CN101566461B - Method for quickly measuring blade of large-sized water turbine - Google Patents

Abstract

The invention discloses a method for quickly measuring a blade of a large-sized water turbine, which is characterized by comprising the following steps: firstly, sticking a mark point to the surface of the blade, then shooting a group of photos from different angles as measured primary data, and inputting the data to a computer to carry out analytic processing; by carrying out image detection for the group of photos in a measurement software, identifying the mark point in each photo and positioning the center of the mark point, and matching the mark points with the same name; finally, reconstructing a three-dimensional coordinate of corresponding object points according to a plurality of two-dimensional coordinates of the mark points, fitting a sparse framework model of a target by the reconstructed three-dimensional mark points, then scanning the surface of the blade block by block by using a binocular grating scanning method to acquire a local dense point cloud, and aligning the dense point cloud to an overall coordinate system according to the local mark points to acquire a dense point model of the blade; and by comparing the dense point model with a CAD designed model after aligning, calculating a three-dimensional processing error of the surface of a workpiece.

Description

Rapid Measuring Method for Large Water Turbine Blades

technical field

The invention relates to a method for rapidly measuring blades of large water turbines, in particular to a method for rapidly measuring the size and profile of blades of medium and large water turbines using visual measurement technology.

Background technique

For expensive large hydro turbine units, the blade is the "heart" of the hydro turbine, and its surface quality and manufacturing precision will directly affect the hydraulic performance, power generation efficiency, operation stability and service life of the entire unit. However, since the profile of the large water turbine blade is a complex free-form surface, its volume and weight are huge, so in the process of manufacturing and processing, the full-scale detection of the overall size and profile of the blade has always been a difficult problem in the industry.

The literature "Research Status and Development Trends of Hydraulic Turbine Runner Maintenance Equipment" (He Qinghua et al., Large Motor Technology, 2003) summarizes several main methods in the current large-scale hydraulic turbine blade inspection. According to the equipment used, there are three-dimensional model method , template method, comb method, coordinate measuring machine method, mechanical arm method, laser interferometer method, photoelectric pathodolite method, laser tracker method, light section method and underwater photography method, etc. The combined template used in the three-dimensional template method is hand-made, has poor rigidity and is easy to deform, and has been gradually eliminated. Although the three-coordinate measuring machine has high precision, it has high requirements on the measurement environment, is not portable, and has a small measurement range; especially the giant three-coordinate measuring machine is extremely expensive, and it is difficult to place and adjust the turbine blades on the three-coordinate machine, which requires time-consuming work in the early stage Establishment of detection coordinate system and measurement programming. The articulated arm, laser tracker, and theodolite can perform high-precision measurement on site, but due to the point-by-point measurement method, the measurement efficiency is low, and it cannot meet the dense point cloud collection requirements required for full-scale reverse design and production inspection. Therefore, the current research hotspots are focused on the non-contact optical measurement method that integrates computer vision technology and photogrammetry technology. Measurement efficiency is a feasible solution to solve the problem of 3D full-scale inspection of medium and large workpieces.

A detection system based on computer vision methods refers to a detection system that uses a CCD camera as an image sensor and comprehensively uses image processing, precision measurement and other technologies for non-contact two-dimensional or three-dimensional coordinate measurement. The document "Visual Measurement Technology and Application" (Academician Ye Shenghua, China Engineering Science, 1999) summarizes the different directions of development of visual measurement technology and its application fields, and proposes several prototype systems of visual inspection. However, due to the limitations of computer computing speed and image acquisition accuracy, no actual detection system has been applied.

In recent years, due to the continuous development of computer technology and optoelectronic technology related to optics, digital image processing, and computer vision theory, it has become possible to study 3D measurement of large and complex surfaces based on optics, digital image processing, and computer vision theory. Therefore, it is of great theoretical significance and great practical value to study a new portable and rapid three-dimensional measurement method for large and complex surfaces based on the fusion of optics, digital images and visual information, and to standardize the detection steps and data processing.

Contents of the invention

Aiming at the detection of the shape and size of large turbine blades during processing, how to improve the speed of detection and reduce the cost of measurement under the premise of ensuring that the accuracy meets the requirements, how to realize on-site detection and avoid vibration, temperature changes, Humidity, dust and other interference. The invention proposes an optical measurement method based on the principle of photogrammetry and stereo vision, and through experimental verification, it can quickly measure the size and profile of large water turbine blades on site.

To achieve the above object, the present invention is achieved by taking the following technical solutions:

A method for quickly measuring blades of large water turbines, comprising the following steps:

The first step is to place circular markers and markers with ring codes on the surface of the blade and its surroundings. The markers include coded points and non-coded points; paste the markers on the magnetic rubber pad and absorb them on the surface of the blade , according to the requirements of the dense point cloud acquisition equipment behind the sticky layout;

The second step is to place a global ruler, which is the basis for restoring the actual size of the blade. The two ends of the ruler are fixed coded or non-coded markers, and the distance between the markers at both ends is strictly calibrated;

The third step is to use a professional digital camera to take photos of the measured blade in all directions, and obtain a set of photos including coded marker points, non-coded marker points, and global scale information;

The fourth step is to calculate the coordinates of the marker points, run the measurement software, import the photo group into the computer, perform image processing, measure the two-dimensional coordinates of the marker points in the photo group, and then calculate the space of all the marker points according to the two-dimensional coordinates. The three-dimensional coordinates in the result, the non-coded landmark point cloud in the result constitutes the global sparse point model of the measured blade, which is used as the basis for global coordinate alignment after the local dense point cloud collection in the next step;

The fifth step is to collect and align dense point clouds on the surface of the measured blade. Using the binocular raster scanning method, two cameras are used to simultaneously capture the coded stripes projected from the projector to the surface of the object. The measurement software automatically performs stereo matching and 3D reconstruction to obtain the blade. The global coordinate alignment of the local dense point cloud is completed according to the position of the local landmark point and the dense point cloud on the surface of the global sparse point model; the above-mentioned binocular raster scanning method is used repeatedly to obtain all the local coordinates aligned with the global coordinates. Dense point cloud.

The sixth step is dense point cloud preprocessing and modeling. Since the obtained blade dense point cloud model has a certain amount of data redundancy, it should be processed before modeling, and the overlapping part of the point cloud will be deleted, and each part of the dense point cloud will be The point clouds are merged to obtain the overall dense point cloud model of the outer surface of the blade, that is, the blade measurement model;

The seventh step is to align the blade measurement model with the coordinate system of the blade CAD model, and perform coordinate transformation on the measurement model;

The eighth step is to compare the blade measurement model with the blade CAD model, and perform a simple vector subtraction operation on the two models to obtain the deviation of each position of the blade casting surface from the design model, which is realized by programming and displayed in 3D, or With the help of commercial comparison software, 2D deviation data is generated according to the profile diagrams of various positions and directions.

In the above method, the third step is to take pictures of the measured blade in all directions, and set the position of the camera station to adopt three successively increasing heights, each of which surrounds 360 degrees, and the shooting requirements are as follows:

1) Take the maximum value in the three-dimensional ruler of the blade as the shooting distance and shoot upright.

2) Try to make each marker point be photographed by more than 4 camera stations in different positions.

3) Try to make the intersection angle of each marker point between 60° and 120°.

4) Make the incident angle of each marker point smaller than 45° as much as possible.

5) Try to make each camera station capture as many landmarks as possible, and keep more than 60% overlap and common points between the photos taken by adjacent camera stations.

The two-dimensional coordinates of the marker points in the measurement photo group of the fourth step, the specific steps are as follows:

1) Use the Canny algorithm to detect the edges in the image to obtain a closed edge set with a single pixel width;

2) Use the gradient amplitude as the weight to calculate the weighted value of the position along the gradient direction, and perform sub-pixel level correction on the edge position along the gradient direction:

$\mathrm{<span\; class="notranslate">\; \&delta;d</span>}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; ,</span>$

Among them, d _i is the distance between a pixel and the detected edge point along the gradient direction, g _i is the gradient magnitude;

3) Use the circularity criterion to identify the ellipses in the edge concentration, and use the following prior information to remove unqualified ellipses, including the area of the ellipse is too small, the ellipse outline is concave, non-closed, and the distance between the ellipse and the adjacent ellipse is too close;

4) The center of the ellipse is fitted twice using the least squares method, and the edges whose distance is greater than 3 times the standard deviation are removed after the first fitting, and then the second fitting is performed;

5) Determine whether there is a ring on the periphery of the ellipse, if not, it is a non-coded point; if there is, make a radial connection between the inner and outer boundaries, and sample 5 times at equal distances on the connection, and use the median value of the 5 samples to match the marked point If it is greater than the threshold, the code of this ring is 1, otherwise it is 0; repeat the above operation every 36° (corresponding to 10-bit code points), and after one rotation, a binary format like "0100100111" is obtained serial number;

6) Look up the table to obtain the ID of the code point, if it cannot be found, it is regarded as a non-code point.

Compared with the three-dimensional template method, laser tracker method, articulated arm method, etc. of the prior art, the present invention has the following advantages based on the visual measurement method:

1. Since CCD is used as the sensor, it is a non-contact measurement method, which will not damage the surface of the blade, and will not be affected by the shape and roughness of the blade surface, and will not damage the measurement equipment.

2. Due to the use of area-by-area scanning and automatic alignment by the software, the measurement time mainly depends on the number of partitions on the blade surface. For large blades, multiple scanning devices can also be used to work in parallel. Compared with the point-by-point measurement method, the on-site measurement efficiency is higher. , the later point cloud processing and model comparison can be processed off-site.

3. Due to the dense point cloud scanning method, it is especially suitable for the task of measuring many points on the surface of the blade. The measurement result is a complete model of the blade represented by the dense point cloud, and any geometric quantity can be measured on this basis.

4. Since the measurement result is a complete triangular surface model of the blade, with the help of the comparison software, it can automatically perform full-scale three-dimensional chromatographic analysis and arbitrarily customize the deviation tolerance for qualification judgment, and the test results are intuitive and fast.

5. The blade point cloud model as the intermediate result of the measurement can be used as the original data for reverse engineering design.

6. Since the CCD is used as the sensor, the measuring equipment used in the method is cheap and the measuring cost is low.

Description of drawings

Fig. 1 is a diagram of a blade measurement and placement base of the present invention.

Fig. 2 is a diagram of a reference scheme of non-coded marker points and coded marker points.

Figure 3 is a photo of the tested blade after pasting the marker points.

Fig. 4 is a diagram of a sparse model of a leaf with three-dimensional marker points generated by a computer.

Fig. 5 is a schematic diagram of the binocular raster scanning acquisition system.

Fig. 6 is a photo of the blade model collected and aligned with dense point clouds on the surface piece by piece.

Figure 7 is a photo of the blade model after dense point cloud preprocessing and modeling.

Figure 8 is a 3D comparison photo of the blade measurement model and the design model.

Figure 9 is a 2D frontal floc photo of a certain section deviation of the blade.

Fig. 10 is a 2D side flock photo of the deviation of the blade section in Fig. 9 .

Fig. 11 is a schematic diagram of the scheme of encoding mark point encoding loop.

Fig. 12 is an arrangement diagram of camera stations.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments.

A rapid detection method for the size and profile of large turbine blades:

The placement of the turbine blades. Due to the large volume of the turbine blades and the thin wall thickness at some positions, the free placement will bring certain deformation and affect the measurement results. Visual measurements, on the other hand, require taking pictures of the measured object from different positions, so the placement of the blades needs to be considered. Follow the principles: first, it is convenient for taking pictures; second, it will not cause deformation, so it is convenient for on-site operation and improves efficiency. The blade measuring base as shown in Figure 1 can be processed. The base can be reused.

Step 1: Arrange marking points on and around the blade surface. Marker points include coded points and non-coded points; high positioning accuracy can be obtained by using circular marker points. Marking points with ring codes can easily identify their numbers, and choose the appropriate marking point pattern and material. Since the blades are mostly made of cast steel, the marking points can be pasted on the magnetic rubber pad for easy arrangement and removal. And can be used repeatedly. as shown in picture 2.

The artificial marker point will have obvious features, and the center of the marker point in the present invention is a solid circle, which is transformed into an ellipse through projection, and is used to locate the marker point; the inside of the coded marker is a non-coded marker, and there is a circle with the center outside Concentric fan rings, as shown in Figure 11, according to different coding rules, the coding signs are divided into 8-bit, 10-bit, 12-bit, 14-bit, 15-bit and other codes, and the n-bit code means that the outer ring Divide into n equal parts, if the color of each bit is the same as the color of the center circle, it is coded as 1, otherwise it is 0. In addition, according to the different colors, coded signs and non-coded signs can be divided into black dot marks on white background and white dot marks on black background. The scale is the basis for recovering the actual size of the blade. Its two ends are fixed coded mark points or non-coded mark points. The distance between the mark points at both ends has been strictly calibrated and the influence of temperature is considered. Before arranging the marking points and scales, the blade surface and its surroundings should be cleaned to remove iron filings and oil stains, so as to ensure that the marking points and scales will not fall or shift.

As shown in Figure 3, non-coded points are evenly pasted on the surface of the blade, and the density of the points should match the single acquisition format of the binocular raster scanning equipment to ensure that there are at least 3 to 5 non-coded points in each scanning area; The coding points are evenly pasted on the surface of the blade, and the density is 1/5 to 1/3 of the density of non-coding points; a number of coding points are placed around the blade to make all the marking points form a spatial distribution.

Step 2: Place rulers around the leaves that are equivalent to the size of the leaves, try to make sure there are rulers in 3 dimensions.

Step 3: Take a sequence of photos of global marker points. Take photos from various horizontal angles around the blades, and the position setting of the camera station can refer to Figure 12. According to the size of the blade to be measured, the camera stations are distributed at three horizontal heights. For example, for a blade with a height of 2m, three horizontal heights of 1m, 2m, and 3m can be selected; each height is rotated 360 degrees to set up the camera station; in order to ensure the three-dimensional coordinates of the marker points For the calculation accuracy, in addition to choosing a professional camera with high resolution and small lens distortion, the principles of photo group shooting are as follows:

3.1 Take the maximum value in the three-dimensional ruler of the blade as the shooting distance and shoot upright.

3.2 Try to make each marker point be photographed by more than 4 camera stations in different positions.

3.3 Try to make the intersection angle of each marker point between 60° and 120°.

3.4 Try to make the incident angle of each marker point less than 45°.

3.5 Try to make each camera station capture as many landmarks as possible, and keep more than 60% overlap and common points between the photos taken by adjacent camera stations.

Step 4: Measure the coordinates of the marker points in the photo group. Perform two-dimensional image processing on the obtained photo group first, identify the marker points and codes in each photo, and measure the center coordinates of each marker point. The specific algorithm is as follows:

4.1 Use the Canny algorithm to detect the edges in the image, and obtain a closed edge set with a single pixel width;

4.2 Use the gradient magnitude as the weight to calculate the weighted value of the position along the gradient direction, and perform sub-pixel level correction on the edge position along the gradient direction:

$\mathrm{<span\; class="notranslate">\; \&delta;d</span>}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; ,</span>$

Among them, d _i is the distance between a pixel and the detected edge point along the gradient direction, and g _i is the gradient magnitude.

4.3 Use the circularity criterion to identify the ellipses in the edge concentration, and use the following prior information to remove unqualified ellipses, including the area of the ellipse is too small, the outline of the ellipse is concave, non-closed, and the distance to the adjacent ellipse is too close.

4.4 Use least squares to fit the center of the ellipse twice, remove the edges whose distance is greater than 3 times the standard deviation after the first fitting, and then perform the second fitting.

4.5 Determine whether there is a ring on the periphery of the ellipse, if not, it is a non-coded point; if there is, make a radial connection between the inner and outer boundaries, sample 5 times at equal distances on the connection, and mark with the median of the 5 samples Point gray threshold for comparison, if it is greater than the threshold, the code of this ring is 1, otherwise it is 0; repeat the above operation every 36° (corresponding to 10-bit code point), and get a shape like "0100100111" after one rotation binary number.

4.6 Look up the table to get the ID of the code point. If not found, it is regarded as a non-coding point.

Step 5: Calculate the three-dimensional coordinates of the marker points and model them. The specific algorithm is as follows:

5.1 The coded markers are matched according to the unique ID number.

5.2 Realize the relative orientation between each photo according to the coplanar equation.

5.3 Use the DLT method to realize the absolute orientation of each photo.

5.4 Use bundle adjustment method to achieve high-precision calibration of the camera.

5.5 Use the forward intersection method to calculate the three-dimensional coordinates of the coded marker points.

5.6 Use the epipolar line constraints of multiple photos to realize the matching of non-coded landmark points.

5.7 Use the forward intersection method to calculate the three-dimensional coordinates of the non-coded marker points.

5.8 Use bundle adjustment to optimize all solution results.

5.9 Deriving all the three-dimensional non-coding marker points, thus obtaining the sparse model of the blade.

Step 6: Acquisition and alignment of local dense point clouds of blades.

The binocular raster scanning method is adopted, and its system principle is shown in Figure 5. The left and right cameras simultaneously shoot the coded stripes projected by the projector onto the surface of the object, and the measurement software automatically performs stereo matching and 3D reconstruction to obtain local landmarks and density on the blade surface. point cloud. The global coordinate alignment of the local dense point cloud is completed according to the position of the local landmark point in the global sparse point model.

Specific steps are as follows:

6.1 Export the blade model data represented by the non-coded point cloud from the photogrammetry software, and import it into the dense point cloud acquisition software as a reference for local dense point cloud alignment.

6.2 Remove the coded marking points pasted on the blade surface, and keep the non-coding marking points.

6.3 Put the binocular grating scanning device facing a certain area on the blade surface, and the left and right cameras take two partial photos. Since the relative positions and directions of the two cameras have been calibrated in advance, they are known data, so the local area can be calculated according to the spatial forward intersection method. The spatial relative positional relationship of the marker points.

6.4 Adjust the relative position between the binocular raster scanning device and the target, so that the calculated number of non-coding marker points is more than three. In the three-dimensional space, use the subgraph isomorphism algorithm to match the regional landmark points and the global key point cloud: the local landmark points are represented as "subgraphs", and the global landmark point cloud to be matched is represented as "big picture", in the "big picture" Search for isomorphic subgraphs.

6.5 After the matching is successful, calculate the rotation matrix R and translation matrix t from the region to the whole.

6.6 The projection device projects coded structured light to the blade part, and the left and right cameras take a sequence of photos synchronously. After image processing, stereo matching, and three-dimensional reconstruction, local dense point cloud data is generated. If the blade reflectivity is too poor, you need to spray developer before scanning.

6.7 The local dense point cloud data is transformed into the global coordinate system according to the rotation matrix and translation matrix [R|t] in the previous step.

6.8 Scan area by area to get the overall dense point cloud model.

6.9 Mobile raster scanning equipment, under the condition of ensuring no gaps, scans the surface of the blade area by area and automatically aligns it to obtain an overall dense point cloud model of the blade with certain data redundancy. See Figure 6.

Step 7: Point cloud preprocessing and triangulation modeling. The K-means clustering method is used to delete the overlapping surfaces of the multi-view point cloud and to process the data fusion. Use general reverse engineering software (such as Geomagic) to perform noise reduction, de-isolation, smoothing, and thinning on the leaf point cloud model, then perform triangulation to generate a triangular network model, and finally fill in the holes generated by the marker point coverage. As a result, the overall triangular network model of the blade is obtained, as shown in Figure 7.

Step 8: Align the coordinates of the measurement model with the CAD design model. If the blade has a processing datum plane during processing, then the design coordinate system of the blade can be determined based on more than three mutually perpendicular surfaces, and the measured mark point cloud model is based on "surface-line-point" or "3-2 -1” mode for coordinate transformation in order to align to the design coordinate system. This coordinate transformation should be completed after obtaining the point cloud model from the landmark points and before collecting the dense point cloud.

If the blade to be tested has no processing datum plane, the measurement model and design model can be read in by commercial comparison software (such as GeomagicQualify), and the "best global registration" command can be executed to complete the optimal matching of the two models.

Step 9: Comparison of the deviation between the blade measurement model and the design model.

After aligning the measurement model with the design model, program or use commercial comparison software (such as GeomagicQualify) to run the "3D comparison" command to obtain the 3D chromatographic deviation of the measurement data and the blade design model, as shown in Figure 8.

According to the process and inspection requirements, any position of interest can be cut, and the 2D color film deficit flocculus of each position of the blade can be easily obtained, as shown in Figure 9 and Figure 10.

Claims (3)

1. A method for quickly measuring a large water turbine blade, is characterized in that, comprises the following steps: The first step is to place circular mark points and mark points with ring codes on the surface of the blade and its surroundings. The mark points include code points and non-code points; paste the mark points on the magnetic rubber pad and absorb them on the surface of the blade. Paste and arrange at intervals according to the requirements of the dense point cloud acquisition equipment behind; The second step is to place a global ruler, which is the basis for restoring the actual size of the blade. The two ends of the ruler are fixed coded or non-coded markers, and the distance between the markers at both ends is strictly calibrated; The third step is to use a professional digital camera to take photos of the measured blade in all directions, and obtain a set of photos including coded marker points, non-coded marker points, and global scale information; The fourth step is to calculate the coordinates of the marker points, run the measurement software, import the photo group into the computer, perform image processing, measure the two-dimensional coordinates of the marker points in the photo group, and then calculate the space of all the marker points according to the two-dimensional coordinates. The three-dimensional coordinates in the result, the non-coded landmark point cloud in the result constitutes the global sparse point model of the measured blade, which is used as the basis for global coordinate alignment after the local dense point cloud collection in the next step; The fifth step is to collect and align dense point clouds on the surface of the measured blade. Using the binocular raster scanning method, two cameras are used to simultaneously capture the coded stripes projected from the projector to the surface of the object. The measurement software automatically performs stereo matching and 3D reconstruction to obtain the blade. The global coordinate alignment of the local dense point cloud is completed according to the position of the local landmark point and the dense point cloud on the surface of the global sparse point model; the above-mentioned binocular raster scanning method is used repeatedly to obtain all the local coordinates aligned with the global coordinates. Dense point cloud; The sixth step is dense point cloud preprocessing and modeling. Since the obtained blade dense point cloud model has certain data redundancy, it should be processed before modeling, and the overlapping part of the point cloud will be deleted, and each part of the dense point cloud will be The point clouds are merged to obtain the overall dense point cloud model of the outer surface of the blade, that is, the blade measurement model; The seventh step is to align the blade measurement model with the coordinate system of the blade CAD model, and perform coordinate transformation on the measurement model; The eighth step is to compare the blade measurement model with the blade CAD model, and perform a simple vector subtraction operation on the two models to obtain the deviation of each position of the blade casting surface from the design model, which is realized by programming and displayed in 3D, or With the help of commercial comparison software, 2D deviation data is generated according to the profile diagrams of various positions and directions. 2. The method for rapidly measuring blades of large water turbines as claimed in claim 1, characterized in that, the third step is to take pictures of the blades to be measured in all directions, and the position of the camera station is set to adopt three successively increasing horizontal heights, each height All surround 360 degrees, and the shooting requirements are as follows: 1) Take the maximum value in the three-dimensional ruler of the blade as the shooting distance and shoot upright; 2) Make each marker point be photographed by more than 4 camera stations in different positions; 3) Make the intersection angle of each marker point between 60° and 120°; 4) Make the incident angle of each marker point less than 45°; 5) Each camera station can take as many landmarks as possible, and keep more than 60% overlap and common points between the photos taken by adjacent camera stations. 3. the large water turbine blade fast measuring method as claimed in claim 1, is characterized in that, the two-dimensional coordinates of marker points in the measurement photo group of described 4th step, concrete method is as follows: 1) Use the Canny algorithm to detect the edges in the image to obtain a closed edge set with a single pixel width; 2) Use the gradient amplitude as the weight to calculate the weighted value of the position along the gradient direction, and perform sub-pixel level correction on the edge position along the gradient direction: $\mathrm{<span\; class="notranslate">\; \&delta;d</span>}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; ,</span>$ Among them, d _i is the distance between a pixel and the detected edge point along the gradient direction, g _i is the gradient magnitude; 3) Use the roundness criterion to identify the ellipses in the edge concentration, and use the following prior information to remove unqualified ellipses, including that the area of the ellipse is too small, the ellipse outline is concave and non-closed, and the distance between the ellipse and the adjacent ellipse is too close; 4) The center of the ellipse is fitted twice using the least squares method, and the edges whose distance is greater than 3 times the standard deviation are removed after the first fitting, and then the second fitting is performed; 5) Determine whether there is a ring on the periphery of the ellipse, if not, it is a non-coded point; if there is, make a radial connection between the inner and outer boundaries, and sample 5 times at equal distances on the connection, and use the median value of the 5 samples to match the marked point If it is greater than the threshold, the code of this ring is 1, otherwise it is 0; every 36° corresponds to a 10-bit code point, repeat the above operation, and get a binary format like "0100100111" after one rotation serial number; 6) Look up the table to obtain the ID of the code point, if it cannot be found, it is regarded as a non-code point.

Images