A method and system for non-destructive testing of the welding quality of a steel structural component

Abstract

The present application relates to the technical field of nondestructive testing of steel structures, and discloses a method and system for nondestructively testing the welding quality of a steel structure. The method applies a pulsed eddy current excitation to the welding area of the steel structure and uses an eddy current sensor array to collect multi-channel electromagnetic response signals. Subsequently, time-frequency analysis is performed on the signals, time-frequency coefficients are extracted, phase angle information is calculated, and a two-dimensional phase angle distribution map is constructed to identify the welding defect boundary. Further, low-dimensional feature vectors are obtained through feature dimension reduction processing, and are matched with a preset defect database to determine the defect type and size. The system is efficient and accurate, and is suitable for use in the field of bridge construction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nondestructive testing of steel structures, and particularly relates to a nondestructive testing method and system for welding quality of a steel structure. BACKGROUND

[0002] Traditional steel structure welding quality detection relies on manual visual inspection or destructive testing, and has problems such as high missed detection rate and low efficiency. Although existing nondestructive testing technologies such as ultrasonic and ray detection can realize internal defect detection, they have insufficient sensitivity to small cracks or surface defects, and are complex and cumbersome to operate.

[0003] Although the pulse eddy current detection technology has the advantages of non-contact and rapid scanning, the single channel signal is easily disturbed by noise, which limits the accuracy of defect positioning and classification. In addition, existing methods are mostly focused on qualitative analysis, and lack the ability of quantitative evaluation of defect size, which is difficult to meet the demand of high-precision manufacturing. SUMMARY

[0004] In view of the above technical deficiencies, the purpose of the present application is to provide a nondestructive testing method and system for welding quality of a steel structure, which solves the problems of low efficiency, poor accuracy and difficult defect recognition in existing steel structure welding quality detection.

[0005] To solve the above technical problems, the present application adopts the following technical solutions:

[0006] In a first aspect, the present application provides a nondestructive testing method for welding quality of a steel structure, which comprises:

[0007] Applying pulse eddy current excitation to the welding area of the steel structure, and collecting electromagnetic response signals generated by the pulse eddy current excitation on the surface of the welding area to obtain multi-channel response data;

[0008] Performing time-frequency analysis on the multi-channel response data to extract time-frequency coefficients representing the electromagnetic properties of the material;

[0009] Calculating phase angle information based on the time-frequency coefficients, and constructing a two-dimensional phase angle distribution map reflecting the distribution of the phase angle on the surface of the welding area;

[0010] Identifying the boundary position of the welding defect through the two-dimensional phase angle distribution map;

[0011] Extracting the feature vector of the two-dimensional phase angle distribution map and performing dimension reduction processing;

[0012] Matching the dimension-reduced feature vector with a preset defect database, and determining the type and size of the defect according to the matching result.

[0013] Preferably, in a possible implementation of the first aspect, collecting the electromagnetic response signals specifically comprises:

[0014] The signal acquisition is performed by using an eddy current sensor array arranged in a two-dimensional grid pattern, each sensor in the eddy current sensor array being connected with an independent signal conditioning channel;

[0015] The multi-channel synchronous acquisition technology is adopted to obtain multi-channel response data aligned in time domain.

[0016] Preferably, in a possible implementation form of the first aspect, the obtaining of the multi-channel response data comprises reconstructing the acquired discrete signals, the reconstructing converting the discrete sampling points into continuous time series by using an interpolation algorithm, the interpolation algorithm satisfying the following relationship:

[0017]

[0018] wherein, is the reconstructed continuous time signal, is the discrete voltage value of the i-th sampling point, is the continuous time variable, is the sampling function, is the total number of sampling points, is the sampling interval.

[0019] Preferably, in a possible implementation form of the first aspect, the time-frequency analysis of the multi-channel response data is performed to extract time-frequency coefficients, which is realized by wavelet packet transform, comprising:

[0020] selecting a wavelet basis function to perform multi-scale decomposition on the response data of each channel;

[0021] extracting, from the wavelet packet tree after the decomposition, the detail coefficients corresponding to a specific frequency band as the time-frequency coefficients.

[0022] Preferably, in a possible implementation form of the first aspect, the calculating of the phase angle information based on the time-frequency coefficients is realized by constructing an analytic signal of the detail coefficients and then calculating the phase angle of the analytic signal, the analytic signal has the expression:

[0023]

[0024] wherein, denotes the i-th detail coefficient, denotes the Hilbert transform, is the imaginary unit.

[0025] Preferably, in a possible implementation form of the first aspect, the phase angle is calculated by the following formula:

[0026] ​​

[0027] in, and They represent analytic signals respectively. The real and imaginary parts, It is the arctangent function in the four quadrants.

[0028] Preferably, in one possible implementation of the first aspect, analyzing the phase angle distribution map to identify the boundary location of welding defects includes:

[0029] Gradient calculation is performed on the phase angle distribution map to obtain a gradient magnitude image;

[0030] Local maxima are located in the gradient magnitude image, and maxima that meet the predetermined continuity conditions are connected to form the boundary contour of the defect.

[0031] Preferably, in one possible implementation of the first aspect, the dimensionality reduction process employs principal component analysis to reduce the dimensionality of the initial feature vector extracted based on the phase angle distribution map;

[0032] The principal component analysis includes:

[0033] Calculation by The matrix formed by the initial feature vectors of each sample covariance matrix Its elements It is given by the following formula:

[0034]

[0035] in, Indicates the first The first sample 1 eigenvalue, Indicates the first The first sample 1 eigenvalue, and For each of the samples, the first... The and the first The mean of the eigenvalues;

[0036] For covariance matrix Perform eigenvalue decomposition and solve the characteristic equation. , to obtain eigenvalues and its corresponding unit eigenvector ,in , The dimension of the initial feature vector;

[0037] Sort the feature values ​​from largest to smallest, and select the top... The eigenvector corresponding to the maximum eigenvalue The projection matrix is constructed ;

[0038] The initial eigenvector of each sample is obtained The dimensionality reduction is performed through the projection transformation to obtain the reduced eigenvector The projection transformation is as follows:

[0039]

[0040] Wherein, is the mean vector of the initial eigenvectors of all training samples, is the transpose of the projection matrix.

[0041] Preferably, in a possible implementation form of the first aspect, the reduced eigenvector is matched with a preset defect database, and the similarity measure between the to-be-tested eigenvector and the eigenvector of each defect template in the database is calculated to realize the matching.

[0042] The size of the defect is determined based on a mapping relationship model between the reduced eigenvector and the defect size.

[0043] In the second aspect, the present application provides a steel structure welding quality nondestructive testing system, which is used to realize the steel structure welding quality nondestructive testing method as described in the first aspect, and comprises:

[0044] The excitation and collection module applies a pulsed eddy current excitation to the welding area of the steel structure and collects electromagnetic response signals generated by the pulsed eddy current excitation on the surface of the welding area to obtain multi-channel response data.

[0045] The signal processing module performs time-frequency analysis on the multi-channel response data to extract time-frequency coefficients representing the electromagnetic properties of the material.

[0046] The phase analysis module calculates phase angle information based on the time-frequency coefficients and constructs a two-dimensional phase angle distribution map reflecting the distribution of the phase angle on the surface of the welding area.

[0047] The defect identification module identifies the boundary position of the welding defect through the two-dimensional phase angle distribution map.

[0048] The feature dimensionality reduction module extracts the features of the two-dimensional phase angle distribution map and performs dimensionality reduction to obtain a low-dimensional eigenvector.

[0049] The defect determination module matches the low-dimensional eigenvector with a preset defect database and determines the type and size of the defect according to the matching result.

[0050] The beneficial effects of the present application are that the present application realizes high-resolution electromagnetic response signal acquisition of the welding area through the pulse eddy current array sensor and the multi-channel synchronous acquisition technology, and significantly improves the micro defect identification sensitivity by combining wavelet packet time-frequency analysis and two-dimensional phase angle imaging.

[0051] The principal component analysis dimension reduction processing effectively compresses the feature dimension, reduces the calculation complexity, and realizes defect type classification and size quantitative prediction through the preset defect database matching and the multivariate regression model.

[0052] The method has high efficiency and accuracy, and can be widely applied to online monitoring of steel structure welding quality in bridge construction and other fields. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0054] Figure 1 A flow chart of a steel structure welding quality nondestructive testing method is provided for the present application.

[0055] Figure 2 A structural diagram of a steel structure welding quality nondestructive testing system is provided for the present application.

[0056] Reference signs: 1-excitation and collection module, 2-signal processing module, 3-phase analysis module, 4-defect identification module, 5-feature dimension reduction module, 6-defect determination module. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0058] Embodiment one: as shown, the present application provides a steel structure welding quality nondestructive testing method, comprising: Figure 1

[0059] Applying pulse eddy current excitation to the welding area of the steel structure, and collecting electromagnetic response signals generated by the pulse eddy current excitation on the surface of the welding area to obtain multi-channel response data.

[0060] ​In this embodiment, the pulsed eddy current excitation is generated by a high frequency pulser with a peak current setting of 5 A, a pulse width of 10 , and a repetition frequency of 1 kHz. The excitation signal is applied to the weld area through an excitation coil with a diameter of 10 mm, which is wound by copper wire and placed 1 mm away from the weld surface. When the excitation coil is subjected to pulsed current, an eddy current field is induced in the weld area, which is distorted by the weld defects and generates a detectable electromagnetic response signal on the surface.

[0061] The collection of electromagnetic response signals is realized by a two-dimensional grid pattern of eddy current sensor array. The sensor array adopts a uniform grid layout of 8 rows and 8 columns, with an adjacent sensor spacing of 5 mm, covering a detection area of 35 mm x 35 mm. Each sensor is a micro eddy current probe with a diameter of 2 mm, composed of a ferrite core and a coil. Each sensor is connected with an independent signal conditioning channel, including a preamplifier, a bandpass filter and an analog-to-digital converter. The preamplifier gain is set to 60 dB, and the bandpass filter passband range is 100 Hz to 100 kHz, suppressing low frequency noise and high frequency interference. A multi-channel synchronous acquisition technology is used, and all 64 channels of analog-to-digital converters are driven by the same clock source, with a sampling clock frequency of 1 MHz, ensuring that the response signals collected by each channel are strictly aligned in time domain, and obtaining multi-channel response data aligned in time domain.

[0062] The collected multi-channel response data is a discrete voltage signal, which is converted to a continuous time sequence through reconstruction processing. The reconstruction processing uses an interpolation algorithm, and in this embodiment, the sinc function interpolation is used to convert the discrete sampling points to continuous time signals. The interpolation algorithm satisfies the following relationship:

[0063]

[0064] wherein, is the reconstructed continuous time signal, is the discrete voltage value of the th sampling point, is the continuous time variable, is the sampling function, is the total number of sampling points, and in this embodiment is set to 1000, is the sampling interval, and in this embodiment is set to 0.1 . The reconstruction process is realized by a digital signal processor, which first performs zero padding on the discrete sequence of each channel to expand to 4096 points, and then applies the above interpolation formula to calculate the continuous signal.

[0065] The time-frequency analysis is performed on the multi-channel response data to extract time-frequency coefficients representing the electromagnetic properties of the material.

[0066] In the embodiment, the wavelet packet transform method is used to process the response data of each channel. First, Daubechies 4 wavelet is selected as the wavelet basis function, which has compact support and regularity, and is suitable for analyzing non-stationary electromagnetic response signals. The response data of each channel is decomposed at multiple scales, and the decomposition layer is set to 3 layers. The signal is decomposed into a wavelet packet tree structure of different frequency bands by wavelet packet transform.

[0067] The detail coefficients and approximation coefficients corresponding to the specific frequency band are extracted from the wavelet packet tree as time-frequency coefficients. Specifically, the high-frequency detail coefficients and low-frequency approximation coefficients after the third layer decomposition are selected, which represent the distribution of the electromagnetic properties of the material in the time-frequency domain. The detail coefficients reflect the high-frequency components of the signal, corresponding to the rapid electromagnetic response changes caused by defects; the approximation coefficients reflect the low-frequency components of the signal, corresponding to the background material properties.

[0068] Based on the time-frequency coefficients, the phase angle information is calculated, and a two-dimensional phase angle distribution map reflecting the distribution of the phase angle on the surface of the welding area is constructed.

[0069] In the embodiment, the phase angle calculation is performed on the detail coefficients extracted from the time-frequency analysis. The detail coefficients are derived from the third layer decomposition of the wavelet packet transform, and each sensor position corresponds to a set of detail coefficients representing the high-frequency electromagnetic response characteristics. First, the Hilbert transform is applied to each detail coefficient to construct an analytic signal. The analytic signal is , wherein is the th detail coefficient, is the Hilbert transform, is the imaginary unit. The Hilbert transform is implemented through convolution operation, and a finite impulse response filter is designed with a filter length of 64 points to ensure accurate extraction of phase information. The real part of the analytic signal is the original detail coefficient , and the imaginary part is the result of the Hilbert transform , forming a complex signal representation.

[0070] Next, the phase angle is calculated based on the analytic signal. The phase angle is derived from the four-quadrant arctangent function, and the calculation formula is , wherein and represent the real part and the imaginary part of the analytic signal , respectively. is the four-quadrant arctangent function, which ensures that the phase angle takes values in toBetween the arcs, avoid angle ambiguity. During the calculation process, the detail coefficients of each sensor position are processed one by one to obtain the corresponding phase angle value. In order to enhance the robustness, for each sensor position, the three coefficients with the largest amplitude in its detail coefficients are selected to calculate the phase angle, and the average value is taken as the final phase angle representative value of the position.

[0071] Then, a two-dimensional phase angle distribution map is constructed. Since the eddy current sensor array is arranged in an 8-row and 8-column two-dimensional grid covering a 35mm x 35mm area, the phase angle value of each sensor position is mapped to the corresponding grid coordinates. The grid point spacing is 5mm, and the phase angle value is directly assigned to the corresponding grid point. In order to form a continuous distribution, a bilinear interpolation method is used to interpolate the grid data, with an interpolation step size of 0.5mm, to generate a high-resolution two-dimensional phase angle matrix. The matrix is represented in the form of an image, where the gray value of each pixel corresponds to the phase angle size, thus directly reflecting the spatial distribution of the phase angle on the surface of the welding area. The phase angle distribution map highlights the local changes in the electromagnetic properties of the material, and the defect area usually exhibits an abnormal gradient of the phase angle.

[0072] The boundary position of the welding defect is identified through the two-dimensional phase angle distribution map.

[0073] In this embodiment, first, the phase angle distribution map is subjected to gradient calculation to obtain a gradient amplitude image. Gradient calculation is realized using the Sobel operator, which includes a horizontal direction template and a vertical direction template, with a template size of 3 by 3 pixels. The horizontal direction template is used to calculate the gradient value of the image in the direction , and the vertical direction template is used to calculate the gradient value of the image in the direction . Specifically, gradient calculation is completed through convolution operation, and the convolution kernel is a preset Sobel operator matrix. The gradient amplitude is calculated as follows: The gradient direction is calculated by the four-quadrant arctangent function: The gradient amplitude image is composed of the gradient amplitude of each pixel point, which highlights the edge region in the phase angle distribution map.

[0074] Local maximum points are located in the gradient amplitude image. The determination of local maximum points is based on the comparison of the gradient amplitude of each pixel point with the gradient amplitudes of its eight neighborhood pixel points. Specifically, a local window of 3 by 3 pixels is defined with the current pixel as the center, and if the gradient amplitude of the current pixel is greater than the gradient amplitudes of all neighborhood pixels, the point is marked as a local maximum point. In order to further eliminate the influence of noise, a gradient amplitude threshold of 0.1 rad / mm is set, and only the local maximum points with a gradient amplitude greater than the threshold are retained.

[0075] The local maximum points that meet the predetermined continuity condition are connected to form the boundary profile of the defect. The continuity condition includes gradient direction consistency and spatial proximity. The gradient direction consistency requires that the difference in gradient direction between the connected maximum points is less than 10 . The spatial proximity requires that the Euclidean distance between the maximum points is less than 2 mm. The connection process adopts an iterative search algorithm, starting from a seed point, tracking adjacent maximum points along the gradient direction, until a closed profile is formed, i.e. the boundary profile of the defect.

[0076] The feature vector of the two-dimensional phase angle distribution map is extracted and dimensionality reduction processing is performed.

[0077] In this embodiment, first, high-dimensional initial features are extracted from the two-dimensional phase angle distribution map. The phase angle distribution map is a matrix represented in the form of an image, with pixel values corresponding to phase angle sizes and a resolution of 0.5 mm. The feature extraction process includes calculating global statistical features and local texture features of the phase angle distribution map. The global statistical features include the mean, standard deviation, skewness, and kurtosis of the phase angle values. The local texture features are calculated by a gray level co-occurrence matrix, and four texture indicators, contrast, correlation, energy, and homogeneity, are selected. The gray level is quantized to 16 levels, and the moving step is 1 pixel. The region of the phase angle distribution map corresponding to each sensor position is divided into multiple overlapping sub-windows, with a sub-window size of 10 pixels x 10 pixels. The above features are extracted from each sub-window. Finally, the high-dimensional initial feature vector of each sample is spliced by the global features and the local features, with a feature dimension of 200.

[0078] Next, principal component analysis is used to reduce the dimensionality of the high-dimensional initial features. Principal component analysis aims to reduce the feature dimension while retaining the main variation information. First, a matrix X consisting of the initial feature vectors of samples is constructed , with a value of 100. Each row of the matrix X corresponds to a sample, and each column corresponds to a feature. The covariance matrix S of the matrix X is calculated , The elements of the covariance matrix S are given by the following formula:

[0079]

[0080] where represents the th feature value of the th sample, represents the th feature value of the th sample, and ​​respectively, are the mean values of the first and the second eigenvalues of all samples. The dimension of the covariance matrix is 200x200.

[0081] The eigenvalues and the corresponding unit eigenvectors are obtained by performing eigenvalue decomposition on the covariance matrix and solving the eigenvalue equation. , where n is the dimension of the initial eigenvector, which is set to 1 to 200 in this embodiment. The eigenvalue decomposition is implemented using the Jacobi iterative algorithm, with the number of iterations set to 100 and the convergence threshold set to 0.0001. The eigenvalues are sorted in descending order, and the eigenvectors corresponding to the top k largest eigenvalues are selected to form the projection matrix , where k is set to 10. The dimension of the projection matrix is 200x10. The initial eigenvector of any sample is projected into the low-dimensional space to obtain the low-dimensional eigenvector . The projection transformation formula is:

[0082]

[0083]

[0084] , where is the mean vector of the initial eigenvectors of all training samples in the pre-set defect database, and is the transpose of the projection matrix. After dimensionality reduction, the feature dimension of each sample is reduced from 200 to 10, and the low-dimensional eigenvector retains more than 90% of the variance information of the original data.

[0085] The dimensionally reduced eigenvector is matched with the pre-set defect database, and the type and size of the defect are determined based on the matching result.

[0086] ​​​​​​​​​​​​​​In the embodiment, the low-dimensional feature vector is matched with a preset defect database, and the matching is realized by calculating the similarity measure between the to-be-tested feature vector and the template feature vectors of each type of defect in the database. The preset defect database contains template feature vectors of 100 known defect types, each template corresponding to a typical defect, such as a pore, a crack, or an incomplete penetration. The similarity measure is realized by calculating the Euclidean distance between the to-be-tested feature vector and the template feature vector, and the smaller the distance value, the higher the similarity. After calculating the Euclidean distance between the to-be-tested vector and all template vectors, the defect type with the smallest distance is selected as the matching result, and a distance threshold of 0.5 is set. If the minimum distance is higher than the threshold, it is determined as an unknown defect.

[0087] The size of the defect is determined based on a mapping relationship model between the low-dimensional feature vector and the defect size. The mapping relationship model is established by using a multivariate linear regression method, and the training data is derived from 50 defect samples with known accurate sizes. The model input is a 10-dimensional low-dimensional feature vector , and the output is a defect size estimate . The expression is:

[0088]

[0089] wherein, is an intercept term, , , , is a regression coefficient. These coefficients are determined by fitting the training data by the least squares method. The to-be-tested low-dimensional feature vector is input into the trained model to calculate the quantitative estimate of the defect size. Finally, the type classification result and the size quantitative result of the defect are output to the system display interface together, and the automatic determination of the welding quality is completed.

[0090] Embodiment two: as shown in Figure 2 , the present application provides a steel structure welding quality nondestructive testing system, comprising:

[0091] The excitation and collection module 1 applies a pulse eddy current excitation to the welding area of the steel structure and collects the electromagnetic response signal generated by the pulse eddy current excitation on the surface of the welding area to obtain multi-channel response data;

[0092] The signal processing module 2 performs time-frequency analysis on the multi-channel response data and extracts time-frequency coefficients representing the electromagnetic properties of the material;

[0093] The phase analysis module 3 calculates the phase angle information based on the time-frequency coefficients and constructs a two-dimensional phase angle distribution map reflecting the distribution of the phase angle on the surface of the welding area;

[0094] The defect recognition module 4 identifies the boundary position of the welding defect through the two-dimensional phase angle distribution map;

[0095] a feature dimension reduction module 5, which extracts features of the two-dimensional phase angle distribution map and performs dimension reduction to obtain a low-dimensional feature vector;

[0096] a defect determination module 6, which matches the low-dimensional feature vector with a preset defect database and determines the type and size of the defect according to the matching result.

[0097] Obviously, various modifications and changes can be made to the present application by those skilled in the art without departing from the spirit and scope of the present application. Thus, it is intended that the present application cover the modifications and changes as long as they come within the scope of the claims of the present application and their equivalents.

Claims

1. A method for non-destructive testing of the quality of a weld of a steel structural member, characterized in that, The method comprises: applying a pulsed eddy current excitation to a welding area of a steel structural member, and collecting electromagnetic response signals generated by the pulsed eddy current excitation on the surface of the welding area to obtain multi-channel response data; The collection of the electromagnetic response signals specifically comprises: signal collection by using an eddy current sensor array arranged in a two-dimensional grid pattern, each sensor in the eddy current sensor array being connected with an independent signal conditioning channel; multi-channel synchronous acquisition technology is adopted to obtain multi-channel response data aligned in time domain; time-frequency analysis is performed on the multi-channel response data to extract time-frequency coefficients representing electromagnetic properties of the material; time-frequency analysis is performed on the multi-channel response data to extract time-frequency coefficients, which is realized by wavelet packet transform, comprising: selecting a wavelet basis function to perform multi-scale decomposition on the response data of each channel; extracting detail coefficients corresponding to a specific frequency band from the wavelet packet tree after decomposition as time-frequency coefficients; calculating phase angle information based on the time-frequency coefficients, and constructing a two-dimensional phase angle distribution map reflecting the distribution of the phase angle on the surface of the welding area; identifying the boundary position of the welding defect through the two-dimensional phase angle distribution map; extracting a feature vector of the two-dimensional phase angle distribution map and performing dimension reduction processing; the dimension reduction processing adopts a principal component analysis method to reduce the dimension of the initial feature vector extracted based on the phase angle distribution map; matching the feature vector after dimension reduction with a preset defect database, and determining the type and size of the defect according to the matching result.

2. The method of claim 1, wherein the steel structural member is a pipe. The acquisition of the multi-channel response data comprises reconstruction processing of the collected discrete signals, the reconstruction processing converts discrete sampling points into continuous time sequences by using an interpolation algorithm, and the interpolation algorithm satisfies the following relationship: wherein, is the reconstructed continuous-time signal, is the discrete voltage value of the th sampling point, is the continuous-time variable, is the sampling function, is the total number of sampling points, is the sampling interval.

3. The non-destructive testing method for welding quality of steel structural components as described in claim 1, characterized in that, The phase angle information is calculated based on the time-frequency coefficients, and is achieved by constructing an analytic signal of the detail coefficients and then calculating the phase angle of the analytic signal The expression is: wherein represents the th detail coefficient, represents the Hilbert transform, is the imaginary unit.

4. The non-destructive testing method for welding quality of steel structural components as described in claim 3, characterized in that, the phase angle is calculated from the formula: wherein and denote the real and imaginary parts of the analysis signal respectively, is the four-quadrant arctangent function.

5. The non-destructive testing method for welding quality of steel structural components as described in claim 1, characterized in that, analyzing the phase angle distribution map to identify the boundary position of the welding defect, comprising: performing gradient calculation on the phase angle distribution map to obtain a gradient amplitude image; locating local maximum points in the gradient amplitude image, and connecting the maximum points meeting a predetermined continuity condition to form a boundary contour of the defect.

6. The method of claim 1, wherein the steel structural member is a pipe. The principal component analysis comprises: The computation is given by a matrix of initial feature vectors of the covariance matrix of the elements of is given by in, Indicates the first The first sample 1 eigenvalue, Indicates the first The first sample 1 eigenvalue, and For each of the samples, the first... The and the first The mean of the eigenvalues; performing eigen decomposition on the covariance matrix solving the eigen equation to obtain eigenvalues and their corresponding unit eigenvectors where , is the dimension of the initial eigenvectors; The characteristic values are sorted from large to small, and the characteristic vectors corresponding to the first maximum characteristic values are selected to constitute a projection matrix ; The initial eigenvector of any sample is obtained Dimension reduction is performed by projection transformation to obtain a reduced eigenvector The projection transformation is wherein, is a mean vector of initial feature vectors of all training samples, is a transpose of the projection matrix.

7. The method of claim 1, wherein the steel structural member is a pipe. matching the feature vector after dimension reduction with a preset defect database, which is realized by calculating the similarity between the to-be-tested feature vector and the feature vectors of the defect templates in the database; ​ determining the size of the defect based on a mapping relationship model between the feature vector after dimension reduction and the size of the defect.

8. A system for non-destructive testing of the quality of a weld of a steel structural member, characterized in that, The system is used to implement a non-destructive testing method for the welding quality of a steel structural member as claimed in any one of claims 1 to 7, comprising: an excitation and collection module for applying a pulsed eddy current excitation to a welding area of a steel structural member, and collecting electromagnetic response signals generated by the pulsed eddy current excitation on the surface of the welding area to obtain multi-channel response data; a signal processing module for performing time-frequency analysis on the multi-channel response data to extract time-frequency coefficients representing electromagnetic properties of the material; a phase analysis module for calculating phase angle information based on the time-frequency coefficients, and constructing a two-dimensional phase angle distribution map reflecting the distribution of the phase angle on the surface of the welding area; a defect identification module for identifying the boundary position of the welding defect through the two-dimensional phase angle distribution map; a feature dimension reduction module for extracting a feature vector of the two-dimensional phase angle distribution map and performing dimension reduction processing. The defect determination module matches the reduced dimension feature vector with a preset defect database, and determines the type and size of the defect according to a matching result.

Images