A method for processing stress-defect aliased magnetic signals

Abstract

The present invention relates to a stress-defect aliasing magnetic signal processing method, which mainly solves the technical problem that the existing magnetic detection technology cannot effectively distinguish between stress concentration and macro defects in a component at the same time. It includes step 1, respectively establishing a stress damage model and a stress-defect aliasing damage model of the same size; step 2, respectively obtaining a stress area magnetic field signal and an aliasing area magnetic field signal according to the stress damage model and the stress-defect aliasing damage model; step 3, respectively performing principal component analysis on the stress area magnetic field signal and the aliasing area magnetic field signal to obtain a first signal sample and a second signal sample; step 4, performing linear discriminant analysis processing on the second signal sample to obtain a third signal sample; step 5, respectively calculating the concentration degree of the point number distribution in the first signal sample and the third signal sample, and comparing the relative error of the concentration degree of the third signal sample relative to the first signal sample.

Description

Technical Field

[0001] The invention relates to a magnetic signal processing method, in particular to a stress-defect aliasing magnetic signal processing method. Background Art

[0002] Internal damage in engineering components (such as stress concentration, macro defects, etc.) can lead to the degradation of the structural mechanical properties of component materials, and even cause sudden safety accidents, which will have a significant impact on production safety. Therefore, it is necessary to conduct effective damage detection on component materials. Ferromagnetic materials are the most widely used type of materials in engineering components. Nondestructive testing of ferromagnetic components is of great significance for maintaining production safety.

[0003] Magnetic detection technology in geomagnetic environment can provide early warning for stress concentration and macro defects. Since ferromagnetic materials have mutual conversion between mechanical energy and magnetic energy, that is, when ferromagnetic materials in a constant environmental magnetic field are subjected to external forces, the magnetic properties of the materials will change due to the material's magnetomechanical coupling effect. Therefore, by measuring the spontaneous magnetic field signal formed on the surface of the component due to stress concentration, the damage location and degree can be determined, and early diagnosis of damage to ferromagnetic components can be achieved. At the same time, when there are macro defects inside the ferromagnetic component, changes in the magnetic field signal will also occur. Magnetic detection technology can be applied under the natural magnetization of the geomagnetic field, and no additional excitation device is required. Compared with leakage magnetic field, ultrasound, X-ray and other detection technologies, the equipment is simple, and there is no need to pre-treat the surface of the component being detected, and there is no effect on the material properties after detection.

[0004] However, since both stress concentration and macro defects in components will produce changes in magnetic field signals, when stress concentration and macro defects exist in the component at the same time, the resulting magnetic detection signal is the result of the aliasing of stress concentration and macro defects. How to use effective methods to distinguish stress concentration and macro defects from the aliased signals and then further evaluate the defects? At present, there is no feasible method application in practice, nor is there any public data reported. Summary of the invention

[0005] The purpose of the present invention is to solve the technical problem that when stress concentration and macro defects exist simultaneously in a component, the magnetic detection signal obtained by the existing magnetic detection technology is the result of the aliasing of stress concentration and macro defects, but cannot be effectively distinguished, and to provide a stress-defect aliasing magnetic signal processing method.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is:

[0007] A stress-defect aliasing magnetic signal processing method, which is special in that it includes the following steps:

[0008] Step 1, respectively establish a stress damage model and a stress-defect aliasing damage model with the same size;

[0009] Step 2, obtaining a stress region magnetic field signal and an aliasing region magnetic field signal according to a stress damage model and a stress-defect aliasing damage model respectively;

[0010] Step 3, performing principal component analysis on the magnetic field signal in the stress area and the magnetic field signal in the aliasing area respectively to obtain a first signal sample and a second signal sample;

[0011] Step 4: Perform linear discriminant analysis on the second signal sample to obtain a third signal sample;

[0012] Step 5, respectively calculating the concentration of the point number distribution in the first signal sample and the third signal sample, and comparing the relative error of the concentration of the third signal sample with respect to the first signal sample;

[0013] If the relative error of the concentration degree between a certain signal in the third signal sample and the first signal sample is ≤20%, it is judged that the signal and the first signal sample are of the same type of signal, that is, the signal is a stress magnetic signal and the other signal is a defect magnetic signal; the distinction processing between the stress magnetic signal and the defect magnetic signal is completed.

[0014] Furthermore, the step 2 is specifically as follows:

[0015] Step 2.1, setting a plurality of first detection paths on the stress damage model that are parallel to each other and have a preset lift-off interval from the upper surface thereof, and setting a plurality of second detection paths on the stress-defect aliasing damage model that are parallel to each other and close to the upper surface thereof; the number and interval of the first detection paths and the second detection paths are the same, and the plane projection positions on the corresponding stress damage model or stress-defect aliasing damage model correspond to each other;

[0016] Step 2.2, extract n magnetic field signals along each first detection path on the stress damage model, and obtain the observation signal matrix X, which is the magnetic field signal in the stress area; extract m magnetic field signals along each second detection path on the stress-defect aliasing damage model, and obtain the observation signal matrix Y, which is the magnetic field signal in the aliasing area.

[0017] Further, in step 2.1, the first detection path and the second detection path are both three;

[0018] The preset lift-off interval is 2 mm.

[0019] Furthermore, the step 3 is specifically as follows:

[0020] Step 3.1, respectively calculating the average value of n magnetic field signals on each first detection path in the observation signal matrix X, and the average value of m magnetic field signals on each second detection path in the observation signal matrix Y;

[0021] De-meaning the observed signal matrices X and Y respectively to obtain de-meaning observed signal matrices X' and Y';

[0022] Step 3.2, respectively calculate the covariance matrices of the mean-removed observation signal matrices X' and Y', and obtain the covariance matrix C1 corresponding to the mean-removed observation signal matrix X' and the covariance matrix C2 corresponding to the mean-removed observation signal matrix Y';

[0023] Step 3.3, calculate the eigenvalues ​​of covariance matrices C1 and C2 and the corresponding eigenvectors, sort the eigenvalues ​​of C1 and C2 from large to small, where C1 retains the largest eigenvalue, and the corresponding eigenvector u constitutes the projection matrix U of the stress signal; C2 retains the first two eigenvalues, and the corresponding two eigenvectors v1 and v2 constitute the projection matrix V of the aliasing signal;

[0024] Step 3.4, project each data of the mean-removed observation signal matrix X' into the projection matrix U to obtain the first signal sample X", and project each data of the mean-removed observation signal matrix Y' into the projection matrix V to obtain the second signal sample Y".

[0025] Furthermore, the step 4 is specifically as follows:

[0026] Step 4.1: Calculate the inter-class discreteness matrix S of the second signal sample Y″ b and the intra-class dispersion matrix S w , and form a discrete matrix S w -1 S b ;

[0027] Step 4.2: Discreteness matrix S w -1 S b Perform eigenvalue decomposition, select the two eigenvectors corresponding to the first two eigenvalues ​​in order from large to small, and obtain the projection matrix W;

[0028] Step 4.3: Project each data of the second signal sample Y″ into the projection matrix W to obtain the third signal sample Z.

[0029] Furthermore, the step 5 is specifically as follows:

[0030] Step 5.1, taking the center of the stress damage model and the stress-defect aliasing damage model as the interval demarcation point, respectively counting the number of points of the first signal sample X″ and the third signal sample Z on the left and right sides of the interval demarcation point, respectively calculating the ratio of the number of points of the first signal sample X″ and the third signal sample Z on the left and right sides of the interval demarcation point, with the number of points on the left side being the numerator and the number of points on the right side being the denominator; the ratio of the first signal sample X″ is recorded as A, and the ratio of the third signal sample Z is recorded as B1 and B2;

[0031] Step 5.2, taking the ratio A of the first signal sample X″ as the standard, respectively calculate the relative error δ of the concentration degree of B1 and B2;

[0032] q = 1 or 2;

[0033] Step 5.3: If the relative error δ between B1 or B2 and the ratio A of the first signal sample X″ is ≤20%, then B1 or B2 is a stress magnetic signal, and B2 or B1 is a defect magnetic signal.

[0034] Furthermore, the stress damage model and the stress-defect aliasing damage model are both ferromagnetic metal component models;

[0035] A stress concentration zone is arranged at the center of the stress damage model, and a stress-defect aliasing zone is arranged at the center of the stress-defect aliasing damage model;

[0036] The stress concentration area and the stress-defect aliasing area are of the same size.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. The present invention uses stress zone magnetic field signals and aliasing zone magnetic field signals as observation samples, and then performs principal component analysis on the stress zone magnetic field signals and the aliasing zone magnetic field signals respectively, performs linear discriminant analysis on the aliasing zone magnetic field signals after the principal component analysis, and finally calculates the concentration degree of the stress zone magnetic field signal after the principal component analysis and the concentration degree of the aliasing zone magnetic field signal after the linear discriminant analysis. By comparing the relative errors of the concentration degrees of the two, effective identification of the aliasing zone magnetic field signal is achieved, which solves the problem that the existing magnetic detection technology cannot effectively distinguish between stress concentration and macro defects in the component at the same time.

[0039] 2. The processing method combining principal component analysis and linear discriminant processing method adopted in the present invention can not only retain the advantages of linear discriminant analysis in maximizing the inter-class distance and minimizing the inter-class distance, so that the degree of separation of samples of different categories after projection is higher, but also correct the linear discriminant analysis through principal component analysis, retain more discrimination information, and finally more accurately identify the defective magnetic field signal from the magnetic field signal in the aliasing area. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 1 is a schematic diagram of the structure of the stress damage model in an embodiment of the present invention, a is an axonometric view, and b is a main view;

[0041] Figure 2 1 is a schematic diagram of the structure of the stress-defect aliasing damage model in an embodiment of the present invention, a is an axonometric view, and b is a main view;

[0042] Figure 3 It is a flowchart of step 3, step 4 and step 5 in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, advantages and features of the present invention clearer, a stress-defect aliasing magnetic signal processing method proposed by the present invention is further described in detail below in combination with the accompanying drawings and specific embodiments. According to the following specific implementation methods, the advantages and features of the present invention will be clearer. It should be noted that: the drawings are all in a very simplified form and use non-precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention; secondly, the structure shown in the drawings is often part of the actual structure.

[0044] This embodiment provides a stress-defect aliasing magnetic signal processing method, comprising the following steps:

[0045] Step 1, respectively establish a stress damage model and a stress-defect aliasing damage model with the same size;

[0046] like Figure 1 and Figure 2 As shown in the figure, both the stress damage model and the stress-defect aliasing damage model are ferromagnetic metal component models; a stress concentration zone is set at the center of the stress damage model, and a stress-defect aliasing zone is set at the center of the stress-defect aliasing damage model; the stress concentration zone and the stress-defect aliasing zone are the same size; for example, the specifications of the stress damage model are: 200mm long, 30mm wide, and 10mm thick; the length, width, and depth of the stress concentration zone are 12mm, 4mm, and 2mm respectively; the specifications of the stress-defect aliasing damage model are the same as those of the aforementioned stress damage model, and the length, width, and depth of the stress concentration zone in the stress-defect aliasing damage model are the same as those of the stress concentration zone in the aforementioned stress damage model, and the length, width, and depth of the defect at the center are 10mm, 2mm, and 1mm respectively. In general, the stress concentration zone is larger than the defect zone at the center;

[0047] Step 2, obtaining a stress region magnetic field signal and an aliasing region magnetic field signal according to a stress damage model and a stress-defect aliasing damage model respectively;

[0048] Step 2.1, setting a plurality of first detection paths which are parallel to each other and have a preset lift-off interval from the upper surface on the stress damage model, and setting a plurality of second detection paths which are parallel to each other and close to the upper surface on the stress-defect aliasing damage model; the number and interval of the first detection paths and the second detection paths are the same, and the plane projection positions on the corresponding stress damage models or stress-defect aliasing damage models correspond to each other; in this embodiment, there are three first detection paths and two second detection paths, each with a length of 20 mm; the interval between two adjacent first detection paths or second detection paths is 2 mm; the preset lift-off interval of the first detection path is 2 mm;

[0049] Step 2.2: extract n magnetic field signals x along each first detection path on the stress damage model in , x in =(x i1 x i2 …x in ), and obtain the observed signal matrix X, X = [x1, x2, x3] T , which is the magnetic field signal of the stress area; along each second detection path on the stress-defect aliasing damage model, m magnetic field signals y are extracted respectively. im ,y im =(y i1 y i2 …y im ), i = 1, 2, 3, and the observed signal matrix Y is obtained, Y = [y1, y2, y3] T , which is the magnetic field signal in the aliasing region. There is no restriction on the specific number of m and n, and they can be plural;

[0050] Step 3: Perform principal component analysis on the magnetic field signal in the stress area and the magnetic field signal in the aliasing area to obtain a first signal sample and a second signal sample; Figure 3 As shown;

[0051] Principal component analysis can reduce the dimensionality of high-dimensional data sets with multiple observed variables, allowing people to find some main aspects from the intricate relationships between things, so that a large amount of statistical data can be used more effectively for quantitative analysis, and can better perform subsequent processing such as visualization and regression. Principal component analysis of the magnetic field signals in the stress area and the aliasing area above can eliminate the interference of noise signals to a certain extent, laying the foundation for the subsequent identification of defect signals;

[0052] Step 3.1, calculate the average value of n magnetic field signals on each first detection path in the observation signal matrix X, denoted as Calculate the average value of m magnetic field signals on each second detection path in the observation signal matrix Y, denoted as

[0053] The observed signal matrices X and Y are respectively subjected to mean removal processing, that is, the magnetic field signals on each first detection path and second detection path in the observed signal matrices X and Y are subtracted from the average values ​​of the magnetic field signals on the corresponding first detection path and second detection path to obtain the observed signal matrices X' and Y' after mean removal. At this time, X'=[x1',x2',x3'] T , each first detection path has n magnetic field signals x in ', x in '=(x i1 'x i2 '…x in '), i=1,2,3; Y'=[y1',y2',y3'] T , each second detection path has m magnetic field signals y im ',y im '=(y i1 'y i2 '…y im '), i = 1, 2, 3;

[0054] Step 3.2, respectively calculate the covariance matrices of the observation signal matrices X' and Y' after the mean is removed, and obtain the covariance matrix C1 corresponding to the observation signal matrix X' after the mean is removed and the covariance matrix C2 corresponding to the observation signal matrix Y' after the mean is removed;

[0055] Take the observed signal matrix X' after removing the mean as an example:

[0056]

[0057] Given a random variable x p ′,x k ′, where p = 1, 2, 3; k = 1, 2, 3;

[0058] but

[0059] Step 3.3, calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrices C1 and C2, sort the eigenvalues ​​of C1 and C2 from large to small, where C1 retains the largest eigenvalue, and the corresponding eigenvector u constitutes the projection matrix U of the stress signal; C2 retains the first two eigenvalues, and the corresponding two eigenvectors v1 and v2 constitute the projection matrix V of the aliasing signal, V = [v1, v2];

[0060] Step 3.4: Project each data of the mean-removed observed signal matrix X' into the projection matrix U to obtain the first signal sample X", X" = u T X; project each data of the mean-removed observed signal matrix Y' into the projection matrix V to obtain the second signal sample Y", Y" = v T Y;

[0061] Step 4: Perform linear discriminant analysis on the second signal sample to obtain a third signal sample;

[0062] Linear discriminant analysis is developed on the basis of principal component analysis. Unlike principal component analysis, the linear discriminant analysis algorithm is essentially a supervised learning algorithm. It can project data belonging to the same category together as much as possible, minimize the variance of data within the category, and project data belonging to different categories as far as possible, maximize the variance between categories, so that after the original data is projected in these directions, different categories can be distinguished as much as possible. After the principal component analysis is used to reduce the dimension of the data, some background noise with a fixed frequency (small variance) is generally filtered out, and the best descriptive features (principal components) are retained instead of classification features. In order to obtain features of different classification attributes, the signal can be first subjected to principal component analysis, and then the linear discriminant analysis algorithm can be used to further classify the signal after principal component analysis, qualitatively identify the defect signal, and finally output it.

[0063] Step 4.1: Calculate the inter-class discreteness matrix S of the second signal sample Y″ b and the intra-class dispersion matrix S w , and form a discrete matrix S w -1 S b ;

[0064] Step 4.2: Discreteness matrix S w -1 S b Perform eigenvalue decomposition, select the two eigenvectors corresponding to the first two eigenvalues ​​in order from large to small, and obtain the projection matrix W;

[0065] Step 4.3, project each data of the second signal sample Y″ into the projection matrix W to obtain a third signal sample Z;

[0066] Step 5, respectively calculating the concentration of the point number distribution in the first signal sample and the third signal sample, and comparing the relative error of the concentration of the third signal sample with respect to the first signal sample, thereby obtaining a comparison result;

[0067] Step 5.1, taking the center of the stress damage model and the stress-defect aliasing damage model as the interval dividing point, count the number of points of the first signal sample X″ and the third signal sample Z on the left and right sides of the interval dividing point respectively, calculate the ratio of the number of points on the left and right sides of the first signal sample X″ and the third signal sample Z respectively (the number of points on the left is the numerator, and the number of points on the right is the denominator), record the ratio of the first signal sample X″ as A, and the ratio of the third signal sample Z as B1 and B2.

[0068] Step 5.2: Taking the ratio A of the first signal sample X″ as the standard, calculate the relative error δ of the concentration degree of B1 and B2.

[0069] q = 1 or 2;

[0070] Step 5.3: If the relative error δ between one of B1 or B2 and the ratio A of the first signal sample X″ is ≤20%, then the signal with a relative error less than 20% is judged to be of the same type as the first signal sample, that is, the signal is a stress magnetic signal, and the other signal is a defect magnetic signal; the distinction between the stress magnetic signal and the defect magnetic signal is completed.

Claims

1. A stress-defect aliasing magnetic signal processing method, characterized in that: The following steps are involved: Step 1, respectively establish a stress damage model and a stress-defect aliasing damage model with the same size; Step 2, obtaining a stress region magnetic field signal and an aliasing region magnetic field signal according to a stress damage model and a stress-defect aliasing damage model respectively; Step 2.1, setting a plurality of first detection paths on the stress damage model that are parallel to each other and have a preset lift-off interval from the upper surface thereof, and setting a plurality of second detection paths on the stress-defect aliasing damage model that are parallel to each other and close to the upper surface thereof; the number and interval of the first detection paths and the second detection paths are the same, and the plane projection positions on the corresponding stress damage model or stress-defect aliasing damage model correspond to each other; Step 2.2, extract n magnetic field signals along each first detection path on the stress damage model to obtain an observation signal matrix X, which is the magnetic field signal of the stress area; Extract m magnetic field signals along each second detection path on the stress-defect aliasing damage model to obtain an observation signal matrix Y, which is the magnetic field signal in the aliasing area; Step 3, performing principal component analysis on the magnetic field signal in the stress area and the magnetic field signal in the aliasing area respectively to obtain a first signal sample and a second signal sample; Step 3.1, respectively calculating the average value of n magnetic field signals on each first detection path in the observation signal matrix X, and the average value of m magnetic field signals on each second detection path in the observation signal matrix Y; De-meaning the observed signal matrices X and Y respectively to obtain de-meaning observed signal matrices X' and Y'; Step 3.2, respectively calculate the covariance matrices of the mean-removed observation signal matrices X' and Y', and obtain the covariance matrix C1 corresponding to the mean-removed observation signal matrix X', and the covariance matrix C2 corresponding to the mean-removed observation signal matrix Y'; Step 3.3, calculate the eigenvalues ​​of covariance matrices C1 and C2 and the corresponding eigenvectors, sort the eigenvalues ​​of C1 and C2 from large to small, where C1 retains the largest eigenvalue, and the corresponding eigenvector u constitutes the projection matrix U of the stress signal; C2 retains the first two eigenvalues, and the corresponding two eigenvectors v1 and v2 constitute the projection matrix V of the aliasing signal; Step 3.4, project each data of the mean-removed observation signal matrix X' into the projection matrix U to obtain the first signal sample X''; project each data of the mean-removed observation signal matrix Y' into the projection matrix V to obtain the second signal sample Y''; Step 4: Perform linear discriminant analysis on the second signal sample to obtain a third signal sample; Step 4.1: Calculate the inter-class discreteness matrix S of the second signal sample Y' b and the intra-class dispersion matrix S w , and form a discrete matrix S w -1 S b ; Step 4.2: Discreteness matrix S w -1 S b Perform eigenvalue decomposition, select the two eigenvectors corresponding to the first two eigenvalues ​​in order from large to small, and obtain the projection matrix W; Step 4.3, project each data of the second signal sample Y" into the projection matrix W to obtain the third signal sample Z; Step 5, respectively calculating the concentration of the point number distribution in the first signal sample and the third signal sample, and comparing the relative error of the concentration of the third signal sample with respect to the first signal sample; Step 5.1, taking the center of the stress damage model and the stress-defect aliasing damage model as the interval demarcation point, respectively counting the number of points of the first signal sample X" and the third signal sample Z on the left and right sides of the interval demarcation point, respectively calculating the ratio of the number of points of the first signal sample X" and the third signal sample Z on the left and right sides of the interval demarcation point, with the number of points on the left as the numerator and the number of points on the right as the denominator; the ratio of the first signal sample X" is recorded as A, and the ratio of the third signal sample Z is recorded as B1 and B2; Step 5.2, taking the ratio A of the first signal sample X" as the standard, calculate the relative error δ of the concentration degree of B1 and B2 respectively; q = 1 or 2; Step 5.3: If the relative error δ between B1 or B2 and the ratio A of the first signal sample X″ is ≤20%, then B1 or B2 is a stress magnetic signal, and B2 or B1 is a defect magnetic signal; the distinction between the stress magnetic signal and the defect magnetic signal is completed.

2. A stress-defect aliasing magnetic signal processing method according to claim 1, characterized in that: In the step 2.1, there are three first detection paths and three second detection paths; The preset lift-off interval is 2 mm.

3. A stress-defect aliasing magnetic signal processing method according to claim 2, characterized in that: The stress damage model and the stress-defect aliasing damage model are both ferromagnetic metal component models; A stress concentration zone is arranged at the center of the stress damage model, and a stress-defect aliasing zone is arranged at the center of the stress-defect aliasing damage model; The stress concentration area and the stress-defect aliasing area are of the same size.

Images