Search

CN-121275789-B - Reflective steel pipe weld defect identification method based on multimode spectrum chromatography technology

CN121275789BCN 121275789 BCN121275789 BCN 121275789BCN-121275789-B

Abstract

The invention discloses a method for identifying defects of a weld joint of a reflective steel pipe based on a multi-mode spectrum chromatography technology, and provides a detection flow of multi-band reflection data acquisition, denoising and normalization, multi-region feature distribution modeling, sparse reconstruction separation abnormal response, multi-scale contrast weighting enhancement, saliency layered imaging and defect classification discrimination. The method realizes effective suppression of high reflection interference, obviously enhances the resolution of defects and background and improves the detection sensitivity and positioning accuracy of small micro defects through measures such as automatic partitioning, dynamic acquisition of environment and materials, self-adaptive parameter optimization and the like.

Inventors

  • FENG HONGWEI
  • YAN GUOBIAO
  • GUO QI

Assignees

  • 广州美亚股份有限公司

Dates

Publication Date
20260508
Application Date
20250923

Claims (10)

  1. 1. A method for identifying defects of weld joints of reflective steel pipes based on a multimode spectrum chromatography technology is characterized by comprising the following steps: s1, collecting multi-channel spectral reflection data of a high-reflection steel pipe welding seam area in multiple wavebands, and recording by combining different positions on the surface of a workpiece with reflection intensity labels; s2, denoising and signal normalization preprocessing are carried out on the multiband spectral reflection data, and a pure multiband spectral reflection data set is generated; S3, extracting multimode spectrum reflection curves from a defect-free weld joint region and a typical defect region based on the recorded different positions and the reflection intensity labels, respectively establishing a normal reflection mode dictionary and a defect sparse response dictionary, and constructing a regional characteristic distribution model; S4, inputting multimode spectrum data of a weld joint region to be detected into the normal reflection mode dictionary with a corresponding reflection intensity label for sparse reconstruction, performing sparse decoding by extracting a reconstruction residual error and utilizing the defect sparse response dictionary, and separating out abnormal sparse components to obtain a defect sensitive response diagram; S5, dynamically weighting the defect sensitive response map based on local contrast and global statistical characteristics by utilizing a multi-scale space analysis method; S6, reconstructing a layered imaging image with prominent defect characteristics by adopting a weighted fusion strategy according to abnormal significance enhancement graphs generated under different wave bands and spatial scales; S7, inputting the weighted and fused layered imaging images into a trained abnormal significance classification model, and automatically carrying out defect positioning and type discrimination according to the abnormal signal characteristics of the region; And S8, monitoring the environmental parameter change and the working condition label of the welding line area in real time, and dynamically and adaptively adjusting the layered imaging parameters and the characteristic dictionary if the system detects the significant change of the reflection intensity or the material background.
  2. 2. The method for identifying defects of weld seams of reflective steel pipes based on the multimode spectrum chromatographic technique according to claim 1, wherein the step S1 specifically comprises: performing multi-space-position automatic partitioning operation on the surface of a weld joint of a target steel pipe, and acquiring axial and radial coordinates of the weld joint by adopting an industrial positioning module to obtain a calibrated space region of the surface to be measured; Synchronously acquiring multi-channel spectral reflection original data of the calibration space region in different spectral bands by using a multi-band spectral acquisition device, and generating a multi-band reflection data matrix corresponding to the space positions one by one as an input of a signal processing chain; based on the real-time illumination measurement module, performing reflection intensity measurement on each spatial acquisition point, and marking the measured reflection background intensity value as an environmental state label on corresponding multichannel spectral reflection data; Judging the actual material type of the surface of the weld joint by means of a workpiece material identification module, and attaching a material identification result to the information concentration of the multi-channel spectral reflection data and the reflection intensity label to form a multi-mode spectral reflection metadata system comprising a space region, reflection intensity and a material three-dimensional label; And carrying out data structuring and structuring on the multimode spectrum reflection metadata to generate a high-dimensional and multi-label structured multimode spectrum reflection data warehouse.
  3. 3. The method for identifying the weld defects of the reflective steel pipe based on the multi-mode spectral chromatography technology according to claim 2, wherein the spatial partition adopts a laser three-dimensional positioning, image processing and automatic weld tracking module, and the step size of the spatial partition is 1 The shielding block is automatically removed through the mask, and the accuracy of the space partition is better than 0.3mm.
  4. 4. The method for identifying defects of weld seams of reflective steel pipes based on the multimode spectrum chromatographic technique according to claim 1, wherein the step S2 specifically comprises: Performing primary reflection background modeling on the original signal of the multiband spectral reflection data in a high reflection environment, and adopting a background fitting algorithm to extract primary reflection components and calibrate regional spectral reflection baselines so as to obtain a primary reflection model of multiband spectral reflection; A wavelet threshold denoising method is applied to the original multiband spectral reflection data based on the main reflection model, and a multiband spectral reflection denoising signal after denoising processing is obtained; Performing a normalization correction algorithm on the multiband spectral reflection denoising signal according to the reflection intensity label and the region segmentation information recorded during acquisition to obtain a multiband spectral reflection standard signal after normalization processing; Performing multi-channel crosstalk removal on the multi-band spectrum reflection standard signal, distinguishing related noise components in synchronous band acquisition, and further stripping mixed signals among channels to obtain a multi-band spectrum reflection confusion-removed signal after channel separation; And performing space-time consistency constraint inspection on the multiband spectral reflection confusion-removed signal, identifying and calibrating possible residual abnormal sample points by using a distribution statistical feature and an autocorrelation detection algorithm, and complementing or correcting an abnormal section by an interpolation resampling mode to output a pure multiband spectral reflection data set.
  5. 5. The method for identifying defects of a weld joint of a reflective steel pipe based on a multi-mode spectral chromatography technique according to claim 4, wherein when modeling a main reflection background of an original signal of multi-band spectral reflection data in a high reflection environment, a main reflection baseline model adopts multi-order polynomial fitting and PCA main component analysis to keep a cumulative variance interpretation rate more than or equal to 98%, and a fusion weight alpha is 0.5 0.8。
  6. 6. The method for identifying defects of weld joints of reflective steel pipes based on multimode spectrum chromatographic technology as set forth in claim 4, wherein said normalization correction algorithm uses z Score method or minimum Maximum normalization method.
  7. 7. The method for identifying defects of weld seams of reflective steel pipes based on the multimode spectrum chromatographic technique according to claim 1, wherein the step S3 specifically comprises: Carrying out specific region subset division on a defect-free weld joint region sample in the multichannel spectral reflection data based on the distributed different positions and reflection intensity labels to obtain a regional spectral reflection data set covering differences of uneven reflection and local materials; On the regional defect-free spectrum reflection data set, a single-spectrum/multi-spectrum combined time sequence statistics and energy spectrum analysis method is adopted to extract reflection curve vectors of each region under a multi-mode spectrum channel, and a region-resolved multi-mode spectrum reflection curve characteristic representation is generated; Based on the defect-free spectral reflection curve characteristics of each subarea, carrying out local characteristic prototype extraction and nonnegative sparse basis representation by using a sparse dictionary learning algorithm, and constructing and outputting a normal reflection mode dictionary; synchronously dividing spectrum data of a weld joint region with obvious defect labels into abnormal region subsets covering different positions and reflection intensity differences, executing abnormal response mode cluster analysis based on multimode spectrum response intensity curve characteristics, and extracting a corresponding defect sparse response characteristic set; The sparse coding and abnormal characteristic enhancement algorithm is applied, the abnormal spectral reflection characteristics of the subareas are input into an abnormal signal sparse decomposition model, the extracted defect sparse response characteristic set is subjected to aggregation training, and a defect sparse response dictionary covering multiple working conditions is established; and carrying out global consistency check on the multi-region and multi-reflection intensity on the normal reflection mode dictionary and the defect sparse response dictionary, and optimizing dictionary parameter configuration through cross reconstruction error analysis.
  8. 8. The method for identifying defects of weld joints of reflective steel pipes based on multimode spectrum chromatographic technology as set forth in claim 7, wherein said sparse response dictionary is prepared by performing K-test on a sample of areas with defective labels The method is characterized in that the method is obtained by extracting central abnormal characteristics through means or Gaussian mixture model clustering, and the category number is set to be 2 according to working conditions 10。
  9. 9. The method for identifying defects of weld seams of reflective steel pipes based on the multimode spectrum chromatographic technique according to claim 1, wherein the step S4 specifically comprises: performing characteristic normalization preprocessing according to the reflection intensity label corresponding to the multimode spectrum data of the weld joint region to be detected to obtain normalized multimode spectrum characteristic parameters; based on the normalized multimode spectrum characteristic parameters, a normal reflection mode dictionary which is trained in advance is adopted, sparse reconstruction operation is carried out on the input characteristics through a sparse coding algorithm, and multimode spectrum sparse representation coefficients and reconstruction signals are obtained; taking the normalized multimode spectrum characteristic parameters and the sparse reconstruction signals as inputs, calculating point-by-point residual vectors of the multimode spectrum characteristic parameters and the sparse reconstruction signals to obtain reconstruction residual distribution; Inputting the reconstructed residual distribution into an abnormal sparse decoding module, performing sparse decoding operation again by adopting an abnormal response sparse dictionary, and separating out an abnormal sparse component under a high reflection noise background; And performing spatial position mapping on the abnormal sparse component, and generating a defect sensitive response graph representing the significance of the abnormal signal according to the index of the workpiece surface area and the reflection intensity label.
  10. 10. The method for identifying defects of weld joints of reflective steel pipes based on multimode spectrum chromatography technology according to claim 1, wherein the automatic defect discrimination is based on a deep convolutional neural network, and the input is layered imaging space The spectral region saliency feature vector is subjected to multi-layer feature extraction and space mapping, and regional defect probability distribution and saliency scoring are output according to categories.

Description

Reflective steel pipe weld defect identification method based on multimode spectrum chromatography technology Technical Field The invention relates to the technical field of weld defect intelligent detection, in particular to a method for identifying weld defects of a reflective steel pipe based on a multimode spectrum chromatography technology. Background Currently, the field of multimode spectral tomography and weld defect intelligent detection is rapidly developed, and the multimode spectral tomography and weld defect intelligent detection become an important link in the quality assurance of metal structures and the engineering management and control of high-end manufacturing production lines. In the nondestructive testing application of the high-reflection steel pipe weld joint, the main flow technical scheme mainly focuses on the following categories: On one hand, the traditional physical detection methods such as ultrasonic, ray, magnetic powder or vortex are limited by the difficult problems of complex surface morphology, serious reflection noise and 'masking' of tiny defects, and the traditional physical detection methods have limited early defect detection sensitivity under the complex welding seam scene such as high-reflection steel pipes and the like, and are difficult to meet the requirements of accurate positioning and quantitative classification. The surface of the high-reflection steel tube has extremely strong main reflection noise in the spectrum signal due to complex influences such as specular reflection, environmental disturbance, uneven material distribution and the like, and is easy to produce a masking effect on a tiny abnormal signal. The conventional background modeling denoising and main component stripping methods are difficult to distinguish micro-scale defect anomalies from main reflection components, so that the detection rate of early anomalies or slight defects is low, and the risks of misjudgment and missed judgment are increased. Disclosure of Invention The invention provides a method for identifying defects of a weld joint of a reflective steel pipe based on a multimode spectrum chromatography technology, aiming at solving the technical problems. The technical scheme of the invention is realized by a method for identifying defects of a weld joint of a reflective steel pipe based on a multimode spectrum chromatography technology, which comprises the following steps: s1, collecting multi-channel spectral reflection data of a high-reflection steel pipe welding seam area in multiple bands, and recording by combining different positions of the surface of a workpiece with reflection intensity labels so as to adapt to different working conditions and material variables. S2, denoising and signal normalization preprocessing are carried out on the multiband spectral reflection data, background interference factors in a high-reflection environment are eliminated, and a pure reflection signal is provided for subsequent diversity distribution analysis. And S3, extracting multimode spectrum reflection curves from the defect-free weld joint region and the typical defect region based on the recorded different positions and reflection intensity labels, and respectively establishing a normal reflection mode dictionary and a defect sparse response dictionary to construct a regional characteristic distribution model. And S4, inputting the multimode spectrum data of the weld joint region to be detected into a normal reflection mode dictionary for sparse reconstruction by a corresponding reflection intensity label, and obtaining a defect sensitive response graph by extracting reconstruction residual errors and performing sparse decoding to separate out abnormal sparse components. S5, dynamically weighting the defect sensitive response graph generated by the abnormal sparse component based on the local contrast and the global statistical characteristic by utilizing a multi-scale space analysis method, so as to realize multi-scale contrast weighting enhancement of abnormal signals and improve the significance distinction between a defect area and a main reflective background. S6, reconstructing a layered imaging image with prominent defect characteristics by adopting a weighted fusion strategy according to abnormal significance enhancement graphs generated under different wave bands and spatial scales so as to consider imaging quality under different positions and reflection levels. And S7, inputting the weighted and fused layered imaging images into a trained abnormal significance classification model, and automatically carrying out defect positioning and type discrimination according to the abnormal signal characteristics of the region to realize high-precision automatic defect output. And S8, monitoring the environmental parameter change of the welding line area and the working condition label in real time, and if the system detects the significant change of the reflection intensity or the mat