Search

CN-122016817-A - Dangerous chemical flaw screening system and method based on computer vision and laser scanning

CN122016817ACN 122016817 ACN122016817 ACN 122016817ACN-122016817-A

Abstract

The invention discloses a dangerous chemical flaw screening system and method based on computer vision and laser scanning, wherein the system comprises an electric portal frame, a camera, a laser 3D scanner, a to-be-detected sample conveying belt, a system control processor and a sample, the screening method is to integrate industrial microscopic imaging, laser structured light scanning, multi-mode data fusion and intelligent flaw classification technologies into the same detection assembly line, realize automatic detection of surface flaws and three-dimensional geometric flaws of dangerous chemicals and perform quality screening on surface features of the dangerous chemicals, and the dangerous chemical flaw screening system and the method can be used for completing dangerous chemical flaw identification, flaw positioning and automatic sorting on the premise of no need of manual intervention, remarkably improve dangerous chemical production quality screening efficiency, reduce exposure of personnel in dangerous environments, and solve the problems of low detection efficiency, large manual discrimination error, high safety risk and incapability of performing refined structural analysis in the existing dangerous chemical quality screening process.

Inventors

  • YU XIAO
  • LOU WENZHONG
  • FENG HENGZHEN
  • MA WENLONG

Assignees

  • 北京理工大学

Dates

Publication Date
20260512
Application Date
20260317

Claims (10)

  1. 1. The hazardous chemical substance flaw screening system based on computer vision and laser scanning is characterized by comprising an electric portal frame (1), a sample conveying belt to be detected (4), a detected sample conveying belt (5), a system control processor (6) and a sample (7); the electric portal frame (1) comprises a Z-direction driving motor (14), an X-direction driving motor (16) and a top beam guide rail (12) provided with an X-direction sliding block assembly (13), wherein a Z-direction lifting module (15) is arranged on the X-direction sliding block assembly (13); a camera (3) and a laser 3D scanner (2) are arranged on the Z-direction lifting module (15); The system control processor (6) comprises a system self-checking module (61), a stepping motor controller (62), an image acquisition module (63), an image recognition parameter adjusting module (64), an image semantic segmentation algorithm module (65), a laser dot matrix three-dimensional scanning grid reconstruction module (66), a model volume slice calculation module (67), a flaw detection classification module (68) and a data recording system (69).
  2. 2. The method for screening the dangerous chemical flaws based on the computer vision and the laser scanning is characterized by being implemented by the dangerous chemical flaw detection system based on the computer vision and the laser scanning as claimed in claim 1, and comprises the following steps: 1) Placing a sample (7) on a sample conveying belt (4) to be detected, and fixing a camera (3) and a laser 3D scanner (2) on a Z-direction lifting module (15) of an electric portal frame (1); 2) The step motor controller (62) controls the X-direction slide block assembly (13) and the sample conveyor belt (4) to be detected to move so that the sample (7) is positioned below the camera (3) and the laser 3D scanner (2), and the image recognition parameter adjusting module (64) adjusts the height of the Z-direction lifting module (15) according to the image of the sample; 3) The image acquisition module (63) sends the two-dimensional image of the sample surface acquired by the camera (3) to the image recognition parameter adjustment module (64); 4) The laser lattice three-dimensional scanning grid reconstruction module (66) triggers the laser 3D scanner (2) to project a structured light array to the surface of the sample (7) and collect point clouds, the laser lattice three-dimensional scanning grid reconstruction module (66) converts the point clouds from a scanner coordinate system to a system unified coordinate system, and the point clouds are rasterized, fitted with local surfaces and interpolated and complemented with missing areas to form continuous surface height distribution; 5) The image semantic segmentation algorithm module (65) performs semantic segmentation on the two-dimensional image, separates a surface area from a background, identifies surface defect characteristics, and outputs a semantic label graph, each class probability graph and types, positions, sizes and confidence levels of defect examples to the defect detection classification module (68); 6) The model volume slice calculation module (67) carries out layering slicing on the three-dimensional grid model generated by the laser dot matrix three-dimensional scanning grid reconstruction module (66) to obtain a section profile corresponding to the slice, the section area is obtained through section closed area integration, the sectional area and the maximum height and the minimum height of each layer of slice are calculated, the sectional area of each layer of slice is integrated to obtain an integral volume, the three-dimensional structure abnormal area is identified, and three-dimensional data of layering sectional area sequences, the integral volume, the maximum height and the minimum height, flaw area, flaw height and flaw volume characteristics are output to the flaw detection classification module (68); 7) The flaw detection classification module (68) takes the two-dimensional flaw distribution data output by the image semantic segmentation module (65) and the three-dimensional data output by the model volume slice calculation module (67) as flaw indexes of input calculation samples, and judges that the sample is unqualified when the flaw indexes exceed a set threshold value, and the sample is qualified otherwise; 8) The stepping motor controller (62) controls the detected sample conveyor belt (5) to classify the samples according to the judging result of the flaw detection classification module (68), and the steps (2) to (7) are repeatedly executed.
  3. 3. The method for screening a hazardous chemical substance flaw based on computer vision and laser scanning according to claim 2, wherein in the step 6), the flaw volume is calculated When the slice direction is set as the normal direction, the slice thickness is set as A k is the cross-sectional area of the defective region on the k-th slice plane, a k+1 is the cross-sectional area of the defective region on the k+1th slice plane, and it is derived that: 。
  4. 4. The method for screening the defects of the hazardous chemical substance based on computer vision and laser scanning according to claim 2, wherein in the step 8), the defect index is calculated by the following steps: 8.1 Set up the first The area ratio of the surface defects is as follows The thickness deviation is measured as the average absolute value of the surface height deviation And the maximum absolute value The volume deviation is measured as the bulge volume And recess/absence volume And normalizing by adopting a flaw detection classification module (68) to obtain: ; Wherein, the 、 、 、 Is a preset quality reference limit value; 8.2 The flaw detection classification module (68) weights according to preset weights to obtain flaw indexes of the sample : Wherein, the As the number of classes of surface flaws, 、 、 、 Is non-negative weight and satisfies ; The flaw detection classification module (68) will And a preset threshold value Comparison when If the test is qualified, and if the test is And the flaw detection classification module (68) controls the detected sample conveyor belt (5) to classify through the stepping motor controller (62), and uploads the judgment result to the data recording system (69).
  5. 5. The method for screening the dangerous chemical defects based on computer vision and laser scanning according to claim 2, wherein in the step 5), an image semantic segmentation algorithm module (65) performs semantic segmentation on the two-dimensional image, separates a surface area from a background, identifies the surface defect characteristics of pits, cracks, pits and unreacted spots, and outputs a semantic label graph, a class probability graph and two-dimensional defect distribution data of types, positions, sizes and confidence of defect instances to a defect detection classification module (68).
  6. 6. The method for screening the dangerous chemical defects based on computer vision and laser scanning according to claim 2, wherein in the step 6), a model volume slice calculation module (67) compares the actual surface height with a design reference plane point by point to form height deviation distribution, obtains a missing area, a maximum missing depth, an average missing depth and a missing volume for missing type defects, obtains a raised area, a maximum raised height, an average raised height and a raised volume for local type defects, and obtains a recessed area, a maximum recessed depth, an average recessed depth and a recessed volume for recessed type defects.
  7. 7. The method for screening hazardous chemical substance flaws based on computer vision and laser scanning according to claim 2, wherein in the step 2), the stepping motor controller (62) controls the X-direction slide block assembly (13) to move to the position above the center of the sample (7), and adjusts the X-direction slide block assembly according to camera picture feedback.
  8. 8. The method for screening hazardous chemical substance flaws based on computer vision and laser scanning according to claim 2, wherein in step 2), the image recognition parameter adjusting module (64) judges focusing according to image definition, brightness and light spot distribution of the sample, so that the stepping motor controller (62) adjusts the height of the Z-direction lifting module (15) to enable the image focal plane and the laser dot matrix projection plane to achieve optimal imaging at the same time.
  9. 9. The method for screening hazardous chemical substance flaws based on computer vision and laser scanning as set forth in claim 2, wherein in step 4), the point cloud comprises a three-dimensional coordinate set composed of a plurality of sampling points 。
  10. 10. The method for screening of hazardous chemical substance flaws based on computer vision and laser scanning as set forth in claim 2, wherein in step 8), the data recording system (69) stores two-dimensional image data, three-dimensional mesh reconstruction data, defect determination data and motion control history record generated during the detection, and generates a quality inspection file.

Description

Dangerous chemical flaw screening system and method based on computer vision and laser scanning Technical Field The invention relates to the technical field of computer vision detection and laser 3D scanning, in particular to a dangerous chemical flaw screening system and method based on computer vision and laser scanning. Background The present dangerous chemical is widely applied to the military and industrial fields due to the characteristics of high sensitivity and easy initiation. However, due to limitations in the manufacturing process, the finished hazardous chemical product often has different levels of defects. These include, but are not limited to, non-uniform surface thickness of the hazardous chemical, surface dishing beyond the loading plane, partial hazardous chemical failure to react completely, and the like. These defects not only affect the performance of the hazardous chemicals, but may also lead to safety hazards during use. After the production of hazardous chemicals is completed, strict quality screening must be performed to reject defective products. However, because the volume of the dangerous chemical is smaller, the surface defect scale is tiny and diversified, defects such as uneven thickness, partial depression, incomplete surface reaction or edge shortage and the like cannot be accurately identified only by manual visual inspection, and the dangerous chemical has inflammable, explosive or high-sensitivity response characteristics, and potential safety hazards exist in the manual contact detection process, so that the existing detection method is difficult to meet the requirements of batch production on real-time performance, stability and safety. In actual production, dangerous chemicals are usually manufactured and put off line in a batch and continuous mode, and due to factors such as efficiency and cost of manual detection, the conventional screening mode is difficult to implement comprehensive detection on each finished product, and mainly adopts sampling detection or performs tight spot inspection on key batches and key time periods. Even if sampling results are qualified in a sampling inspection mode, random defects of all products in the same batch caused by production processes cannot be effectively avoided, and a small amount of finished products with the problems of uneven thickness, exceeding a loading plane, surface pits or incomplete reaction and the like are easily mixed into a qualified product circulation link, so that potential risks are brought to subsequent storage, transportation, assembly and use, and quality control requirements of dangerous chemicals on high consistency and high reliability are difficult to meet. The existing manual detection has strong dependence on personnel experience, and standard execution is easy to be inconsistent. For the defects of 'small scale, various forms and unclear boundaries' of surface flaws of dangerous chemicals, the visual or simple visual inspection means are difficult to identify stably, especially under the influence of factors such as illumination change, observation angle difference, personnel fatigue and the like, the visibility of the defects and the fluctuation of judging results are obvious, different inspectors understand different thresholds such as 'whether slight potholes are unqualified or not', 'whether local thickness difference is over-limited', 'how the edge material shortage degree is graded', and the like, thus easily causing missed inspection and misjudgment, causing poor consistency of detection results among batches and shifts, and being difficult to form a quantifiable and traceable quality detection system. In addition, because hazardous chemicals generally have inflammable, explosive and high-sensitivity response characteristics, manual detection inevitably involves operations such as close-range observation, picking, placing, carrying, turning and checking, increases the probability of occurrence of dangerous factors such as friction, collision, extrusion, static accumulation and the like, and has certain potential safety hazards. Especially, under the conditions that the detection rate is required to be improved, the residence time is prolonged, the repeated detection frequency is increased, the personnel exposure time and the contact frequency are further increased, the safety management of dangerous chemical production is not facilitated, and the safety production requirement of the modern dangerous chemical production on personnel far away from a dangerous source and the human intervention is reduced is not matched. Therefore, an automatic and standardized detection method for large-batch screening of dangerous chemical finished products is needed, and on the premise of minimizing manual close contact and subjective judgment, the rapid and consistent judgment of each product is realized, and meanwhile, the detection result recording, tracing and statistical analysis capabilities are provide