Ultrasonic guided wave visual fusion small-diameter pipe defect detection and reconstruction method and system

Abstract

The application discloses a kind of ultrasonic guided wave visual fusion small pipe diameter pipeline defect detection and reconstruction method and system, it is related to nondestructive testing, pipeline defect detection, ultrasonic guided wave detection, visual detection and three-dimensional reconstruction technical field, including obtaining guided wave detection data and visual image data, establishing cylindrical coordinate reference and completing mapping, based on cylindrical coordinate reference generation pipeline inner surface model and obtain visual defect characterization information, according to the difference of guided wave detection data and reference guided wave response identifies guided wave abnormal support area.The method described in the application achieves better results in terms of small pipe diameter pipeline defect positioning accuracy, defect depth quantification ability, guided wave and visual cross-modal fusion effect, and pipeline inner surface three-dimensional defect reconstruction quality.

Description

Technical Field

[0001] This invention relates to the fields of non-destructive testing, pipeline defect detection, ultrasonic guided wave testing, visual inspection and three-dimensional reconstruction, specifically to a method and system for detecting and reconstructing defects in small-diameter pipelines using ultrasonic guided wave and visual fusion. Background Technology

[0002] Small-diameter pipes are widely used in petrochemical, urban gas, water supply and drainage, pharmaceutical, food processing and heat exchange equipment and other fields. Due to their narrow internal space and complex structure, they are susceptible to corrosion, erosion, deposition, local deformation and stress concentration during long-term service, resulting in defects such as wall thinning, corrosion pits and cracks. If these defects are not detected and quantitatively assessed in a timely manner, they may lead to safety risks such as leakage, blockage or bursting.

[0003] Existing visual inspection methods can intuitively obtain images of the inner surface of pipes and display the appearance and boundaries of defects well. However, they are easily affected by uneven lighting inside the pipe, reflection, occlusion, liquid accumulation, dirt and image blurring, making it difficult to accurately reflect the depth of defects and wall thickness loss.

[0004] Ultrasonic guided wave testing can propagate along the pipe wall and respond to anomalies such as corrosion, cracks, and thinning. It is suitable for detecting pipe wall anomalies and performing axial coarse positioning. However, its spatial resolution is limited by the guided wave wavelength, excitation pulse, propagation mode, and receiver channel arrangement, making it difficult to directly correspond to the fine defect boundaries and three-dimensional morphology of the inner surface of the pipe.

[0005] Existing solutions that combine visual inspection and ultrasonic inspection mostly remain at the stage of displaying results side by side or manual comparison, lacking a unified spatial coordinate benchmark. It is difficult to accurately map the abnormal response of guided waves to the visual defect boundary, and the visual inspection results are also difficult to effectively constrain the abnormal area of ​​guided waves.

[0006] Meanwhile, existing 3D reconstruction methods rely heavily on visual geometric information and lack quantitative support for defect depth based on guided wave anomaly response, which can easily lead to inconsistencies between the reconstructed morphology and the actual wall thickness loss. Summary of the Invention

[0007] In view of the above-mentioned problems, the present invention is proposed.

[0008] Therefore, the technical problem solved by this invention is as follows: In existing methods for detecting defects in small-diameter pipes, visual inspection methods are easily affected by factors such as uneven lighting inside the pipe, reflection, occlusion, liquid accumulation, and image blurring, resulting in difficulties in quantifying defect depth and accurately characterizing wall thickness loss; although ultrasonic guided wave detection methods can detect pipe wall anomalies, their axial and circumferential positioning resolution is relatively limited, resulting in coarse abnormal response areas and difficulty in directly corresponding to the defect boundaries and three-dimensional morphology of the inner surface of the pipe; existing multi-source detection methods also suffer from inconsistent spatial references between guided wave data and visual image data, low degree of cross-modal information fusion, difficulty in effectively redistributing guided wave abnormal responses based on visual defect boundaries, and the problem of how to determine defect depth and generate a three-dimensional defect model of the inner surface of the pipe based on the fused defect support information.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an ultrasonic guided wave and visual fusion method for defect detection and reconstruction of small-diameter pipes, including acquiring guided wave detection data and visual image data, establishing a cylindrical coordinate reference and completing the mapping, generating a pipe inner surface model based on the cylindrical coordinate reference and acquiring visual defect characterization information, and identifying abnormal guided wave support areas based on the difference between guided wave detection data and reference guided wave response.

[0010] Based on the abnormal support region and visual defect characterization information of the guided wave, the abnormal response of the guided wave is redistributed to form cross-modal defect support information. The defect depth is determined based on the cross-modal defect support information, and the inner surface model of the pipe is updated along the local normal to obtain the three-dimensional defect model of the inner surface of the pipe.

[0011] As a preferred embodiment of the ultrasonic guided wave vision fusion method for small-diameter pipe defect detection and reconstruction described in this invention, the step of establishing a cylindrical coordinate reference and completing the mapping includes: acquiring the axial scanning position, circumferential reference position and visual calibration parameters of the vision acquisition device, and determining the axial position and circumferential angle of the visual image pixels or visual three-dimensional points.

[0012] The axial candidate interval of the guided wave response is determined based on the propagation time and group velocity of the guided wave, and the circumferential candidate interval of the guided wave response is determined based on the circumferential arrangement of the guided wave receiving channel and the channel response distribution.

[0013] Visual image pixels or visual 3D points and waveguide responses are written into cylindrical meshes divided by axial position and circumferential angle, respectively.

[0014] As a preferred embodiment of the ultrasonic guided wave vision fusion method for small-diameter pipe defect detection and reconstruction described in this invention, the acquisition of visual defect characterization information includes:

[0015] Multi-frame geometric registration is performed on the visual image data mapped to the cylindrical coordinate reference to generate a model of the inner surface of the pipe.

[0016] Defect feature lines are extracted based on the surface geometric changes in the pipe inner surface model and the grayscale, texture or color changes in the visual image. The connectivity of the defect feature lines is then processed to obtain the visual defect boundary.

[0017] Boundary attribution information is generated based on the position and distance of cylindrical mesh cells relative to the visual defect boundary, and visually observable states are generated based on local brightness saturation, image sharpness, or multi-frame matching residuals.

[0018] Visual defect boundaries, boundary attribution information, surface geometric residuals, and visually observable states are used as visual defect characterization information.

[0019] As a preferred embodiment of the ultrasonic guided wave visual fusion method for small-diameter pipe defect detection and reconstruction described in this invention, the method for identifying abnormal support regions of guided waves includes: acquiring the guided wave reference response of a defect-free calibrated pipe section or a stable region of the pipe section to be inspected, and performing time zero-point correction, amplitude correction, and channel consistency correction on the guided wave detection data and the guided wave reference response.

[0020] An abnormal guided wave response is generated based on at least one of the amplitude difference, energy difference, phase difference, propagation time difference, or time-frequency characteristic difference between the corrected guided wave detection data and the guided wave reference response.

[0021] The guided wave anomaly response that meets the anomaly conditions determined by the calibrated pipe section anomaly samples is mapped to a cylindrical mesh, and the connected anomaly cylindrical mesh cells are identified as guided wave anomaly support regions.

[0022] As a preferred embodiment of the ultrasonic guided wave visual fusion method for small-diameter pipe defect detection and reconstruction described in this invention, the redistribution of the guided wave abnormal response includes: determining the overlapping area between the guided wave abnormal support area and the visual defect boundary, and determining the unit size of the local cylindrical mesh based on the magnitude and spatial variation of the guided wave abnormal response within the overlapping area.

[0023] Based on the boundary assignment information, surface geometric residuals, and visually observable state of the local cylindrical mesh elements, the allocation coefficients of the local cylindrical mesh elements for the guided wave anomaly response are determined.

[0024] The distribution coefficient is used to normalize the spatial redistribution of the guided wave anomaly response in the overlapping region, so that the total amount of guided wave anomaly response in the same guided wave anomaly support region remains consistent before and after redistribution.

[0025] The redistributed guided wave anomaly response is written into the corresponding local cylindrical mesh element to obtain the redistributed guided wave anomaly response distribution.

[0026] As a preferred embodiment of the ultrasonic guided wave visual fusion method for small-diameter pipe defect detection and reconstruction described in this invention, the method for forming cross-modal defect support information includes: normalizing the guided wave abnormal response, surface geometric residual, and distance to the visual defect boundary after redistribution of each local cylindrical grid cell.

[0027] Based on the spatial correspondence between the normalized guided wave anomaly response and the visual defect boundary attribution information, each local cylindrical mesh unit is divided into evidence consistency unit, guided wave evidence priority unit, or visual evidence priority unit.

[0028] For evidence-consistent units, correlated normalized guided wave anomaly response, surface geometric residual, boundary distance, and visually observable state.

[0029] For guided wave evidence priority units, the normalized guided wave abnormal response is retained and written into the local resampling identifier.

[0030] For the visual evidence priority unit, the surface geometric residual and visual defect boundary attribution information are retained and written into the waveguide verification identifier.

[0031] The unit identifiers and corresponding data of the evidence consistency unit, the waveguide evidence priority unit, and the visual evidence priority unit are used as cross-modal defect support information.

[0032] As a preferred embodiment of the ultrasonic guided wave and visual fusion method for small-diameter pipe defect detection and reconstruction described in this invention, the step of determining the defect depth based on cross-modal defect support information includes: acquiring a calibrated pipe segment sample with a known defect depth or wall thickness loss, using the normalized guided wave abnormal response, surface geometric residual, boundary distance, and visually observable state corresponding to the calibrated pipe segment sample as input, and the known defect depth or wall thickness loss as output, to train a support vector regression model.

[0033] Acquire the resampled visual image of the region corresponding to the waveguide evidence priority cell with the local resampled identifier, and update the surface geometric residual, boundary distance, and visual observable state.

[0034] By inputting the evidence consistency unit and the updated waveguide evidence priority unit into the trained support vector regression model, the defect depth of the corresponding local cylindrical mesh unit is obtained.

[0035] Output the waveguide verification position based on the waveguide verification identifier, and update the defect depth of the visual evidence priority unit after obtaining the waveguide verification result.

[0036] As a preferred embodiment of the ultrasonic guided wave vision fusion method for small-diameter pipe defect detection and reconstruction described in this invention, the step of updating the pipe inner surface model along the local normal includes: determining the local normal of the local cylindrical mesh vertices of the pipe inner surface model pointing to the inside of the pipe wall.

[0037] Based on the defect depth of the local cylindrical mesh element, the mesh vertices are displaced along the local normal. Based on the distance from the corresponding local cylindrical mesh element to the visual defect boundary, continuous transition constraints are applied to the displacement of the mesh vertices on both sides of the visual defect boundary.

[0038] Apply gradient constraints to the defect depth gradient of adjacent mesh cells that match the size of the local cylindrical mesh cells.

[0039] While maintaining the mesh topology connections of the pipe's inner surface model, the three-dimensional defect model of the pipe's inner surface is obtained by using the constrained mesh vertex positions.

[0040] Another objective of this invention is to provide an ultrasonic guided wave and visual fusion system for defect detection and reconstruction in small-diameter pipes. This system can coordinate data acquisition and mapping, defect characterization generation, cross-modal fusion, and depth determination and 3D reconstruction to map guided wave detection data and visual image data to the same cylindrical coordinate reference. It can also redistribute the abnormal response of guided waves based on visual defect characterization information. This solves the problems in current small-diameter pipe detection technology, such as the independence of guided wave detection results and visual detection results, unclear spatial correspondence of defects, difficulty in quantifying defect depth, and lack of cross-modal support for 3D reconstruction results.

[0041] As a preferred embodiment of the ultrasonic guided wave vision fusion small-diameter pipe defect detection and reconstruction system of the present invention, it includes: a data acquisition and mapping module, a defect characterization generation module, a cross-modal fusion module, and a depth determination and three-dimensional reconstruction module.

[0042] The data acquisition and mapping module is used to acquire guided wave detection data and visual image data, establish cylindrical coordinate references, and complete the mapping.

[0043] The defect characterization generation module is used to generate a model of the inner surface of the pipe based on the cylindrical coordinate reference and obtain visual defect characterization information, as well as to identify abnormal support areas of the guided wave based on the difference between the guided wave detection data and the reference guided wave response.

[0044] The cross-modal fusion module is used to redistribute the guided wave anomaly response based on the guided wave anomaly support region and visual defect characterization information to form cross-modal defect support information.

[0045] The depth determination and 3D reconstruction module is used to determine the defect depth based on cross-modal defect support information, update the inner surface model of the pipe along the local normal, and obtain a 3D defect model of the inner surface of the pipe.

[0046] As a preferred embodiment of the ultrasonic guided wave vision fusion small-diameter pipe defect detection and reconstruction system described in this invention, the data acquisition and mapping module includes: an ultrasonic guided wave detection device, a vision acquisition device capable of moving along the inside of the pipe, and a position acquisition device.

[0047] The position acquisition device is communicatively connected to both the ultrasonic guided wave detection device and the visual acquisition device, so that the guided wave detection data and visual image data are associated with the same cylindrical coordinate reference.

[0048] The cross-modal fusion module is also used to generate evidence consistency status, local resampling identifiers, or guided wave verification identifiers for local cylindrical mesh elements.

[0049] The visual acquisition device includes at least one of the following: a motion control terminal, an observation pose control terminal, an illumination control terminal, or a scanning step size control terminal.

[0050] The cross-modal fusion module is connected to the corresponding control terminal, converts the local resampling identifier into a resampling instruction including the target axial interval, the target circumferential interval and the scanning resolution, and writes the visual image data obtained from the resampling back to the cross-modal defect support information of the corresponding local cylindrical mesh unit.

[0051] The beneficial effects of this invention are as follows: The ultrasonic guided wave and visual fusion method for defect detection and reconstruction in small-diameter pipes provided by this invention establishes a cylindrical coordinate reference and maps guided wave detection data and visual image data to the same cylindrical grid, achieving a unified expression of guided wave anomaly response and visual defect location, thus improving the consistency of spatial positioning of defects in small-diameter pipes. By extracting visual defect boundaries, boundary attribution information, surface geometric residuals, and visually observable states, the visual detection results can not only represent the defect appearance boundary but also provide a reliable basis for the redistribution of guided wave anomaly response, improving the accuracy of defect boundary identification and the utilization of visual evidence. Furthermore, by considering the guided wave anomaly support area... The invention redistributes guided wave anomaly responses using domain and visual defect characterization information, ensuring that the anomalous energy obtained from guided wave detection falls more reasonably within or near the visual defect boundary, thus improving the correspondence accuracy between the guided wave anomaly response and the actual defect area. By forming cross-modal defect support information and determining the defect depth, the defect depth calculation is simultaneously constrained by guided wave anomalies, visual geometric residuals, boundary distances, and visually observable states, improving the reliability of quantitative defect depth results. By updating the pipe inner surface model along the local normal and applying boundary continuity transition constraints and gradient constraints, the generated three-dimensional defect model of the pipe inner surface more closely matches the actual defect morphology. This invention achieves significantly better results in terms of defect location accuracy in small-diameter pipes, quantitative defect depth capability, cross-modal fusion effect of guided wave and vision, and reconstruction quality of three-dimensional defects on the pipe inner surface. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] It should be noted that the accompanying drawings of this invention are mainly used to illustrate the technical relationships and data flow between various detection data, processing objects, judgment processes and processing results.

[0054] To maintain the simplicity and clarity of the accompanying drawings and to highlight the main technical processes, some of the textual descriptions in the drawings have adopted abbreviations of terms, generalizations of relationships, simplified descriptions of conditions, or simplified expressions of calculation processes without changing their technical meaning. The complete calculation processes, parameter conditions, judgment thresholds, and limiting relationships described in each specific implementation method are not listed one by one in the accompanying drawings.

[0055] All textual descriptions in the accompanying drawings correspond to the respective processing objects and technical processes in the specific embodiments. The complete meanings, calculation methods, judgment conditions, and parameter limitations of the relevant terms shall be based on the descriptions in the specific embodiments.

[0056] Figure 1 is a schematic diagram of the principle of multimodal data acquisition and cylindrical coordinate grid mapping in a method for defect detection and reconstruction of small-diameter pipes using ultrasonic guided wave vision fusion.

[0057] Figure 2 is a schematic diagram of the spatial redistribution, local mesh refinement and feature evolution process of the guided wave anomaly response in this invention;

[0058] Figure 3 is a decision logic tree diagram of the cross-modal defect evidence state division and closed-loop feedback mechanism in this invention;

[0059] Figure 4 is a schematic diagram of the physical constraint profile for quantifying defect depth and reconstructing three-dimensional surfaces based on support vector regression (SVR) in this invention.

[0060] Figure 5 shows the hardware and software architecture and feedback data flow diagram of a small-diameter pipe defect detection and 3D reconstruction system based on the fusion of ultrasonic guided waves and visual images. Detailed Implementation

[0061] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0062] Example 1

[0063] Reference Figure 1 This invention provides a specific implementation method for the acquisition of guided wave detection data and visual image data, the establishment and mapping of cylindrical coordinate references, the acquisition of visual defect characterization information, and the identification of abnormal support areas of guided waves in a method for defect detection and reconstruction of small-diameter pipes using ultrasonic guided wave and visual fusion.

[0064] First, guided wave detection data is collected by multiple receiving channels arranged along the circumference of the pipeline, with each receiving channel corresponding to a known circumferential installation angle.

[0065] Specifically: Each waveguide receiving channel has a preset or calibrated circumferential arrangement position.

[0066] The guided wave detection data includes: excitation time, time sequence signals of each receiving channel, and guided wave sampling frequency.

[0067] Visual image data is acquired by moving along the inside of the pipe.

[0068] The visual image data includes: continuous intra-frame surface images, axial position at the time of acquisition, circumferential reference position, and visual calibration parameters.

[0069] Visual calibration parameters include: pixel scale, image center position, and lens distortion correction parameters.

[0070] Furthermore, the axial zero point is taken as the pipe inlet end face, weld end face, or pre-set axial mark, and the radial position is determined by the distance from the pipe center axis to the pipe center axis based on the visual three-dimensional point or the pipe inner surface model point. The circumferential zero point is taken as the circumferential direction corresponding to the first receiving channel in the guided wave detection data, the circumferential mark on the inner wall of the pipe, or the initial observation direction in the visual acquisition process.

[0071] Cylindrical coordinate references include: axial position, circumferential angle, and radial position.

[0072] Axial position indicates the location of the defect along the length of the pipe, circumferential angle indicates which circumferential direction the defect is located on the inner wall of the pipe, and radial position indicates the spatial shape of the inner surface model of the pipe.

[0073] Furthermore, the inner surface of the pipe to be inspected is divided into cylindrical grids.

[0074] Specifically: the axial grid length is set to 2mm-10mm, and the circumferential grid angle is set to 1-5 degrees.

[0075] In typical small-diameter pipe inspection scenarios, the axial grid length is set to 5mm and the circumferential grid angle is set to 2 degrees.

[0076] Each cylindrical mesh element corresponds to an axial interval and a circumferential angle interval. Subsequently, the guided wave response, visual image features, surface geometric residuals, and defect states are all written into the corresponding cylindrical mesh element.

[0077] For each frame of visual image, distortion correction is first performed according to the visual calibration parameters. Then, based on the axial acquisition position, circumferential reference position, and pixel scale of the frame, the pixels in the image are mapped to the corresponding axial position and circumferential angle.

[0078] If the visual acquisition process can obtain a three-dimensional point cloud, then the axial position, circumferential angle and radial distance of the three-dimensional point are determined based on its spatial position relative to the central axis of the pipe.

[0079] If multiple frames of visual data fall within the same cylindrical grid cell, the data with a Laplacian sharpness value that meets the preset sharpness benchmark, an overexposed and underexposed pixel ratio that does not exceed the preset exposure anomaly ratio, and a registration error with adjacent frames that does not exceed 1mm will be selected first.

[0080] If the difference in Laplacian sharpness value of multiple frames does not exceed 10%, the difference in the proportion of overexposed and underexposed pixels does not exceed 10%, and the difference in registration error of multiple frames does not exceed 0.2mm, the quality of multiple frames is considered to be similar. The median value of grayscale, color, or three-dimensional point radius within the grid cell is taken as the representative value to reduce the influence of local noise.

[0081] For a cylindrical mesh element with valid three-dimensional points, the median distance from the valid three-dimensional points within the mesh element to the central axis of the pipe is taken as the inner surface radius of the mesh element.

[0082] If the number of valid visual points in a certain grid cell is less than 3, the visual 3D information of the grid cell is considered insufficient. Instead of directly using its 3D points to generate a surface, interpolation is performed based on adjacent valid grid cells to complete the data, and the visual observability of the grid cell is reduced.

[0083] The initial model of the pipe's inner surface is formed by the axial position, circumferential angle, and inner surface radius of each cylindrical mesh element.

[0084] For each cylindrical mesh element, calculate the difference between its inner surface radius and the smooth radius of its local neighborhood, and use the difference as the surface geometric residual.

[0085] The local neighborhood is selected by taking 5×5 grid cells around the current grid cell and using the median method to obtain the smoothing radius.

[0086] If the current grid cell shows a significant indentation into the pipe wall relative to the neighborhood smooth radius, the surface geometric residual increases.

[0087] If there is no obvious depression, the surface geometric residual is close to 0.

[0088] Visual defect boundaries are extracted based on grayscale abrupt changes, texture abrupt changes, and surface geometric residuals in visual images.

[0089] Specifically, the local gray-level gradient is calculated for the mapped cylindrical image. When the gray-level gradient of a grid cell is higher than the average gradient of the effective region plus 1.5 times the standard deviation, the grid cell is considered to have an image anomaly.

[0090] When the surface geometric residual of a grid cell is greater than 0.1 mm and is more than three times the standard deviation of the geometric residual of the effective region, the grid cell is considered to have a geometric anomaly.

[0091] Mesh cells that satisfy image anomalies or geometric anomalies are considered as candidate cells for visual defects.

[0092] Connectivity processing is performed on candidate units of visual defects, and isolated noise regions with fewer than 3 grid units are deleted. The outer contour of the remaining connected regions is used as the boundary of the visual defects.

[0093] If a grid cell is located inside the boundary of a visual defect, its boundary attribution information is recorded as strong attribution.

[0094] If a grid cell is located outside the visual defect boundary but is no more than two axial grid lengths away from the boundary, its boundary attribution information is recorded as weak attribution.

[0095] If a grid cell is farther from the boundary of a visual defect, its boundary attribution information is recorded as non-attributed.

[0096] The visual observable state is used to indicate whether the visual data of a cylindrical grid cell is reliable, and its value ranges from 0 to 1.

[0097] The closer the value is to 1, the clearer the visual image, the normal exposure, and the stable multi-frame registration.

[0098] The closer the value is to 0, the more likely it is to be overexposed, underexposed, blurry, obscured, or registration failed.

[0099] A preferred approach to calculating visually observable states is as follows:

[0100]

[0101] in, Indicates the visually observable state. This indicates the proportion of overexposed and underexposed pixels within the grid cell. This represents the image sharpness coefficient. It is set to 1 when the Laplacian sharpness of the image reaches a preset sharpness benchmark, and decreases proportionally when it falls below the benchmark. This represents the multi-frame registration stability coefficient. It is set to 1 when the multi-frame matching error does not exceed 1mm, and decreases as the error increases when it exceeds 1mm.

[0102] If the calculation result is less than 0, it is treated as 0; if it is greater than 1, it is treated as 1.

[0103] Visual defect characterization information includes: visual defect boundary, boundary attribution information, surface geometric residuals, and visually observable state.

[0104] It should be noted that the visual defect boundary is the outline of the suspected defect on the inner wall of the pipe, determined by the outer contour of the connected region of candidate units of image anomalies or geometric anomalies.

[0105] Boundary attribution information is used to characterize whether cylindrical mesh cells are located inside, near, or outside the visual defect boundary.

[0106] Surface geometric residuals are used to characterize the degree of local concavity of the mesh cell relative to the neighborhood smooth radius.

[0107] Visually observable states are used to characterize whether the grid cell is overexposed, underexposed, blurred, occluded, or has failed to register.

[0108] For each guided wave receiving channel, first obtain the reference guided wave response of the defect-free calibration pipe section, or extract the reference guided wave response from a stable region in the pipe section to be inspected that is visually confirmed to be free of obvious defects.

[0109] Then, time zero-point correction, amplitude correction, and channel consistency correction are performed on the guided wave detection data.

[0110] Zero-time correction is performed by comparing the peak positions of the detected signal and the reference signal within the direct wave window.

[0111] Amplitude correction is achieved by comparing the root mean square amplitudes of the detected signal and the reference signal within the direct wave window.

[0112] Channel consistency correction is used to eliminate amplitude deviations caused by differences in installation coupling between different receiving channels.

[0113] The corrected guided wave detection data is compared with the reference guided wave response.

[0114] The differences include: amplitude difference, energy difference, and arrival time difference.

[0115] If the difference response within a certain time window is higher than the average difference response of the reference stable region plus 3 times the standard deviation, then the guided wave anomaly response is considered to exist in that time window.

[0116] The duration of the abnormal time window should be no less than half the duration of the guided wave excitation pulse; abnormal peak values ​​shorter than this duration should be discarded as noise.

[0117] For pulse-echo guided wave detection, the axial candidate position is determined based on the arrival time of the abnormal response peak, the excitation time, and the guided wave group velocity.

[0118] Specifically, the difference between the peak arrival time and the excitation time is multiplied by the waveguide group velocity and converted into axial candidate positions according to the round-trip propagation path.

[0119] The width of the axial candidate interval is not less than one axial grid length and not less than the axial resolution range corresponding to the duration of the guided wave pulse.

[0120] The circumferential candidate interval is determined based on the abnormal energy distribution of the receiving channel.

[0121] Specifically, the percentage of abnormal energy in each receiving channel within the abnormal time window is calculated, and the channel with a higher percentage of abnormal energy contributes more to the circumferential position.

[0122] The circumferential center of the waveguide anomaly is determined based on the installation circumferential angle and the proportion of abnormal energy in each channel, and the circumferential candidate width is determined based on the degree of energy distribution dispersion.

[0123] If the number of consecutive adjacent channels with the highest percentage of abnormal energy does not exceed 2, and the cumulative percentage of abnormal energy in these consecutive adjacent channels is not less than 70%, then the range covering these consecutive adjacent channels with a circumferential angle installed and extended outward by one circumferential grid angle to each side will be used as the circumferential candidate interval.

[0124] If there are more than two channels with an abnormal energy percentage of not less than 10%, or if the circumferential standard deviation of the abnormal energy distribution is greater than the circumferential angle interval between adjacent channels, then the range covering the above channels with the circumferential angle installed and extending outward by at least one circumferential grid angle to each side will be taken as the circumferential candidate interval.

[0125] Cylindrical mesh elements falling into the axial and circumferential candidate intervals are written into the guided wave anomaly response values.

[0126] If the same grid cell corresponds to multiple anomalous responses, then retain the anomalous response value with the largest amplitude or energy value, or retain the cumulative energy value of multiple anomalous responses.

[0127] Connectivity processing is performed on the grid cells containing the abnormal response values ​​of the guided wave.

[0128] The connection processing specifically involves:

[0129] In the cylindrical unfolded mesh, axially adjacent, circumferentially adjacent, or diagonally adjacent mesh cells that are all written with guided wave anomaly response values ​​are divided into the same connected region. The abnormal mesh cells located in the same connected region constitute the guided wave anomaly support region.

[0130] If the connected region has fewer than 2 grid cells and the abnormal response does not exceed 1.5 times the threshold, it is discarded as isolated noise.

[0131] If the axial spacing between two abnormal support regions is less than half of the axial resolution range of the waveguide, and the circumferential candidate regions overlap, they are merged into the same abnormal support region of the waveguide.

[0132] Finally, visual defect characterization information and waveguide anomaly support area are output. The visual defect characterization information is used to represent the defect boundary, geometric indentation degree and visual data reliability found on the visual side.

[0133] The waveguide anomaly support region is used to indicate the spatial range and intensity of the anomaly response found on the waveguide side.

[0134] Example 2

[0135] Reference Figure 2 and Figure 3 This invention provides a specific implementation method for detecting and reconstructing defects in small-diameter pipes using ultrasonic guided wave and visual fusion, in which the guided wave abnormal response is redistributed based on the guided wave abnormal support region and visual defect characterization information to form cross-modal defect support information. Firstly, as... Figure 2 As shown in sub-figure (1), the information on the abnormal support area and visual defect characterization of the guided wave is read.

[0136] The information on the waveguide anomaly support region includes: the axial candidate interval, the circumferential candidate interval, the set of cylindrical mesh elements corresponding to the anomaly region, and the waveguide anomaly response intensity within each cylindrical mesh element.

[0137] Visual defect characterization information includes: visual defect boundary, boundary attribution information, surface geometric residuals, and visually observable state.

[0138] The information on the waveguide anomaly support area and the visual defect characterization is all located under the same cylindrical coordinate reference, and spatial overlap can be directly judged.

[0139] Furthermore, the overlap between the waveguide anomaly support region and the visual defect boundary was determined.

[0140] For each waveguide anomalous support region, determine whether there are cylindrical mesh elements within or near the visual defect boundary in that region.

[0141] If it exists, it is considered that the waveguide anomaly support region has a spatial correspondence with the visual defect characterization information, and enters the waveguide anomaly response redistribution process.

[0142] If it does not exist, it means that an anomaly was found on the waveguide side but no effective defect boundary was formed on the visual side. This area is temporarily designated as the priority area for waveguide evidence and is marked with a local resampling identifier.

[0143] Furthermore, such as Figure 2 As shown in subgraph (2), the need for local mesh refinement is determined based on the degree of spatial variation of the guided wave abnormal response.

[0144] Specifically, in the region where the waveguide anomaly support area overlaps with the visual defect boundary, the dispersion of the waveguide anomaly response intensity of each grid cell is statistically analyzed.

[0145] If the standard deviation of the abnormal response intensity does not exceed 30% of its average value, it indicates that the guided wave response in this region is relatively smooth, and the original cylindrical mesh size should be maintained.

[0146] If the standard deviation of the abnormal response intensity exceeds 30% of its average value, it indicates that the guided wave response space in this region has changed significantly, and the original cylindrical mesh unit is divided into smaller local cylindrical mesh units.

[0147] Preferably, each original grid cell is divided into 2×2 local cylindrical grid cells.

[0148] Increase the redistribution resolution in regions where the guided wave response changes significantly, and avoid introducing too many invalid meshes in regions with gentle response.

[0149] Determine the necessary data for each local cylindrical mesh cell to participate in the redistribution.

[0150] Each local cylindrical mesh element acquires at least four types of data:

[0151] First, the intensity of the guided wave anomaly response at the location; second, the boundary attribution information at the location; third, the surface geometric residual at the location; and fourth, the visually observable state at the location.

[0152] If a local cylindrical mesh cell is obtained by subdividing the original mesh, and there is no independent visual data at the local location, then it inherits the visual representation information of the original mesh cell to which it belongs.

[0153] If the original grid cell also lacks visual information, the visual observable state of the local cylindrical grid cell is set to 0, and the weight of visual evidence is reduced in subsequent evidence state division.

[0154] Furthermore, the allocation coefficients of each local cylindrical mesh element are determined.

[0155] The allocation coefficient is used to represent the proportion of the local cylindrical mesh element that obtains the guided wave anomaly response from the guided wave anomaly support region.

[0156] The assignment coefficients are determined by boundary attribution information, surface geometric residuals, and visually observable states.

[0157] Specifically, local cylindrical mesh cells located inside the visual defect boundary preferentially receive higher allocation coefficients.

[0158] Local cylindrical mesh cells located near the visual defect boundary receive a moderate allocation coefficient, while local cylindrical mesh cells far from the visual defect boundary receive a lower allocation coefficient.

[0159] Under the same boundary assignment conditions, the larger the surface geometric residual, the higher the assignment coefficient; the worse the visually observable state, the lower the assignment coefficient.

[0160] Specifically, the allocation coefficients will be implemented according to the following rules:

[0161] Each local cylindrical mesh cell is first set with a base assignment value of 0.001 to prevent all weights from being 0.

[0162] If the cell is located inside the visual defect boundary, the boundary contribution is 1; if it is located outside the boundary but at a distance of no more than two axial grid lengths, the boundary contribution decreases with increasing distance, ranging from 0 to 1; if the distance is even greater, the boundary contribution is 0.

[0163] The surface geometric residuals are normalized to the maximum residual within the current waveguide anomalous support region. The closer the normalization result is to 1, the more obvious the depression.

[0164] In the allocation coefficients, the boundary contribution is the main constraint, while the surface geometric residual and the visually observable state are the adjustment terms, with the contribution of the surface geometric residual being higher than that of the visually observable state.

[0165] Furthermore, such as Figure 2 As shown in subgraph (3), the normalized spatial redistribution of the guided wave anomaly response is performed.

[0166] For the same guided wave anomaly support region, first calculate the total guided wave anomaly response in the region before redistribution, and then redistribute it according to the distribution coefficient of each local cylindrical mesh element.

[0167] A preferred method for calculating the guided wave anomaly response after local cylindrical mesh element redistribution is as follows:

[0168]

[0169] in, Indicates the first The guided wave anomaly response after redistribution of local cylindrical mesh elements This represents the total amount of guided wave anomaly response before redistribution within the same guided wave anomaly support region. Indicates the first The distribution coefficient of each local cylindrical mesh element. This indicates the number of local cylindrical mesh elements participating in the redistribution within the waveguide anomaly support region.

[0170] If the waveguide anomalous support area overlaps with multiple visual defect boundaries, the area is assigned according to the distance between the local cylindrical mesh element and each visual defect boundary.

[0171] Prioritize assigning visual defect boundaries to the nearest and highest boundary assignment information. If two visual defect boundaries are close in distance and the local cylindrical mesh element is located between them, then the local cylindrical mesh element is marked as a composite defect transition element. During subsequent 3D reconstruction, the transition is maintained continuously, and it is not forcibly split into two discontinuous defects.

[0172] Furthermore, the guided wave anomaly response, surface geometric residuals, and boundary distances after redistribution are normalized.

[0173] Normalization does not introduce complex formulas and is performed according to the following rules:

[0174] The guided wave anomaly response is based on the maximum redistribution response within the current guided wave anomaly support region, with the maximum value corresponding to 1, and other elements scaled proportionally. The surface geometric residual is based on the maximum geometric residual within the current overlapping region, with the maximum value corresponding to 1. Boundary distance is converted to boundary proximity, with proximity at the visual defect boundary being 1, proximity exceeding two axial grid lengths from the boundary being 0, and the proximity decreasing linearly with distance in the intermediate region. The normalized result serves as the input feature for subsequent defect depth determination.

[0175] The evidence state of a local cylindrical grid cell is defined as follows: if the normalized guided wave anomaly response of a local cylindrical grid cell is not less than 0.2, the boundary attribution information is not less than 0.5, and the visual observable state is not less than 0.4, then the cell is classified as an evidence-consistent cell.

[0176] The evidence-consistent unit indicates that the guided wave anomaly and the visual defect support each other in space, and can directly enter the defect depth calculation.

[0177] If the normalized guided wave anomaly response of a local cylindrical mesh cell is not less than 0.2, but the boundary attribution information is less than 0.5, or the visual observable state is less than 0.4, then the cell is classified as a guided wave evidence priority cell and a local resampling identifier is written.

[0178] Local resampling identifiers include: target axial interval, target circumferential interval, and suggested scan resolution.

[0179] The target axial interval is the length of two local grids before and after the center of the element, and the target circumferential interval is the angle of two local grids before and after the center of the element.

[0180] During resampling, the axial scanning step size should not exceed half the axial length of the local mesh, and the circumferential sampling interval should not exceed half the circumferential angle of the local mesh.

[0181] If the boundary attribution information of a local cylindrical mesh cell is not less than 0.5, the normalized value of the surface geometric residual is not less than 0.3, but the normalized guided wave anomaly response is less than 0.2, the local cylindrical mesh cell is classified as a visual evidence priority cell and written with a guided wave verification identifier.

[0182] The waveguide verification criteria include: the axial position of the unit center, the circumferential angle of the center, and the recommended verification range.

[0183] The axial position of the unit center is the midpoint of the corresponding axial interval of the local cylindrical mesh unit, and the circumferential angle of the center is the midpoint of the corresponding circumferential angle interval of the local cylindrical mesh unit.

[0184] The recommended scope of review is the area formed by extending two local mesh lengths forward and backward along the axial direction and two local mesh angles forward and backward along the circumferential direction, with the axial position and circumferential angle of the element center as the center.

[0185] The guided wave verification marker is used to prompt the subsequent re-acquisition or re-analysis of the guided wave response at this location to determine whether the visual anomaly corresponds to the actual wall thickness loss.

[0186] Finally, cross-modal defect support information is generated.

[0187] The cross-modal defect support information for each local cylindrical mesh element includes: the element's axial position, circumferential position, normalized guided wave anomaly response, normalized surface geometric residual, boundary proximity, visually observable state, evidence state, and resampling or verification identifier.

[0188] Example 3

[0189] Reference Figure 4 This is an embodiment of the present invention, which provides a specific implementation method for detecting and reconstructing defects in small-diameter pipes using ultrasonic guided wave vision fusion, wherein the defect depth is determined based on cross-modal defect support information, and the pipe inner surface model is updated along the local normal to obtain a three-dimensional defect model of the pipe inner surface.

[0190] First, establish calibration samples for defect depth. The calibration samples are derived from calibration pipe sections with known defect depth or known wall thickness loss.

[0191] The calibration samples include: guided wave detection data, visual image data, cylindrical grid position information, and labels for defect depth or wall thickness loss corresponding to the calibration pipe section.

[0192] Calibration pipe sections can be made by machining pits, thinning zones, or localized corrosion of different depths to simulate defects, or real defective pipe sections with completed wall thickness measurements can be used.

[0193] Each calibration sample is processed according to the steps of cylindrical coordinate mapping, visual defect characterization information acquisition, guided wave abnormal support region identification, guided wave abnormal response redistribution, and normalization.

[0194] The corresponding normalized guided wave anomaly response, normalized surface geometric residual, boundary proximity and visually observable state are obtained, and the known defect depth or wall thickness loss is used as the output label.

[0195] Furthermore, a support vector regression model is trained. The input to the support vector regression model includes four features: normalized guided wave anomaly response, normalized surface geometric residual, boundary proximity, and visually observable state; the output is the defect depth. A radial basis function kernel is used during training. The kernel function parameter is preferably the reciprocal of the number of input features, i.e., 0.25 when the input has four features; the penalty coefficient is preferably 10; and the insensitive loss width is preferably 2% of the maximum defect depth of the calibrated sample.

[0196] The model training adopts a sequential minimum optimization method. Training stops when the objective function changes by less than 0.0001 in two consecutive iterations, or when the number of iterations reaches 500.

[0197] After training, save the feature normalization baseline, support vectors, kernel function parameters, penalty coefficients, and bias terms.

[0198] Handling situations where there are insufficient calibration samples. If the number of calibration samples is no less than 30 and covers at least three depth levels of defects—shallow, medium, and deep—then training or updating the support vector regression model is allowed.

[0199] If the number of calibration samples is less than 30, but models with the same material, pipe diameter range, and wall thickness range have been saved in advance, the saved models will be called to calculate the defect depth, and the results will be marked as inherited calibration models in the output.

[0200] If the number of calibrated samples is less than 30 and there is no saved model available for reference, the final quantitative defect depth will not be output. Only the defect location, evidence status, and calibration prompts will be output to avoid incorrect conclusions due to model extrapolation.

[0201] Local visual resampling is performed on the priority cells for waveguide evidence. For cells with local resampling identifiers, a resampling region is generated based on their central axial position and central circumferential angle.

[0202] The resampling region preferably covers the lengths of two local grids before and after the unit, as well as the angles of two local grids before and after the circumference.

[0203] During resampling, the axial scan step size should not exceed half the axial length of the local mesh, and the circumferential sampling interval should not exceed half the circumferential angle of the local mesh.

[0204] The resampled image is remapped to the cylindrical coordinate reference, and the surface geometric residuals, boundary proximity, and visual observability are recalculated.

[0205] If the visually observable state after resampling is not lower than 0.3, then the updated visual features are used to participate in the defect depth calculation.

[0206] If the visually observable state is still below 0.3 after resampling, it indicates that there is continuous occlusion, severe reflection, liquid accumulation, or blurred imaging in the area. In this case, the abnormal waveguide response is retained, the visual side feature is marked as low confidence, and the defect depth output result is accompanied by a low visual confidence label.

[0207] Furthermore, the defect depth is calculated for the evidence consistency unit and the guided wave evidence priority unit after resampling.

[0208] For the evidence-consistent unit, the normalized guided wave anomaly response, normalized surface geometric residual, boundary proximity, and visually observable state are directly input into the trained support vector regression model to obtain the defect depth. For the resampled guided wave evidence-priority unit, the updated surface geometric residual, boundary proximity, and visually observable state are used to replace the original visual features before being input into the model to obtain the defect depth.

[0209] A boundary constraint is applied to the model output. If the defect depth in the model output is less than 0, it is corrected to 0; if the defect depth in the model output is greater than 90% of the pipe wall thickness, it is corrected to 90% of the pipe wall thickness, and the local cylindrical mesh element is marked as a deep defect core element. This boundary constraint is used to prevent the model from outputting depth values ​​exceeding the physically reasonable range under conditions of insufficient sample coverage or abnormal input.

[0210] For visual evidence priority cells, the axial position and circumferential angle of the cell's center are read, and a suggested verification range is generated based on the axial length and circumferential angle of the local cylindrical mesh where the cell is located. The axial position, circumferential angle, and suggested verification range are output as the waveguide verification position.

[0211] It is recommended that the scope of the review be preferably the area covered by two local grid lengths before and after the unit center, and two local grid angles before and after the circumference.

[0212] If the guided wave verification has been completed before the 3D model is output, the verified guided wave anomaly response is rewritten into the unit, and the cross-modal defect support information is regenerated and then input into the support vector regression model to calculate the defect depth.

[0213] If the waveguide verification has not yet been completed, the unit will only generate a temporary topography depth based on the surface geometric residuals, and will be marked with a waveguide verification mark.

[0214] The temporary morphology depth is only used to display the morphology of suspected defects and is not used as a final quantitative conclusion on wall thickness loss.

[0215] The local normal is determined based on the spatial orientation of the adjacent faces of the mesh vertices in the inner surface model of the pipe.

[0216] The inner surface model of the pipe includes: mesh vertices and mesh patches.

[0217] For each mesh vertex, calculate the face normals of each mesh patch adjacent to the mesh vertex, and average the face normals to obtain the average normal of the mesh vertex.

[0218] The average normal is uniformly adjusted to point inwards towards the inside of the pipe wall, and the adjusted average normal is used as the local normal of the grid vertex.

[0219] For an ideal circular tube, the local normal is basically radially directed towards the tube wall.

[0220] For pipes with bends or local deformations, the local normal is adaptively determined based on the orientation of the local mesh patch.

[0221] The defect depth of the local cylindrical mesh element is assigned to the corresponding mesh vertex.

[0222] If a grid vertex is adjacent to multiple local cylindrical grid cells, the depth contribution is determined based on the distance between the vertex and the center of each cell; the closer the cells are, the greater the contribution.

[0223] If an adjacent cell has a low visual confidence marker or a marker to be verified by the waveguide, the cell's contribution to the vertex displacement is reduced to avoid model mutation caused by low confidence regions.

[0224] Apply continuous transition constraints to the displacement of mesh vertices on both sides of the visual defect boundary.

[0225] The mesh vertices located inside the visual defect boundary are displaced according to the calculated defect depth.

[0226] The displacement of a grid vertex located near the boundary of a visual defect but on the outside gradually decreases as the distance increases.

[0227] External vertices that are more than two axial grid lengths away from the visual defect boundary are not subject to defect depth displacement.

[0228] Apply gradient constraints to the defect depth between adjacent local cylindrical mesh elements.

[0229] Specifically, for every 1mm distance between two adjacent local cylindrical mesh elements, the allowable difference in defect depth shall not exceed 0.5mm. If the depth difference between two adjacent elements exceeds this limit, the larger depth value will be adjusted back to the allowable range of the adjacent elements.

[0230] The correction needs to be executed iteratively, for 5 rounds, or it can be stopped early when all adjacent units satisfy the gradient constraint.

[0231] Gradient constraints are used to suppress unreasonable spikes caused by noise, local occlusion, or fluctuations in model predictions.

[0232] Finally, the inner surface model of the pipe is updated along the local normal.

[0233] A preferred method for calculating the updated mesh vertex positions is:

[0234]

[0235] in, This indicates the position of the mesh vertices before the update. Indicates the updated mesh vertex positions. This represents the vertex displacement after being processed by boundary continuity transition constraints and gradient constraints. This indicates a local normal pointing towards the inside of the pipe wall.

[0236] During the update, the original mesh topology and connection relationship are kept unchanged, and only the position of the mesh vertices is changed, so as to obtain a three-dimensional defect model of the inner surface of the pipe.

[0237] Finally, the output includes a three-dimensional defect model of the inner surface of the pipe, a defect depth distribution map, an evidence status map, the location of the guided wave verification, and the identification of low-confidence areas.

[0238] Among them, the three-dimensional defect model of the inner surface of the pipe is used to display the spatial morphology of the defect; the defect depth distribution map is used to quantitatively assess the wall thickness loss; the evidence status map is used to distinguish the areas of consistent evidence, the areas of priority for guided wave evidence, and the areas of priority for visual evidence; the guided wave verification location is used to guide subsequent re-inspection; and the low confidence area marker is used to indicate that the test results need to be verified manually or by equipment.

[0239] Example 4

[0240] Reference Figure 5 As an embodiment of the present invention, a small-diameter pipe defect detection and three-dimensional reconstruction system based on ultrasonic guided wave and visual image fusion is provided, including: a data acquisition and mapping module 100, a defect characterization generation module 200, a cross-modal fusion module 300, and a depth determination and three-dimensional reconstruction module 400.

[0241] The data acquisition and mapping module 100 is used to acquire guided wave detection data and visual image data, establish cylindrical coordinate references, and complete mapping.

[0242] The defect characterization generation module 200 is used to generate a model of the inner surface of the pipe based on the cylindrical coordinate reference and obtain visual defect characterization information, as well as to identify the abnormal support area of ​​the guided wave based on the difference between the guided wave detection data and the reference guided wave response.

[0243] The cross-modal fusion module 300 is used to redistribute the guided wave anomaly response based on the guided wave anomaly support region and visual defect characterization information to form cross-modal defect support information.

[0244] The depth determination and 3D reconstruction module 400 is used to determine the defect depth based on the cross-modal defect support information, update the inner surface model of the pipe along the local normal, and obtain the 3D defect model of the inner surface of the pipe.

[0245] Furthermore, the data acquisition and mapping module 100 includes: an ultrasonic guided wave detection device 101, a visual acquisition device 102 capable of moving along the inside of the pipe, and a position acquisition device 103.

[0246] The position acquisition device 103 is communicatively connected to the ultrasonic guided wave detection device 101 and the visual acquisition device 102, respectively, so that the guided wave detection data and visual image data are associated with the same cylindrical coordinate reference.

[0247] The cross-modal fusion module 300 is also used to generate evidence consistency status, local resampling identifier, or guided wave verification identifier for local cylindrical mesh elements.

[0248] The visual acquisition device 102 includes at least one of the following: a motion control terminal 102A, an observation pose control terminal 102B, an illumination control terminal 102C, or a scanning step size control terminal 102D.

[0249] The cross-modal fusion module 300 is connected to the corresponding control terminal, converts the local resampling identifier into a resampling instruction including the target axial interval, the target circumferential interval and the scanning resolution, and writes the visual image data obtained by resampling back to the cross-modal defect support information of the corresponding local cylindrical mesh unit.

[0250] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0251] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for defect detection and reconstruction of small-diameter pipes using ultrasonic guided wave and visual fusion, characterized in that, include: Acquire guided wave detection data and visual image data, establish a cylindrical coordinate reference and complete the mapping, generate a pipe inner surface model based on the cylindrical coordinate reference and acquire visual defect characterization information, and identify abnormal waveguide support areas based on the difference between the guided wave detection data and the reference guided wave response. The acquisition of visual defect characterization information includes: Multi-frame geometric registration is performed on the visual image data mapped to the cylindrical coordinate reference to generate a model of the inner surface of the pipe; Defect feature lines are extracted based on surface geometric changes in the pipe inner surface model and grayscale, texture or color changes in the visual image. The connectivity of the defect feature lines is then processed to obtain the visual defect boundary. Boundary attribution information is generated based on the position and distance of cylindrical mesh cells relative to the visual defect boundary, and visually observable states are generated based on local brightness saturation, image sharpness, or multi-frame matching residuals. Visual defect boundaries, boundary attribution information, surface geometric residuals, and visually observable states are used as visual defect characterization information. Based on the abnormal support region and visual defect characterization information of the guided wave, the abnormal response of the guided wave is redistributed to form cross-modal defect support information. The defect depth is determined based on the cross-modal defect support information, and the inner surface model of the pipe is updated along the local normal to obtain the three-dimensional defect model of the inner surface of the pipe.

2. The ultrasonic guided wave vision fusion method for defect detection and reconstruction of small-diameter pipes as described in claim 1, characterized in that, The establishment of the cylindrical coordinate datum and the completion of mapping include: Acquire the axial scanning position, circumferential reference position, and visual calibration parameters of the visual acquisition device to determine the axial position and circumferential angle of the visual image pixels or visual three-dimensional points; The axial candidate interval of the guided wave response is determined based on the propagation time and group velocity of the guided wave, and the circumferential candidate interval of the guided wave response is determined based on the circumferential arrangement of the guided wave receiving channel and the channel response distribution. Visual image pixels or visual 3D points and waveguide responses are written into cylindrical meshes divided by axial position and circumferential angle, respectively.

3. The ultrasonic guided wave vision fusion method for defect detection and reconstruction of small-diameter pipes as described in claim 1, characterized in that, The identified waveguide anomalous support region includes: Obtain the guided wave reference response of the defect-free calibration pipe section or the stable region of the pipe section to be inspected, and perform time zero-point correction, amplitude correction and channel consistency correction on the guided wave test data and the guided wave reference response; An abnormal guided wave response is generated based on at least one of the amplitude difference, energy difference, phase difference, propagation time difference, or time-frequency characteristic difference between the corrected guided wave detection data and the guided wave reference response; The guided wave anomaly response that meets the anomaly conditions determined by the calibrated pipe section anomaly samples is mapped to a cylindrical mesh, and the connected anomaly cylindrical mesh cells are identified as guided wave anomaly support regions.

4. The ultrasonic guided wave vision fusion method for defect detection and reconstruction of small-diameter pipes as described in claim 3, characterized in that, The redistribution of the abnormal response of the guided wave includes: Determine the overlapping area between the waveguide anomaly support region and the visual defect boundary, and determine the element size of the local cylindrical mesh based on the magnitude and spatial variation of the waveguide anomaly response within the overlapping area; Based on the boundary assignment information, surface geometric residuals, and visually observable state of the local cylindrical mesh elements, the allocation coefficients of the local cylindrical mesh elements for the guided wave anomaly response are determined. The distribution coefficient is used to normalize the spatial redistribution of the guided wave anomaly response in the overlapping area, so that the total amount of guided wave anomaly response in the same guided wave anomaly support area remains consistent before and after redistribution. The redistributed guided wave anomaly response is written into the corresponding local cylindrical mesh element to obtain the redistributed guided wave anomaly response distribution.

5. The ultrasonic guided wave and visual fusion method for defect detection and reconstruction of small-diameter pipes as described in claim 4, characterized in that, The formation of cross-modal defect support information includes: After redistribution, the guided wave anomaly response, surface geometric residual, and distance to the visual defect boundary of each local cylindrical mesh element are normalized. Based on the spatial correspondence between the normalized guided wave abnormal response and the visual defect boundary attribution information, each local cylindrical mesh unit is divided into evidence consistency unit, guided wave evidence priority unit, or visual evidence priority unit. For evidence-consistent units, correlate normalized guided wave anomaly response, surface geometric residual, boundary distance, and visually observable state; For guided wave evidence priority units, the normalized guided wave abnormal response is retained and written into the local resampling identifier; For the visual evidence priority unit, retain the surface geometric residual and visual defect boundary attribution information and write it into the waveguide verification identifier; The unit identifiers and corresponding data of the evidence consistency unit, the waveguide evidence priority unit, and the visual evidence priority unit are used as cross-modal defect support information.

6. The ultrasonic guided wave and visual fusion method for defect detection and reconstruction of small-diameter pipes as described in claim 5, characterized in that, The step of determining the defect depth based on cross-modal defect support information includes: Obtain calibration pipe segment samples with known defect depth or wall thickness loss, and use the normalized guided wave anomaly response, surface geometric residual, boundary distance and visually observable state corresponding to the calibration pipe segment samples as inputs and the known defect depth or wall thickness loss as outputs to train a support vector regression model. Acquire the resampled visual image of the region corresponding to the waveguide evidence priority cell with the local resampled identifier, and update the surface geometric residual, boundary distance and visual observable state. By inputting the evidence consistency unit and the updated waveguide evidence priority unit into the trained support vector regression model, the defect depth of the corresponding local cylindrical mesh unit is obtained. Output the waveguide verification position based on the waveguide verification identifier, and update the defect depth of the visual evidence priority unit after obtaining the waveguide verification result.

7. The ultrasonic guided wave vision fusion method for defect detection and reconstruction of small-diameter pipes as described in claim 6, characterized in that, The process of updating the pipe inner surface model along the local normal includes: Determine the local normals pointing from the vertices of the local cylindrical mesh on the inner surface of the pipe to the inside of the pipe wall; Based on the defect depth of the local cylindrical mesh element, the mesh vertices are displaced along the local normal. Based on the distance from the corresponding local cylindrical mesh element to the visual defect boundary, a continuous transition constraint is applied to the displacement of the mesh vertices on both sides of the visual defect boundary. Apply gradient constraints to the defect depth gradient of adjacent mesh cells that match the size of the local cylindrical mesh cells; While maintaining the mesh topology connections of the pipe's inner surface model, the three-dimensional defect model of the pipe's inner surface is obtained by using the constrained mesh vertex positions.

8. A system for detecting and reconstructing defects in small-diameter pipes based on the fusion of ultrasonic guided waves and visual images, employing the ultrasonic guided wave and visual fusion method for detecting and reconstructing defects in small-diameter pipes as described in any one of claims 1 to 7, characterized in that: It includes a data acquisition and mapping module (100), a defect characterization generation module (200), a cross-modal fusion module (300), and a depth determination and 3D reconstruction module (400). The data acquisition and mapping module (100) is used to acquire guided wave detection data and visual image data, establish cylindrical coordinate references and complete mapping; The defect characterization generation module (200) is used to generate a pipe inner surface model based on a cylindrical coordinate reference and obtain visual defect characterization information, and to identify abnormal waveguide support areas based on the difference between the waveguide detection data and the reference waveguide response. The acquisition of visual defect characterization information includes: Multi-frame geometric registration is performed on the visual image data mapped to the cylindrical coordinate reference to generate a model of the inner surface of the pipe; Defect feature lines are extracted based on surface geometric changes in the pipe inner surface model and grayscale, texture or color changes in the visual image. The connectivity of the defect feature lines is then processed to obtain the visual defect boundary. Boundary attribution information is generated based on the position and distance of cylindrical mesh cells relative to the visual defect boundary, and visually observable states are generated based on local brightness saturation, image sharpness, or multi-frame matching residuals. Visual defect boundaries, boundary attribution information, surface geometric residuals, and visually observable states are used as visual defect characterization information. The cross-modal fusion module (300) is used to redistribute the guided wave anomaly response based on the guided wave anomaly support region and visual defect characterization information to form cross-modal defect support information; The depth determination and 3D reconstruction module (400) is used to determine the defect depth based on the cross-modal defect support information, update the inner surface model of the pipe along the local normal, and obtain the 3D defect model of the inner surface of the pipe.

9. The small-diameter pipe defect detection and three-dimensional reconstruction system based on ultrasonic guided wave and visual image fusion as described in claim 8, characterized in that: The data acquisition and mapping module (100) includes an ultrasonic guided wave detection device (101), a visual acquisition device (102) capable of moving along the inside of the pipe, and a position acquisition device (103). The position acquisition device (103) is connected to the ultrasonic guided wave detection device (101) and the visual acquisition device (102) respectively, so that the guided wave detection data and visual image data are associated with the same cylindrical coordinate reference. The cross-modal fusion module (300) is also used to generate evidence consistency status, local resampling identifier, or guided wave verification identifier for local cylindrical mesh elements; The visual acquisition device (102) includes at least one of a motion control terminal (102A), an observation pose control terminal (102B), an illumination control terminal (102C), or a scanning step size control terminal (102D); The cross-modal fusion module (300) is connected to the corresponding control terminal, converts the local resampling identifier into a resampling instruction including the target axial interval, the target circumferential interval and the scanning resolution, and writes back the visual image data obtained by resampling to the cross-modal defect support information of the corresponding local cylindrical mesh unit.

Images