Quantitative Detection Method for Internal Defects in Nuclear Power Pipeline Forgings

By constructing a three-dimensional tensor field of acoustic properties on nuclear power pipeline forgings and reconstructing the acoustic impedance distribution map using a nonlinear inversion algorithm, the problem of distinguishing between defects and non-uniform regions of the microstructure in complex microstructure backgrounds was solved. This enabled the accurate extraction of the spatial orientation and size of defects, improving the accuracy and precision of detection.

CN122651895BActive Publication Date: 2026-11-10ZHEJIANG JIEDE MASCH TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611165362.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-08-03
Publication Date
2026-11-10
Estimated Expiration
2046-08-03

AI Technical Summary

Technical Problem

Existing technologies struggle to accurately distinguish between defects and areas of uneven microstructure in the complex microstructure of nuclear power pipeline forgings, and cannot precisely extract the spatial orientation and equivalent size of defects, leading to uncertainty and misjudgment in quantitative detection results.

Method used

By deploying an ultrasonic phased array probe array on the surface of the forging, a virtual acoustic beam focusing scanning grid covering the entire thickness is generated, a three-dimensional tensor field of acoustic properties is constructed, the acoustic impedance distribution map is reconstructed using a nonlinear inversion algorithm, and the abnormal impedance region is separated by boundary curvature morphology analysis, and the spatial azimuth angle and equivalent size of the defect are extracted.

Benefits of technology

It enables accurate differentiation between defects and areas of non-uniformity in complex tissue backgrounds, improving the accuracy of defect detection and the precision of quantitative analysis, and reducing the false positive rate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122651895B_ABST
    Figure CN122651895B_ABST
Patent Text Reader

Abstract

This invention discloses a quantitative detection method for internal defects in nuclear power pipeline forgings, belonging to the field of non-destructive testing technology for nuclear power pipeline forgings. The method includes: deploying an ultrasonic phased array probe array on the surface of the forging to be tested; generating a virtual acoustic beam focusing scanning grid covering the entire thickness of the forging through time-series control; extracting multi-angle scattering characteristics of each focal point based on the echo signal to construct a three-dimensional tensor field of acoustic properties inside the forging; iteratively reconstructing the acoustic impedance distribution map and calibrating abnormal impedance regions based on this tensor field using a nonlinear inversion algorithm; performing curvature morphology analysis on the boundaries of the abnormal impedance regions to separate defect candidate areas and areas of non-uniform structure; and calculating the spatial azimuth and equivalent size of the defect in the defect candidate areas by combining multi-angle scattering characteristics and scanning geometry to generate quantitative defect detection results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for nuclear power pipeline forgings, specifically to a quantitative detection method for internal defects in nuclear power pipeline forgings. Background Technology

[0002] During manufacturing and service, nuclear power pipeline forgings may develop internal defects such as cracks, inclusions, and porosity, threatening the integrity and operational safety of the equipment. Therefore, accurate quantitative detection of internal defects in forgings is crucial. Conventional ultrasonic phased array testing methods typically rely on single or limited-angle acoustic beam scanning, identifying defects by analyzing changes in the amplitude of the echo signal. However, these methods extract low-dimensional defect feature information, primarily reflecting the reflection intensity of the defect in a specific incident direction, making it difficult to comprehensively describe the three-dimensional morphology and orientation characteristics of the defect. In quantitative defect characterization, the lack of sufficient acquisition of scattering directionality makes it impossible to accurately obtain the spatial azimuth of the defect, leading to significant uncertainty in the quantitative detection results. Furthermore, due to the characteristics of the forging process, nuclear power pipeline forgings often exhibit internal structural inhomogeneities such as coarse grains or texture. Conventional detection methods often rely on simple threshold segmentation to determine areas of abnormal acoustic impedance, making it difficult to effectively distinguish between genuine hazardous defects and benign structural inhomogeneities morphologically. Structurally inhomogeneous areas resemble minor defects in acoustic response, easily leading to misjudgments and increasing unnecessary rework or scrap costs. To meet the high safety requirements of nuclear power pipeline forgings, accurately distinguishing between defects and uneven microstructures in complex microstructures, and precisely extracting the spatial orientation of defects, are bottlenecks that current quantitative detection technologies urgently need to overcome. Summary of the Invention

[0003] The purpose of this invention is to provide a quantitative detection method for internal defects in nuclear power pipeline forgings. This method can accurately distinguish between defects and areas of uneven structure in complex microstructure, and precisely extract the spatial orientation and equivalent size of defects, thereby achieving high-precision quantitative detection of defects.

[0004] To achieve the above objectives, this invention provides the following technical solution: This invention relates to a quantitative detection method for internal defects in nuclear power pipeline forgings. This method generates a virtual acoustic beam focusing scanning grid covering the entire thickness of the forging by deploying an ultrasonic phased array probe array on the surface of the forging and using timing control to regulate the excitation and reception phases of each array element. Based on this, multi-angle scattering features of each focal point are extracted from the echo signal of this virtual acoustic beam focusing scanning grid to construct a three-dimensional tensor field reflecting the internal acoustic properties of the forging. Subsequently, based on this acoustic property three-dimensional tensor field, a nonlinear inversion algorithm is used to iteratively reconstruct the acoustic impedance distribution map inside the forging, and the abnormal impedance region is calibrated on the reconstructed acoustic impedance distribution map. This invention, by performing curvature morphology analysis on the boundaries of the abnormal impedance region, can separate the abnormal impedance region into a defect candidate region and a region of non-uniform microstructure, thereby avoiding interference from microstructure non-uniformity in defect determination and improving the accuracy of defect identification. For the separated defect candidate regions, this invention extracts defect orientation parameters based on multi-angle scattering characteristics, and combines the scanning geometry of the virtual acoustic beam focusing scanning grid to accurately calculate the spatial azimuth and equivalent size of the defect, and finally generates a quantitative defect detection result containing the above quantitative information.

[0005] As a technical solution of this invention, in the process of constructing a virtual acoustic beam focusing scanning grid, the ultrasonic phased array probe array is flexibly divided into multiple sub-aperture groups along the circumference and axial direction of the outer wall of the forging, based on the geometric shape and actual thickness distribution of the forging under test. Each sub-aperture group contains at least four adjacent array elements. A corresponding focusing depth range is independently set for each sub-aperture group, and the required excitation delay time and receiving window offset for each array element are calculated based on this focusing depth range. By sequentially performing transmit focusing and receive dynamic focusing operations on each sub-aperture group, focused acoustic beam scanning lines corresponding to each sub-aperture group are generated. These focused acoustic beam scanning lines are then precisely stitched together according to their spatial positions, ultimately forming a complete, blind-zone-free virtual acoustic beam focusing scanning grid covering the entire thickness of the forging, ensuring full coverage and resolution of the defect area by the sound field.

[0006] In a preferred embodiment of the present invention, the process of extracting multi-angle scattering features and constructing a three-dimensional tensor field of acoustic properties includes: for each focal point in the virtual acoustic beam focusing scanning grid, systematically collecting at least three echo amplitude sequences and echo phase sequences under different incident angle directions, and arranging these sequences in order of incident angle to construct a vector characterizing the scattering direction response characteristics of the focal point. By performing covariance matrix decomposition on the scattering direction response vector, eigenvalues ​​and their corresponding eigenvectors in the main scattering direction are extracted. The eigenvalues ​​are used as indicators to measure the anisotropy intensity of the material's microstructure, and the eigenvectors are used as spatial codes for the main scattering direction. Finally, the anisotropy intensity indicators and main scattering direction codes of all focal points are filled into a three-dimensional matrix according to their real spatial coordinates to form an information-rich three-dimensional tensor field of acoustic properties inside the forging, providing a multi-dimensional data foundation for subsequent accurate defect analysis.

[0007] In reconstructing the acoustic impedance distribution map, one technical solution of this invention is to discretize the internal space of the forging into numerous voxel units and assign an initial acoustic impedance estimate to each voxel unit to form an initial acoustic impedance distribution map. Preferably, the initial value is obtained by first retrieving the median value of the standard acoustic impedance range based on the forging material grade as a reference, then selecting multiple calibration points on the surface of the forging, measuring their acoustic impedance values ​​using the pulse-echo method, and calculating the correction coefficients for each calibration point. By spatially interpolating these correction coefficients, a correction coefficient distribution field covering the entire internal space of the forging is generated. Finally, this distribution field is multiplied point by point with the reference initial acoustic impedance value to obtain a more accurate initial acoustic impedance estimate for each voxel unit. Based on this, the forward sound field propagation prediction echo data is calculated, and the acoustic impedance values ​​of the voxel units are iteratively updated along the gradient descent direction by continuously calculating the residual between the prediction data and the measured data until the residual meets the requirements, thereby obtaining a high-precision reconstructed acoustic impedance distribution map. During the abnormal impedance region calibration stage, the set of voxels whose acoustic impedance values ​​deviate significantly from the average value of adjacent voxels are calibrated as abnormal impedance regions, effectively capturing the discontinuous features inside the material.

[0008] For the separation of anomalous impedance regions, this invention employs refined morphological processing. First, the anomalous impedance regions are labeled as three-dimensional connected components, and a triangular surface mesh is extracted for each connected component. The average curvature and Gaussian curvature of each vertex on this surface mesh are calculated, generating a curvature morphology histogram. After multi-peak decomposition of this histogram, the curvature intervals corresponding to the main peaks are extracted, and the portions of the connected component whose curvature values ​​fall within these intervals are marked as smooth regions. When the volume proportion of smooth regions within a connected component exceeds a preset threshold, the connected component is determined to be a region of non-uniform structure; otherwise, it is determined to be a candidate defect region. This boundary curvature morphology analysis method can, based on the sharpness characteristics of the geometric morphology, fundamentally distinguish between non-uniform structures such as compositional segregation caused by metallurgical processes and actual defects such as cracks and inclusions, significantly reducing the false positive rate.

[0009] In the process of quantitative defect detection, the method of this invention extracts the direction component with the largest amplitude in the main scattering direction encoding of the focal point within each defect candidate region, and uses the tilt angle and azimuth angle of this direction component in three-dimensional space as the initial value of defect orientation. Simultaneously, a geometric mapping relationship table between the incident and receiving directions of the focused sound beam in the virtual sound beam focusing scanning grid and the defect reflection response is pre-established. By finding the entry in this relationship table that is closest to the initial value of defect orientation, the initial value is corrected using the corresponding response intensity weighting value, thereby obtaining an accurate defect spatial azimuth angle. Further, along the direction of the defect spatial azimuth angle and its orthogonal directions, the boundary distance of the region with a continuous abrupt drop in acoustic impedance value is statistically analyzed. Preferably, this boundary distance refers to the maximum span distance between the nearest neighbor boundary points where the acoustic impedance gradient modulus exceeds a preset gradient threshold, and this distance is converted into the projected width of the defect in the direction perpendicular to the azimuth angle, as the equivalent size of the defect. Combining the defect spatial azimuth angle and the equivalent size for output constitutes a comprehensive and accurate quantitative defect detection result.

[0010] As an improved technical solution of this invention, after constructing the three-dimensional tensor field of acoustic properties, it can be further enhanced. By performing global normalization and histogram equalization on the anisotropic intensity index in the three-dimensional tensor field, an enhanced anisotropic distribution map is generated, thereby effectively enhancing the intensity difference in low-contrast regions. This enhanced anisotropic distribution map is then fused point-by-point with the scattering principal direction encoding to generate a multimodal acoustic property fusion tensor field, which is used in subsequent nonlinear inversion iterative reconstruction. This optimization method can strengthen the acoustic feature expression of weak defects, making the defect boundary outline clearer in the reconstructed acoustic impedance distribution map, further improving the detection capability of small defects and the stability of quantitative measurement.

[0011] The technical effects and advantages provided by the present invention in the above technical solution are as follows: By collecting echo amplitude and phase sequences from at least three different incident angles at each focal point, a scattering direction response vector is constructed. This vector is then decomposed using its covariance matrix to extract anisotropic intensity indices and scattering direction codes along the main scattering direction, forming a three-dimensional tensor field of acoustic properties. Unlike conventional single-dimensional analysis that relies solely on echo amplitude, this scheme analyzes the directional distribution characteristics of acoustic properties from multi-angle scattering responses, allowing the orientation information of defects to be quantified and encoded within the three-dimensional tensor field. Subsequently, based on the multi-dimensional constraints provided by this tensor field, the spatial azimuth angle of the defect can be more accurately inverted, overcoming the shortcomings of traditional methods that cannot accurately calculate defect orientation due to a lack of scattering anisotropic information. Furthermore, by marking the boundaries of the anomalous impedance region with three-dimensional connected components and extracting the surface mesh, the average curvature and Gaussian curvature of the mesh vertices are calculated. A curvature morphology histogram is generated and multi-peak decomposition is performed. Based on the curvature intervals corresponding to the main peaks, the connected components are divided into smooth and sharp regions. The volume ratio of the smooth region determines whether the connected component belongs to an inhomogeneous region or a defect candidate region. Unlike conventional simple segmentation methods based on acoustic impedance thresholds, this scheme introduces boundary curvature morphology as a criterion for distinguishing defects from tissue. By utilizing the morphological difference between the drastic curvature changes at defect boundaries and the relatively smooth boundaries of tissue inhomogeneous regions, reliable separation of the two types of abnormal regions is achieved at the geometric level. This approach reduces the probability of misclassifying tissue inhomogeneity as defects, improving the specificity of defect detection and the accuracy of quantitative analysis. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0013] Figure 1 This is a flowchart of a quantitative detection method for internal defects in nuclear power pipeline forgings. Figure 2 This is a flowchart of the abnormal impedance region separation based on curvature morphology analysis; Figure 3 This is a flowchart for generating quantitative defect detection results; Figure 4 This is a flowchart of the multimodal acoustic property fusion tensor field generation process; Figure 5 It is a curve showing the relationship between the normalized echo amplitude and the incident angle direction at different focusing depths; Figure 6 It is the histogram of the curvature shape of the connected domain of the abnormal impedance region and the multimodal decomposition result of the Gaussian mixture model. Figure 7It is the normalized histogram equilibrium mapping curve of enhanced anisotropy intensity. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] See Figure 1 This invention provides a method for quantitative detection of internal defects in nuclear power pipeline forgings. The method includes: deploying an ultrasonic phased array probe array on the surface of the forging to be tested; generating a virtual acoustic beam focusing scanning grid covering the entire thickness of the forging by controlling the excitation and reception phases of each array element in a timing sequence; extracting multi-angle scattering features of each focal point based on the echo signals of the virtual acoustic beam focusing scanning grid to construct a three-dimensional tensor field of acoustic properties inside the forging; iteratively reconstructing the acoustic impedance distribution map inside the forging using a nonlinear inversion algorithm based on the acoustic property three-dimensional tensor field, and marking abnormal impedance regions on the acoustic impedance distribution map; performing curvature morphology analysis on the boundaries of the abnormal impedance regions to separate the abnormal impedance regions into defect candidate regions and non-uniform structure regions; for the defect candidate regions, extracting defect orientation parameters based on the multi-angle scattering features, and calculating the spatial azimuth angle and equivalent size of the defect by combining the scanning geometry of the virtual acoustic beam focusing scanning grid, thereby generating quantitative defect detection results.

[0016] Example 1

[0017] In a specific implementation, in Example 1 of the method for quantitative detection of internal defects in nuclear power pipeline forgings, the method for generating a virtual acoustic beam focusing scanning grid covering the full thickness of the forging is as follows: An ultrasonic phased array probe array is deployed on the surface of the forging under test, and the excitation and reception phases of each array element are controlled by timing. Based on the geometry and thickness distribution of the forging under test, the ultrasonic phased array probe array is divided into multiple sub-aperture groups along the circumference and axial direction of the forging's outer wall. Each sub-aperture group contains at least four adjacent array elements. During division, based on the three-dimensional contour scanning data of the forging's outer wall, the spatial distribution density of the sub-aperture groups is increased in sections where the curvature change exceeds a preset threshold, ensuring that the thickness variation range of the forging wall covered by each sub-aperture group does not exceed the focal spot diameter corresponding to the beam half-diffusion angle. A focusing depth range is independently set for each sub-aperture group. The lower limit of the focusing depth range is the minimum remaining thickness between the outer and inner walls of the forging below that sub-aperture group, and the upper limit of the focusing depth range is the maximum remaining thickness between the outer and inner walls of the forging below that sub-aperture group. The excitation delay time and receiving window offset of each array element are calculated according to the focusing depth range. The excitation delay time is calculated based on the propagation speed of the sound wave in the forging material, the geometric distance from the spatial coordinates of the array element to the focal point, and the beam deflection angle, combined with the relative path difference between the array element position and the focal point. The receiving window offset is set according to the echo arrival time difference at different depths within the focusing depth range, enabling dynamic tracking of scattered signals at different depths during reception.

[0018] Based on the excitation delay time and the receiving window offset, transmit focusing and receive dynamic focusing are sequentially performed on each sub-aperture group to generate the corresponding focused acoustic beam scan line. During transmit focusing, a pulse excitation signal with a corresponding time offset is applied to each array element according to the excitation delay time, so that the sound waves emitted by each array element are superimposed in phase at the predetermined focal point. During receive dynamic focusing, a time-varying delay compensation is applied to the echo signal received by each array element according to the receiving window offset, so that the receiving focus at each moment along the sound beam propagation path moves continuously along the depth direction, forming a complete focused acoustic beam scan line. The focused acoustic beam scan lines of all sub-aperture groups are spliced ​​according to their spatial positions to form a virtual acoustic beam focused scan grid covering the entire thickness of the forging. During the splicing process, spatial coordinate interpolation and amplitude transition smoothing are performed on the focused acoustic beam scan lines in the boundary area of ​​adjacent sub-aperture groups to eliminate signal discontinuities caused by grating lobe effects.

[0019] In some embodiments, when there are eight array elements in a sub-aperture group and two array elements overlap between two adjacent sub-aperture groups, the focused acoustic beam scan line in the overlapping area is obtained by weighted averaging of the data generated by the two sub-aperture groups respectively, and the weighting coefficient is based on the linear change of the distance from the center of the focused acoustic beam to the boundary of the overlapping area.

[0020] Based on the echo signal of the virtual acoustic beam focused scanning grid, the multi-angle scattering features of each focal point are extracted, and the following is the implementation method for constructing the three-dimensional tensor field of acoustic properties inside the forging.

[0021] For each focal point in the virtual acoustic beam focusing scanning grid, at least three echo amplitude and phase sequences under different incident angle directions are collected. Different incident angle directions are obtained by decomposing the acoustic ray paths corresponding to multiple different transmit and receive combinations in the virtual acoustic beam focusing scanning grid of the same focal point. Each incident angle direction corresponds to the vector direction from the midpoint of the line connecting the center of the transmit array element and the center of the receive array element to the focal point.

[0022] The echo amplitude and echo phase sequences at the same focal point are arranged in order of incident angle to construct the scattering direction response vector for that focal point. The scattering direction response vector includes amplitude and phase dimensions. For the case of collecting N incident angles, the dimension of the scattering direction response vector is 2N. The first N elements are the normalized echo amplitudes under the N incident angles, and the last N elements are the echo phases under the N incident angles. The phase values ​​are represented in radians and folded to the interval [0, 2π).

[0023] The covariance matrix of the scattering direction response vector is decomposed to extract eigenvalues ​​and corresponding eigenvectors along the main scattering direction. The eigenvalues ​​are used as an anisotropy intensity index, and the eigenvectors are used as the main scattering direction encoding. The covariance matrix is ​​constructed from the amplitude components of the scattering direction response vector. This is done by treating the normalized echo amplitudes at N incident angles as a point in N-dimensional space, where each dimension of the point represents the amplitude at the corresponding incident angle. The second-order statistical characteristics reflecting the energy of the scattering direction are obtained by calculating the sample covariance matrix of this point in N-dimensional space. Eigenvalue decomposition of the covariance matrix yields N eigenvalues ​​and N corresponding eigenvectors. The largest eigenvalue is selected as the anisotropy intensity index, and the eigenvector corresponding to the largest eigenvalue is used as the main scattering direction encoding. Each component in the main scattering direction encoding corresponds to a weight representing the scattering energy contribution at an incident angle.

[0024] The formula used for eigenvalue decomposition of the covariance matrix is:

[0025] in, This represents the normalized echo amplitudes composed of N incident angles. The covariance matrix, matrix elements Indicates the first The direction of the incident angle is related to the first... Covariance between the amplitudes of each incident angle direction and The values ​​of all are integers between 1 and N; Indicates the first 1 eigenvector The value of is an integer between 1 and N, and the eigenvector is an N-dimensional column vector with a magnitude of 1. Indicates the first There are eigenvalues, which are real numbers and non-negative; among all eigenvalues, the eigenvalue with the largest value is denoted as . , The corresponding large eigenvector is denoted as , That is, as an index of anisotropy intensity, That is, as the encoding of the main scattering direction. The Middle The component represents the direction of the incident angle. The projection intensity of the scattering principal direction on the surface, The value of is an integer between 1 and N.

[0026] The anisotropic intensity index and scattering principal direction code of each focal point are filled into a three-dimensional matrix according to spatial coordinates to form a three-dimensional tensor field of acoustic properties inside the forging. The three dimensions of the three-dimensional matrix correspond to the spatial discrete coordinates of the forging along the axial, circumferential, and wall thickness directions, respectively. Each matrix element contains a scalar anisotropic intensity index and a vector scattering principal direction code. During the filling process, for focal points that do not fall into grid nodes, the anisotropic intensity index and scattering principal direction code are mapped to the eight nearest neighbor grid nodes using a trilinear interpolation method. This filling process is a scattering operation that assigns the properties of discrete focal points to regular grid nodes. First, the grid cell where each focal point is located is determined, and the normalized coordinates of the focal point along the axial, circumferential, and wall thickness directions within that cell are calculated. The values ​​of the normalized coordinates in each direction are all in the range of zero to one. For the anisotropy intensity index, the weights assigned to the eight grid nodes are determined by the normalized coordinates: if a node is located in the opposite direction to the focal point relative to the cell corner in a certain direction, the weight for that direction is the complementary value of the normalized coordinates; if the node is in the same direction, the weight for that direction is the normalized coordinate value itself. Multiplying the weights of the three directions yields the final assigned weight for that node. The same weights are used for assigning the scattering principal direction encoding vectors. The scattering principal direction encodings on the eight nodes are respectively summed with the scattering principal direction encoding of the focal point multiplied by its corresponding weight. This process requires the focal point to be strictly located inside the grid cell to ensure that the sum of the weights remains 1, thus conserving the total amount of the attribute before and after assignment.

[0027] See Figure 5 In the graph, the horizontal axis represents the incident angle direction number, ranging from 1 to 50°, and the vertical axis represents the normalized echo amplitude, ranging from 0 to 1. The legend labels curves for multiple focal point depths, from 5mm to 50mm, spaced 5mm apart, distinguished by different line types and markings. Each curve reflects the distribution characteristics of the normalized echo amplitude at the corresponding focal point along different incident angle directions.

[0028] The curves exhibit a distinct local peak shape, and as the depth of the focal point increases, the peak positions shift sequentially to the right along the incident angle direction. For example, the normalized echo amplitude of the focal point at a depth of 5 mm is mainly concentrated in the interval numbered 10 to 15 along the incident angle direction, with the peak value close to 1; the peak value of the focal point at a depth of 15 mm is located near the interval numbered 0 to 10; the peak value of the focal point at a depth of 25 mm moves to the interval numbered 35 to 40; and the peak values ​​of the focal points at depths of 35 mm, 40 mm, and 50 mm are located in the intervals numbered 40 to 45, 30 to 35, and 20 to 30, respectively. The echo amplitude distributions of the focal points at different depths show a certain degree of overlap, but the peak values ​​are clearly separated, indicating that the incident angle directions corresponding to the focal points at different depths are distinguishable.

[0029] There is a certain correlation between the peak width of the curve and the depth. For shallower depths (such as 10mm and 15mm), the peak width of the curve is narrower, and the variation of the normalized echo amplitude is more concentrated. For deeper depths (such as 40mm and 45mm), the peak width of the curve is relatively wider, and the normalized amplitude remains at a high level over a wider range of incident angles. In some depths, the normalized echo amplitude of the curve is close to zero in the non-peak range, reflecting that the scattering energy at the focal point corresponding to that incident angle is weak.

[0030] Example 2

[0031] In practical implementation, the acoustic impedance distribution map inside the forging is iteratively reconstructed using a nonlinear inversion algorithm based on the three-dimensional tensor field of acoustic properties, and the abnormal impedance region is marked on the acoustic impedance distribution map as follows: The internal space of the forging is discretized into multiple voxel elements. An initial acoustic impedance estimate is assigned to each voxel element, generating an initial acoustic impedance distribution map. Based on the material grade of the forging under test, the standard acoustic impedance range corresponding to that material grade is retrieved from the material property database. The material property database stores standard acoustic impedance parameters for various commonly used material grades in nuclear power pipeline forgings. The standard acoustic impedance range is obtained by statistically analyzing the acoustic impedance values ​​obtained after performing multiple acoustic impedance measurements on standard samples of the same material grade. The median value of the standard acoustic impedance range is used as the initial reference acoustic impedance value for all voxel elements within the forging. The median value is the arithmetic mean of the upper and lower limits of the standard acoustic impedance range.

[0032] At least three calibration points are selected at different locations on the surface of the forging to be tested. These calibration points cover regions with different curvatures and wall thicknesses on the outer wall of the forging. The spatial distance between each calibration point is no less than five times the width of a single element in the ultrasonic phased array probe array. The measured acoustic impedance value at each calibration point is measured using the pulse-echo method. The pulse-echo method involves vertically emitting an ultrasonic pulse at the calibration point and receiving the reflected echo from the bottom surface. The measured acoustic impedance value is calculated based on the ratio of the bottom surface echo amplitude to the emitted pulse amplitude and the sound path distance. The ratio of the measured acoustic impedance value to the initial reference acoustic impedance value is calculated to generate a correction coefficient for each calibration point. The correction coefficient reflects the degree of deviation of the actual acoustic impedance at the calibration point relative to the median value of the material's standard acoustic impedance.

[0033] Spatial interpolation is performed on the correction coefficients at each calibration point to generate a correction coefficient distribution field covering the internal space of the forging. The spatial interpolation uses the radial basis function interpolation method, with the spatial coordinates of the calibration points as interpolation nodes and the correction coefficients as node values. A Gaussian radial basis function is selected in the radial basis function interpolation method, and its shape parameter is set to be proportional to the reciprocal of the average distance between calibration points, ensuring a smooth spatial transition in the interpolation results. The correction coefficient distribution field is a scalar field defined on all voxel elements within the internal space of the forging. The correction coefficient for each voxel element is obtained by weighted summation of the correction coefficients at each calibration point using the radial basis function.

[0034] The initial acoustic impedance estimate for each voxel is obtained by multiplying the correction coefficient distribution field point by point with the initial reference acoustic impedance value. The result of multiplying the value of the correction coefficient distribution field at each voxel position with the initial reference acoustic impedance value is the initial acoustic impedance estimate for that voxel. The initial acoustic impedance distribution map is generated by filling all voxel units with the initial acoustic impedance estimates. The initial acoustic impedance distribution map is a three-dimensional array, with the three dimensions corresponding to the voxel indices along the axial, circumferential, and wall thickness directions within the forging's internal space, respectively.

[0035] Based on the initial acoustic impedance distribution map and the propagation path of the virtual acoustic beam focusing scanning grid, the predicted echo data for forward sound field propagation is calculated. The calculation of the predicted echo data for forward sound field propagation employs a ray tracing method. The transmission and reception paths corresponding to each focused acoustic beam scanning line in the virtual acoustic beam focusing scanning grid are discretized into several ray segments. For each ray segment, the reflection and transmission coefficients of the sound wave at the interface are calculated based on the initial acoustic impedance estimates of the voxel elements through which the segment passes. The energy attenuation and phase change during sound wave propagation are recursively accumulated to finally obtain the predicted echo signal amplitude and phase at the receiving array element location.

[0036] The residuals between the predicted echo data and the measured echo data for forward sound field propagation are calculated. The measured echo data are the echo signals actually acquired at the corresponding focused sound beam scan lines in the virtual sound beam focusing scan grid. The residual is the joint norm of the difference between the amplitude and the measured echo signal, and the difference between the phase and the phase of the predicted echo signal. The gradient descent direction of the objective function is constructed based on the residuals. The objective function is defined as the sum of squared residuals for all focused sound beam scan lines.

[0037] In some embodiments, the update rule for updating the acoustic impedance estimate of each voxel unit along the gradient descent direction of the objective function is as follows:

[0038] in, Indicates the first The acoustic impedance vector formed by the acoustic impedance estimates of all voxel units at the next iteration. Index for iteration count, The initial value is 0. The vector is obtained by expanding the estimated initial acoustic impedance of each voxel element in the initial acoustic impedance distribution diagram in voxel order. Indicates the first The acoustic impedance vector updated in the next iteration; This represents the iteration step size, with a value range of 1. to , The value is determined based on the magnitude of the acoustic impedance vector and the current gradient vector. The second-norm value is determined when the ratio of the second-norm of the gradient vector to the second-norm of the acoustic impedance vector is greater than... hour, Take the smaller value ,otherwise Take the larger value ; Describe the objective function exist The gradient vector at the point, the objective function Defined as the sum of squared residuals between the predicted echo data and the measured echo data of the forward sound field propagation, each component in the gradient vector corresponds to the rate of change of the acoustic impedance estimate of a voxel unit with respect to the objective function, and this component is calculated by the adjoint state method.

[0039] The process is repeated iteratively. After each iteration, the acoustic impedance vector is updated, and the predicted echo data for the forward sound field propagation is recalculated, along with a new residual. This continues until the norm of the residual is less than a preset residual threshold. The preset residual threshold is set to a fraction of the initial residual norm. The initial residual norm is calculated from the predicted echo data and measured echo data corresponding to the initial acoustic impedance distribution map before the first iteration. When the residual norm is less than the preset residual threshold, the iteration stops, and the updated acoustic impedance distribution map is used as the reconstructed acoustic impedance distribution map.

[0040] On the reconstructed acoustic impedance distribution map, the set of voxels whose acoustic impedance values ​​deviate from the average acoustic impedance values ​​of their adjacent voxels by more than a preset deviation multiple is marked as an abnormal impedance region. For each voxel in the reconstructed acoustic impedance distribution map, its six neighboring voxels are identified, and the arithmetic mean of the acoustic impedance values ​​of these neighboring voxels is calculated as the average acoustic impedance value of the neighboring voxels. The absolute value of the difference between the acoustic impedance value of this voxel and the average acoustic impedance value of its neighboring voxels is calculated. When this absolute value exceeds the average acoustic impedance value of the neighboring voxels multiplied by the preset deviation multiple, the voxel is marked as abnormal. The preset deviation multiple is set to 0.15, which is based on the tolerance range of acoustic impedance uniformity of nuclear power pipeline forging materials. The set of all voxel units marked as abnormal constitutes the abnormal impedance region.

[0041] Example 3

[0042] In specific implementation, please refer to Figure 2 The following is a method for separating the abnormal impedance region into a defect candidate region and an inhomogeneous tissue region by performing curvature morphology analysis on the boundary of the abnormal impedance region: For each voxel element in the abnormal impedance region, adjacent voxels are marked as belonging to the same connected region according to the six-neighbor connectivity criterion. This process continues until all voxels satisfying the connectivity condition have been marked, and the voxel index set of each connected region is output. The six-neighbor connectivity criterion is defined as follows: for a voxel element in the abnormal impedance region, if there are voxel elements belonging to the abnormal impedance region among its six directly adjacent voxel elements along the three orthogonal coordinate axes of the discretized mesh inside the forging space, then the two adjacent voxel elements are assigned to the same connected region. The connected region marking is performed using a three-dimensional seed filling algorithm. Starting from any unmarked voxel element in the abnormal impedance region, this voxel element is used as the seed voxel and assigned a new connected region label. The six neighbors of the seed voxel are searched, and unmarked voxel elements belonging to the abnormal impedance region in the neighborhood are added to the current connected region and used as new seed voxels to continue the search. This continues until no new voxel elements can be added to the current connected region. Then, the next unmarked voxel element is selected to start a new connected region construction process. This process is repeated until all voxel elements in the abnormal impedance region are assigned to their corresponding connected regions. The output of each connected domain is a set of voxel indices, which records the coordinates of all voxel elements in the three-dimensional space grid.

[0043] An isosurface extraction algorithm is performed on the voxel index set of each connected component to generate a triangulated surface mesh for that component. The triangulated surface mesh contains vertex coordinate sequences and triangular facet topological relationships. The isosurface extraction algorithm employs the moving cube algorithm, which treats each voxel cell in the connected component's voxel index set as a cube. The topological configuration of the isosurfaces inside the cube is determined based on the binary states of the cube's eight vertices, where the binary states are determined by whether the spatial coordinates corresponding to the cube's vertices belong to an anomalous impedance region. All facets from the cubes traversed by the surface of the connected component are combined to form the triangulated surface mesh for that component. The vertex coordinate sequence of the triangulated surface mesh records the three-dimensional coordinates of each triangular facet vertex in the forging's internal spatial coordinate system, and the triangular facet topological relationships record which three vertices constitute each triangular facet and the adjacency relationships between the triangular facets.

[0044] For each vertex in the triangulated surface mesh, extract the set of first-order neighborhood triangles for that vertex. Calculate the average curvature and Gaussian curvature of that vertex based on the normal vectors of each triangle in the first-order neighborhood triangle set. The first-order neighborhood triangle set is defined as the set of all triangles in the triangulated surface mesh that use the current vertex as one vertex. The normal vector of each triangle is calculated by the cross product of its two edge vectors, and the normal vectors all point outwards from the connected region.

[0045] Mean curvature of vertices and Gaussian curvature The calculation formula is:

[0046] in, Represents the vertex index in the triangulated surface mesh where the curvature is currently being calculated; Represents vertices The average curvature at that point is a scalar value, with the unit being the reciprocal of the unit of length; Represents vertices The Gaussian curvature at that point is a scalar value, expressed in units of length squared. Represents vertices The set of first-order neighborhood vertices, where each vertex in the set of first-order neighborhood vertices is connected to a vertex in the set of first-order neighborhood vertices. They are directly connected by an edge in the triangulated surface mesh; Represents the set of first-order neighborhood vertices A neighboring vertex; and They represent the edges respectively In two triangular facets sharing a common edge, the edge... The interior angles at the two opposite vertices, Represents the cotangent function; Represents vertices Three-dimensional coordinate vector in the internal spatial coordinate system of the forging; Representing neighborhood vertices Three-dimensional coordinate vector in the internal spatial coordinate system of the forging; Indicates the edge It is one-third of the sum of the areas of the two triangular faces sharing a common edge, i.e., the edge. The associated local area weight; Represents vertices The set of first-order neighborhood triangular patches; express A triangular facet in; Represents a triangular facet At the apex The interior angle at the location; Represents triangular facets One-third of the area.

[0047] Based on the distribution of the ratio of mean curvature to Gaussian curvature, a curvature shape histogram is generated for each connected region. For each vertex on the triangulated surface mesh of the connected region, the ratio of mean curvature to Gaussian curvature is calculated. When the absolute value of the Gaussian curvature is less than... When the ratio is zero, it is set to zero to avoid division by zero anomalies. The range of the ratio distribution is divided into several equal-width intervals. The proportion of vertices falling into each equal-width interval is counted out of the total number of vertices. A histogram of the curvature shape of the connected components is generated with the equal-width intervals as the x-axis and the proportion of vertices as the y-axis. The width of the equal-width intervals is determined based on the standard deviation of the ratio distribution, and the number of intervals is taken as an integer value that is directly proportional to the square root of the total number of vertices.

[0048] Multimodal decomposition is performed on the curvature shape histogram to extract the curvature intervals corresponding to the main peaks. The portions of the connected components whose curvature values ​​fall within these intervals are marked as smooth regions. The multimodal decomposition uses a Gaussian mixture model to fit the numerical curves of the curvature shape histogram. The number of Gaussian components in the Gaussian mixture model is determined by the Akaike information criterion. The Akaike information criterion calculates information criterion values ​​for different numbers of Gaussian components as the number of Gaussian components increases from 1 to a preset upper limit. The number of Gaussian components that minimizes the information criterion value is selected as the number of peaks in the multimodal decomposition. In the Gaussian mixture model fitting results of the curvature shape histogram, the Gaussian components are sorted from largest to smallest weight. The curvature interval corresponding to the Gaussian component with the largest weight is taken as the curvature interval corresponding to the main peak. The surface region formed by all vertices in the connected component whose curvature ratios fall within the curvature interval corresponding to the main peak is marked as a smooth region. The volume of the smooth region is obtained by multiplying the number of voxel units covered by the smooth region by the volume of a single voxel unit.

[0049] When the proportion of smooth regions to the total volume of connected regions exceeds a preset threshold, the connected region is classified as a region of non-uniform microstructure; otherwise, it is classified as a candidate defect region. The preset threshold is set to 0.6. This value is based on the fact that non-uniform microstructure regions in nuclear power pipeline forgings typically present as large areas with gradually changing internal acoustic impedance and gentle changes in surface curvature, while defect regions exhibit local anomalies with drastic changes in boundary curvature and a small proportion of smooth regions. The total volume of the connected region is obtained by multiplying the number of all voxel units within the connected region by the volume of a single voxel unit. The ratio of the smooth region volume to the total volume of the connected region is the proportion of the smooth region to the total volume of the connected region.

[0050] See Figure 6 The figure shows the distribution of the ratio of average curvature to Gaussian curvature of the vertices on the surface of each connected domain within the anomalous impedance region. This distribution is presented as a curvature shape histogram, with the horizontal axis representing the ratio of average curvature to Gaussian curvature and the vertical axis representing the proportion of the corresponding vertex to the total number of vertices. The histogram is represented by gray bars, showing the frequency distribution characteristics of this ratio in different intervals. The solid black line in the figure is the fitting curve of the curvature shape histogram based on the Gaussian mixture model, which clearly depicts the multi-peak structure of the curvature ratio. The dashed line represents the main peak component in the fitting curve, and the dotted line represents the secondary peak component, reflecting the main and secondary curvature intervals in the distribution, respectively. The shaded area in the figure is marked as a smooth area, corresponding to the interval where the curvature ratio is relatively concentrated, indicating that the vertex distribution in this range is relatively uniform and the curvature changes are gradual.

[0051] The curve trend shows that the main peak component is concentrated in the range of curvature ratio close to 0 to 0.7, with the highest proportion of vertices at the peak, indicating that most vertices are located within this smooth region, consistent with the characteristics of uneven tissue. The secondary peak component is distributed in the range of curvature ratio approximately 1.3 to 2.5, with a lower proportion of vertices, representing regions with more drastic curvature changes, potentially corresponding to candidate defect areas. Curvature ratios with an absolute value less than 1 × 10⁻ 8 The Gaussian curvature is set to zero to avoid anomalies and ensure the stability of the statistical results.

[0052] This figure illustrates the curvature morphology analysis of the abnormal impedance region boundary as described in Example 3. By decomposing the curvature morphology histogram using a multi-peak decomposition method and extracting the curvature intervals corresponding to the main peaks, smooth regions are identified. When the proportion of smooth regions exceeds a set threshold of 0.6, the connected region is determined to be a region of non-uniform structure; otherwise, it is determined to be a candidate region for defects. The current example histogram and fitted curve clearly reflect the bimodal distribution characteristics of the curvature morphology within the connected region, providing a quantitative basis for further distinguishing between candidate regions for defects and regions of non-uniform structure.

[0053] Example 4

[0054] In specific implementation, please refer to Figure 3For the defect candidate region, defect orientation parameters are extracted based on multi-angle scattering characteristics. Combined with the scanning geometry of the virtual acoustic beam focusing scanning grid, the spatial azimuth angle and equivalent size of the defect are calculated, and the quantitative defect detection results are generated as follows: For each candidate defect region, the focal point is selected, and the direction component with the largest amplitude in the principal scattering direction encoding is extracted. The tilt angle and azimuth angle of the direction component in three-dimensional space are used as the initial values ​​for defect orientation. The principal scattering direction encoding is the feature vector obtained when constructing the three-dimensional tensor field of acoustic properties inside the forging in Example 1. The principal scattering direction encoding is an N-dimensional vector, where N is the number of incident angle directions used when acquiring echo signals from the focal point. The component represents the first... The projection intensity of the scattering principal direction along each incident angle, The value of is an integer between 1 and N. All components in the main scattering direction encoding are traversed, and the absolute values ​​of each component are compared. The component with the largest absolute value is recorded as the direction component with the largest amplitude. The incident angle direction corresponding to the direction component with the largest amplitude has a definite geometric direction in the virtual acoustic beam focusing scanning grid. This geometric direction is represented by a spatial vector pointing from the midpoint of the line connecting the center of the transmitting element and the center of the receiving element to the focal point. This spatial vector is decomposed in the forging coordinate system to obtain the tilt angle relative to the forging wall thickness direction and the azimuth angle relative to the forging circumferential reference direction. The tilt angle and azimuth angle are used as the initial values ​​of the defect orientation at the focal point. A vector average is performed on the initial values ​​of the defect orientation at all focal points within the defect candidate region. During vector averaging, the initial value of the defect orientation at each focal point is converted into a unit spatial vector. The sum of all unit spatial vectors is calculated and normalized. The tilt angle and azimuth angle corresponding to the normalized composite vector are used as the initial values ​​of the defect orientation for the defect candidate region.

[0055] Based on the incident and receiving directions of the focused acoustic beam in the virtual acoustic beam focusing scanning grid, a geometric mapping table is established between the defect reflection response and the defect orientation. The geometric mapping table is a two-dimensional data table. The rows of the data table correspond to different defect orientations, characterized by two angular parameters: tilt angle and azimuth angle. The tilt angle is discretized within a preset angular step size from 0 to 90 degrees, and the azimuth angle is discretized within a preset angular step size from 0 to 360 degrees. The columns of the data table correspond to all available combinations of the incident and receiving directions of the focused acoustic beam in the virtual acoustic beam focusing scanning grid. Each combination of the incident and receiving directions is determined by the spatial vectors of the center coordinates of the transmitting sub-aperture group and the center coordinates of the receiving sub-aperture group relative to the focal point.

[0056] The response intensity weighted value stored in each cell of the geometric mapping table is calculated from the acoustic reflection intensity coefficient based on the sine relationship between the angle between the defect orientation and the incident direction of the focused sound beam. The acoustic reflection intensity coefficient is defined as the ratio of reflected energy to incident energy when the normal vector direction of the defect surface and the incident sound beam direction satisfy the specular reflection condition. At that time, the response intensity weighted value Calculated using the following formula:

[0057] in, This represents the weighted value of the response intensity, a dimensionless scalar, with a value range of 0 to 1. ; The maximum response intensity weighting value is set to 1.0, representing the normalized response intensity when the direction of the defect surface normal vector is completely parallel to the direction of the incident sound beam. This represents the angle between the direction of the normal vector of the defect surface and the direction of the incident sound beam, expressed in degrees, ranging from 0 to 90 degrees. It is calculated by the inverse cosine of the dot product of the spatial vector corresponding to the defect orientation and the spatial vector corresponding to the incident direction of the focused sound beam; This indicates the peak angle of specular reflection. Setting it to 0 degrees means that the response intensity reaches its maximum when the defect surface is perpendicular to the incident direction of the sound beam. This indicates the angular response width parameter, set to 15 degrees. The value is determined based on the half-power beam angle of the array elements in the ultrasonic phased array probe array. When the half-power beam angle of the array element is 30 degrees, Take half of the half-power beam angle, which is 15 degrees.

[0058] The geometric mapping table is searched for the mapping entry closest to the initial defect orientation value. The search method involves finding the row with the smallest absolute difference between the tilt angle and azimuth angle in the initial defect orientation value, respectively, and selecting this row as the closest mapping entry. From the closest mapping entry, the response intensity weighted values ​​corresponding to each combination of focused beam incident and receiving directions are extracted. These response intensity weighted values ​​are then used to correct the initial defect orientation value. The correction method is as follows: the three largest combinations of focused beam incident and receiving directions with the highest response intensity weighted values ​​are selected, and the measured echo amplitude values ​​corresponding to these three combinations are obtained. The ratio of the measured echo amplitude value to the response intensity weighted value is used as a correction factor. The average of the three correction factors is calculated, and this average is multiplied by the tilt angle and azimuth angle in the initial defect orientation value. The multiplication result is used as the corrected tilt angle and azimuth angle. The combination of the corrected tilt angle and azimuth angle constitutes the defect spatial azimuth angle.

[0059] Along the defect spatial azimuth direction, the boundary distances of regions with continuously decreasing acoustic impedance values ​​are statistically analyzed within the defect candidate region. These boundary distances are the maximum span between the nearest neighbor boundary points where the acoustic impedance gradient modulus exceeds a preset gradient threshold along the defect spatial azimuth direction and its orthogonal directions. A region with continuously decreasing acoustic impedance is defined as a sequence of continuous voxel units within the defect candidate region where the acoustic impedance value monotonically decreases along the defect spatial azimuth direction and the rate of change of acoustic impedance between adjacent voxels exceeds three times the average rate of change. The acoustic impedance gradient modulus is obtained by calculating the ratio of the acoustic impedance difference between the current voxel unit and its adjacent voxel units to the voxel unit size. The preset gradient threshold is five times the average gradient modulus of the acoustic impedance values ​​in the normal region outside the defect candidate region. The average gradient modulus of the acoustic impedance values ​​in the normal region is statistically obtained from the non-abnormal impedance regions in the reconstructed acoustic impedance distribution map.

[0060] In the direction perpendicular to the defect's spatial azimuth, the boundary distance is converted into a projected width. This conversion is achieved by multiplying the maximum span distance along the defect's spatial azimuth and the maximum span distance along the orthogonal direction by the sine of the angle between the defect's spatial azimuth and each direction, respectively. The larger of these products is taken as the projected width of the defect in the direction perpendicular to the spatial azimuth, and this projected width is used as the equivalent size of the defect. The defect's spatial azimuth and equivalent size are combined to form a quantitative defect detection result. This result includes the tilt angle value, azimuth angle value, and equivalent size value, recorded in millimeters.

[0061] Example 5

[0062] In specific implementation, please refer to Figure 4 Based on the echo signal of the virtual acoustic beam focusing scanning grid, the multi-angle scattering features of each focal point are extracted. After constructing a three-dimensional tensor field of acoustic properties inside the forging, the three-dimensional tensor field of acoustic properties is enhanced and a multi-modal acoustic property fusion tensor field is generated. The implementation method is as follows: The anisotropic intensity indices of all voxel elements in the three-dimensional tensor field of acoustic properties are traversed, and the maximum and minimum values ​​of the anisotropic intensity indices are extracted. A linearly normalized mapping interval is constructed based on the maximum and minimum values. The traversal process is carried out sequentially along the three dimensions of the three-dimensional tensor field of acoustic properties, which correspond to the voxel coordinate indices along the axial, circumferential, and wall thickness directions of the forging's internal space, respectively. During the traversal, the anisotropy intensity index value of the current voxel is compared with the recorded maximum anisotropy intensity index value. If the current voxel's anisotropy intensity index value is greater than the recorded maximum anisotropy intensity index value, the recorded maximum anisotropy intensity index value is updated to the current voxel's anisotropy intensity index value. Simultaneously, the current voxel's anisotropy intensity index value is compared with the recorded minimum anisotropy intensity index value. If the current voxel's anisotropy intensity index value is less than the recorded minimum anisotropy intensity index value, the recorded minimum anisotropy intensity index value is updated to the current voxel's anisotropy intensity index value. After the traversal is complete, the final recorded global maximum anisotropy intensity index value is denoted as... The minimum value of the final recorded global anisotropy intensity index is denoted as... The lower bound of the linearly normalized mapping interval is The upper bound of the linearly normalized mapping interval is .

[0063] The anisotropy intensity index of each voxel is projected onto a numerical range of zero to one according to a linearly normalized mapping interval, generating the normalized intensity value of each voxel. For voxel coordinates... The voxel unit is used to obtain the original value of the anisotropic intensity index of the voxel unit. ,in This represents the index coordinates of the voxel element along the axial direction of the forging. This represents the index coordinates of the voxel element along the circumferential direction of the forging. This represents the index coordinates of the voxel element along the wall thickness direction of the forging. This represents the anisotropic intensity index of the voxel unit extracted from the acoustic property three-dimensional tensor field. Normalized intensity value. The calculation formula is:

[0064] in, Voxel unit The normalized intensity value is a dimensionless scalar with a value range of 0 to 1; Voxel unit The original values ​​of the anisotropy intensity index are real numbers greater than or equal to zero, obtained from the three-dimensional tensor field of acoustic properties; The minimum global anisotropy intensity index is the minimum anisotropy intensity index obtained after traversing all voxel elements in the three-dimensional tensor field of acoustic properties. Represents the maximum global anisotropy intensity index, which is the maximum anisotropy intensity index obtained after traversing all voxel elements in the three-dimensional tensor field of acoustic properties; when and The difference is less than At that time, directly The value is set to 0 to avoid division by zero.

[0065] The normalized intensity values ​​of all voxel units are arranged according to their original spatial coordinates to form an anisotropic intensity distribution map. The anisotropic intensity distribution map is a three-dimensional array, the dimensions of which are consistent with the number of voxel units in the acoustic property three-dimensional tensor field in the axial, circumferential, and wall thickness directions. Each element in the three-dimensional array stores the normalized intensity value of the corresponding voxel unit. .

[0066] Histogram equalization is performed on the anisotropic intensity distribution map to enhance the intensity differences in low-contrast regions, generating an enhanced anisotropic distribution map. The histogram equalization process involves: statistically analyzing the grayscale histogram of the normalized intensity values ​​of all voxel units in the anisotropic intensity distribution map. Normalized intensity values ​​are considered grayscale levels, with a grayscale resolution of 256 levels. This means the continuous interval of normalized intensity values ​​from 0 to 1 is uniformly divided into 256 grayscale levels, with each grayscale level covering a normalized intensity value interval width of 1 / 256. The number of voxel units falling into each grayscale level is counted to generate the frequency distribution for each grayscale level. The cumulative distribution function is calculated based on the frequency distribution of each grayscale level, with the grayscale level index as follows: cumulative distribution function value Equal to the gray level index from 0 to The sum of frequencies divided by the total number of voxel units, where The value of is an integer between 0 and 255; the normalized intensity value of each voxel element in the anisotropic intensity distribution map. The original gray level index is mapped to a new gray level index using the cumulative distribution function. The mapping rule is that the new gray level index is equal to the cumulative distribution function value. Multiply by 255 and round down; use the normalized value corresponding to the new gray level index as the normalized intensity value of the corresponding voxel unit in the enhanced anisotropy distribution map. Enhanced normalized intensity value The method for determining the enhanced anisotropy distribution map is to divide the new gray level index by 255 to obtain the enhanced normalized intensity value in the range of 0 to 1. After generating the enhanced anisotropy distribution map, the voxel units in the low contrast region whose original normalized intensity values ​​were concentrated in narrow cells are stretched to a wider range of intensity values.

[0067] The enhanced anisotropy distribution map and the scattering principal direction encoding are fused point-by-point to generate a multimodal acoustic property fusion tensor field. The scattering principal direction encoding is extracted from the three-dimensional acoustic property tensor field, for voxel units. The principal scattering direction is encoded as a vector. , The dimension is consistent with the number of incident angle directions used when constructing the three-dimensional tensor field of acoustic properties. The component represents the weight of the scattering energy contribution of the voxel unit in different incident angle directions. The point-by-point weighted fusion method is as follows: for each voxel unit... The normalized enhancement intensity value of the voxel unit in the enhanced anisotropy distribution map. As a weighting coefficient, it is used in conjunction with the encoding vector of the main scattering direction. Multiplying the individual components together yields the fusion vector. ,in Voxel unit The multimodal fusion vector at the location, the dimension of the fusion vector and the encoding vector of the scattering principal direction. Same. Arrange the multimodal fusion vectors of all voxel units according to spatial coordinates, and compare them with the enhanced normalized intensity values. The data are stored together in the multimodal acoustic property fusion tensor field. Each voxel in the multimodal acoustic property fusion tensor field contains two parts of data: an enhanced normalized intensity scalar and a multimodal fusion vector.

[0068] In some embodiments, when the scattering main direction encoding vector The sum of the absolute values ​​of all components is less than At that time, the multimodal fusion vector is directly used. Set it to a zero vector to avoid introducing numerical noise into the fusion operation when the encoding of the main scattering direction is approximately zero.

[0069] A multimodal acoustic property fusion tensor field replaces the acoustic property 3D tensor field in subsequent nonlinear inversion iterative reconstruction. The replacement method is as follows: during the nonlinear inversion iterative reconstruction process, when it is necessary to call the anisotropic intensity index of the voxel element in the acoustic property 3D tensor field, the enhanced normalized intensity value of the corresponding voxel element is extracted from the multimodal acoustic property fusion tensor field. As an alternative, when it is necessary to call the scattering principal direction encoding of voxel units in the three-dimensional tensor field of acoustic properties, the multimodal fusion vector of the corresponding voxel unit is extracted from the multimodal acoustic property fusion tensor field. Alternatively, the remaining steps of the nonlinear inversion iterative reconstruction are consistent with those described in Example 2, including the generation of the initial acoustic impedance distribution map, the calculation of the predicted echo data for forward sound field propagation, the calculation of the residuals and the construction of the gradient descent direction of the objective function, the iterative update of the acoustic impedance estimate, and the calibration of the anomalous impedance region.

[0070] See Figure 7 In the graph, the horizontal axis represents the original normalized intensity value, ranging from 0 to 1, and the vertical axis represents the enhanced normalized intensity value after histogram equalization, also ranging from 0 to 1. The solid line in the curve represents the histogram equalization mapping function, and the dashed line represents the linear reference mapping function. This is used to compare the effect of histogram equalization on the distribution of intensity values.

[0071] As can be seen from the figure, the histogram equalization mapping function exhibits an overall "S"-shaped nonlinear curve characteristic. The curve is significantly higher than the linear reference line in both the lower range (approximately 0 to 0.3) and the higher range (approximately 0.7 to 1.0) of the original normalized intensity value, indicating that the intensity value in this range is stretched, enhancing the contrast between the low-intensity and high-intensity regions.

[0072] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A quantitative detection method for internal defects in nuclear power plant pipeline forgings, characterized in that, The method includes: An ultrasonic phased array probe array is deployed on the surface of the forging to be tested. By controlling the excitation and reception phases of each array element in a timing sequence, a virtual acoustic beam focusing scanning grid covering the full thickness of the forging is generated. Based on the echo signals from the virtual acoustic beam focusing scanning grid, multi-angle scattering features of each focal point are extracted to construct a three-dimensional tensor field of acoustic properties inside the forging, including: For each focal point in the virtual acoustic beam focusing scanning grid, at least three echo amplitude sequences and echo phase sequences under different incident angle directions are collected; Arrange the echo amplitude sequence and echo phase sequence of the same focal point in order of incident angle to construct the scattering direction response vector of that focal point; The covariance matrix of the scattering direction response vector is decomposed to extract the eigenvalues ​​and corresponding eigenvectors in the main scattering direction. The eigenvalues ​​are used as an anisotropy intensity index and the eigenvectors are used as the main scattering direction encoding. The anisotropic intensity index and the scattering principal direction code of each focal point are filled into a three-dimensional matrix according to spatial coordinates to form a three-dimensional tensor field of acoustic properties inside the forging. Based on the three-dimensional tensor field of the acoustic properties, a nonlinear inversion algorithm is used to iteratively reconstruct the acoustic impedance distribution map inside the forging, and abnormal impedance regions are marked on the acoustic impedance distribution map. Curvature morphology analysis is performed on the boundary of the abnormal impedance region to separate the abnormal impedance region into a defect candidate region and a tissue inhomogeneity region, including: The abnormal impedance region is labeled with three-dimensional connected components, the surface mesh of each connected component is extracted, and the average curvature and Gaussian curvature of each vertex on the surface mesh are calculated. Based on the distribution of the ratio of the average curvature to the Gaussian curvature, a curvature shape histogram for each connected region is generated; The curvature shape histogram is decomposed into multiple peaks, the curvature intervals corresponding to the main peaks are extracted, and the parts of the connected domain where the curvature values ​​are located in the curvature intervals corresponding to the main peaks are marked as smooth regions. When the proportion of the smooth region to the total volume of the connected region is greater than a preset proportion threshold, the connected region is determined to be a region of uneven organization; otherwise, it is determined to be a candidate region of defect. For the defect candidate region, the defect orientation parameters are extracted based on the multi-angle scattering characteristics. Combined with the scanning geometry of the virtual acoustic beam focusing scanning grid, the spatial azimuth angle and equivalent size of the defect are calculated to generate quantitative defect detection results.

2. The quantitative detection method for internal defects in nuclear power pipeline forgings according to claim 1, characterized in that, The step of arranging an ultrasonic phased array probe array on the surface of the forging to be tested, and generating a virtual acoustic beam focusing scanning grid covering the full thickness of the forging by timing the excitation and reception phases of each array element, includes: Based on the geometric shape and thickness distribution of the forging to be tested, the ultrasonic phased array probe array is divided into multiple sub-aperture groups along the circumferential and axial directions of the outer wall of the forging, and each sub-aperture group contains at least four adjacent array elements. For each sub-aperture group, an independent focusing depth range is set, and the excitation delay time and receiving window offset of each array element are calculated according to the focusing depth range. Based on the excitation delay time and the receiving window offset, transmit focusing and receive dynamic focusing are performed sequentially on each sub-aperture group to generate the focused acoustic beam scan line of the corresponding sub-aperture group. By splicing the focused acoustic beam scanning lines of all sub-aperture groups according to their spatial positions, a virtual acoustic beam focused scanning grid covering the entire thickness of the forging is formed.

3. The quantitative detection method for internal defects in nuclear power pipeline forgings according to claim 2, characterized in that, Based on the three-dimensional tensor field of the acoustic properties, a nonlinear inversion algorithm is used to iteratively reconstruct the acoustic impedance distribution map inside the forging, and abnormal impedance regions are marked on the acoustic impedance distribution map, including: The internal space of the forging is discretized into multiple voxel elements, and an initial acoustic impedance estimate is assigned to each voxel element to generate an initial acoustic impedance distribution map. Based on the initial acoustic impedance distribution map and the propagation path of the virtual sound beam focusing scanning grid, the predicted echo data of the forward sound field propagation is calculated. Calculate the residual between the predicted echo data and the measured echo data of the forward sound field propagation, and construct the gradient descent direction of the objective function based on the residual; The acoustic impedance estimate of each voxel unit is updated along the gradient descent direction of the objective function. The iteration is repeated until the norm of the residual is less than a preset residual threshold. The updated acoustic impedance distribution map is then used as the reconstructed acoustic impedance distribution map. On the reconstructed acoustic impedance distribution map, the set of voxels whose acoustic impedance values ​​deviate from the average acoustic impedance values ​​of adjacent voxels by more than a preset deviation multiple is marked as an abnormal impedance region.

4. The quantitative detection method for internal defects in nuclear power pipeline forgings according to claim 3, characterized in that, For the candidate defect region, defect orientation parameters are extracted based on the multi-angle scattering characteristics. Combined with the scanning geometry of the virtual acoustic beam focusing scanning grid, the spatial azimuth and equivalent size of the defect are calculated to generate quantitative defect detection results, including: For each candidate defect region, the directional component with the largest amplitude in the main scattering direction encoding is extracted, and the tilt angle and azimuth angle of this directional component in three-dimensional space are used as the initial value of defect orientation. Based on the incident and receiving directions of the focused sound beam in the virtual sound beam focusing scanning grid, a geometric mapping relationship table between defect reflection response and defect orientation is established. The mapping entry closest to the initial value of the defect orientation is found in the geometric mapping table. The initial value of the defect orientation is then corrected according to the response intensity weighting value corresponding to the mapping entry to obtain the spatial azimuth of the defect. Along the direction of the spatial azimuth angle of the defect, the boundary distance of the region where the acoustic impedance value drops continuously within the defect candidate area is statistically analyzed, and the boundary distance is converted into the projection width of the defect in the direction perpendicular to the spatial azimuth angle to generate the equivalent size of the defect. The defect spatial azimuth angle and the defect equivalent size are combined to form the quantitative defect detection result.

5. The quantitative detection method for internal defects in nuclear power pipeline forgings according to claim 4, characterized in that, The boundary distance is the maximum span between the nearest neighbor boundary points whose acoustic impedance gradient modulus exceeds a preset gradient threshold along the azimuth direction of the defect space and its orthogonal direction.

6. The quantitative detection method for internal defects in nuclear power pipeline forgings according to claim 1, characterized in that, The abnormal impedance region is labeled with three-dimensional connected components, the surface mesh of each connected component is extracted, and the average curvature and Gaussian curvature of each vertex on the surface mesh are calculated, including: For each voxel unit in the abnormal impedance region, adjacent voxels are marked as the same connected domain according to the six-neighbor connectivity criterion until all voxels that meet the connectivity conditions are marked, and the voxel index set of each connected domain is output. An isosurface extraction algorithm is performed on the voxel index set of each connected component to generate a triangulated surface mesh for that connected component. The triangulated surface mesh contains a vertex coordinate sequence and a triangular facet topological relationship. For each vertex in the triangulated surface mesh, extract the set of first-order neighborhood triangles of that vertex, and calculate the average curvature and Gaussian curvature of that vertex based on the normal vectors of each triangle in the set of first-order neighborhood triangles.

7. The quantitative detection method for internal defects in nuclear power pipeline forgings according to claim 3, characterized in that, The process of discretizing the internal space of the forging into multiple voxel units, assigning an initial acoustic impedance estimate to each voxel unit, and generating an initial acoustic impedance distribution map includes: Based on the material grade of the forging to be tested, the standard acoustic impedance range corresponding to the material grade is retrieved from the material property database, and the median value of the standard acoustic impedance range is used as the initial reference acoustic impedance value of all voxel units inside the forging. At least three calibration points at different locations are selected on the surface of the forging to be tested. The measured acoustic impedance value of each calibration point is measured using the pulse echo method. The ratio of the measured acoustic impedance value to the initial value of the reference acoustic impedance is calculated to generate the correction coefficient for each calibration point. Spatial interpolation is performed on the correction coefficients of each calibration point to generate a correction coefficient distribution field covering the internal space of the forging; The correction coefficient distribution field is multiplied point by point with the initial value of the reference acoustic impedance to obtain the initial acoustic impedance estimate of each voxel unit. The initial acoustic impedance estimate is then used to fill all voxel units to generate the initial acoustic impedance distribution map.

8. The quantitative detection method for internal defects in nuclear power pipeline forgings according to claim 7, characterized in that, The spatial interpolation process employs the radial basis function interpolation method, using the spatial coordinates of the calibration point as interpolation nodes and the correction coefficient as node values.

Citation Information

Patent Citations

  • Multi-mode ultrasonic fusion pressure vessel welding seam defect nondestructive testing method and multi-mode ultrasonic fusion pressure vessel welding seam defect nondestructive testing system

    CN121577750A

  • Stainless steel pipeline welding seam internal defect detection system

    CN122150394A