Rail end face flash grinding device and method based on 3D vision
Through the rail end face flush grinding device based on 3D vision, the automatic polishing of RGB-D cameras and industrial robots is used to solve the problems of low efficiency and safety hazards of rail end face flush grinding in the existing technology, achieving high precision and high efficiency Grinding effect.
Patent Information
- Application Number
- CN202411221179.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-11-08
AI Technical Summary
The existing rail end face flush grinding methods are inefficient and rely on manual labor, making it difficult to ensure accuracy and safety, especially long-term high-strength operation increases operation risks and error rates.
Using a rail end face flash polishing device based on 3D vision, the rail end face point cloud data is collected using an RGB-D camera, and the polishing path is calculated through algorithms. The industrial robot automatically polishes it to reduce manual participation and improve accuracy and efficiency.
It significantly improves the accuracy and efficiency of the end face of the rail, reduces the operating risks and error rates, reduces the demand for highly skilled workers, and improves the safety of the workplace.
Smart Images

Figure CN118905786A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of automatic rail grinding, and in particular to a rail end face flash grinding device and method based on 3D vision. Background Art
[0002] Rails play a vital role in the railway transportation system. The quality of their manufacturing and maintenance is directly related to the safety and efficiency of train operation. During the manufacturing process of rails, the rail ends often produce burrs, which are excess materials or burrs generated during the cutting or processing process. These burrs not only affect the performance of the rails, but also cause safety hazards during the connection and use of the rails. Therefore, burr grinding on the rail ends is an indispensable part of rail manufacturing and maintenance.
[0003] At present, the methods for grinding the flash on the end faces of rails mainly include manual grinding, semi-automatic grinding and grinding methods based on CAD models. The manual grinding method is inefficient and relies on the experience and skills of the operator. It is difficult to ensure that every flash is completely removed, resulting in unstable processing quality. In addition, long-term high-intensity manual grinding can easily cause operator fatigue, increase operational risks and error rates. Although semi-automatic grinding has improved the level of automation to a certain extent, the operation is still complicated and still relies on manual labor, and operators need to have higher skills. Although the grinding method based on CAD models can reduce manual participation and improve efficiency, due to the casting allowance between the actual rails and the CAD model, it is difficult to accurately correspond the shape of the flash on the end faces of the actual rails. Summary of the invention
[0004] According to the technical problem that the grinding method of the rail end face flash is not polished with high quality, a rail end face flash grinding device and method based on 3D vision is provided. The present invention can use 3D vision technology to guide industrial robots to automatically grind the flash of the end faces of rails of different models, effectively improving the grinding accuracy and efficiency, reducing the dependence on manual labor, and reducing the operation risk and error rate, and has good application prospects.
[0005] The technical means adopted by the present invention are as follows:
[0006] A rail end face flash grinding device based on 3D vision comprises a first conveyor belt arranged behind a lifting conveyor belt and a second conveyor belt arranged in front of the lifting conveyor belt, an automatic grinding device is arranged on the side of the lifting conveyor belt, first infrared sensors are arranged on both sides of the front end of the first conveyor belt, a first rail clamping device is arranged on the upper part of the front end of the first conveyor belt, second infrared sensors are arranged on both sides of the rear end of the second conveyor belt, a second rail clamping device is arranged on the upper part of the rear end of the second conveyor belt, and the target rail passes through the first conveyor belt, the lifting conveyor belt and the second conveyor belt in sequence.
[0007] Furthermore, the first rail clamping device and the second rail clamping device have the same structure, the first rail clamping device includes a metal plate, the metal plate is fixed under the bracket, a first micro linear guide is fixed to the upper surface of the front end of the metal plate through a guide rail slider, a second micro linear guide is fixed to the upper surface of the rear end of the metal plate through a guide rail slider, a first cylinder and a second cylinder are fixed to the middle of the upper surface of the metal plate, the piston rods of the first cylinder and the second cylinder are respectively connected to floating joints, the first clamping arm is connected to the floating joint of the first cylinder, the second clamping arm is connected to the floating joint of the second cylinder, the first clamping arm is fixedly connected to the guide rail slider, the second clamping arm is fixedly connected to the guide rail slider, the first cylinder drives the first clamping arm to slide on the first micro linear guide, and the second cylinder drives the second clamping arm to slide on the second micro linear guide, thereby clamping the target rail.
[0008] Furthermore, the automatic grinding device includes an industrial robot, the TCP end of the industrial robot is connected to the back of the triangular prism connector, an RGB-D camera is connected below the bottom of the triangular prism connector, a grinder is connected to the left side of the front of the triangular prism connector, and the RGB-D camera is used to collect point cloud data of the end face of the target rail; a pneumatic guide rail is connected to the right side of the triangular prism connector, and a slider on the pneumatic guide rail is connected to the right end of the metal baffle.
[0009] Furthermore, when the automatic grinding device grinds the end face of the target rail, the slider on the pneumatic guide rail drives the metal baffle to move downward to prevent the dust generated during the grinding process from affecting the camera lens. When the grinding is completed, the pneumatic guide rail drives the metal baffle to reset upward and prepare for the next point cloud data collection.
[0010] Furthermore, the first clamping arm is fixedly connected to the guide rail slider via a hexagon socket screw, and the second clamping arm is fixedly connected to the guide rail slider via a hexagon socket screw.
[0011] The present invention also provides a method for grinding rail end face burrs based on 3D vision, which is implemented based on any one of the above-mentioned rail end face burrs grinding devices based on 3D vision, and comprises the following steps:
[0012] S1, placing the target rail on the first conveyor belt. When the front end of the target rail is conveyed to the position of the first infrared sensor, the first conveyor belt stops conveying, the first rail clamping device clamps the rail, and the industrial robot enters the grinding area;
[0013] S2, collect the point cloud data of the front end face of the target rail through the RGB-D camera, and use the algorithm to calculate the grinding path of the rail end face flash;
[0014] S3, the industrial robot automatically grinds the flash of the end face of the target rail according to the grinding path of S2. After completing the grinding of the front end face of the target rail, the industrial robot resets and leaves the grinding area;
[0015] S4, the first rail clamping device releases the rail, the lifting conveyor belt rises to the same height as the first conveyor belt, the first conveyor belt and the lifting conveyor belt convey the target rail to the second conveyor belt, and when the end of the target rail leaves the position of the first infrared sensor, the first conveyor belt stops conveying;
[0016] S5. When the end of the target rail reaches the position of the second infrared sensor, the second conveyor belt stops conveying, the second rail clamping device clamps the target rail, and the lifting conveyor belt starts to descend;
[0017] S6, the industrial robot re-enters the grinding area, collects point cloud data of the end face of the target rail through the RGB-D camera, and uses the algorithm to calculate the grinding path of the flash of the end face of the rail;
[0018] S7, guiding the industrial robot to automatically grind the end surface of the target rail according to the grinding path of S6;
[0019] S8, grinding is completed, and the target rail is sent out from the front end of the second conveyor belt.
[0020] Furthermore, the method of collecting the target rail end face point cloud data and calculating the grinding path in S2 and S6 is the same, including the following steps:
[0021] S21, using an RGB-D camera to collect point cloud data of the end face of a standard rail, and preprocessing it to extract contour points of the end face of the preprocessed standard rail;
[0022] S22, using an RGB-D camera to collect point cloud data of the end surface of the rail to be polished with flash, and performing preprocessing;
[0023] S23, extracting FPFH feature points of the standard rail end face contour point cloud and the rail end face contour point cloud to be polished respectively, and using the SAC-IA algorithm to perform coarse registration on the standard rail end face contour point cloud and the rail end face contour point cloud to be polished;
[0024] S24, based on the rough registration in S23, further optimize the rough registration result by using the ICP algorithm to obtain an accurate registration matrix;
[0025] S25, transforming the contour points of the end face of the standard rail using the precise registration matrix to obtain the flash grinding points of the end face of the rail to be ground;
[0026] S26. Fitting the contour points using a cubic non-uniform B-spline curve to obtain a grinding path for the rail end face to be ground.
[0027] Furthermore, the preprocessing methods in S21 and S22 are the same, and include the following steps:
[0028] The point cloud is streamlined using a voxel grid downsampling algorithm;
[0029] Statistical filtering is used to denoise the streamlined point cloud, and straight-through filtering is used to remove irrelevant background;
[0030] Clustering segmentation is used to extract the point cloud data of the rail end surface, and bilateral filtering is used to further smooth the point cloud data of the rail end surface.
[0031] Furthermore, the process of extracting the pre-processed standard rail end face contour points in S21 is as follows:
[0032] S211, for any point P in the point cloud P of the standard rail end surface i , set the search radius r, and record the neighborhood points within the search radius r as set H(P i ), for each point P i , find its nearest k neighboring points, and calculate the average normal vector of these neighboring points as point P i The normal vector
[0033] S212, generate perpendicular to P i Point Normal The tangent plane β of the set H(P i ) are projected onto the tangent plane β, and the projected point set H(P i '), and then the neighboring points after projection are P i ' is a corner point, and each of them is connected to P in a counterclockwise order. i 'Connect to form a series of angles θ=(θ1,θ2,…,θ n ), find the largest angle θ among θ max ;
[0034] S213, set the angle threshold ε, if θ max >ε, then P i Add the boundary points to the set T;
[0035] S214. Repeat S211 to S213 until all boundary points are extracted.
[0036] Furthermore, in S23, the method of extracting FPFH feature points and performing coarse registration using the SAC-IA algorithm is as follows:
[0037] S231, use the ISS algorithm to extract the standard rail end face point cloud, and add the key points to the set G1, G1 = {g1, g2, ..., g n};
[0038] S232, construct a simplified point feature histogram SPFH of G1, and re-determine the k neighborhood points of each point in the set G1, and use the SPFH of the neighboring points to calculate the final histogram FPFH of each key point in G1:
[0039]
[0040] In the formula, w k is the weight, g k g i Neighboring points;
[0041] S233, calculate the Euclidean distance d between the FPFH value of the query point in its neighborhood and its average value FPFH :
[0042]
[0043] Among them, p i represents the FPFH value of the ith interval, and m i represents the average FPFH value of the ith interval;
[0044] S234, set the distance threshold d by FPFH Compared with the threshold d, when d FPFH When it is greater than the threshold, the point is regarded as an FPFH feature point and added to the set C1;
[0045] S235. Repeat S231-S234 for the point cloud data of the end face of the rail to be polished, extract the FPFH feature points of the point cloud of the end face of the rail to be polished, and add the FPFH feature points to the set C2.
[0046] S236. For each feature point in the standard rail end face point cloud C1, find the most similar feature point in the rail end face point cloud C2 to be polished, use the Euclidean distance as a similarity measure, and find the closest matching point pair.
[0047] S237. Randomly select a group of matching pairs as samples, estimate the rigid transformation matrix between the source point cloud and the target point cloud through the RANSAC (random sampling consensus) method, and use the Huber formula to determine whether it is the optimal transformation matrix:
[0048]
[0049] Where: a is the distance difference of the sample data point pair after transformation; ω d is the set distance threshold.
[0050] S238, repeat step S237 until the error function converges to a minimum value or reaches a preset maximum number of iterations. At this point, the transformation matrix obtained is the coarse registration transformation matrix
[0051] S239, using the coarse registration matrix to perform rigid body transformation on the standard rail point cloud to obtain the transformed standard rail point cloud P.
[0052] Compared with the prior art, the present invention has the following advantages:
[0053] The invention can reduce the chances of workers directly contacting grinding equipment and flash debris, thereby improving the safety of the workplace. At the same time, the device significantly improves grinding efficiency and reduces the need for highly skilled workers, thereby reducing labor costs.
[0054] The invention can adapt to rail end faces of different types and specifications, has good adaptability and flexibility, and can effectively ensure the accuracy and consistency during the grinding process. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0056] Figure 1 It is a schematic diagram of the overall structure of the present invention.
[0057] Figure 2 It is a schematic diagram of the rail clamping device of the present invention.
[0058] Figure 3 This is an exploded view of the rail clamping device of the present invention.
[0059] Figure 4 It is a schematic diagram of the automatic grinding device of the present invention.
[0060] Figure 5 The figure is a flow chart of the method of the present invention.
[0061] In the figure: 1, target rail; 2, first conveyor belt; 3, first rail clamping device; 4, first infrared sensor; 5, lifting conveyor belt; 6, automatic grinding device; 7, second infrared sensor; 8, second rail clamping device; 9, second conveyor belt;
[0062] 301. Bracket; 302. Metal plate; 303. First micro linear guide; 304. First micro linear guide; 305. Second micro linear guide; 306. Guide slider; 307. First cylinder; 308. Second cylinder; 309. Floating joint; 3010. First clamping arm; 3011. Second clamping arm; 3012. Hexagon socket screw; 601. Industrial robot; 602. Prism connector; 603. Pneumatic guide; 604. Metal baffle; 605. Grinder; 606. RGB-D camera. DETAILED DESCRIPTION
[0063] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0064] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0065] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0066] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, the numerical expressions and numerical values do not limit the scope of the present invention. At the same time, it should be clear that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The technology, methods and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, methods and equipment should be regarded as part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.
[0067] In the description of the present invention, it is necessary to understand that the directions or positional relationships indicated by directional words such as "front, back, up, down, left, right", "lateral, vertical, perpendicular, horizontal" and "top, bottom" are usually based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description. Unless otherwise specified, these directional words do not indicate or imply that the device or element referred to must have a specific direction or be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the scope of protection of the present invention: the directional words "inside and outside" refer to the inside and outside relative to the contours of each component itself.
[0068] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below their position devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.
[0069] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. If not otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of the present invention.
[0070] like Figure 1As shown, the present invention provides a rail end face flash grinding device based on 3D vision, comprising: a target rail 1, a first conveyor belt 2, a first rail clamping device 3, a first infrared sensor 4, a lifting conveyor belt 5, an automatic grinding device 6, a second infrared sensor 7, a second rail clamping device 8, and a second conveyor belt 9. The first infrared sensor 4 is fixed on both sides of the right end of the first conveyor belt 2; the first rail clamping device 3 is installed above the right end of the first conveyor belt 4. The second infrared sensor 7 is fixed on both sides of the left end of the second conveyor belt 9; the second rail clamping device 8 is installed above the left end of the second conveyor belt 9; the lifting conveyor belt 5 is placed between the first conveyor belt 2 and the second conveyor belt 9.
[0071] The target rail 1 is placed on the first conveyor belt 2. When the front end of the target rail 1 is conveyed to the position of the first infrared sensor 4, the first conveyor belt 2 stops conveying, and the first rail clamping device 3 clamps the rail. The industrial robot 601 in the automatic polishing device 6 enters the polishing area, collects the point cloud data of the front end face of the target rail 1 through the RGB-D camera 606, and calculates the polishing path of the rail end face burr using an algorithm. At the same time, the pneumatic guide rail 603 drives the metal baffle 604 to move downward to prevent the dust generated during the polishing process from affecting the camera lens. Subsequently, the industrial robot 601 automatically polishes the end face burr of the target rail 1. After completing the polishing of the front end face of the target rail 1, the industrial robot 601 resets and leaves the polishing area. At the same time, the pneumatic guide rail 603 drives the metal baffle 604 to reset upward. The first rail clamping device 3 releases the rail, and the lifting conveyor belt 5 rises to the same height as the first conveyor belt 2. The first conveyor belt 2 and the lifting conveyor belt 5 convey the target rail 1 to the second conveyor belt 9. When the end of the rail leaves the position of the first infrared sensor 4, the first conveyor belt 2 stops conveying. When the end of the rail reaches the position of the second infrared sensor 7, the second conveyor belt 9 stops conveying, the second rail clamping device 8 clamps the target rail 1, and the lifting conveyor belt 5 begins to descend. The industrial robot 601 re-enters the polishing area, and collects the end face point cloud data of the target rail 1 through the RGB-D camera 606, and uses the algorithm to calculate the polishing path of the end face flash of the rail. At the same time, the pneumatic guide rail 603 drives the metal baffle 604 to move downward. Then guide the industrial robot 601 to automatically polish the end face of the target rail 1.
[0072] like Figure 2 and Figure 3As shown, the first rail clamping device 3 mainly includes: a bracket 301, a metal plate 302, a first micro linear guide 303, a first micro linear guide 304, a second micro linear guide 305, a guide slider 306, a first cylinder 307, a second cylinder 308, a floating joint 309, a first clamping arm 3010, a second clamping arm 3011, and a hexagon socket screw 3012. The metal plate 302 is fixed below 301, and the first micro linear guide 303 and the first micro linear guide 304 are respectively fixed at the left and right ends of the metal plate 302. The first cylinder 307 and the second cylinder 308 are fixed in the middle of the metal plate 302, and the floating joint 309 is respectively connected to the piston rods of the first cylinder 307 and the second cylinder 308. The first clamping arm 3010 and the second clamping arm 3011 are respectively connected to the floating joints 309 on the first cylinder 307 and the second cylinder 308. At the same time, the first clamping arm 3010 and the second clamping arm 3011 are fixed together with the slider 306 on the first micro linear guide 304 and the second micro linear guide 305 by hexagon socket screws. The first cylinder 307 and the second cylinder 308 can drive the first clamping arm 3010 and the second clamping arm 3011 to clamp the rail, thereby maintaining the grinding stability.
[0073] The second rail clamping device 8 and the first rail clamping device 3 are the same device.
[0074] like Figure 4 As shown, the automatic grinding device 6 mainly includes: an industrial robot 601, a triangular prism connector 602, a pneumatic guide rail 603, a metal baffle 604, a grinder 605, and an RGB-D camera 606. The back of the triangular prism connector 602 is connected to the TCP end of the industrial robot; the RGB-D camera 606 and the grinder 606 are respectively installed below the bottom surface and on the left side of the front of the triangular prism connector 602; the RGB-D camera 606 is used to collect the end face point cloud data of the target rail 1; the pneumatic guide rail 603 is installed on the right side of the triangular prism connector 602; the right end of the metal baffle 604 is connected to the slider on the pneumatic guide rail 603; when the automatic grinding device 6 grinds the end face of the target rail 1, the slider on the pneumatic guide rail 603 drives the metal baffle 604 to move downward to prevent the dust generated during the grinding process from affecting the camera lens. When the grinding is completed, the pneumatic guide rail 603 drives the metal baffle 604 to reset upward and prepare for the next point cloud data collection.
[0075] The present invention also provides a method for grinding rail end face flash based on 3D vision, such as Figure 5 As shown, the following steps are included:
[0076] S1, placing the target rail on the first conveyor belt. When the front end of the target rail is conveyed to the position of the first infrared sensor, the first conveyor belt stops conveying, the first rail clamping device clamps the rail, and the industrial robot enters the grinding area;
[0077] S2, collect the point cloud data of the front end face of the target rail through the RGB-D camera, and use the algorithm to calculate the grinding path of the rail end face flash;
[0078] S3, the industrial robot automatically grinds the flash of the end face of the target rail according to the grinding path of S2. After completing the grinding of the front end face of the target rail, the industrial robot resets and leaves the grinding area;
[0079] S4, the first rail clamping device releases the rail, the lifting conveyor belt rises to the same height as the first conveyor belt, the first conveyor belt and the lifting conveyor belt convey the target rail to the second conveyor belt, and when the end of the target rail leaves the position of the first infrared sensor, the first conveyor belt stops conveying;
[0080] S5. When the end of the target rail reaches the position of the second infrared sensor, the second conveyor belt stops conveying, the second rail clamping device clamps the target rail, and the lifting conveyor belt starts to descend;
[0081] S6, the industrial robot re-enters the grinding area, collects point cloud data of the end face of the target rail through the RGB-D camera, and uses the algorithm to calculate the grinding path of the flash of the end face of the rail;
[0082] S7, guiding the industrial robot to automatically grind the end surface of the target rail according to the grinding path of S6;
[0083] S8, grinding is completed, and the target rail is sent out from the front end of the second conveyor belt.
[0084] The method of collecting the target rail end face point cloud data and calculating the grinding path in S2 and S6 is the same, including the following steps:
[0085] S21, using an RGB-D camera to collect point cloud data of the end face of a standard rail, and preprocessing it to extract contour points of the end face of the preprocessed standard rail;
[0086] The preprocessing method in S21 and S22 is the same, including the following steps:
[0087] The point cloud is streamlined using a voxel grid downsampling algorithm;
[0088] Statistical filtering is used to denoise the streamlined point cloud, and straight-through filtering is used to remove irrelevant background;
[0089] Clustering segmentation is used to extract the point cloud data of the rail end surface, and bilateral filtering is used to further smooth the point cloud data of the rail end surface.
[0090] The process of extracting the pre-processed standard rail end face contour points in S21 is as follows:
[0091] S211, for any point P in the point cloud P of the standard rail end surface i , set the search radius r, and record the neighborhood points within the search radius r as set H(P i ), for each point P i , find its nearest k neighboring points, and calculate the average normal vector of these neighboring points as point P i The normal vector
[0092] S212, generate perpendicular to P i Point Normal The tangent plane β of the set H(P i ) are projected onto the tangent plane β, and the projected point set H(P i '), and then the neighboring points after projection are P i ' is a corner point, and each of them is connected to P in a counterclockwise order. i 'Connect to form a series of angles θ=(θ1,θ2,…,θ n ), find the largest angle θ among θ max ;
[0093] S213, set the angle threshold ε, if θ max >ε, then P i Add the boundary points to the set T;
[0094] S214. Repeat S211 to S213 until all boundary points are extracted.
[0095] S22, using an RGB-D camera to collect point cloud data of the end surface of the rail to be polished with flash, and performing preprocessing;
[0096] S23, extracting FPFH feature points of the standard rail end face contour point cloud and the rail end face contour point cloud to be polished respectively, and using the SAC-IA algorithm to perform coarse registration on the standard rail end face contour point cloud and the rail end face contour point cloud to be polished;
[0097] The method of extracting FPFH feature points and using the SAC-IA algorithm to roughly align the standard rail end face contour point cloud and the rail end face contour point cloud to be polished is as follows:
[0098] S231, use the ISS algorithm to extract the standard rail end face point cloud, and add the key points to the set G1, G1 = {g1, g2, ..., g n};
[0099] S232, construct a simplified point feature histogram SPFH of G1, and re-determine the k neighborhood points of each point in the set G1, and use the SPFH of the neighboring points to calculate the final histogram FPFH of each key point in G1:
[0100]
[0101] In the formula, w k is the weight, g k g i Neighboring points;
[0102] S233, calculate the Euclidean distance d between the FPFH value of the query point in its neighborhood and its average value FPFH :
[0103]
[0104] Among them, p i represents the FPFH value of the ith interval, and m i represents the average FPFH value of the ith interval;
[0105] S234, set the distance threshold d by FPFH Compared with the threshold d, when d FPFH When it is greater than the threshold, the point is regarded as an FPFH feature point and added to the set C1;
[0106] S235. Repeat S231-S234 for the point cloud data of the end face of the rail to be polished, extract the FPFH feature points of the point cloud of the end face of the rail to be polished, and add the FPFH feature points to the set C2.
[0107] S236. For each feature point in the standard rail end face point cloud C1, find the most similar feature point in the rail end face point cloud C2 to be polished, use the Euclidean distance as a similarity measure, and find the closest matching point pair.
[0108] S237, randomly select a group of matching pairs as samples, estimate the rigid transformation matrix between the source point cloud and the target point cloud through the RANSAC (random sampling consensus) method, and use the Huber formula to determine whether it is the optimal transformation matrix:
[0109]
[0110] Where: a is the distance difference of the sample data point pair after transformation; ω d is the set distance threshold.
[0111] S238, repeat step S237 until the error function converges to a minimum value or reaches a preset maximum number of iterations. At this point, the transformation matrix obtained is the coarse registration transformation matrix
[0112] S239, using the coarse registration matrix to perform rigid body transformation on the standard rail point cloud to obtain the transformed standard rail point cloud P.
[0113] S24, based on the rough registration in S23, further optimize the rough registration result by using the ICP algorithm to obtain an accurate registration matrix;
[0114] The specific steps to further optimize the rough registration results using the ICP algorithm are as follows:
[0115] S241, perform closest point pair matching on the standard rail point cloud P after the rough registration matrix transformation and the rail end surface point cloud Q to be polished. i , find the point in Q that is closest to As its corresponding point, that is:
[0116]
[0117] S242, calculate the rigid transformation matrix T * , so that the distance error between corresponding points is minimized, that is
[0118]
[0119] S243, transform the matrix T * Act on the standard rail point cloud P after the rough registration matrix transformation and update it to P';
[0120] S244, calculate the distance error between P' and Q. If the error is less than the set value or the number of iterations is greater than the set value, the iteration ends. Otherwise, repeat the above steps until the convergence condition is met.
[0121] S25, transforming the contour points of the end face of the standard rail using the precise registration matrix to obtain the flash grinding points of the end face of the rail to be ground;
[0122] S26. Fitting the contour points using a cubic non-uniform B-spline curve to obtain a grinding path for the rail end face to be ground.
[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A rail end face flash grinding device based on 3D vision, characterized by: The invention comprises a first conveyor belt (2) arranged behind a lifting conveyor belt (5) and a second conveyor belt (9) arranged in front of the lifting conveyor belt (5); an automatic grinding device (6) is arranged on the side of the lifting conveyor belt (5); first infrared sensors (4) are arranged on both sides of the front end of the first conveyor belt (2); a first rail clamping device (3) is arranged on the upper part of the front end of the first conveyor belt (2); second infrared sensors (7) are arranged on both sides of the rear end of the second conveyor belt (9); a second rail clamping device (8) is arranged on the upper part of the rear end of the second conveyor belt (9); and a target rail (1) passes through the first conveyor belt (2), the lifting conveyor belt (5) and the second conveyor belt (9) in sequence.
2. The rail end face flash grinding device based on 3D vision according to claim 1 is characterized in that: The first rail clamping device (3) and the second rail clamping device (8) have the same structure. The first rail clamping device (3) comprises a metal plate (302). The metal plate (302) is fixed below the bracket (301). A first micro linear guide (303) is fixed to the upper surface of the front end of the metal plate (302) via a guide slider (306). A second micro linear guide (305) is fixed to the upper surface of the rear end of the metal plate (302) via a guide slider (306). A first cylinder (307) and a second cylinder (308) are fixed to the middle of the upper surface of the metal plate (302). The piston rods of the first cylinder (307) and the second cylinder (308) are respectively connected to floating joints (309). The arm (3010) is connected to the floating joint (309) of the first cylinder (307), the second clamping arm (3011) is connected to the floating joint (309) of the second cylinder (308), the first clamping arm (3010) is fixedly connected to the guide slider (306) of the first micro linear guide (304), the second clamping arm (3011) is fixedly connected to the guide slider (306) of the second micro linear guide (305), the first cylinder (307) drives the first clamping arm (3010) to slide on the first micro linear guide (303), and the second cylinder (308) drives the second clamping arm (3011) to slide on the second micro linear guide (305), thereby clamping the target rail (1).
3. The rail end face flash grinding device based on 3D vision according to claim 1, characterized in that: The automatic grinding device (6) comprises an industrial robot (601), wherein the TCP end of the industrial robot (601) is connected to the back of a triangular prism connector (602), an RGB-D camera (606) is connected below the bottom surface of the triangular prism connector (602), a grinder (605) is connected to the left side of the front of the triangular prism connector (602), and the RGB-D camera (606) is used to collect end face point cloud data of the target rail (1); a pneumatic guide rail (603) is connected to the right side of the triangular prism connector (602), and a slider on the pneumatic guide rail (603) is connected to the right end of a metal baffle (604).
4. The rail end face flash grinding device based on 3D vision according to claim 1, characterized in that: When the automatic grinding device (6) grinds the end face of the target rail (1), the slider on the pneumatic guide rail (603) drives the metal baffle (604) to move downward to prevent the dust generated during the grinding process from affecting the camera (606) lens. When the grinding is completed, the pneumatic guide rail (603) drives the metal baffle (604) to reset upward and prepare for the next point cloud data collection.
5. The rail end face flash grinding device based on 3D vision according to claim 2, characterized in that: The first clamping arm (3010) is fixedly connected to the guide rail slider (306) via a hexagon socket screw (3012), and the second clamping arm (3011) is fixedly connected to the guide rail slider (306) via a hexagon socket screw (3012).
6. A rail end face burr grinding method based on 3D vision, implemented based on the rail end face burr grinding device based on 3D vision as claimed in any one of claims 1 to 5, characterized in that: The steps include: S1, placing the target rail on the first conveyor belt. When the front end of the target rail is conveyed to the position of the first infrared sensor, the first conveyor belt stops conveying, the first rail clamping device clamps the rail, and the industrial robot enters the grinding area; S2, collect the point cloud data of the front end face of the target rail through the RGB-D camera, and use the algorithm to calculate the grinding path of the rail end face flash; S3, the industrial robot automatically grinds the flash of the end face of the target rail according to the grinding path of S2. After completing the grinding of the front end face of the target rail, the industrial robot resets and leaves the grinding area; S4, the first rail clamping device releases the rail, the lifting conveyor belt rises to the same height as the first conveyor belt, the first conveyor belt and the lifting conveyor belt convey the target rail to the second conveyor belt, and when the end of the target rail leaves the position of the first infrared sensor, the first conveyor belt stops conveying; S5. When the end of the target rail reaches the position of the second infrared sensor, the second conveyor belt stops conveying, the second rail clamping device clamps the target rail, and the lifting conveyor belt starts to descend; S6, the industrial robot re-enters the grinding area, collects point cloud data of the end face of the target rail through the RGB-D camera, and uses the algorithm to calculate the grinding path of the flash edge of the end face of the rail; S7, guiding the industrial robot to automatically grind the end surface of the target rail according to the grinding path of S6; S8, grinding is completed, and the target rail is sent out from the front end of the second conveyor belt.
7. The rail end face flash grinding method based on 3D vision according to claim 6, characterized in that: The method of collecting the target rail end face point cloud data and calculating the grinding path in S2 and S6 is the same, including the following steps: S21, using an RGB-D camera to collect point cloud data of the end face of a standard rail, and preprocessing it to extract contour points of the end face of the preprocessed standard rail; S22, using an RGB-D camera to collect point cloud data of the end surface of the rail to be polished with flash, and performing preprocessing; S23, extracting FPFH feature points of the standard rail end face contour point cloud and the rail end face contour point cloud to be polished respectively, and using the SAC-IA algorithm to perform coarse registration on the standard rail end face contour point cloud and the rail end face contour point cloud to be polished; S24, based on the rough registration in S23, further optimize the rough registration result by using the ICP algorithm to obtain an accurate registration matrix; S25, transforming the contour points of the end face of the standard rail using the precise registration matrix to obtain the flash grinding points of the end face of the rail to be ground; S26. Fitting the contour points using a cubic non-uniform B-spline curve to obtain a grinding path for the rail end face to be ground.
8. The rail end face flash grinding method based on 3D vision according to claim 7, characterized in that: The preprocessing method in S21 and S22 is the same, including the following steps: The point cloud is streamlined using a voxel grid downsampling algorithm; Statistical filtering is used to denoise the streamlined point cloud, and straight-through filtering is used to remove irrelevant background; Clustering segmentation is used to extract the point cloud data of the rail end surface, and bilateral filtering is used to further smooth the point cloud data of the rail end surface.
9. The rail end face flash grinding method based on 3D vision according to claim 7, characterized in that: The process of extracting the pre-processed standard rail end face contour points in S21 is as follows: S211, for any point P in the point cloud P of the standard rail end surface i , set the search radius r, and record the neighborhood points within the search radius r as set H(P i ), for each point P i , find its nearest k neighboring points, and calculate the average normal vector of these neighboring points as point P i The normal vector S212, generate perpendicular to P i Point Normal The tangent plane β of the set H(P i ) are projected onto the tangent plane β, and the projected point set H(P i '), and then the neighboring points after projection are P i ' is a corner point, and each of them is connected to P in a counterclockwise order. i 'Connect to form a series of angles θ=(θ1,θ2,…,θ n ), find the largest angle θ among θ max ; S213, set the angle threshold ε, if θ max >ε, then P i Add the boundary points to the set T; S214. Repeat S211 to S213 until all boundary points are extracted.
10. The rail end face flash grinding method based on 3D vision according to claim 7, characterized in that: In S23, the method of extracting FPFH feature points and using the SAC-IA algorithm for rough registration is as follows: S231, use the ISS algorithm to extract the point cloud of the standard rail end surface, and add the key points to the set G1, G1 = {g1, g2, ..., g n }; S232, construct a simplified point feature histogram SPFH of G1, and re-determine the k neighborhood points of each point in the set G1, and use the SPFH of the neighboring points to calculate the final histogram FPFH of each key point in G1: In the formula, w k is the weight, g k g i Neighboring points; S233, calculate the Euclidean distance d between the FPFH value of the query point in its neighborhood and its average value FPFH : Among them, p i represents the FPFH value of the ith interval, and m i represents the average FPFH value of the ith interval; S234, set the distance threshold d by FPFH Compared with the threshold d, when d FPFH When it is greater than the threshold, the point is regarded as an FPFH feature point and added to the set C1; S235. Repeat S231-S234 for the point cloud data of the end face of the rail to be polished, extract the FPFH feature points of the point cloud of the end face of the rail to be polished, and add the FPFH feature points to the set C2. S236, for each feature point in the standard rail end face point cloud C1, find the most similar feature point in the rail end face point cloud C2 to be polished, use the Euclidean distance as a similarity measure, and find the closest matching point pair; S237. Randomly select a group of matching pairs as samples, estimate the rigid transformation matrix between the source point cloud and the target point cloud through the RANSAC method, and use the Huber formula to determine whether it is the optimal transformation matrix: Where: a is the distance difference of the sample data point pair after transformation; ω d is the set distance threshold; S238, repeating step S237 until the error function converges to a minimum value or reaches a preset maximum number of iterations, at which point the obtained transformation matrix is the coarse registration transformation matrix; S239, using the coarse registration matrix to perform rigid body transformation on the standard rail point cloud to obtain the transformed standard rail point cloud P.
Citation Information
Patent Citations
Robot trajectory planning method for dynamic deslagging in ingot casting process
CN111014594A
Grinding machine capable of automatically adjusting grinding area according to surface smoothness of workpiece
CN111906609A
Workpiece polishing system and method based on 3D vision camera
CN112318320A
Bathroom cabinet visual intelligent spraying and color changing implementation method
CN115205314A
Target object contour point cloud extraction method and device, equipment and storage medium
CN116168056A
Cited By
Polishing device and method
CN120095685A
Training method of ovarian cancer lesion segmentation model
CN120374647A
