Search

CN-121061408-B - Tubular pile inner wall welding method based on visual perception

CN121061408BCN 121061408 BCN121061408 BCN 121061408BCN-121061408-B

Abstract

The invention discloses a tubular pile inner wall welding method based on visual perception, which belongs to the technical field of visual detection and specifically comprises the steps of firstly installing inner wall micro-texture light shielding rings and root light guide wedges on two sides of a welding line, calibrating the geometry of a camera and a light field, establishing and locking a local coordinate system of the welding line, then starting annular polarized light fields to be lighted in a segmented mode along the welding line, guiding glancing light to be close to the root so as to be stably incident, obtaining a polarization separation image pair through a double-channel orthogonal polarization acquisition unit, synchronously recording an azimuth mark and a segmentation number, generating a polarization reflection image and a brightness reference image through polarization unmixing and curvature compensation, corresponding to the local coordinate system, constructing a feature dictionary, extracting weld toe contours, root crack textures and heat affected zone boundaries from the feature dictionary, finally inputting the feature vector into an identifier, combining the brightness reference image to obtain a defect mask and a tag, and projecting the weld back local coordinate system to complete detection.

Inventors

  • HAN HONGSHUAI
  • Yu Zhenjin

Assignees

  • 天津潘庄建材有限公司

Dates

Publication Date
20260508
Application Date
20250925

Claims (7)

  1. 1. The tubular pile inner wall welding method based on visual perception is characterized by comprising the following steps of: s1, mounting inner wall micro-texture light shielding rings and weld root light guide wedges on two sides of a weld, calibrating the geometry of a camera and a light field, and establishing and locking a local coordinate system of the weld, wherein the specific process is as follows: Mounting inner wall micro-texture shading rings on the inner walls of two sides of a welding line by using clamps, attaching and mounting a welding line root light guide wedge on the welding line root, and attaching azimuth marks on the outer sides of the shading rings; the camera is fixed on the inner wall positioning bracket, the azimuth mark and the tip of the light guide wedge are aligned, a checkerboard mark piece is placed for collecting images, and camera parameters and imaging geometry are calculated; Establishing and locking a local coordinate system of the welding seam according to camera parameters and imaging geometry, taking the tip of the light guide wedge as an original point, taking the trend of the welding seam as a trend axis and taking the normal direction of the pipe wall as a normal axis; S2, starting an annular polarized light field module to light in a sectional mode along the trend of the welding seam, leading the glancing light to be close to the root of the welding seam by a light guide wedge, limiting the range of a reflection angle and stabilizing the incident direction; s3, the double-channel polarization acquisition unit acquires images by means of mutual orthogonal polarization to obtain polarization separation image pairs, and simultaneously records the orientation marks and the segment numbers of the micro-texture shading rings on the inner wall; s4, performing polarization unmixing and curvature compensation mapping according to the polarization separation image pair, generating a welding seam region polarization reflection image and a brightness reference image, and corresponding the result to a welding seam local coordinate system; S5, constructing a weld joint region feature dictionary, and extracting weld toe contours, weld joint root crack textures and heat affected zone texture boundaries according to the polarized reflection diagram to form a weld joint region feature vector; S6, inputting the feature vector into a weld joint region image feature identifier, combining the brightness reference image to obtain a mask and a label of the defect part, and projecting the mask and the label back to a weld joint local coordinate system.
  2. 2. The method for welding the inner wall of the tubular pile based on visual perception according to claim 1, wherein in the step S2, the starting annular polarized light field module is lighted in a segmented manner along the trend of the welding seam, the light guiding wedge guides the glancing light to be close to the root of the welding seam, and the process of limiting the reflection angle range and stabilizing the incident direction is as follows: Loading weld strike segment numbers in a controller, starting an annular polarized light field module, and setting a polarizer azimuth angle and a segment lighting sequence; the current section light source is lightened according to the sectional sequence, so that the light guide wedge is attached to the inner wall of the welding seam, the position of the incidence surface of the light guide wedge and the position of the light field are adjusted, and the glancing light is attached to the root of the welding seam; and fixing the incidence direction according to the incidence surface of the light guide wedge and the azimuth angle of the polaroid, and shielding lateral reflection by using an inner wall micro-texture shading ring to limit the range of a reflection angle.
  3. 3. The method for welding the inner wall of the tubular pile based on visual perception according to claim 1, wherein in the step S3, the two-channel polarization acquisition unit acquires a polarization separation image pair by using mutually orthogonal polarization, and the process of recording the orientation mark and the segment number of the inner wall micro-texture shading ring comprises the following steps: loading a weld trend segmentation numbering table in a controller, configuring a synchronous trigger signal and a time stamp source, and setting two channel polarizers to be mutually orthogonal; triggering the current section acquisition according to the section number, aligning the double-channel polarization acquisition unit with the weld joint area, and synchronously acquiring images by mutual orthogonal polarization; acquiring two-channel original images, generating a polarization separation image pair according to channel pairing, and binding a time stamp and a segment number; identifying and recording the azimuth marks and the segment numbers of the micro-texture shading rings on the inner wall, and writing the results into metadata fields of the polarization separation image pairs.
  4. 4. The method for welding the inner wall of the tubular pile based on visual perception according to claim 1, wherein in the step S4, the process of performing polarization unmixing and curvature compensation mapping according to the polarization separation image pair to generate a weld joint region polarized reflection map and a brightness reference map, and mapping the result to a local coordinate system of the weld joint is as follows: Importing a polarization separation image pair and calibration parameters, reading segment numbers, generating an inner wall curvature grid according to the shape of the pipe wall, and associating a welding seam local coordinate system; A polarization unmixing model is built according to the mutual orthogonal polarization, and the polarization separation image pair is unmixed pixel by pixel to obtain the initial values of the polarization reflection component and the brightness component; Performing curvature compensation mapping according to the inner wall curvature grid and the incidence direction, and expanding the polarized reflection component to a weld joint local coordinate system plane; Generating a weld joint region polarized reflection diagram by using the unfolded polarized reflection components, generating a brightness reference diagram by using a brightness component initial value, and reserving a segment number and a time stamp; Registering the polarized reflection map of the welding seam area and the brightness reference map to a local coordinate system of the welding seam according to the segment numbers and the time stamps, and completing the correspondence of the results.
  5. 5. The method for welding the inner wall of the tubular pile based on visual perception according to claim 4, wherein the method is characterized in that a polarization unmixing model is built according to mutual orthogonal polarization, and the polarization separation image is unmixed pixel by pixel to obtain the specific contents of initial values of a polarization reflection component and a brightness component: Reading azimuth angles of the polarization separation image pair and the polaroid, determining mutually orthogonal polarization directions, and defining pixel observation vectors and imaging response coefficients; Establishing a polarization unmixing model according to the mutually orthogonal polarization relation, forming a coefficient matrix by using the azimuth angle of the polaroid and imaging response, and forming an observation vector by using pixel intensity; Inputting an observation vector to each pixel, and solving a polarization unmixing model by using least square to obtain initial values of a polarization reflection component and a brightness component; And generating corresponding images by using the initial values of the polarized reflection components and the brightness components, enabling indexes to be consistent with the polarized separation image pairs, and keeping the pixel positions in one-to-one correspondence.
  6. 6. The method for welding the inner wall of the tubular pile based on visual perception according to claim 1, wherein in the step S5, the process of constructing the weld joint region feature dictionary, extracting the weld toe contour, the weld joint root crack texture and the heat affected zone texture boundary according to the polarization reflection diagram, and forming the weld joint region feature vector is as follows: reading a polarized reflection diagram of a welding seam area under a welding seam local coordinate system, and establishing a characteristic dictionary field of the welding seam area, wherein the characteristic dictionary field comprises a welding toe contour, a crack texture and a heat affected zone texture boundary; generating a contour searching band along the trend of the weld joint, extracting a toe contour curve according to the polarized reflection gradient and the intensity transition tracking edge, and registering the toe contour curve to a characteristic dictionary of the weld joint region; setting a strip window in the neighborhood of the root of the welding seam, calculating the response and direction consistency of the linear texture, and connecting the thin line segments to obtain crack textures of the root of the welding seam and registering; Scanning polarization reflection changes line by line along the normal direction of the pipe wall, positioning texture mutation trend, drawing texture boundaries of a heat affected zone and registering the texture boundaries to a characteristic dictionary of a weld joint area; And encoding the weld toe contour, the crack texture and the heat affected zone texture boundary into weld joint region feature vectors according to the weld joint region feature dictionary, and reserving the segment numbers and the time stamp indexes.
  7. 7. The method for welding the inner wall of the tubular pile based on visual perception according to claim 1, wherein in the step S6, the process of inputting the feature vector into the image feature identifier of the weld region, combining the brightness reference map to obtain the mask and the label of the defect part, and projecting the mask and the label back to the local coordinate system of the weld is as follows: Loading a welding line region feature vector and a brightness reference graph, and setting a welding line region image feature identifier input format and a label set in a processing unit; matching the feature vector with the brightness reference graph according to the pixel position and the segment number to form a feature matrix and a brightness channel, and completing the recognition input assembly; Starting a weld joint region image feature identifier, outputting a defect part mask and a label according to a feature matrix and a brightness channel, and recording an index mapping table and a time stamp; And calling registration parameters according to the segment numbers and the time stamps, registering the mask and the label of the defect part to a local coordinate system of the welding seam, and keeping the pixel positions in one-to-one correspondence.

Description

Tubular pile inner wall welding method based on visual perception Technical Field The invention relates to the technical field of visual detection, in particular to a tubular pile inner wall welding method based on visual perception. Background The quality evaluation of the welded inner wall of the pipe pile usually depends on machine vision and endoscopic detection. The existing scheme mainly adopts ring light, coaxial light, dark field side light or area array LED light sources, acquires a weld joint region image by matching with an industrial camera and an endoscope lens, estimates the outline and height change of the weld joint by combining line laser triangulation, structural light stripe, luminosity three-dimensional imaging means and the like, and is assisted with a defect detection network of filtering, edge enhancement, threshold segmentation, morphology, template matching and depth learning, so that defects such as cracks, air holes, incomplete welding and the like are identified. However, the inner wall of the pipe pile is mostly an elongated metal curved surface cavity, the space is limited, the curvature is continuously changed, and the surface reflectivity is high. In the visual inspection after welding, the traditional annular light source, LED surface light and other illumination modes have insufficient suitability under the geometric and material conditions, and are easy to generate strong specular reflection or shadow shielding in key areas such as weld roots, joint edges and the like, so that high dynamic range scenes with overexposure and underexposure exist locally, the texture details are lost and the boundaries are virtual, meanwhile, the weld areas show pseudo textures and bright spots which interfere due to uneven reflection distribution, the characteristic expression is disordered, the key defect information is difficult to accurately capture, misleading is generated on the algorithm based on templates or learning, the accuracy and stability of defect identification are reduced, and the effective judgment of welding quality is interfered. Therefore, there is a need for a tubular pile inner wall welding method facing a metal curved surface cavity and having adaptive visual perception and illumination control, which can stably obtain an interpretable image and reliably extract key features under a complex reflection condition. Disclosure of Invention The invention aims to provide a tubular pile inner wall welding method based on visual perception, which solves the problems in the background technology. The aim of the invention can be achieved by the following technical scheme: a tubular pile inner wall welding method based on visual perception comprises the following steps: S1, installing an inner wall micro-texture shading ring and a weld root light guide wedge on two sides of a weld, calibrating the geometry of a camera and a light field, and establishing and locking a local coordinate system of the weld; S2, starting an annular polarized light field module to light in a sectional mode along the trend of the welding seam, leading the glancing light to be close to the root of the welding seam by a light guide wedge, limiting the range of a reflection angle and stabilizing the incident direction; s3, the double-channel polarization acquisition unit acquires images by means of mutual orthogonal polarization to obtain polarization separation image pairs, and simultaneously records the orientation marks and the segment numbers of the micro-texture shading rings on the inner wall; s4, performing polarization unmixing and curvature compensation mapping according to the polarization separation image pair, generating a welding seam region polarization reflection image and a brightness reference image, and corresponding the result to a welding seam local coordinate system; S5, constructing a weld joint region feature dictionary, and extracting weld toe contours, weld joint root crack textures and heat affected zone texture boundaries according to the polarized reflection diagram to form a weld joint region feature vector; S6, inputting the feature vector into a weld joint region image feature identifier, combining the brightness reference image to obtain a mask and a label of the defect part, and projecting the mask and the label back to a weld joint local coordinate system. In the step S1, the inner wall micro-texture light shielding rings and the weld root light guide wedges are arranged at the two sides of the weld joint, the geometry of a camera and a light field is calibrated, and the process of establishing and locking a local coordinate system of the weld joint is as follows: Mounting inner wall micro-texture shading rings on the inner walls of two sides of a welding line by using clamps, attaching and mounting a welding line root light guide wedge on the welding line root, and attaching azimuth marks on the outer sides of the shading rings; the camera is fixed on the inner wall posit