Multi-station robot photographing task distribution method for hub surface defect detection

Abstract

The invention discloses a multi-station robot photographing task allocation method for hub surface defect detection, and the method comprises the steps: forming a hub triangular patch model, and carrying out the Gaussian mapping of the hub triangular patch model; finding out the task area boundary of each station robot based on a simulated annealing algorithm, wherein the included angle between the normal vector of the triangular patch in the same task area and the Z axis is in a set range; generating a camera viewpoint of each station robot for completing task area shooting, and adjusting the viewpoint with collision into a collision-free viewpoint; planning a local optimal path passing through all collision-free camera viewpoints based on a greedy algorithm; and carrying out collision detection on the local optimal path, and re-planning the road section with collision. According to the multi-station wheel hub defect detection method, the task area corresponding to each station is determined according to the wheel hub curved surface features by adopting the simulated annealing algorithm, finally, automatic balanced distribution of the multi-station robot operation tasks is achieved, collision detection is conducted on the generated shooting path, and the reliability of multi-station wheel hub defect detection is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of defect detection, and more specifically, the present invention relates to a multi-station robot photographing task allocation method for detecting defects on the surface of a wheel hub. Background Art

[0002] As a load-bearing component of an automobile, the wheel hub is an important part of the automobile. For safety considerations, manufacturers have very strict requirements for the surface quality inspection of the wheel hub to avoid subtle defects damaging the safety of the product. However, the manufacturing process of the wheel hub is complex, and in each process, the wheel hub may face machine and manual damage, which makes the defect characteristics on the surface of the wheel hub complex and poses a great challenge to the surface quality inspection of the wheel hub.

[0003] In recent years, with the rapid development of industrial manufacturing, industrial robots have been able to replace humans to perform repetitive and labor-intensive operations such as handling, spraying, and welding, and can replace humans to work in high-temperature, dangerous, and harmful working environments, and have been widely used in the field of industrial automation. In the field of quality inspection automation, the cycle of a single robot to complete photographing and inspection is relatively long, which is difficult to meet the requirements of factories for efficiency in actual applications. It is often necessary to set up multi-station robots to complete the overall surface quality inspection task through the mutual cooperation between multi-station robots.

[0004] The previous task division methods for multiple robots mainly allocate tasks according to the production line layout and the relative position relationship between the workpiece and the robot. This allocation method can be used for simple curved surface workpieces, but for complex surface workpieces like wheel hubs, it will cause the posture of a single robot to change too much in its own working area, prone to singular points, and increase the risk of the robot colliding. Summary of the Invention

[0005] The present invention provides a multi-station robot photographing task allocation method for detecting defects on the surface of a wheel hub, and the method is as follows:

[0006] (1) Divide the curved surface model of the wheel hub to be measured into a series of triangular patches to form a wheel hub triangular patch model, and perform Gaussian mapping on the wheel hub triangular patch model to obtain a Gaussian sphere;

[0007] (2) Based on the simulated annealing algorithm, find the task area boundaries of each station robot, and the included angle between the normal vector of the triangular patches in the same task area and the Z-axis is within a set range;

[0008] (3) Generate camera viewpoints for each station robot to complete photographing in the task area, perform collision detection on all camera viewpoints, and adjust the viewpoints with collisions to collision-free viewpoints;

[0009] (4) Plan the local optimal path passing through all collision-free camera viewpoints based on the greedy algorithm;

[0010] (5) Conduct collision detection on the local optimal path, re-plan the sections with collisions, and complete the shooting path planning for all station cameras.

[0011] Furthermore, the number of task regions divided by the hub triangular patch model is the same as the number of station robots, and one station robot corresponds to one task region.

[0012] Furthermore, the process of determining the task region boundary based on the simulated annealing algorithm is as follows:

[0013] (21) Set the initial temperature T, the attenuation rate α, and the termination temperature T end ;

[0014] (22) Input the set composed of the normal vectors of the triangular patches on the hub model. The task region boundary X = [x 1 , x 2 , …, x n , where x n represents the boundary between task region n and task region n + 1, and randomly generate a set of initial solutions X initial ;

[0015] (23) The energy function E(X) of the region division under the current solution X cur is expressed as: E(X) = (S 1 - S 2 ) 2 + (S 1 - S 3 ) 2 + (S 2 - S 3 ) 2 + … (S m - S n ) 2 , where S n represents the total area of the triangular patches contained in region n;

[0016] (24) Randomly generate a set of new solutions X cur within the neighborhood range of the current solution X new , X new = (1 - δ) · X cur + δ · (rand · (X max - X min ) + X min ), where δ represents the step size, rand represents a random number, and X max , X min are the set maximum and minimum values;

[0017] (25) Judge the new solution X new and the current solution X cur to calculate the energy difference ΔE = E(X new ) - E(X cur ). If ΔE < 0, then accept the new solution X new . If ΔE > 0, then accept the current solution X -ΔE / T with a probability of e cur ;

[0018] (26) Decay the temperature according to T new = αT cur , where T cur is the current temperature, and repeat steps (24)-(26) until the temperature decays to the termination temperature T end or the interpolation between the new solution X new and the current solution X cur is less than the threshold ξ, and the current solution is the optimal solution.

[0019] Furthermore, the path planning method based on the greedy algorithm is specifically as follows:

[0020] (41) Take the origin p h of the robot as the current point p cur , and add it to the set P;

[0021] (42) Find the view point closest to the current point p cur in the view point set as the next point p next ;

[0022] (43) Set the found next point p next as the current point p cur , add it to the set P, and remove the point p next from the view point set, then return to step (42) until the view point set is empty, and set the origin of the robot as the path end point and add it to the set P.

[0023] Furthermore, the collision detection method for the local optimal path is specifically as follows:

[0024] (51) Take out two adjacent points P c and P n from the set P in sequence, and discretize the section P c P n between the two points with a step size of t to obtain multiple discrete points;

[0025] (52) Calculate the inverse solution of the robot at the discrete points in sequence to obtain the angles of each joint of the robot, and judge whether the robot collides. If a collision occurs, then search from the starting point P c to the end point P nFor the collision-free path, insert the searched intermediate points between two points P in set P c and P n to complete the update of path P c P n ;

[0026] (53) After traversing all adjacent pairs of points in set P, the path in set P is the collision-free path starting from the robot's origin, passing through all viewpoints in sequence, and returning to the robot's origin.

[0027] Furthermore, the process of obtaining the collision-free path from the starting point P c to the ending point P n is as follows:

[0028] (521) Set the maximum number of sampling points, the maximum connection distance ρ, and initialize the undirected graph G(V, E), where the vertex set V represents the set of collision-free sampling points, and the connection set E represents the collision-free paths between vertices;

[0029] (522) Randomly sample a collision-free sampling point P, add this sampling point P to the vertex set V, calculate the distances between the sampling point P and other vertices in the vertex set. If there exists a distance less than or equal to ρ, detect whether the connection line between the sampling point P and the corresponding point collides. If no collision occurs, add this connection line to the connection set E;

[0030] (523) Return to step (522) until the number of generated sampling points reaches the maximum number of sampling points, and complete the construction of the undirected graph G(V, E);

[0031] (524) Use the A* algorithm to search for the shortest path from the starting point P c to the ending point P n in the undirected graph G(V, E). The shortest path is the collision-free path from the starting point P c to the ending point P n .

[0032] Furthermore, use a hierarchical bounding box in the form of a top-down binary tree for collision detection;

[0033] Construct an AABB hierarchical bounding box for components that do not move relatively, including: workbench, workpiece, robot base;

[0034] Construct an OBB hierarchical bounding box for objects with relative motion, including: robot.

[0035] Furthermore, the construction process of the AABB hierarchical bounding box is as follows:

[0036] (61) Calculate the minimum and maximum values of the vertex coordinates of all triangular patches on the hub triangular patch model in the X, Y, and Z directions. Determine the 8 vertices of the bounding box based on the minimum and maximum values on the three coordinate axes, and construct an AABB bounding box containing the component model as the root node of the AABB hierarchical bounding box;

[0037] (62) Select the longest axis of the AABB bounding box corresponding to the current node as the splitting axis, and perform splitting with a plane passing through the center of the bounding box and perpendicular to the longest axis. Split the model corresponding to the current node into two parts, construct the AABB bounding boxes of the two parts respectively, and use these two parts as the two child nodes of the current node;

[0038] (63) Take these two nodes as the current node and return to step (62) until each triangular patch is divided into leaf nodes, and the AABB hierarchical bounding box is established.

[0039] Further, the construction process of the OBB bounding box is specifically as follows:

[0040] Calculate the total area S of all triangular patches within the robot triangular patch model;

[0041] Let the vertices of the triangular patch be u i 、v i 、w i , calculate the center coordinate c i of the triangular patch and the triangle area s i . The center coordinate c i = (u i + v i + w i ) / 3, and the triangle area s i = (u i - v i ) · (u i - w i ) / 2. Then the total area of all triangular patches within the robot triangular patch model where n is the number of triangular patches within the robot triangular patch model.

[0042] Calculate the area-weighted center point C i of the robot triangular patch model, and its specific calculation formula is as follows;

[0043]

[0044] Construct a covariance matrix, perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors, use the three unitized eigenvectors as coordinate axes, and use the area-weighted center point C iConstruct a new coordinate system with the origin of coordinates, and obtain the transformation matrix T of the new coordinate system relative to the model coordinate system where the triangular facet model is located. Calculate the AABB bounding box of the robot's triangular facet model in the new coordinate system, and transform it to the model coordinate system to obtain the position, orientation, and size of the OBB bounding box.

[0045] Further, the covariance matrix is specifically expressed as follows:

[0046]

[0047] In the formula,

[0048] The multi-station robot photographing task allocation method provided by the present invention takes into account the hub surface characteristics and also considers the task balance problem among multiple stations. The areas corresponding to the robots at each station are determined through the simulated annealing algorithm, and finally the automatic allocation of the operation tasks of the multi-station robots is realized. At the same time, collision detection is performed on the paths generated after the task allocation, which greatly improves the working efficiency and reliability of the multi-station hub defect detection and enhances the degree of industrial automation in hub defect detection. Brief Description of the Drawings

[0049] Figure 1 It is a flowchart of the station robot photographing task allocation method for hub surface defect detection provided by an embodiment of the present invention;

[0050] Figure 2 It is a schematic diagram of the process of obtaining a Gaussian sphere based on the hub surface model provided by an embodiment of the present invention. Among them, (a) is the surface model of the hub, (b) is the hub triangular facet model after triangular facet division, and (c) is the Gaussian sphere formed by Gaussian mapping of the hub triangular facet model;

[0051] Figure 3 It is a schematic diagram of the division of the three task areas S1, S2, and S3 corresponding to the three-station robots provided by an embodiment of the present invention;

[0052] Figure 4 It is a schematic diagram of the positions of the three task areas in the surface model of the hub provided by an embodiment of the present invention. Among them, (a) is the position of area S1 in the surface model of the hub, (b) is the position of area S2 in the surface model of the hub, and (c) is the position of area S3 in the surface model of the hub;

[0053] Figure 5 It is the division result of the task area boundary searched by the simulated annealing algorithm on the Gaussian sphere provided by an embodiment of the present invention;

[0054] Figure 6 It is the division result of the task area boundary on the hub triangular facet model provided by an embodiment of the present invention;

[0055] Figure 7 The variation curve of the energy function value of the simulated annealing algorithm provided by the embodiment of the present invention with the number of iterations;

[0056] Figure 8 It is a schematic diagram of the scenario of the three-station robot provided by the embodiment of the present invention. Specific embodiments

[0057] The following is a further detailed description of the specific embodiments of the present invention by referring to the accompanying drawings and describing the embodiments, so as to help those skilled in the art have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.

[0058] The robot photographing task allocation method of the present invention takes into account the hub surface characteristics and task balance, and can simultaneously perform collision detection and collision avoidance on the path generated after task allocation, improving the stability and reliability of the generated path, and realizing the automatic allocation of multi-station robot tasks to improve efficiency.

[0059] Figure 1 It is a flowchart of the photographing task allocation method for the station robot used for hub surface defect detection provided by the embodiment of the present invention. The method is as follows:

[0060] (1) Divide the surface model of the hub to be measured into a series of triangular patches to form a hub triangular patch model, and perform Gaussian mapping on the hub triangular patch model to obtain a Gaussian sphere;

[0061] The surface model of the hub is divided into a series of triangular patches to form a hub triangular patch model. The triangular patch model approximately represents the hub surface with triangular patches. Gaussian mapping is performed on the hub triangular patch model to fully represent the normal vector characteristics of the triangular surfaces in the hub triangular patch model. Figure 2 (a) is the surface model of the hub, Figure 2 (b) is the hub triangular patch model after triangular patch division, Figure 2 (c) is the Gaussian sphere formed by Gaussian mapping of the hub triangular patch model.

[0062] Gaussian mapping refers to a mapping from a surface in the Euclidean space R 3 to the unit sphere S 2 This unit sphere is called a Gaussian sphere. The specific process of Gaussian mapping is to move the starting point of the normal vector at each point of the surface to the center of the Gaussian sphere. Each normal vector will have an intersection point with the Gaussian sphere surface. The set of all these intersection points is called the Gaussian mapping of the surface.

[0063] (2) Based on the simulated annealing algorithm, determine the task area boundaries of each station robot. The angle between the normal vector of the triangular patches in the same task area and the Z-axis is within a set range;

[0064] Observation wheel Figure 2 As can be seen from the Gaussian sphere in (c), an important surface feature of the wheel hub is that the normal vectors of the triangular patches on the upper surface of the wheel hub are located in the upper hemisphere, which indicates that the orientation of the upper surface is upward, and the same is true for other wheel hubs. Therefore, the number of divided regions of the Gaussian sphere is determined according to the number of inspection stations (there is a station robot at each inspection station to collect the images of the wheel hub surface). If three-station robots are used for wheel hub defect detection, as Figure 8 shown, the Gaussian sphere can be divided into three regions to match the photographing tasks of the three station robots.

[0065] After determining the number of divided regions of the Gaussian sphere and the wheel hub triangular patch model according to the number of stations, it is necessary to determine a reasonable region division method from the perspective of avoiding collisions, so as to minimize the collision risks between the station robot and the workbench robot, between the station robot and the workpiece and surrounding equipment. By observing the shape of the wheel hub, it can be found that the wheel hub is a rotationally symmetric figure. Therefore, within the same region, the attitude of the viewing point should be as rotationally symmetric as possible. Therefore, when dividing the Gaussian sphere, ensure that each region is also rotationally symmetric. The division results of the wheel hub in the three-station robot workstation example are as follows Figure 3 shown. Figure 3 In, region S1 represents the set of triangular patches on the upper surface of the wheel hub with a small angle between the normal vector and the z-axis, region S3 represents the set of triangular patches on the upper surface of the wheel hub with a normal vector nearly perpendicular to the z-axis, region S2 represents the set of triangular patches with an angle between the normal vector and the z-axis between region S1 and region S3. The representative triangular patches of regions S1, S2, and S3 are as Figure 4 (a), 4(b), and the red highlighted regions in 4(c).

[0066] After determining the number of divided regions of the wheel hub according to the wheel hub surface normal vector characteristics and the number of stations, it is also necessary to determine the boundaries of each region, that is, to determine the specific triangular patch indexes and quantities in each region to facilitate subsequent viewing point planning. In order to balance the operation time of each station and avoid too long waiting time at a certain station, it is necessary to further determine the specific boundaries of each region according to the task balance strategy.

[0067] Determining the region boundary according to task balance belongs to the random search problem in the field of global optimization. The simulated annealing algorithm is a probability-based global optimization algorithm and has been widely used in engineering practice. The present invention uses the simulated annealing algorithm to solve the region boundary. The specific steps of the simulated annealing algorithm for searching the region boundary are as follows:

[0068] (21) Set the initial temperature T, attenuation rate α, and termination temperature T end ;

[0069] (22) Input the set composed of the normal vectors of the triangular patches on the hub model, set the regional boundary as the target to be solved by the algorithm, and define the form of the solution calculated by the algorithm as X = [x 1 , x 2 , …, x n , where x n represents the boundary between regions n and n + 1, and randomly generate a set of initial solutions X initial within the solution interval;

[0070] (23) The energy function E(X) of the regional division under the current solution X cur is expressed as: E(X) = (S 1 - S 2 ) 2 + (S 1 - S 3 ) 2 + (S 2 - S 3 ) 2 + … (S m - S n ) 2 , where S n represents the total area of the triangular patches contained in region n;

[0071] (24) Randomly generate a set of new solutions X cur within the neighborhood range of the current solution X new , X new = (1 - δ)·X cur + δ·(rand·(X max - X min ) + X min ), where δ represents the step size, rand represents a random number, and X max , X min are the set maximum and minimum values;

[0072] (25) Judge the energy difference ΔE = E(X new ) - E(X cur ) between the new solution X new and the current solution X cur . If ΔE < 0, then accept the new solution X new . If ΔE > 0, then accept the current solution X -ΔE / T with a probability of e cur ;

[0073] (26) Decay the temperature according to T new = αT cur , where T cur is the current temperature, and repeat steps (24) - (26) until the temperature decays to the termination temperature T end or the new solution Xnew The interpolation with the current solution X cur is less than the threshold ξ, the search is completed, and the current solution is output as the optimal solution.

[0074] The division result of the search area boundary on the Gaussian sphere by the simulated annealing algorithm is as Figure 5 shown, and the division result of the search area boundary on the hub triangular facet model by the simulated annealing algorithm is as Figure 6 shown. The change curve of the energy function value of the simulated annealing algorithm with the number of iterations is as Figure 7 shown.

[0075] (3) Generate the camera viewpoints for each station robot to complete the task area shooting, put them into the corresponding viewpoint set, and perform collision detection on all camera viewpoints in the viewpoint set;

[0076] Segment the task area obtained for each station. For each segmented area, establish a minimum bounding box. According to parameters such as the camera's field of view, depth of view, and viewing angle, establish a visible cone model of the camera. Use the visible cone model of the camera to constrain the bounding box, and generate initial viewpoints based on the position of the bounding box. Use the ray tracing algorithm to determine whether these viewpoints are occluded. For the occluded viewpoints, apply the line-of-sight offset strategy to update them to obtain a viewpoint set;

[0077] Perform collision detection on each viewpoint in the viewpoint set. If the viewpoint does not have collision interference, add the viewpoint to the final viewpoint set. If the viewpoint has collision interference, correct the viewpoint. The correction method is to rotate the coordinate system of the viewpoint around its z-axis with a certain search step until a pose without collision interference is found, and add this pose as the corrected pose of the viewpoint to the final viewpoint set.

[0078] (4) Plan the local optimal path passing through all collision-free camera viewpoints based on the greedy algorithm;

[0079] Since a tool coordinate system is established on the camera and the camera viewpoint is the robot pose, after obtaining the collision-free final viewpoint set for each station, apply the greedy algorithm to the viewpoint set of each station to search for: a driving path starting from the set robot origin, passing through all viewpoints and then returning to the robot origin. This path determines the order of the robot passing through all viewpoints in the viewpoint set. Among them, the path planning method based on the greedy algorithm is as follows:

[0080] (41) Take the robot origin p h as the current point p cur , and add it to the set P;

[0081] (42) Find the viewpoint closest to the current point p cur in the viewpoint set as the next point p next ;

[0082] (43) The next point p will be found next Set it as the current point p cur Add it to the set P and remove the point p from the set of viewpoints next , return to step (42) until the set of viewpoints is empty, set the robot origin as the end point of the path and add it to the set P. The set P is the locally optimal path found that starts from the robot origin, passes through all viewpoints in sequence, and then returns to the robot origin.

[0083] (5) Perform collision detection on the planned locally optimal path. After passing the collision detection, complete the shooting path planning for all station cameras.

[0084] After finding the locally optimal path, if the robot moves directly along the locally optimal path, it may cause the robot to collide with other devices. Therefore, it is also necessary to check and avoid collisions for the locally optimal path. The collision detection and collision avoidance operations of the locally optimal path need to be implemented by means of the probabilistic roadmap algorithm and the graph search algorithm.

[0085] In the embodiment of the present invention, the method for performing collision detection on the locally optimal path is specifically as follows:

[0086] (51) Take out two adjacent points P in the set P in sequence c 、P n , the line segment P c P n is the section searched from the point P c to the point P n by the greedy algorithm. Discretize this section with a step size of t to obtain multiple discrete points P = (1 - t)·P c + t·P n ;

[0087] (52) Calculate the inverse kinematics of the robot at the discrete points in sequence. According to the calculated joint angles of the robot, update the hierarchical bounding box of the robot, and judge whether the robot collides according to the updated hierarchical bounding box; if there is no collision, the path P c P n is valid, and no change is made to the path point set P. If there is a collision, the path P c P n is invalid, and it is necessary to search for a collision-free path from the starting point P c to the end point P n in the established roadmap, and insert the searched intermediate points between the two points P c and P n in the set P to complete the update of the path P c P n ;

[0088] (53) After traversing all adjacent pairs of points in set P, the path in set P is a collision-free path that starts from the robot's origin, passes through all viewpoints in sequence, and returns to the robot's origin.

[0089] The probabilistic roadmap is a sampling-based planning algorithm that does not require an accurate model of the robot configuration space. Given a sufficient number of sampled points, a collision-free path can always be found, effectively solving the difficulty of robot path planning in high-dimensional spaces. Moreover, the probabilistic roadmap method only needs to construct the roadmap once and can be reused, making it very suitable for this offline planning scenario. The method for forming a collision-free path with P c as the starting point and P n as the ending point is as follows:

[0090] (521) Set the maximum number of sampled points Max_Nodes_Number and the maximum connection distance ρ, and initialize the undirected graph G(V, E). Here, the vertex set V represents the set of collision-free sampled points, and the connection set E represents the collision-free paths between vertices;

[0091] (522) Randomly sample a collision-free sampled point P, add this sampled point P to the vertex set V, calculate the distances between the sampled point P and other vertices in the vertex set. If there exists a distance less than or equal to ρ, check whether the connection between the sampled point P and the corresponding point collides. If no collision occurs, add this connection to the connection set E;

[0092] (523) Return to step (522) until the number of generated sampled points reaches Max_Nodes_Number, at which point the construction of the undirected graph G(V, E) is completed;

[0093] (524) Use the A* algorithm to search for the shortest path from the starting point P c to the ending point P n in the undirected graph G(V, E). This shortest path is the collision-free path with P c as the starting point and P n as the ending point.

[0094] In the embodiments of the present invention, hierarchical bounding boxes are used to accelerate collision detection. For objects such as workbenches, workpieces, and robot bases that are fixed and do not move relative to each other, AABB hierarchical bounding boxes are constructed, and for objects with relative motion such as robots, OBB hierarchical bounding boxes are constructed.

[0095] The hierarchical bounding boxes in the form of a top-down binary tree and the OBB hierarchical bounding boxes adopted by the present invention have the following specific construction process:

[0096] (1) Determine the root node of the hierarchical bounding box:

[0097] For the AABB bounding box, calculate the minimum and maximum values of the vertex coordinates of all triangular facets on the hub triangular facet model in the X, Y, and Z directions. Determine the 8 vertices of the bounding box based on the minimum and maximum values on the three coordinate axes, thereby constructing an AABB bounding box that encloses the component model. Take this bounding box as the root node of the AABB hierarchical bounding box, and the component model is an object that is fixed and does not move relatively.

[0098] Since the OBB bounding box can establish a bounding box along any direction according to the geometric characteristics of the object, and in order to ensure that all parts can be photographed, the bounding box to be established needs to tightly enclose the robot triangular facet model as much as possible. Therefore, an OBB bounding box is established for the robot triangular facet model. The three main directions of the point set contained in each independent region are calculated through the principal component analysis (PCA), and these three main directions are used as the main axes of the OBB bounding box to establish the OBB bounding box. The specific process of establishing the OBB bounding box is as follows:

[0099] Calculate the total area S of all triangular facets within the robot triangular facet model;

[0100] Let the vertices of the triangular facet be u i 、v i 、w i , calculate the center coordinate c i and the triangle area s i . The center coordinate c i = (u i + v i + w i ) / 3, and the triangle area s i = (u i - v i ) · (u i - w i ) / 2. Then the total area of all triangular facets within the robot triangular facet model where n is the number of triangular facets within the robot triangular facet model.

[0101] Calculate the area-weighted center point C i of the robot triangular facet model, and its calculation formula is as follows;

[0102]

[0103] Construct the covariance matrix, perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors, use the three unitized eigenvectors as the coordinate axes, and use the area-weighted center point C iConstruct a new coordinate system with the origin as the coordinate origin, and obtain the transformation matrix T of the new coordinate system relative to the model coordinate system where the triangular facet model is located. Calculate the AABB bounding box of the robot's triangular facet model in the new coordinate system, and transform it to the model coordinate system to obtain the position, orientation, and size of the OBB bounding box.

[0104] In the embodiment of the present invention, the covariance matrix is represented as follows:

[0105]

[0106] In the formula,

[0107] Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors, Cv k = λ k v k , where v k represents the eigenvector, and λ k represents the eigenvalue, k ∈ {1, 2, 3}. Since the covariance matrix is a real symmetric matrix, the corresponding eigenvectors are pairwise orthogonal. Use the three unitized eigenvectors of the covariance matrix as the coordinate axes, and establish a new coordinate system with the area-weighted center point C i as the coordinate origin and obtain the transformation matrix T between the two coordinate systems.

[0108] (2) Divide the bounding box;

[0109] For the AABB bounding box, select the longest axis of the bounding box as the splitting axis, and perform splitting with a plane passing through the center of the bounding box and perpendicular to the longest axis. Divide the current model into two parts, and use these two parts as the two child nodes of the current node;

[0110] For the OBB bounding box, select the direction of the largest principal component of PCA as the splitting axis, and perform splitting with a plane passing through the center of the bounding box and perpendicular to the largest principal component. Divide the current model into two parts, and use these two parts as the two child nodes of the current node;

[0111] (3) Take these two nodes as the current node, and return to step (2) until each triangular facet is divided into leaf nodes, then the construction of the hierarchical bounding box in the form of a top-down balanced binary tree is completed.

[0112] In the embodiment of the present invention, it is necessary to detect whether the robot will collide with other objects in the environment, such as other robots and the hub to be measured, during the movement process. The collision detection process based on the hierarchical bounding box is specifically as follows:

[0113] (1) Read the hierarchical bounding boxes of the two collision detection objects, and use the top-level bounding boxes of the two collision detection objects as the current-level bounding boxes of the two collision detection objects respectively;

[0114] (2) Detect whether the current layer bounding boxes of the two collision detection objects intersect. If they do not intersect, it means that there will be no collision between the two collision detection objects. If they intersect, then execute step (2);

[0115] (3) Take one collision detection object as the first collision detection object, and the next layer bounding box of the first collision detection object as the current layer bounding box, and execute step (2) until traversing to the bottom layer bounding box of the first collision detection object. If the current layer bounding boxes of the two collision detection objects still intersect, then execute step (4);

[0116] (4) Take the other collision detection object as the first collision detection object and execute step (3).

[0117] The method for assigning photographing tasks of the station robot for detecting surface defects of the wheel hub provided by the present invention has the following beneficial technical effects:

[0118] (a) The wheel hub is divided into regions according to the normal vector characteristics of the triangular patch model of the wheel hub, which improves the traditional task division of complex curved surfaces according to the production line layout and proximity principle, ignoring the workpiece characteristics, resulting in excessive changes in the robot posture, greatly increasing the risk of collision with surrounding components. At the same time, assigning task regions according to the normal vector characteristics enables as many regions as possible to be photographed from one viewpoint, avoiding redundant viewpoints generated in the subsequent process;

[0119] (b) Use the simulated annealing algorithm to search and determine the boundaries of each region, without the need to input initial values, and the energy function has a fast descent speed, good algorithm convergence, and is suitable for this kind of global optimization problem;

[0120] (c) Adopt the probabilistic roadmap algorithm and the hierarchical bounding box technology to perform collision avoidance operations on the path. The combination of the probabilistic roadmap method and the hierarchical bounding box method can perform collision detection on the path while realizing collision avoidance operations, greatly improving the stability and feasibility of automatically generating the path.

[0121] The present invention has been described exemplarily. Obviously, the specific implementation of the present invention is not limited by the above methods. As long as various non-substantive improvements are made by adopting the method concept and technical solution of the present invention, or the concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A multi-station robot photo-taking task allocation method for wheel hub surface defect detection, characterized in that: The method is specifically as follows: (1) Divide the surface model of the wheel hub to be tested into a series of triangular patches to form a wheel hub triangular patch model, and perform Gaussian mapping on the wheel hub triangular patch model; (2) Based on the simulated annealing algorithm, the task area boundary of each robot station is found, and the angle between the normal vector of the triangle in the same task area and the Z axis is within the set range; (3) Generate camera viewpoints for each workstation robot to complete the task area, perform collision detection on all camera viewpoints, and adjust the viewpoints with collision to non-collision viewpoints; (4) Planning the local optimal path through all collision-free camera viewpoints based on the greedy algorithm; (5) Perform collision detection on the local optimal path, replan the road sections with collisions, and complete the shooting path planning of all workstation cameras.

2. The multi-station robot photography task allocation method for wheel hub surface defect detection according to claim 1, characterized in that: The number of task areas divided by the wheel hub triangle patch model is the same as the number of workstation robots, and one workstation robot corresponds to one task area.

3. The multi-station robot photo taking task allocation method for wheel hub surface defect detection according to claim 2, characterized in that: The specific process of determining the boundary of the task area based on the simulated annealing algorithm is as follows: (21) Set the initial temperature T, decay rate α, and end temperature T end ; (22) Input the set of normal vectors of the triangular facets on the wheel hub model, and the task area boundary X = [x1, x2, …, x n ], where x n represents the boundary between task area n and task area n+1, and randomly generates a set of initial solutions X initial ; (23) Current solution X cur The energy function E(X) of the lower region division is expressed as: E(X) = (S1-S2) 2 +(S1-S3) 2 +(S2-S3) 2 +…(S m -S n ) 2 , where S n Represents the sum of the areas of the triangles contained in region n; (24) In the current solution X cur A set of new solutions X is randomly generated within the neighborhood of new , X new =(1-δ)·X cur +δ·(rand·(X max -X min )+X min ), where δ represents the step size, rand represents a random number, and X max , X min is the set maximum and minimum value; (25) Determine the new solution X new With the current solution X cur The energy difference ΔE=E(X new )-E(X cur ), if ΔE<0, then accept the new solution X new , if ΔE>0, then with probability e -ΔE / T Accept the current solution X cur ; (26) Set the temperature to T new =αT cur Attenuation, T cur is the current temperature, and steps (24)-(26) are repeated until the temperature decays to the end temperature T end Or a new solution X new With the current solution X cur The interpolation value between is less than the threshold ξ, and the current solution is the optimal solution.

4. The multi-station robot photography task allocation method for wheel hub surface defect detection according to claim 1, characterized in that: The path planning method based on the greedy algorithm is as follows: (41) Set the robot origin p h As the current point p cur , added to the set P; (42) Find the distance from the current point p in the viewpoint set cur The nearest viewpoint is taken as the next point p next ; (43) will find the next point p next Set to current point p cur Add to set P and remove point p from the viewpoint set next , return to step (42) until the viewpoint set is empty, set the robot origin as the path end point and add it to the set P.

5. The method for allocating photographing tasks of a multi-station robot for wheel hub surface defect detection according to claim 1, characterized in that: The specific method for collision detection of local optimal paths is as follows: (51) Take two adjacent points P from the set P in turn c , P n , for the road section P between two points c P n Discretize with a step size of t to obtain multiple discrete points; (52) Calculate the robot inverse solution at discrete points in turn, obtain the angles of each joint of the robot, and determine whether the robot collides. If a collision occurs, search for the starting point P c To the end point P n The collision-free path is inserted into the two points P of the set P. c and P n Between, complete the path P c P n Updates; (53) After completing the traversal of all two adjacent points in the set P, the path in the set P is a collision-free path starting from the robot origin, passing through all viewpoints in sequence and returning to the robot origin.

6. The method for allocating photographing tasks of a multi-station robot for wheel hub surface defect detection according to claim 5, characterized in that the starting point P c To the end point P n The collision-free path acquisition process is as follows: (521) Set the maximum number of sampling points, the maximum connection distance ρ, and initialize the undirected graph G(V,E), where the vertex set V represents the set of collision-free sampling points, and the link set E represents the collision-free paths between vertices; (522) Randomly sample a collision-free sampling point P, add the sampling point P to the vertex set V, calculate the distance between the sampling point P and other vertices in the vertex set, if there is a distance less than or equal to ρ, check whether the line connecting the sampling point P and the corresponding point collides, if no collision occurs, add the line to the line set E; (523) Return to step (522) until the number of generated sampling points reaches the maximum number of sampling points, and the construction of the undirected graph G(V,E) is completed; (524) Use the A* algorithm to search for the starting point P in the undirected graph G(V,E) c To the end point P n The shortest path is the starting point P c To the end point P n collision-free path.

7. The multi-station robot photography task allocation method for wheel hub surface defect detection according to claim 1, characterized in that: A top-down binary tree of hierarchical bounding boxes is used for collision detection; Construct AABB hierarchical bounding boxes for fixed parts that do not move relative to each other, including: workbench, workpiece, and robot base; Construct OBB hierarchical bounding boxes for objects with relative motion, including robots.

8. The method for allocating photographing tasks of a multi-station robot for wheel hub surface defect detection according to claim 1, characterized in that: The construction process of the AABB hierarchical bounding box is as follows: (61) Calculate the minimum and maximum values ​​of the vertex coordinates of all triangular facets on the wheel hub triangular facet model in the X, Y, and Z directions, determine the eight vertices of the bounding box according to the minimum and maximum values ​​on the three coordinate axes, and construct an AABB bounding box containing the component model as the root node of the AABB hierarchical bounding box; (62) Select the longest axis of the AABB bounding box corresponding to the current node as the segmentation axis, and segment the model corresponding to the current node with a plane passing through the center of the bounding box and perpendicular to the longest axis, and segment the model into two parts. Construct AABB bounding boxes for the two parts respectively, and use the two parts as two child nodes of the current node; (63) These two nodes are taken as current nodes and the process returns to step (62) until each triangle face is divided into leaf nodes and the AABB hierarchical bounding box is established.

9. The method for allocating photographing tasks of a multi-station robot for wheel hub surface defect detection according to claim 1, characterized in that: The construction process of the OBB bounding box is as follows: Calculate the total area S of all triangles in the robot triangle model; Assume that the vertices of the triangle are u i 、v i 、w i , calculate the center coordinate c of the triangle i and the area of ​​the triangle s i , center coordinate c i =(u i +v i +w i ) / 3, the area of ​​the triangle s i =(u i -v i )·(u i -w i ) / 2, then the sum of the areas of all triangles in the robot triangle model Where n is the number of triangles in the robot triangle model; Calculate the area weighted center point C of the robot triangle patch model i , the calculation formula is as follows; Construct a covariance matrix, perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors, use the three normalized eigenvectors as coordinate axes, and use the area weighted center point C as the coordinate axis. i Construct a new coordinate system for the coordinate origin, and obtain the transformation matrix T of the new coordinate system relative to the model coordinate system where the triangular patch model is located, calculate the AABB bounding box of the robot triangular patch model in the new coordinate system, and convert it to the model coordinate system to obtain the position, posture and size of the OBB bounding box.

10. The method for allocating photographing tasks of a multi-station robot for wheel hub surface defect detection according to claim 9, characterized in that: The covariance matrix is ​​expressed as follows: In the formula,

Images