System and method for detecting large-curvature surface bulge defect of wind power blade mold based on multi-angle oblique illumination

Abstract

The invention discloses a system and a method for detecting a large-curvature surface bulge defect of a wind power blade mold based on multi-angle oblique illumination, and belongs to the field of automatic quality detection. The system comprises an image acquisition unit arranged along the normal direction of the surface, an annular inclined light source group and a control processing unit. The method comprises the steps of controlling light sources to be lightened in sequence; the image acquisition unit is used for synchronously acquiring a surface reflection image when being lightened every time; the processing unit extracts a highlight area formed by strong reflection of the convex defect edge in each image; and finally, synthesizing all the highlight areas to form a complete defect map. According to the optical principle that the flat surface is in a dark state under oblique light and the edge of the convex defect is in a bright state, through multi-angle sequential irradiation and image synthesis, the complete contour of the micro convex defect on the large-curvature surface of the mold can be quickly and accurately recognized, the height of the micro convex defect can be deduced, and accurate guidance is provided for automatic grinding.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automated manufacturing and quality detection, and in particular to an automatic optical detection system for micro-protruding defects of large irregular curved surfaces (such as the inner surface of a wind turbine blade forming mold) and a detection method thereof. BACKGROUND

[0002] Large wind turbine blades are usually formed by composite materials in a large mold. The surface quality of the inner surface of the mold is crucial to the surface quality of the formed blade, and even to the aerodynamic performance and service life of the blade. After the blade is demolded, the inner surface of the mold (cavity) will have micro-protruding defects such as solidified release agent, adhesive drops, and composite material dust randomly distributed.

[0003] In order to ensure the forming quality of the next blade, the mold must be maintained before the next mold closing, and the key process is to remove these protruding defects by polishing. At present, this polishing process highly depends on manual work. For example, a hundred-meter wind turbine blade mold usually needs to be equipped with about 50 workers to perform the work by hand-holding pneumatic polishers, and the single maintenance time can be up to 24 hours.

[0004] The existing manual polishing method has the following major defects: Low efficiency: It consumes a lot of manpower and time, and seriously affects the turnover rate of the mold and the production rhythm of the wind turbine blade; 1. Unstable quality: The mold cavity is a large-curvature free-form surface, and it is difficult for workers to accurately control the polishing force and angle by hand-holding the polisher. The polishing quality highly depends on the skill and responsibility of the workers; 2. High risk of mold damage: It is easy to cause "over-polishing", that is, the defects are polished away, but the smooth surface of the mold is also damaged, resulting in concave pits or scratches on the surface of the mold. These damages will be copied to the subsequent produced blades, causing appearance defects of the blades; 3. Poor working environment: A large amount of dust is generated during polishing, and workers need to work in a closed cavity for a long time, which is not conducive to health.

[0005] Therefore, to realize the automatic and intelligent polishing maintenance of the wind turbine blade mold is a technical problem to be solved in the field. The first prerequisite for automatic polishing is to quickly, accurately and non-contactly identify and locate all randomly distributed micro-protruding defects on the surface of the mold. The existing machine vision detection method often has difficulty in achieving high precision and high efficiency at the same time when facing such "large-curvature background" and "micro-protruding target". SUMMARY

[0006] To achieve the above purpose, the present application provides a wind turbine blade mold large-curvature surface protruding defect detection system based on multi-angle oblique light, characterized by comprising: one image acquisition unit configured to have its optical axis approximately along the local normal direction of the surface of the mold to be measured, for acquiring the reflected light image of the surface to be measured; one ring-shaped light source group comprising at least two light sources arranged in a ring-shaped array around the image acquisition unit, each light source configured to obliquely illuminate the surface to be measured at a large angle; one control and processing unit electrically connected to the image acquisition unit and the ring-shaped light source group, configured to: (a) control the light sources in the ring-shaped light source group to be sequentially and individually turned on; (b) synchronously trigger the image acquisition unit to take a surface reflection image when each light source is turned on; (c) receive and process the multiple images taken by the image acquisition unit, and extract the highlight area (i.e. the local profile of the defect) formed by the strong reflection of the protruding defect in each image through image processing algorithm; (d) synthesize all highlight areas from the multiple images to generate a defect map containing the complete profile of all protruding defects.

[0007] Preferably, the control and processing unit is further configured to infer the height distribution information of the protruding defect based on the pixel gray value (reflection intensity) in the highlight area.

[0008] Correspondingly, the present application also provides a method for detecting protruding defects on large-curvature surface of wind turbine blade mold based on multi-angle oblique illumination, characterized in that it comprises the following steps: Step one (positioning): move a detection head containing a normal image acquisition unit and a ring-shaped oblique light source group above a local area of the mold to be measured, and approximately align the optical axis of the image acquisition unit with the surface normal of the local area; Step two (sequential illumination and acquisition): sequentially and individually turn on each light source in the ring-shaped oblique light source group; during the illumination of each light source, use the image acquisition unit to acquire a surface reflection image of the local area, and obtain N images (N is the number of light sources) in total; Step three (image processing and extraction): process each of the N images, identify and extract the highlight pixel area formed by the strong reflection of the protruding defect in the image, and obtain N defect local profile maps; Step four (synthesis and output): synthesize the N defect local profile maps (for example, through pixel-level logical "or" operation), and generate a defect map that can reflect the complete profile of all protruding defects in the local area; Step 5 (Feedback): The defect map is fed back to the control system of the robotic arm's grinding head to guide the grinding head to perform targeted local grinding.

[0009] Preferably, between steps three and four, the following is also included: Step 3.5 (Height Inference): Analyze the grayscale intensity within the bright pixel area. Since the more prominent the defect, the smaller the angle between its surface normal and the oblique light source, the stronger the light reflected back to the image acquisition unit. Therefore, the height distribution of the defect is inferred based on the grayscale values ​​of the bright area. Beneficial effects

[0010] Compared with the prior art, the present invention has the following significant advantages: High detection accuracy: Utilizing the dark-field optical principle of "oblique illumination and normal reception," the smooth mold surface appears dark, while the edges of tiny raised defects appear bright, resulting in an extremely high signal-to-noise ratio and easy segmentation and extraction; Good contour integrity: Through sequential illumination by the ring light source group, the edge contours of the protruding defects in different directions can be illuminated separately, and finally combined into a complete defect contour, avoiding the contour loss caused by a single-direction light source; Strong resistance to background interference: This method detects the abrupt change in reflectance caused by the geometric feature of the "protrusion," rather than detecting the color or texture of the material itself. Therefore, it can effectively overcome uneven lighting caused by large curvature changes on the mold surface, as well as interference from potential oil stains, color differences, etc. It can acquire three-dimensional information: not only can it obtain the 2D contour location of defects, but it can also qualitatively or quantitatively infer the height information of defects by analyzing the intensity of reflected light, providing a basis for the grinding system to control the grinding depth and intensity; High efficiency: The switching speed of multiple light sources and the image acquisition speed are extremely fast (millisecond level). It can be mounted on a robotic arm for "flying shooting" or "stop and shoot immediately", realizing rapid full inspection of large-area molds and making automated grinding possible. Attached Figure Description

[0011] Figure 1 : A flowchart of the defect detection method described in this invention.

[0012] Figure 2 : A schematic diagram of the detection principle described in this invention.

[0013] Figure 3 : A schematic diagram of the defect contour synthesis described in this invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] In a typical application scenario, the detection system described in the present invention is integrated into a detection head (10) which is mounted together with a polishing head device at the end of an execution arm of a numerical control device. The execution arm moves along a predetermined trajectory on the inner surface of a large wind turbine blade mold (50) under the control of the control system.

[0016] The detailed structure of the core of the present invention - the detection head (10) - includes: An image acquisition unit (11), such as an industrial CCD or CMOS camera, whose optical axis (L) of the lens (12) is approximately perpendicular to (i.e. along the normal direction) the local surface of the mold (50) during operation; A ring-shaped light source group (13) composed of multiple (e.g. 8 or 16) independent light sources (14) (preferably high-brightness LEDs) uniformly distributed in a ring array around the camera lens (12); All light sources (14) are configured to illuminate the central field of view area (F) of the mold surface (50) at a large oblique incidence angle (a). The oblique incidence angle (a) (referring to the angle between the light ray and the surface normal L) is preferably between 60 degrees and 85 degrees; The detection head (10) is built-in or externally connected with a control and processing unit (such as FPGA, DSP or industrial computer), which controls the on-off of each light source (14) in the light source group (13) through circuit connection, and synchronously receives image data from the image acquisition unit (11).

[0017] The detection principle of the present invention is based on structured light dark field illumination.

[0018] As shown in Figure 2 When the light ray (G1) emitted by any oblique light source (14) in the ring-shaped light source group (13) illuminates the flat surface (S1) of the mold (50), due to the large incidence angle (a), its main reflected light ray (R1) (whether specular reflection or diffuse reflection) will be reflected in a direction away from the normal (L). Therefore, the image acquisition unit (11) arranged along the normal (L) receives very weak light intensity, and the image presents a dark background. When the light ray (G2) illuminates a convex defect (such as a solidified release agent particle) on the surface of the mold (50), the normal (L2) of the local surface (S2) is deflected. This causes the local incidence angle (b) of the oblique incident light (G2) at the local surface (S2) to be much smaller than (a). Therefore, a large part of the reflected light ray (R2) will be concentrated in the normal (L) direction, which is captured by the image acquisition unit (11) and forms a high-brightness pixel area.

[0019] In short, the flat surface is a dark field, and the local defect surface is a bright field, achieving high signal-to-noise ratio detection of defects.

[0020] Figure 1 The complete flow of the method is shown. The method is automatically executed by the control and processing unit on the CNC system.

[0021] Step 501: The CNC system moves the probe (10) to the surface of the mold (50) to be measured. The control system ensures that the normal line (L) of the probe (10) is aligned with the normal line of the current local surface according to the CAD model of the mold or the pre-scanned data.

[0022] Step 502: Initialization. The control and processing unit initializes a light source index i = 1, and creates a blank "defect composite map" M composite (for example, a completely black image).

[0023] Step 503: Activate the light source. The control unit turns on the i th light source (14) in the ring-shaped light source group (13).

[0024] Step 504: Capture the image. Synchronously trigger the image acquisition unit (11) to take a surface reflection image I i .

[0025] Step 505: Image processing and extraction (the focus of the invention).

[0026] 505a (preprocessing): Preprocess the collected raw image I i , such as using Gaussian filtering or median filtering, to remove random noise that may be introduced by the CMOS / CCD sensor.

[0027] 505b (threshold segmentation): Since the defects appear as highlights against a high-contrast background, set a global or adaptive grayscale threshold T. Convert I i to a binary image B i : all pixel points with a grayscale value greater than T (highlight area) are set to 1 (white), and the rest are set to 0 (black).

[0028] 505c (morphological processing): (optional) Perform morphological "opening operation" (erosion first, then dilation) on the binary image B i to eliminate isolated noise points, and then perform morphological "closing operation" to connect disconnected defect outlines.

[0029] 505d (contour extraction): B i at this time is the local contour map of all defects facing the side of the i th light source under the illumination of the light source. For example, Figure 3If the light source is on the left, the left bright edge of the defect is extracted.

[0030] Step 506: Image synthesis.

[0031] The control and processing unit will synthesize the current obtained binary local profile map B i into the total defect synthesis map M composite . The simplest way is to use pixel-level logical OR operation: M composite = M composite OR B i .

[0032] This means that as long as a pixel point is detected as a defect (white) in any one angle image, it will be marked as a defect (white) in the synthesis map.

[0033] Step 507: Loop judgment.

[0034] Light source index i is incremented i = i + 1.

[0035] Determine i whether it is greater than the total number of light sources N (for example, N = 8). If not, return to step 503, turn on the next light source and repeat the collection and processing (as Figure 3 shown, turn on the upper light source and extract the upper edge of the defect).

[0036] Step 508: Generate defect map.

[0037] When all N light sources have been turned on in a loop (N i >N), the loop ends.

[0038] At this time, the M composite image contains the complete profile of all defects synthesized from the local profiles of all directions. This image is the final "defect map".

[0039] Step 509: Height information inference (optional but preferred).

[0040] In step 506, in addition to binary synthesis, the processing unit can also create a "height map" H.

[0041] For each pixel (x, y) on H, its value can be defined as the maximum gray value of all N original gray images (I1 to I N ) at that pixel point: H(x, y) = max(I1(x, y), I2(x, y),..., IN(x, y). N(x, y) ).

[0042] Since the higher the height of the protrusion, the greater the slope of its edge, the stronger the light intensity reflected back to the camera. Therefore, the brightness of a pixel in the H image (a gray-scale image) is positively correlated with the height or steepness of the defect at that point.

[0043] Step 510: Feedback and polishing.

[0044] The control and processing unit sends the final defect map M composite (containing the XY position and profile of the defect) and (optionally) the height map H (containing the Z-direction information of the defect) to the control system.

[0045] Step 511: According to the map, the control system plans the motion trajectory of the polishing head, so that it only performs accurate and local polishing on the white areas (i.e. the areas with defects) in M composite , and completely skips the flat dark areas, thereby achieving efficient, accurate and non-damaging automatic polishing of the mold.

[0046] Those skilled in the art should understand that the above embodiments are only to illustrate the present application, not to limit the present application, and any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A system for detecting protrusion defects on the large curvature surface of wind turbine blade molds based on multi-angle oblique illumination, characterized in that, include: An image acquisition unit is configured to have its optical axis set approximately along the local normal direction of the large curvature surface, for acquiring images of reflected light from the surface; At least two light sources are arranged in an array around the image acquisition unit, and each light source is configured to illuminate the surface at an oblique angle; A control and processing unit, electrically connected to the image acquisition unit and the at least two light sources, is configured to: (a) Control the at least two light sources to be lit sequentially and one by one; (b) When each light source is lit, the image acquisition unit is simultaneously triggered to capture a surface reflection image; (c) Extract the bright areas formed by strong reflections from the protruding defects in each of the captured images; (d) Synthesize the highlighted areas from all images to generate a defect map that reflects the complete outline of the raised defect.

2. The system according to claim 1, characterized in that: The at least two light sources are evenly distributed in a ring array.

3. The system according to claim 1, characterized in that: The oblique angle, that is, the angle between the light source ray and the local normal direction, is between 60 degrees and 85 degrees.

4. The system according to claim 1, characterized in that: The control and processing unit is further configured to: (e) Estimate the height distribution of the protrusion defect based on the pixel grayscale intensity within the bright area.

5. The system according to claim 1, characterized in that: The large curvature surface is the inner surface of the wind turbine blade forming mold.

6. A method for detecting protrusion defects on the large curvature surface of wind turbine blade molds based on multi-angle oblique illumination, characterized in that, Includes the following steps: (S1) Position a detection system, which includes a normal image acquisition unit and at least two oblique light sources, above a local area of ​​the large curvature surface; (S2) Illuminate the at least two oblique light sources one by one in sequence; (S3) During each period when the light source is lit, the image acquisition unit acquires a surface reflection image of the local area, and a total of multiple images are obtained; (S4) Process each of the multiple images to extract the bright areas formed by strong reflection from the protruding defects in the image, and obtain multiple local contour maps; (S5) Synthesize the multiple local contour maps to generate a defect map that reflects the complete contour of all protruding defects in the local area.

7. The method according to claim 6, characterized in that: The process described in step (S4) includes: (S4.1) Perform grayscale thresholding on the image to generate a binary image, wherein the bright area is the foreground; (S4.2) (Optional) Perform morphological operations on the binary image to remove noise or broken contours.

8. The method according to claim 6, characterized in that... The synthesis process described in step (S5) includes performing a pixel-level logical "OR" operation on the plurality of local contour maps (binary images).

9. The method according to claim 6, characterized in that, It also includes the following steps: (S6) Based on the pixel grayscale intensity in the bright area, infer the height distribution of the protrusion defect.

10. The method according to claim 6 or 9, characterized in that, It also includes the following steps: (S7) Send the defect map and / or the height distribution information to the controller of a robotic arm grinding system; (S8) Control the robotic arm grinding system to grind the protruding defect according to the information.

Images