Search

CN-122023238-A - Automatic defect detection method and system for industrial DR equipment based on deep learning

CN122023238ACN 122023238 ACN122023238 ACN 122023238ACN-122023238-A

Abstract

The invention discloses an automatic defect detection method and system of industrial DR equipment based on deep learning, which relate to the technical field of industrial nondestructive detection and comprise the steps of acquiring industrial DR imaging data of a workpiece to be detected in a preset imaging posture, and constructing an original image set according to an imaging sequence; for industrial DR imaging data in an original image set, performing spatial continuity analysis based on the consistency of the variation of pixel gray levels in spatial neighborhoods, forming an abnormal region in which gray level variation is continuously distributed spatially, and determining the abnormal region as a continuous gray level abnormal region. According to the invention, through carrying out space continuity analysis on industrial DR imaging data, a continuous gray scale abnormal region is formed and identified, and the main extension direction is further determined based on the space distribution form of the continuous gray scale abnormal region in an image coordinate system, so that directivity description parameters are introduced at the abnormal region level, and the continuous gray scale abnormal region is not treated as a gray scale abnormal object.

Inventors

  • REN BOQI
  • Hua Deqiao

Assignees

  • 苏州锐迪安机械设备有限公司

Dates

Publication Date
20260512
Application Date
20251216

Claims (9)

  1. 1. An automatic defect detection method for industrial DR equipment based on deep learning is characterized by comprising the following steps: step one, acquiring industrial DR imaging data of a workpiece to be detected in a preset imaging posture, and constructing an original image set according to an imaging sequence; Step two, aiming at industrial DR imaging data in an original image set, based on the consistency of the change of pixel gray scales in a space adjacent domain, performing space continuity analysis to form an abnormal region with gray scale changes continuously distributed in space, and determining the abnormal region as a continuous gray scale abnormal region; Determining a main extension direction of each continuous gray abnormal region based on the spatial distribution form of the continuous gray abnormal region in an image coordinate system, and representing the main extension direction as a directional description parameter; step four, extracting gray level change characteristics along a main extension direction and perpendicular to the main extension direction respectively based on the directivity description parameters, and combining the gray level change characteristics in different directions to construct a structure projection characteristic parameter set for describing the structure projection superposition characteristics; Inputting the structural projection characteristic parameter set into a defect discrimination parameter set obtained through training, and performing discrimination analysis on the continuous gray scale abnormal region to distinguish continuous gray scale abnormality formed by structural projection from continuous gray scale abnormality formed by crack defect; and step six, generating defect identification output information as a defect detection result of the industrial DR equipment according to the result of the discriminant analysis.
  2. 2. The automatic defect detection method for the industrial DR equipment based on the deep learning according to claim 1 is characterized in that the original image set comprises a continuous gray scale abnormal region formed by overlapping projection of a multi-layer structure, reinforcing ribs and a pore canal or an inclined inner cavity forming an included angle with the radial direction in the industrial DR imaging process.
  3. 3. The automatic defect detection method of an industrial DR apparatus based on deep learning according to claim 1, wherein the step of acquiring the continuous gray scale abnormal region comprises: in an original image set, aiming at each group of industrial DR imaging data, traversing the gray level change of each pixel in the industrial DR imaging data by taking a preset space neighborhood as an analysis unit to obtain a gray level difference value set of the pixel gray level in the space neighborhood; calculating a pixel gray change consistency index based on the gray difference value set, comparing the pixel gray change consistency index with a preset consistency threshold value, and determining a candidate pixel set meeting the preset consistency threshold value; performing connectivity aggregation based on the pixel adjacency relation of the candidate pixel set to form an abnormal region with gray level change distributed continuously in space; calculating a region length parameter and a region width parameter for the abnormal region, and determining the abnormal region as a continuous gray scale abnormal region when the region length parameter and the region width parameter meet a preset morphological threshold.
  4. 4. The method for automatic defect detection of an industrial DR device based on deep learning according to claim 1, wherein the step of processing the direction description parameter comprises: For each continuous gray scale abnormal region, acquiring a pixel coordinate set contained in the continuous gray scale abnormal region, and calculating a second order center moment of the continuous gray scale abnormal region based on the pixel coordinate set; Determining the main axis direction of the continuous gray scale abnormal region based on the second order central moment, and determining the main axis direction as the main extension direction of the continuous gray scale abnormal region; The main extending direction is expressed as a directivity description parameter in an image coordinate system, and the directivity description parameter at least comprises an included angle parameter of the main extending direction and an image transverse axis and a corresponding direction unit vector parameter; And calculating a direction saliency index of the continuous gray scale abnormal region based on the characteristic value of the second-order central moment, comparing the direction saliency index with a preset direction threshold value, and retaining the directivity description parameter when the direction saliency index meets the preset direction threshold value.
  5. 5. The method for automatic defect detection of an industrial DR device based on deep learning according to claim 1, wherein the step of constructing a set of structural projection feature parameters comprises: Determining a region center path distributed along the main extending direction in each continuous gradation abnormal region based on the directivity description parameter, the region center path being constituted by pixels located inside the continuous gradation abnormal region; acquiring a gray sampling point sequence in the continuous gray abnormal region along a region center path at preset sampling intervals, and calculating gray change characteristics of the main extending direction based on the gray sampling point sequence; Constructing orthogonal cross-lines at a plurality of positions of the regional center path along the direction perpendicular to the main extension direction, and counting only sampling points falling into the continuous gray level abnormal region to generate orthogonal direction gray level change characteristics; and combining the main extension direction gray level change characteristic and the orthogonal direction gray level change characteristic to construct a structure projection characteristic parameter set for describing the structure projection superposition characteristic.
  6. 6. The method for automatically detecting defects of an industrial DR equipment based on deep learning according to claim 5, wherein the step four is to perform stability constraint processing on the projected feature parameter set of the structure when constructing the feature parameter set, and includes: calculating a characteristic stability index based on parameter distribution conditions of gray scale change characteristics in a main extending direction and gray scale change characteristics in an orthogonal direction in a structural projection characteristic parameter set; the characteristic stability index at least comprises a characteristic parameter quantity index and a characteristic parameter dispersion index; And according to the characteristic stability index, performing weight adjustment or parameter screening on the main extension direction gray scale change characteristic and the orthogonal direction gray scale change characteristic in the structural projection characteristic parameter set to form a stabilized structural projection characteristic parameter set, and taking the stabilized structural projection characteristic parameter set as the input of subsequent discriminant analysis.
  7. 7. The automatic defect detection method of an industrial DR apparatus based on deep learning according to claim 1, wherein in step five, performing discriminant analysis on the continuous gray-scale abnormal region comprises: Generating a distinguishing input vector from the structure projection characteristic parameter set according to a preset field sequence; The preset field sequence at least comprises a change amplitude parameter, a change frequency parameter and a change amplitude parameter of the main extension direction gray scale change characteristic and a change frequency parameter of the orthogonal direction gray scale change characteristic; inputting the discrimination input vector into a defect discrimination parameter set obtained through training, and outputting a structure projection discrimination score and a crack defect discrimination score; calculating a score difference parameter of the structure projection discrimination score and the crack defect discrimination score; Comparing the score difference parameter with a preset score threshold, judging the continuous gray scale abnormal region as continuous gray scale abnormal formed by structural projection when the score difference parameter meets the preset score threshold, and judging the continuous gray scale abnormal region as continuous gray scale abnormal formed by crack defects when the score difference parameter does not meet the preset score threshold.
  8. 8. The method for automatically detecting defects in an industrial DR device based on deep learning according to claim 7, further comprising, when performing a discriminant analysis of the continuous gray scale anomaly region: before the judging input vector is input into the defect judging parameter set obtained through training, calculating a direction consistency index based on the parameter corresponding relation between the main extension direction gray level change characteristic and the orthogonal direction gray level change characteristic in the structural projection characteristic parameter set; taking the direction consistency index as a discrimination constraint parameter to participate in the calculation process of the structure projection discrimination score and the crack defect discrimination score; When the direction consistency index meets the preset consistency condition, the score difference parameter is adopted to judge, and when the direction consistency index does not meet the preset consistency condition, consistency correction is carried out on the score difference parameter and then the judgment is carried out, so that continuous gray scale abnormality formed by structural projection and continuous gray scale abnormality formed by crack defect are distinguished.
  9. 9. An automatic defect detection system of industrial DR equipment based on deep learning, characterized in that the detection system adopts the detection method of any one of claims 1-8.

Description

Automatic defect detection method and system for industrial DR equipment based on deep learning Technical Field The invention relates to the technical field of industrial nondestructive testing, in particular to an automatic defect detection method and system for industrial DR equipment based on deep learning. Background The industrial digital ray imaging technology can acquire the internal information of the workpiece under the condition of not damaging the structure of the workpiece to be detected, has been widely applied to the quality detection field of castings, welding structural members and complex mechanical parts, can acquire two-dimensional imaging data reflecting the internal density change and structural distribution characteristics of the workpiece through industrial DR imaging equipment, and provides a basic data source for the identification and judgment of internal defects; The existing industrial DR automatic defect detection method is generally used for analyzing and judging abnormal areas in imaging data based on image gray features, morphological features or statistical features, wherein one type of method is used for extracting suspected defect areas by carrying out area segmentation or gray abnormality detection on images, and the other type of method is used for classifying and identifying the extracted areas by combining machine learning or deep learning technology; However, in practical industrial application, the detected workpiece often has a multi-layer structure, reinforcing ribs, a pore canal forming an included angle with the ray direction, an inclined inner cavity and other complex internal structures, different spatial structures can be overlapped in the projection direction in the industrial DR imaging process to form a gray scale abnormal region which is continuously distributed in a two-dimensional imaging result, the continuous gray scale abnormal region has the characteristics of structural continuity and crack ductility in visual characteristics, so that the conventional method has a certain technical challenge in distinguishing the continuous gray scale abnormal formed by overlapping structural projections from the continuous gray scale abnormal formed by real crack defects, and therefore, the invention provides an automatic defect detection method and an automatic defect detection system for industrial DR equipment based on deep learning. Disclosure of Invention The invention aims to provide an automatic defect detection method and system for industrial DR equipment based on deep learning, so as to solve the problems in the background art. The invention can be realized by the following technical scheme that the automatic defect detection method of the industrial DR equipment based on deep learning comprises the following steps: step one, acquiring industrial DR imaging data of a workpiece to be detected in a preset imaging posture, and constructing an original image set according to an imaging sequence; Step two, aiming at industrial DR imaging data in an original image set, based on the consistency of the change of pixel gray scales in a space adjacent domain, performing space continuity analysis to form an abnormal region with gray scale changes continuously distributed in space, and determining the abnormal region as a continuous gray scale abnormal region; Determining a main extension direction of each continuous gray abnormal region based on the spatial distribution form of the continuous gray abnormal region in an image coordinate system, and representing the main extension direction as a directional description parameter; step four, extracting gray level change characteristics along a main extension direction and perpendicular to the main extension direction respectively based on the directivity description parameters, and combining the gray level change characteristics in different directions to construct a structure projection characteristic parameter set for describing the structure projection superposition characteristics; Inputting the structural projection characteristic parameter set into a defect discrimination parameter set obtained through training, and performing discrimination analysis on the continuous gray scale abnormal region to distinguish continuous gray scale abnormality formed by structural projection from continuous gray scale abnormality formed by crack defect; And step six, generating defect identification output information according to the result of discriminant analysis, and outputting the defect identification output information as an automatic defect detection result of industrial DR equipment. The invention further technically improves that the original image set comprises a continuous gray scale abnormal region formed by overlapping and projecting a multilayer structure, reinforcing ribs and a pore canal or an inclined inner cavity forming an included angle with the ray direction in the industrial DR imaging process. The invention