Fingerprint assisted metal surface crack detection method based on passive RFID

Abstract

The application relates to a passive RFID-based fingerprint-assisted metal surface crack detection method, which comprises the following steps: arranging a measurement system and constructing a fingerprint library; for a metal surface needing to be detected whether cracks exist, arranging a label array on the metal surface, collecting phase information, and obtaining a phase measurement matrix; performing a Meanshift clustering algorithm, classifying the labels into two categories, and preliminarily determining the position of metal cracks; performing geometric method-based coarse positioning; and performing fingerprint matching fine-grained positioning based on distance weight.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of RFID structural health monitoring. It relates to a fingerprint-assisted metal surface crack detection method based on passive RFID. BACKGROUND

[0002] In recent years, with the development of science and technology, the infrastructure in various fields has been greatly improved. Facilities located in harsh environments, such as railways, bridges, pipelines, and aircraft, will have a significantly increased probability of crack damage to their welded metal structures due to frequent stress. Small-scale damage can lead to performance degradation, while large-scale damage can cause sudden failure of the metal structure, which will be a great potential threat in engineering projects. We need to monitor facilities in operation in real time without damaging the target structure, so monitoring technology is crucial.

[0003] Structural health monitoring (SHM) is a modern method that uses a combination of various sensor technologies and specific algorithms to extract characteristic parameters of the monitored structure, and analyzes whether the structure is damaged and the extent of the damage through collected data. This technology is commonly used in bridge monitoring, large-scale industrial equipment monitoring, and aerospace equipment monitoring. Traditional SHM systems transmit signals through wired systems such as optical fibers or coaxial cables, which have the advantages of high resolution and accuracy. However, when applying traditional SHM, people need to lay cables to power the monitoring system. These requirements increase the installation and maintenance costs of SHM systems working in harsh environments.

[0004] Radio frequency identification (RFID) systems have developed rapidly in recent years due to their passive, low-cost, and wireless characteristics, and are applied to identify and track objects. RFID systems receive radio frequency signals transmitted by tags, form an induced current on the tag, and reflect the current information of the tag back to the reader for data processing. Therefore, RFID systems can monitor targets for a long time and achieve SHM. We can attach tags to the metal surface to be monitored, and analyze the crack occurrence in the area according to the changes in the tag return data during the monitoring process.

[0005] Currently, metal surface crack monitoring based on RFID is mainly divided into two categories. The first category is to analyze the occurrence of cracks by changing the physical structure of a single tag due to cracks, thereby affecting the radiation characteristics of the tag antenna. The second category is to determine the characteristics of cracks by changing the coupling relationship between two tags due to cracks. SUMMARY

[0006] This invention provides a fingerprint-assisted crack detection method for metal surfaces based on passive RFID. It aims to utilize the phase change generated by electromagnetic wave propagation and, by measuring the phase information of the tag's backscattered signal, to provide crack information on the target metal surface. The specific technical approach is as follows:

[0007] A fingerprint-assisted method for detecting cracks on metal surfaces based on passive RFID includes the following steps:

[0008] The first step is to deploy a measurement system and build a fingerprint database.

[0009] (1) Design an integral metal plate according to the size of the metal surface to detect whether there are cracks, and arrange the tag array on the integral metal plate; build a passive RFID measurement system, which also includes an antenna and a reader;

[0010] (2) Design M sampling points on the integral metal plate, with each sampling point m evenly distributed around the perimeter of the integral metal plate; for each sampling point m, the direction of the sampled crack is represented by the angle θ, and the angle θ increments from π / K to π in steps of π / K, for a total of K angles, which are represented as θ1→θ K When a crack passes through a sampling point m and its direction is θ, the crack is defined as crack (m, θ), and the data collected by the antenna from the tag array is defined as fingerprint (m, θ), where θ = θ j j = 1, ..., K;

[0011] (3) For each crack (m, θ), a pair of corresponding metal plates are made; the gap between the pair of metal plates is used to simulate the crack (m, θ); when the pair of metal plates are completely merged, they form the integral metal plate, at which point the simulated crack does not exist; the gap between each crack is the same, that is, the gap between the two metal plates is a constant value, so the crack width simulated for each crack is a constant value.

[0012] (4) For each crack (m, θ), using the same gap, the phase information of the tag array is collected using a measurement system as follows: The phase information of the tag array on the whole metal plate without cracks is read by the antenna to obtain the phase information matrix P0 when there is no crack; the pair of metal plates corresponding to this crack are separated according to the preset gap width, and the phase information of the tag array is read by the antenna to obtain the phase information matrix P when there is a simulated crack. m,θ Furthermore, the phase measurement matrix ΔP of the crack (m,θ) is obtained. m,θ =|P m,θ -P0|, thus obtaining the fingerprint (m,θ) of the crack (m,θ);

[0013] (4) Traverse each collection point and the crack direction to be collected to obtain a total of M*K fingerprints, and construct a fingerprint database accordingly;

[0014] Second step, for the metal surface which needs to detect whether there is a crack, according to the method of the first step, arrange the measurement system, arrange the tag array on the metal surface, collect the phase information, obtain the phase measurement matrix; execute Meanshift clustering algorithm, if there is a crack, use this algorithm to divide the tags into two categories, to preliminarily determine the position of the metal crack, continue to execute the third step;

[0015] Third step, coarse positioning based on geometric method: after the tag nodes are divided into two categories, coarse positioning is carried out by using the geometric method, coarse positioning nodes are obtained, the generated coarse positioning straight line s0 is taken as the result of coarse positioning of the crack by using the least square method;

[0016] Fourth step, fine-grained positioning based on distance weight fingerprint matching, the method is as follows:

[0017] Let the actually measured phase measurement matrix be ΔP, and the actually measured phase measurement matrix ΔP and the fingerprint library ΔP m,θ are compared and matched after normalization, let ΔP be normalized by column as ΔP norm ; each fingerprint ΔP m,θ in the fingerprint library is also normalized by column according to the same rule to obtain ΔP m,θ_norm ;

[0018] The obtained coarse positioning straight line s0 is matched with the fingerprint with similar position and direction, the fingerprint with a distance less than a preset threshold d t from the coarse positioning crack straight line s0 is screened in the normalized fingerprint library, the reciprocal of the Euclidean distance of each screened fingerprint from the actually measured phase measurement matrix is taken as the weight, and the distance weight matching algorithm is used to realize fine-grained positioning of the fingerprint matching, and the actual crack width w is obtained.

[0019] Further, the method for positioning the tags by using the Meanshift clustering algorithm in the second step is as follows:

[0020] (1) let there be L antennas, and an L-dimensional coordinate system is established by using the tag phase information collected by the antennas, wherein the n-th tag node coordinate is set as

[0021] (2) a random tag node is set as the center of the initial cluster S c , and the radius of the cluster is set as R c ;

[0022] The Meanshift migration vector is defined as:

[0023]

[0024] Wherein: k is the cluster S c The number of tag nodes in the cluster S c Is a set of tag nodes, and needs to meet S c All tag nodes in the cluster S The distance to the initial cluster center point Is less than R c ;

[0025] (3) Meanshift clustering processing is performed, in each tag node set, the two sets with the largest number of tag nodes are selected as two parts of the tag array separated by cracks, and are used to preliminarily locate the actual crack position.

[0026] The present application proposes a metal surface crack detection method based on RFID for the problem of multi-feature detection of metal surface cracks. First, an RFID tag array is designed, the crack feature is perceived by the phase characteristics of the tag reflected signal affected by the crack, and a fingerprint library is constructed. Then, a crack coarse positioning algorithm based on the geometric method is designed, and the coarse positioning result is combined with the fingerprint method for fine-grained positioning, and the direction, position and width of the crack are calculated. Simulation and experimental results show that, compared with the traditional RFID fingerprint method, the present method can effectively reduce the crack multi-feature monitoring error. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 Fingerprint library construction schematic diagram

[0028] Figure 2 Antenna-to-tag distance schematic diagram without cracks

[0029] Figure 3 Antenna-to-tag distance schematic diagram with cracks

[0030] Figure 4 Experimental scene and equipment diagram

[0031] Figure 5 Tag node clustering result

[0032] Figure 6 Crack coarse positioning result

[0033] Figure 7 Fine-grained positioning result and actual crack comparison

[0034] Figure 8 Fingerprint library construction flowchart

[0035] Figure 9 Total flowchart of detecting whether there is a crack on the metal surface. DETAILED DESCRIPTION

[0036] A fingerprint assisted multi-feature monitoring method for metal surface crack based on passive RFID is described below in combination with the accompanying drawings.

[0037] 1. Construction of fingerprint library

[0038] The construction of fingerprint library is completed in offline stage, and the flow is shown as Figure 8

[0039] As shown in Figure 1 , the round dots in the figure represent the main monitoring tag array; the square dots represent the crack passing points of the fingerprint to be collected, and the number of points is represented by m, a total of M points; the dashed line represents the direction of the crack developed at the current square point; the direction is represented by angle θ, and the angle is stepped by π / 30 from π / 30 to π, a total of 30 angles, respectively represented as θ1→θ 30 ; the crack width is w=1cm. When the crack passes through the square point m and the crack direction is θ, the crack is defined as crack (m, θ), and the data collected by the antenna from the RFID tag array is defined as fingerprint (m, θ).

[0040] When there is no crack on the metal surface of the fingerprint collection area, the initial phase measurement matrix received by the antenna of each tag is defined as where represents the phase vector measured by the lth RFID antenna, and n represents the number of tag nodes. When the crack (m, θ) appears, the phase measurement matrix is where When different cracks appear, the relative distance between each node in the RFID tag array and the RFID antenna will change, and then according to the phase received by the RFID antenna:

[0041]

[0042] In the formula: λ is the wavelength of the radio frequency wave emitted by the antenna. Since the frequency of the radio frequency wave generated by the RFID reader is constant in the system, λ is a fixed value; r is the distance from the antenna to the RFID tag; p0 is the initial phase, which is affected by the hardware circuit. In the system, since the hardware device does not change, p0 is considered as a constant; p noise is the error caused by multipath effect and Gaussian white noise, where the Gaussian white noise follows a Gaussian distribution with a mean of 0 and a standard deviation of σ, i.e. p noise ~ N(0, σ 2 ).

[0043] Since the system takes the average of 50 phases read by the antenna to the tag as the final phase p in each crack condition, the influence of Gaussian white noise can be greatly reduced, so the system can be considered​

[0044] p~r(2)

[0045] The change of the phase measurement matrix at crack (m, θ) can be expressed as:

[0046]

[0047] In the formula: represents the phase change amount of the link (l, n) between the antenna l and the tag n at the crack (m, θ). According to formula (2), the position change of the tag n when the crack (m, θ) occurs can be calculated by the measured ΔP m,θ , so ΔP m,θ can be constructed as the fingerprint of the crack (m, θ) to build the entire fingerprint library.

[0048] In summary, the specific implementation steps of the fingerprint library establishment can be summarized as:

[0049] The overall metal plate is designed, and the tag array is arranged on the overall metal plate; the measurement system further includes an antenna and a reader

[0050] M acquisition points are designed on the overall metal plate, and each acquisition point m is uniformly distributed on the periphery of the overall metal plate; for each acquisition point m, the acquired crack direction is represented by an angle θ, and the angle θ is stepped by π / 30 from π / 30 to π, a total of 30 angles, which are represented as θ1→θ 30 ; when the crack passes through the acquisition point m and the crack direction is θ, the crack is defined as crack (m, θ), and the data collected by the antenna from the tag array is defined as fingerprint (m, θ), θ=θ j , j=1, …, 30;

[0051] For each acquisition point, a pair of metal plates is made for each crack direction that needs to be acquired, that is, for each crack (m, θ), a pair of metal plates is made according to the acquisition point position m and the crack direction θ; the gap between the pair of metal plates is used to simulate the crack (m, θ); the pair of metal plates forms the overall metal plate when they are completely merged, at this time, the simulated crack does not exist;

[0052] For each crack (m, θ), the corresponding metal plate and tag array are used, and the phase information of the tag array is collected by using the measurement system, and the method is as follows:

[0053] (1) The tag array reads the phase information once on the overall metal plate without the crack by using the antenna pair, and P0 is obtained.

[0054] (2) The corresponding pair of metal plates is separated according to the preset crack width and direction, and the tag array reads the phase information by using the antenna pair, and P m,θ is obtained, and further ΔPm,θ = |P m,θ - P0.

[0055] (3) Obtain the fingerprint (m, θ) of the crack (m, θ).

[0056] Traverse each collection point m and the crack direction θ to be collected, and obtain M*30 fingerprints in total, to thereby construct the fingerprint library.

[0057] 2. Crack feature extraction

[0058] This step and the subsequent steps correspond to the online phase in Figure 8 . It should be noted that the number of antennas, antenna placement, number of tags, and tag array arrangement in the online phase need to be consistent with those in the offline phase. That is, for the metal plate to be detected for the presence of a crack, the phase information needs to be collected under the same conditions as in the fingerprint construction phase, and this part of the work is implemented in the offline phase. The positioning calculation can be implemented online.

[0059] This embodiment is not an actual detection example, but a simulation example. This embodiment uses the measurement system used when constructing the fingerprint library and a pair of metal plates to simulate the detection of metal surface cracks.

[0060] Since a certain brittle area of a metal will develop cracks in the form of an approximate straight line along the direction of stress when subjected to stress for a long time, the present application represents a crack region on the metal surface located at point (x0, y0) with a width of w and an angle of θ0 with the X-axis in the plane XOY coordinate system of the tag array as:

[0061]

[0062] Due to the presence of the crack, the tags on both sides of the tag array will produce relative displacement due to the development of the crack, as shown in Figure 3 the distance of a certain tag in the tag array to the antenna before the crack appears is where h represents the distance of the antenna from the plane of the tag array, and d represents the distance of the projection point of the antenna on the plane of the tag array from the tag. After the crack appears, the distance to the antenna is r ′ = r + Δr. The first and second order derivatives of r with respect to d at the tag point are:

[0063]

[0064]

[0065] Since in this system, the change value Δd of d caused by the crack is much smaller than the initial d0 and h0 of the system, therefore is close to 0 and Since it remains almost unchanged, this system can be considered as r ~ d. Combining equation (2), we can obtain...

[0066] p~d(6)

[0067] Therefore, based on the received phase measurement matrix ΔP m,θ To calculate Δd for each tag, further analysis of the changes in the entire tag array is used to infer the characteristics of the crack. Figure 2 and Figure 3 It is known that when a single crack appears in the metal area to be detected, it will definitely divide the tag array into two parts. Tags of the same type in each part have the same movement trend during the crack generation process. Therefore, the location of the metal crack can be determined by dividing the tags into two categories through a clustering algorithm. This system uses a density-based clustering algorithm—Meanshift clustering algorithm (Fukunaga K, Hostetler L. The Estimation of the Gradient of a Density Function, with Applications in Pattern Recognition[J].IEEE Transactions on Information Theory, 1975, 21(1):32–40.)—to classify the tags. The specific process of the Meanshift clustering algorithm in this system is as follows:

[0068] ① Based on the tag phase information collected by a total of L antennas, a new L-dimensional coordinate system is established, where the coordinates of the nth tag node are set as

[0069] ② Add a random label node Let S be the initial cluster. c The center of the cluster is set to R. c .

[0070] ③Meanshift vector is defined as:

[0071]

[0072] In the formula: k is the cluster S c The number of nodes with labels in cluster S. c It is a set of tag nodes, and it needs to satisfy S c All tag nodes To the initial cluster center point The distance is less than R c

[0073] ④ Cluster S c The coordinates of the new cluster center point can be represented as:

[0074]

[0075] ⑤ Repeat steps 3 and 4 until the cluster center point no longer moves. After completing the iteration process, all processed label nodes in this iteration are assigned to set Γ. i middle.

[0076] ⑥ Randomly select a label node that has not been used in previous iterations as the set Γ j Repeat steps 2 through 5 until all tag nodes have been processed, starting from the initial center.

[0077] After Meanshift clustering, the two sets with the most tag nodes in each tag node set are selected as the two parts of the tag array separated by the crack, which are used to roughly locate the actual crack location and further compare it with the fingerprint database to calculate the direction and width of the crack.

[0078] 3. Coarse positioning based on geometric methods

[0079] After the label nodes are classified into two sets, coarse localization can be performed using a geometric method. Based on the label node classification results from the previous step, an N×N dimensional coarse localization matrix A can be generated to represent the two separated label classes, where the elements a of the matrix are... i,j Corresponding to the j-th label in the i-th row, a i,j The value can only be 1 or 0 to represent two types of tags, such as Figure 3 The crack situation in the middle corresponds to

[0080]

[0081] After obtaining the coarse localization matrix A, the following steps are performed to obtain the results of the coarse crack localization:

[0082] ① Starting from i=1, according to a i,1 to a i,N Iterate through each element in the i-th row of A in sequence, whenever there is an element a i,n With the previous element a i,n-1 If the values ​​are different, stop the traversal and generate a coordinate. The point is found and placed into set B, where (x) i,n ,y i,n () represents the actual coordinates of the label in the i-th row and j-th column of the label array. If no element in this row has a different value from the previous one, proceed directly to the next step.

[0083] ②Increase the value of i by 1, and repeat step ① until i = N.

[0084] ③ Starting from j=1, according to a 1,j to aN,j traverses each element in the jth column of A in order, and whenever there is an element a n,j different from the previous element a n-1,j , a point with coordinates is generated and put into set B. If there is no element in the current row whose value is different from the previous one, go to the next step directly.

[0085] (4) Increase the value of j by 1 and repeat step (3) until j = N.

[0086] (5) Fit the points in set B as coarse localization nodes with least square method, and the resulting straight line is the result of coarse localization of the crack. The expression of the straight line s0 is y = a0x + b0.

[0087] 4. Fine-grained localization based on distance-weighted fingerprint matching

[0088] The actual phase measurement matrix can be written as P = ΔP + P0, where P0is the actual phase measurement matrix of the reference tag.

[0089]

[0090] wherein represents the actual measured phase change value of the lth antenna to the n th tag. Since the measured fingerprint library is corresponding to the crack with a width of w = 1 cm, while the actual measured crack width is uncertain, in order to better perform fingerprint matching to obtain more accurate crack direction estimation value, the actual measured phase measurement matrix ΔP and the fingerprint library ΔP m,θ both need to be normalized before comparison and matching. Now ΔP is column-normalized as ΔP norm , that is,

[0091]

[0092] Each fingerprint ΔP m,θ in the fingerprint library is also column-normalized according to the same rule as formula (11) to obtain ΔP m,θ_norm . In order to reduce the number of fingerprints for comparison to improve the matching accuracy and system response time, the system does not need to compare and match all the fingerprints in the fingerprint library with ΔP norm , but only the fingerprints with similar position and direction to the coarse localization straight line s0 obtained in the previous step are used for matching. Since each fingerprint corresponds to a crack (m, θ) and uniquely corresponds to a crack straight line s i : y = a i x + b i , each straight line s i corresponds to a feature vector (a i , b i). In the normalized fingerprint library, select the fingerprint with the distance between the characteristic vector (a0, b0) of the rough positioning crack straight line s0 and the characteristic vector (a, b) of the fingerprint less than the preset threshold d t into the set C, and according to the reciprocal of the Euclidean distance between each fingerprint and the measured phase measurement matrix as the weight, a virtual fingerprint closest to the actual crack is fitted from the fingerprint library as the width reference matrix ΔP sta

[0093]

[0094] wherein q is a factor added to ensure the normalization of ΔP v . Finally, the measured phase measurement matrix ΔP norm is compared with the width reference matrix ΔP v , and the width w of the actual crack can be obtained:

[0095]

[0096] At the same time, by using the same distance weight matching algorithm as formula (12), the straight line equation s of the actual crack: y = ax + b can also be calculated. Wherein a and b can be represented as:

[0097]

[0098]

[0099] At this point, the fine-grained positioning of the crack is completed, the direction and position of the crack are obtained, and the width of the crack is also calculated, that is, the multi-feature monitoring of the metal surface crack is completed.

[0100] 5. Experimental results

[0101] We verified the effectiveness of the method through experiments. The experimental scene is shown in Figure 4 , which includes an Impinj Speedway R420 reader, 2 Laird S9028PCL directional polarization antennas, and 16 WYUAN PF5515 anti-metal RFID tags. The tag array is located on a metal plate 0.33 meters from the ground, and the distance between each tag is 0.1 meters. The tag closest to the left antenna is in Figure 7 ​The coordinates in the mid-axis coordinate system are (0.55, 1, 0.33), with the unit being meters. The antenna height is 1.1 meters, the spacing between the two antennas is 1.2 meters, the transmit power of each antenna is 32.5 dBm, and the gain is 9 dBic. The reader's operating frequency is set to 925.875 MHz. A ThinkPad computer, configured with an Intel Core i5-5200U CPU and 8.00 GB of memory, drives the RFID reader and processes the returned data through a C# software platform.

[0102] In the experiment, with Figure 4 For example, the generation of a metal crack is simulated by reading the phase of the tag array when two metal plates are tightly fitted and when they are parallel and separated by 1 cm, respectively, and the angle between the crack and the X-axis is... A computer-controlled RFID reader communicates with RFID tags by transmitting radio frequency waves through an antenna. It records the phase information of each tag read by each antenna in both cases of no crack and cracked conditions. The phase measurement matrix ΔP is calculated using equation (10), and the direction angle, location, and width of the crack are then calculated using the method proposed in this invention. The effects of each stage in the entire multi-feature crack monitoring process are determined by… Figure 5 , Figure 6 and Figure 7 exhibit.

[0103] Figure 5 The diagram illustrates the clustering results of tags in a tag array in a new coordinate system based on the phase values ​​read from each antenna using the Meanshift algorithm. The points in the upper and lower sections of the diagram represent tag nodes clustered into two classes, and the number next to each point represents the tag node's ID. Based on the clustering results, the geometric method proposed in this invention is used for coarse crack localization, as shown in the diagram. Figure 6 As shown. The points on the left and right sides of the straight line in the figure correspond to... Figure 6 The two types of label nodes are: points on the straight line represent coarse positioning nodes, and straight lines passing through the metal surface represent coarse positioning lines of cracks obtained by fitting the coarse positioning nodes. Figure 7 This is a comparison chart of the final fine-grained localization results and the actual crack. The crack direction error calculated by the method proposed in this invention is 0.133 rad, and the crack width error is 0.814 cm. This error is larger than the error obtained from simulation, which may be due to experimental errors and environmental interference. Figure 7 It can still be seen that the crack obtained by the method proposed in this invention has a small difference from the actual crack, and has good multi-feature crack monitoring performance.

[0104] 6. Summary of Process Steps

[0105] ① According to the steps of "1, building the fingerprint library", use 30*M to artificially manufacture different cracks on the metal plate to build the fingerprint library

[0106] ② In actual monitoring, according to the same antenna and tag arrangement as in the first step, deploy the antenna and tag on the actual monitoring scene and the metal surface to be monitored.

[0107] ③ Use the antenna to read the phase information of the tag array in real time to obtain the phase matrix P, and use the method of "2, crack feature extraction" to perform clustering processing on the tag array

[0108] ④ According to the clustering result in the third step, cooperate with the method of "3, coarse positioning based on geometric method" to obtain the coarse positioning straight line s0.

[0109] ⑤ According to the method of "4, fingerprint matching fine-grained positioning based on distance weight", the measured crack width w and the measured crack straight line equation s: y = ax + b can be obtained by using the phase matrix P obtained in the third step and the coarse positioning straight line s0 obtained in the fourth step. Thus, the entire detection process is completed.

Claims

1. A passive RFID-based fingerprint-assisted metal surface crack detection method, comprising the following steps: Step 1: Arranging a measurement system and constructing a fingerprint library (1) Designing an overall metal plate according to the size of the metal surface to be detected for cracks, and arranging a tag array on the overall metal plate; building a passive RFID-based measurement system, which further comprises an antenna and a reader; (2) Design M sampling points on the integral metal plate, with each sampling point m evenly distributed around the perimeter of the integral metal plate; for each sampling point m, the direction of the sampled crack is represented by the angle θ, and the angle θ increments from π / K to π in steps of π / K, for a total of K angles, which are represented as θ1→θ K When a crack passes through a sampling point m and its direction is θ, the crack is defined as crack (m, θ), and the data collected by the antenna from the tag array is defined as fingerprint (m, θ), where θ = θ j j = 1, ..., K; (3) For each crack (m, θ), a corresponding pair of metal plates is made; the gap between the pair of metal plates is used to simulate the crack (m, θ); the pair of metal plates forms the overall metal plate when they are completely merged, at which time the simulated crack does not exist; the gap of each crack is the same, i.e., the gap between the two metal plates is a constant value, so that the simulated crack width of each crack is a constant value; (4) For each crack (m, θ), using the same gap, the phase information of the tag array is collected using a measurement system as follows: The phase information of the tag array on the whole metal plate without cracks is read by the antenna to obtain the phase information matrix P0 when there is no crack; the pair of metal plates corresponding to this crack are separated according to the preset gap width, and the phase information of the tag array is read by the antenna to obtain the phase information matrix P when there is a simulated crack. m,θ Furthermore, the phase measurement matrix ΔP of the crack (m,θ) is obtained. m,θ =|P m,θ -P0|, thus obtaining the fingerprint (m,θ) of the crack (m,θ); (4) Traverse each collection point and the crack direction to be collected, a total of M*K fingerprints are obtained to construct a fingerprint library; Step 2: For the metal surface to be detected for cracks, arrange the measurement system according to the method of Step 1, arrange the tag array on the metal surface, collect the phase information, and obtain a phase measurement matrix; Perform Meanshift clustering algorithm, if there is a crack, use this algorithm to divide the tags into two categories to preliminarily determine the position of the metal crack, and continue to Step 3; Step 3: Coarse positioning based on geometric method: after the tag nodes are divided into two categories, coarse positioning is performed using the geometric method to obtain coarse positioning nodes, and the least square method is used for fitting, and the generated coarse positioning straight line s0 is taken as the result of crack coarse positioning; Step 4: Fingerprint matching fine-grained positioning based on distance weight, the method is as follows: Let the actual measured phase measurement matrix be ΔP, and the actual measured phase measurement matrix ΔP and the fingerprint library ΔP m,θ After normalization, comparison and matching are performed, and let ΔP be normalized by column as ΔP norm ; each fingerprint ΔP in the fingerprint library m,θ is also normalized by column according to the same rule to obtain ΔP m,θ_norm ; The obtained rough positioning straight line s0 and the fingerprint with similar position and direction are matched, and the fingerprint with the distance less than a preset threshold d from the rough positioning crack straight line s0 is screened in the normalized fingerprint library t The inverse of the Euclidean distance between each screened fingerprint and the measured phase measurement matrix is taken as a weight, the distance weight matching algorithm is used, the fine-grained positioning of the fingerprint matching is realized, and the actual crack width w is obtained.

2. The fingerprint-assisted metal surface crack detection method of claim 1, wherein, The method for positioning the tags in Step 2 using Meanshift clustering algorithm is as follows: (1) Set the antenna to L, use the antenna to collect the label phase information, establish an L-dimensional coordinate system, where the n-th label node coordinate is set as (2) Give a random label node Let S be the initial cluster. c The center of the cluster is set to R. c ; The Meanshift migration vector is defined as: wherein: k is the cluster S c the number of tag nodes in the cluster S c is a set of tag nodes, and needs to satisfy S c all tag nodes in the cluster S the distance from the initial cluster center point is less than R c ; (3) Meanshift clustering processing is performed, in each tag node set, the two sets with the largest number of tag nodes are selected as the two parts of the tag array separated by the crack to preliminarily locate the actual crack position.

Images