methods, and eddy current methods. Since strain measurement technologies such as electrical strain measurement and fiber optic strain measurement are relatively mature and strain measurements can directly reflect the structural strength state, this method has good application prospects in the field of aerospace structural health monitoring. However, directly using strain measurements at a single point in the structure for damage monitoring has the problems of a small monitorable area and low sensitivity. Damage monitoring can generally only be achieved through abnormal strain values ​​when structural damage occurs at the measurement point, making it difficult to reliably monitor structural damage over a certain area. Therefore, a method based on strain measurement that can monitor structural damage over a specific area is needed. Summary of the Invention

[0004] To address the limitation of traditional single-point strain measurement methods in monitoring structural damage within a given area, this invention proposes a structural damage clustering monitoring method based on strain distribution similarity metrics. Based on the linear elastic mechanical characteristics of aircraft structures, this invention considers measuring the multi-point distributed strain within the monitoring area. Combining typical operational loading conditions of the aircraft structure, it quantitatively describes the structural strain distribution using similarity metrics algorithms for both damage-free and damage monitoring stages. Furthermore, a density clustering analysis algorithm is employed to categorize the structural strain distribution states and identify strain distribution states caused by structural anomalies.

[0005] The technical solution is as follows: Firstly, a structural damage clustering monitoring method based on strain distribution similarity measurement is provided. Multiple monitoring paths are set up around the damage monitoring area, and strain measurement sensors are deployed on the monitoring paths. Strain distribution data are collected under the same loading conditions, in both the undamaged and damage monitoring states of the structure. The Fréchet distance of the strain distribution is calculated based on the strain distribution benchmark of each monitoring path, and this distance is used as the coordinates of high-dimensional spatial points for DBSCAN density clustering. The clustering results are used to determine whether the structure has abnormal strain distribution caused by damage, thereby realizing damage monitoring.

[0006] The method specifically includes: Step S1: Determine multiple monitoring paths at the monitoring locations of the aircraft structure and determine the direction of each monitoring path. The number of monitoring paths N is greater than or equal to 2. Step S2: Deploy multiple strain measurement sensors on each monitoring path. The number of sensors should be greater than or equal to 3. Taking 3 is to facilitate the formation of strain distribution, so that the strain distribution similarity measurement method can be used for regional damage monitoring in the future. Step S3: Under the condition of no structural damage and under external load, synchronously collect the monitoring data of each strain measurement sensor on each monitoring path of the structure to obtain the no-damage strain distribution dataset for each monitoring path. For any monitoring path i, its no-damage strain distribution dataset is m. H ×n i matrix H i Each row of the matrix represents a strain distribution sample at the same time along the monitoring path, therefore m H n represents the number of samples in the undamaged strain distribution. i This represents the number of strain measurement sensors along the monitoring path. Step S4: For any monitoring path i, calculate the strain distribution H when the structure is undamaged and not subjected to external loads. i0 Based on the matrix H i The strain distribution similarity metric matrix was calculated using the Friesian distance method, and the similarity metric matrices of each path were horizontally stacked to form the strain distribution similarity metric matrix SH of the structure in the undamaged state. SH is m H A matrix of size ×N; Step S5: Structural damage monitoring stage. Under the same external load as in step S3, synchronously collect monitoring data from strain monitoring sensors along each monitoring path of the structure to obtain the damage monitoring strain distribution dataset for each monitoring path. For any monitoring path i, its damage monitoring strain distribution dataset is m. D ×n i matrix D i , where m D The number of strain distribution samples for damage monitoring; Step S6: For any monitoring path i, use H i0Based on matrix D i The strain distribution similarity metric matrix was calculated using the Fraser distance method, and the similarity metric matrices of each path were horizontally stacked to form the structural damage monitoring strain distribution similarity metric matrix SD. SD is m D A matrix of size ×N; Step S7: Vertically stack the strain distribution similarity measurement matrix SH of the undamaged state and the strain distribution similarity measurement matrix SD of the damage monitoring stage to form matrix ST. Treat each row of matrix ST as the coordinates of N-dimensional spatial points and use the DBSCAN density clustering method to cluster all N-dimensional spatial points in matrix ST. Step S8: Count the cluster categories and the number of points in each category corresponding to the high-dimensional spatial points in the structurally undamaged stage and the damage monitoring stage, respectively. When a category that does not exist in the structurally undamaged stage appears in the damage monitoring stage, and the number of high-dimensional spatial points (PD) contained in that category accounts for a certain percentage (m) of the total number of high-dimensional spatial points in the damage monitoring stage, the clustering is calculated. D When the proportion exceeds the threshold T, it is determined that the structure monitoring part is damaged. T is selected in the range of (0,1), and it is recommended not to be lower than 0.6.

[0007] Wherein, matrix H i Sum matrix D i Each row of data is arranged in the order of strain monitoring data from each strain sensor at the same moment along the monitoring path.

[0008] The calculation process of the structural strain distribution similarity metric matrix SH in step S4 includes: Step S41, establish the strain distribution benchmark for each path. Normalization is performed, where the j-th element h i0,j Normalize according to the following formula:

[0009] Among them, h i0,j,norm for h i0,j The normalized value, h i0,min For H i0 The minimum value in, h i0,max For H i0 The maximum value in; The normalized strain distribution baseline H for each path is obtained after normalization. i0,norm ; Step S42, for the strain distribution dataset matrix H of each monitoring path i The data in each row is normalized using the same method as in step S41, resulting in the normalized dataset matrix H. i,norm ; Step S43, calculate H using the Fraser distance calculation method.i,norm Data for each row and the normalized baseline H i0,norm Frescher distance matrix ,in For H i,norm The j-th row is relative to H i0,norm The strain distribution and the Fréchet distance value; Step S44: Horizontally stack the Friesian distance matrices of the N monitoring paths to form a non-destructive strain distribution similarity metric matrix. ; The calculation process of the structural damage monitoring strain distribution similarity measurement matrix SD in step S6 is the same as the calculation process of matrix SH.

[0010] The process of clustering using the DBSCAN density clustering method in step S7 includes: Step S71: Set the minimum number of neighborhood points Minpts according to the size of the number of monitoring paths N. It is recommended that Minpts be greater than or equal to N+1. Step S72: Treat each row of matrix ST as the coordinates of a point in N-dimensional space, and calculate the distance d from each data point to its k-th nearest neighbor. k The Euclidean distance is used for calculation, where k = Minpts-1; Step S73, d of all coordinate points k The values ​​are sorted in ascending order, and d is plotted. k Value curve; Step S74, calculate d according to the following formula k The slope k of each point i on the value curve i :

[0011] Where d k,i d corresponding to point i k value; For all k i Value, determine its maximum value k max When k i / k max The value first exceeds the inflection point threshold T k When, then the inflection point i is in d k The corresponding d on the value curve k The value is used as the neighborhood radius Eps, T k Select within the range (0,1), and it is recommended not to exceed 0.1; Step S75: Perform DBSCAN clustering using the Minpts and Eps as parameters.

[0012] In step S75, the distance is calculated using Euclidean distance during the clustering process. Euclidean distance has a smaller computational load and higher computational efficiency.

[0013] Secondly, a structural damage clustering monitoring system based on strain distribution similarity measurement is provided for executing any of the structural damage clustering monitoring methods based on strain distribution similarity measurement described in the first aspect, comprising multiple strain measurement sensors and a processor. Strain measurement sensors are arranged on the monitoring path around the damage monitoring area to collect strain distribution data under the same loading conditions, both in the case of no structural damage and in the case of damage monitoring. The processor calculates the Fréchet distance of the strain distribution based on the strain distribution benchmark of each monitoring path, and uses this as the coordinates of high-dimensional spatial points to perform DBSCAN density clustering. The clustering results are used to determine whether the structure has abnormal strain distribution caused by damage, thereby realizing damage monitoring.

[0014] Among them, the strain measurement sensor is a fiber optic grating strain sensor, which is relatively simple to set up.

[0015] The beneficial effects of this invention are at least as follows: This invention utilizes strain distribution measurements along various monitoring paths within an aircraft structural monitoring area. By combining strain distribution similarity quantification with the DBSCAN clustering method, it achieves damage monitoring within the aircraft structural area. Compared to traditional single-point strain measurement methods for structural damage monitoring, this method covers a larger monitoring area, eliminates the need for precise pre-determination of monitoring points, and features a simple and efficient algorithm. Furthermore, this method fully leverages the quasi-distributed measurement networking capabilities and lightweight, small size of fiber optic strain sensors. Compared to using resistance strain gauges for strain distribution measurement, it does not add excessive mass to structural damage monitoring and is easy to install and implement. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram illustrating the example structure and the arrangement of fiber optic strain sensors.

[0018] Figure 2 d is calculated based on structural health status and damage monitoring data. k Value curve.

[0019] Figure 3 This is a bar chart showing the damage monitoring results based on DBSCAN cluster analysis. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The features and illustrative embodiments of various aspects of the present invention will now be described in detail. Numerous specific details are set forth in the following detailed description to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention may be practiced without requiring some of these specific details. The following description of embodiments is merely intended to provide a better understanding of the invention by illustrating examples of the invention. The invention is by no means limited to any specific setups and methods set forth below, but covers any improvements, substitutions, and modifications to structures, methods, and devices without departing from the spirit of the invention. Well-known structures and techniques are not shown in the drawings and the following description to avoid unnecessarily obscuring the invention.

[0022] This invention takes a critical connection point in an aircraft structure (the connection between the skin and the beam) as an example, using it as a monitoring structure. Fiber optic strain sensors are arranged in the monitoring area. The structure is subjected to the same load conditions, with and without damage at the edges of the connection holes. The specific implementation process of the method is illustrated using hole edge damage monitoring as an example. The specific implementation process is as follows: Step 1: Establish four monitoring paths within a 65mm x 40mm area around the structural connection point being monitored, and determine the direction of each path. The arrangement and path direction of the fiber optic strain sensors for each path are as follows: Figure 1 As shown, in this embodiment, four monitoring paths are set within a 65mm*40mm range around the structural connection part. Eight strain measurement points are set on monitoring path 1 and monitoring path 3, and five measurement points are set on monitoring path 2 and monitoring path 4.

[0023] Step 2: Set up 8 strain measurement points on monitoring path 1 and monitoring path 3, and 5 strain measurement points on monitoring path 2 and monitoring path 4.

[0024] Step 3: During the structural damage-free stage, load conditions are applied to the structure. During this loading process, data from strain sensors along each monitoring path are simultaneously collected to obtain a dataset of the damage-free strain distribution along each monitoring path, totaling 12382 samples. Then m... H=12382, forming a non-destructive strain distribution dataset for four monitoring paths: H1 is a 12382×8 matrix, H2 is a 12382×5 matrix, H3 is a 12382×8 matrix, and H4 is a 12382×5 matrix.

[0025] Step 4: Select the strain distribution H along each path when the structure is unloaded. 10 H 20 H 30 and H 40 Using this as a benchmark, and after normalization, the normalized strain distribution benchmark H is determined. 10,norm H 20,norm H 30,norm and H 40,norm Simultaneously, the non-destructive strain distribution datasets for each monitoring path are normalized to obtain the normalized dataset H for each path. 1,norm H 2,norm H 3,norm and H 4,norm For each monitoring path, the Fréchet distance between the normalized strain distribution represented by each row of its normalized dataset and the corresponding benchmark is calculated to obtain the Fréchet distance matrix for each monitoring path. Based on this, the Friesian distance matrices of the four monitoring paths are horizontally stacked to form a non-destructive strain distribution similarity metric matrix. .

[0026] Step 5: A 3mm crack damage sample was taken on the side of the connecting hole 1. The same load condition as in Step 3 was applied to the structure. During the loading process, the monitoring data of each strain sensor on each monitoring path of the structure were collected synchronously to obtain the damage monitoring strain distribution dataset of each monitoring path, with a total of 9906 samples. Therefore, mD=9906, which constitutes the damage monitoring strain distribution dataset of 4 monitoring paths. D1 is a 9906×8 matrix, D2 is a 9906×5 matrix, D3 is a 9906×8 matrix, and D4 is a 9906×5 matrix.

[0027] Step 6: Normalize the damage monitoring strain distribution datasets for each path to obtain normalized datasets D1,norm, D2,norm, D3,norm, and D4,norm. Calculate the Fréchet distance for the corresponding monitoring path using benchmarks H10,norm, H20,norm, H30,norm, and H40,norm, respectively, to obtain the Fréchet distance matrix for each monitoring path. Based on this, the Friesian distance matrices of the four monitoring paths are horizontally stacked to form a similarity measurement matrix for strain distribution in damage monitoring. .

[0028] Step 7: Vertically stack the non-destructive strain distribution similarity matrix SH and the damage monitoring strain distribution similarity matrix SD. For the newly formed matrix [SH SD]T, treat each row as a 4-dimensional spatial point coordinate and perform clustering using the DBSCAN density clustering method. Select the clustering parameter Minpts as 5, then k = Minpts - 1, which is 4. Calculate the distance dk from the 4th nearest point to each point, sort all dk values ​​in ascending order, and plot the dk value curve as shown below. Figure 2 As shown. The slope of the curve is calculated. In this embodiment, Tk is set to 0.1, the i value corresponding to the abrupt inflection point is 22271, the ki / kmax value is 0.153, and the corresponding dk value is 0.038. This value is used as the neighborhood radius Eps. The determined parameters Minpts and Eps are used to cluster the 4-dimensional spatial points represented by the matrix [SH SD]T. The clustering results are statistically shown below. Figure 3 As shown, the horizontal axis represents the cluster category number, and the vertical axis represents the number of data samples contained in each category. If the categories are the same, they are in the same category number. Considering the large number of data samples, the vertical axis is a logarithmic coordinate axis with base 10. In this embodiment, the threshold T is set to 0.8. After calculation, in the damage monitoring data, there is a category that does not exist in the structural no-damage stage and the number of high-dimensional spatial points PD contained in the category is 9893. This accounts for 0.998% of the total number of high-dimensional spatial points mD=9906 in the damage monitoring stage. Since this exceeds the threshold, it can be determined that damage has occurred in the structural monitoring part.

Claims

1. A structural damage clustering monitoring method based on strain distribution similarity measurement, characterized in that, Multiple monitoring paths are set up around the damage monitoring area, and strain measurement sensors are deployed along the monitoring paths. Strain distribution data are collected under the same loading conditions, both in the undamaged and damage monitoring states. The Fréchet distance of the strain distribution is calculated based on the strain distribution benchmark of each monitoring path, and this distance is used as the coordinates of high-dimensional spatial points for DBSCAN density clustering. The clustering results are used to determine whether the structure has abnormal strain distribution caused by damage, thereby realizing damage monitoring.

2. The structural damage clustering monitoring method based on strain distribution similarity measurement according to claim 1, characterized in that, The method specifically includes: Step S1: Determine multiple monitoring paths at the monitoring locations of the aircraft structure and determine the direction of each monitoring path. The number of monitoring paths N is greater than or equal to 2. Step S2: Deploy multiple strain measurement sensors on each monitoring path. The number of sensors should be greater than or equal to 3. Taking 3 is to facilitate the formation of strain distribution, so that the strain distribution similarity measurement method can be used for regional damage monitoring in the future. Step S3: Under the condition of no structural damage and under external load, synchronously collect the monitoring data of each strain measurement sensor on each monitoring path of the structure to obtain the no-damage strain distribution dataset for each monitoring path. For any monitoring path i, its no-damage strain distribution dataset is m. H ×n i matrix H i , where m H n represents the number of samples in the non-damage strain distribution. i This represents the number of strain measurement sensors along the monitoring path. Step S4: For any monitoring path i, calculate the strain distribution H when the structure is undamaged and not subjected to external loads. i0 Based on the matrix H i The strain distribution similarity metric matrix was calculated using the Friesian distance method, and the similarity metric matrices of each path were horizontally stacked to form the strain distribution similarity metric matrix SH of the structure in the undamaged state. SH is m H A matrix of size ×N; Step S5: Structural damage monitoring stage. Under the same external load as in step S3, synchronously collect monitoring data from strain monitoring sensors along each monitoring path of the structure to obtain the damage monitoring strain distribution dataset for each monitoring path. For any monitoring path i, its damage monitoring strain distribution dataset is m. D ×n i matrix D i , where m D The number of strain distribution samples for damage monitoring; Step S6: For any monitoring path i, use H i0 Based on matrix D i The strain distribution similarity metric matrix was calculated using the Fraser distance method, and the similarity metric matrices of each path were horizontally stacked to form the structural damage monitoring strain distribution similarity metric matrix SD. SD is m D A matrix of size ×N; Step S7: Vertically stack the strain distribution similarity measurement matrix SH of the undamaged state and the strain distribution similarity measurement matrix SD of the damage monitoring stage to form matrix ST. Treat each row of matrix ST as the coordinates of N-dimensional spatial points and use the DBSCAN density clustering method to cluster all N-dimensional spatial points in matrix ST. Step S8: Count the cluster categories and the number of points in each category corresponding to the high-dimensional spatial points in the structurally undamaged stage and the damage monitoring stage, respectively. When a category that does not exist in the structurally undamaged stage appears in the damage monitoring stage, and the number of high-dimensional spatial points (PD) contained in that category accounts for a certain percentage (m) of the total number of high-dimensional spatial points in the damage monitoring stage, the clustering is calculated. D When the proportion exceeds the threshold T, it is determined that damage has occurred in the structural monitoring area.

3. The structural damage clustering monitoring method based on strain distribution similarity measurement according to claim 1, characterized in that, Matrix H i Sum matrix D i Each row of data is arranged in the order of strain monitoring data from each strain sensor at the same moment along the monitoring path.

4. The structural damage clustering monitoring method based on strain distribution similarity measurement according to claim 1, characterized in that, The calculation process of the strain distribution similarity metric matrix SH in the undamaged state of the structure in step S4 includes: Step S41, establish the strain distribution benchmark for each path. Normalization is performed, where the j-th element h i0,j Normalize according to the following formula: Among them, h i0,j,norm for h i0,j The normalized value, h i0,min For H i0 The minimum value in, h i0,max For H i0 The maximum value in; The normalized strain distribution baseline H for each path is obtained after normalization. i0,norm ; Step S42, for the strain distribution dataset matrix H of each monitoring path i The data in each row is normalized using the same method as in step S41, resulting in the normalized dataset matrix H. i,norm ; Step S43, calculate H using the Fraser distance calculation method. i,norm Data for each row and the normalized baseline H i0,norm Frescher distance matrix ,in For H i,norm The j-th row is relative to H i0,norm The strain distribution and the Fréchet distance value; Step S44: Horizontally stack the Friesian distance matrices of the N monitoring paths to form a non-destructive strain distribution similarity metric matrix. ; The calculation process of the structural damage monitoring strain distribution similarity measurement matrix SD in step S6 is the same as the calculation process of matrix SH.

5. The structural damage clustering monitoring method based on strain distribution similarity measurement according to claim 1, characterized in that, Step S7 involves clustering using the DBSCAN density clustering method, which includes: Step S71: Set the minimum number of neighborhood points (Minpts) based on the size of the number of monitoring paths (N). Step S72: Treat each row of matrix ST as the coordinates of a point in N-dimensional space, and calculate the distance d from each data point to its k-th nearest neighbor. k The Euclidean distance is used for calculation, where k = Minpts-1; Step S73, d of all coordinate points k The values ​​are sorted in ascending order, and d is plotted. k Value curve; Step S74, calculate d according to the following formula k The slope k of each point i on the value curve i : Where d k,i d corresponding to point i k value; For all k i Value, determine its maximum value k max When k i / k max The value exceeds the inflection point threshold T for the first time k When, then the inflection point i is in d k The corresponding d on the value curve k The value is used as the neighborhood radius Eps; Step S75: Perform DBSCAN clustering with Minpts and Eps as parameters.

6. The structural damage clustering monitoring method based on strain distribution similarity measurement according to claim 5, characterized in that, In step S75, the distance is calculated using Euclidean distance during the clustering process.

7. A structural damage clustering monitoring system based on strain distribution similarity measurement, characterized in that, The system for performing the structural damage clustering monitoring method based on strain distribution similarity measurement as described in any one of claims 1 to 6 includes multiple strain measurement sensors and a processor. Strain measurement sensors are arranged on the monitoring path around the damage monitoring area to collect strain distribution data under the same loading conditions, both when the structure is undamaged and under damage monitoring conditions. The processor calculates the Fréchet distance of the strain distribution based on the strain distribution benchmark of each monitoring path, and uses this as the coordinates of high-dimensional spatial points to perform DBSCAN density clustering. The clustering results are used to determine whether the structure has abnormal strain distribution caused by damage, thereby realizing damage monitoring.

8. The structural damage clustering monitoring system based on strain distribution similarity measurement according to claim 7, characterized in that, The strain measurement sensor is a fiber optic grating strain sensor.

Images