Ship group automatic assembling and welding method based on global point cloud matching

Abstract

The invention provides an automatic assembling and welding method for a ship assembly based on global point cloud matching. The method specifically comprises the following steps: generating theoretical point cloud data based on a theoretical model of a part; scanning the entity part and preprocessing the scanning data to obtain actually measured point cloud data; performing multi-stage registration on the theoretical point cloud data and the actually measured point cloud data; after the theoretical model and the actual model are registered, a grabbing posture and an assembling posture are calculated; and grabbing the part for assembling and welding based on the calculated grabbing posture and assembling posture. The method is high in registration precision, and automatic assembly can be realized.

Description

Technical Field

[0001] This invention relates to the field of shipbuilding, and in particular to an automatic assembly and welding method for ship groups based on global point cloud matching. Background Technology

[0002] Currently, the assembly process of ship assembly groups suffers from low assembly accuracy and poor reliability. To address these issues, patent CN119989518A discloses an assembly position analysis method and system based on automated welding equipment in a ship assembly group. The method includes acquiring a 3D model of the completed assembly group and exporting it as an XML file; parsing the model coordinates of key vertices of the base plate and stiffeners from the XML file; determining the model coordinates of the gripping points of each stiffener based on its key vertices; acquiring the robot coordinates of the key vertices on the base plate identified by the automated equipment; calculating the coordinate transformation matrix between the model coordinates of the base plate and the robot coordinates; and using the coordinate transformation matrix, transforming the model coordinates of the stiffeners to the robot coordinate system based on the model coordinates of the key vertices and the gripping points to obtain the assembly position of the stiffeners. However, this invention does not provide a detailed solution for the selection of key vertices in the base plate model and the positioning and acquisition of key vertices of the stiffener plate and the base plate in the robot coordinate system; the method lacks a comprehensive description of the selection and positioning of key vertices, and cannot guarantee the effective acquisition of the coordinates of key vertices with appropriate accuracy, thus failing to automatically obtain the correct coordinate transformation matrix and assembly coordinates, and failing to guarantee assembly accuracy.

[0003] Patent publication number CN111275747A discloses a virtual assembly method, apparatus, equipment, and medium, including determining point cloud data to be registered and reference point cloud data based on the original point cloud data of the part to be assembled; determining the registration weight between the point to be registered and the associated reference point based on the allowable error of the part to be assembled; and performing point cloud registration on the point cloud data to be registered and the reference point cloud data based on the registration weight to achieve virtual assembly of the part to be assembled. However, this invention relies on manually placing marker points and manually preset registration weights, which cannot achieve automated acquisition and registration. The coarse registration uses a 4PCS algorithm, which is prone to producing poor coarse registration results when applied to ship plate parts with indistinct three-dimensional features, leading to failure of subsequent fine registration and making automation difficult. Summary of the Invention

[0004] In view of the shortcomings of the above-mentioned related technologies, the purpose of this invention is to provide an automatic assembly and welding method for ship crews based on global point cloud matching, so as to solve the problem of difficult automation of assembly and welding in the related technologies.

[0005] To achieve the above and other related objectives, this invention provides an automated assembly and welding method for ship subgroups based on global point cloud matching, the specific steps of which include:

[0006] Theoretical point cloud data is generated based on the theoretical model of the parts;

[0007] Scan the physical part and preprocess the scanned data to obtain measured point cloud data;

[0008] The theoretical point cloud data and the measured point cloud data are registered in multiple stages.

[0009] After the theoretical model and the actual model are registered, the grasping posture and assembly posture are calculated.

[0010] The part is grasped and assembled based on computational gripping and assembly postures.

[0011] Optionally, the step of generating theoretical point cloud data based on the theoretical model of the part is as follows: obtain the Gen file of the part, extract the part contour information, and generate theoretical model point cloud data.

[0012] Optionally, assembly points are defined based on the dimensions of the assembly tool, and coordinate transformation is performed on the assembly points to obtain the actual assembly points. The assembly posture is then calculated based on the actual assembly points.

[0013] Optionally, the step of preprocessing the scanned data is as follows: after scanning the physical part, obtain the measured point cloud data of the part, first downsample the actual point cloud, and then denoise the background of the actual point cloud and extract the contour.

[0014] Optionally, the step of registering theoretical point cloud data with measured point cloud data includes coarse registration and fine registration.

[0015] Optionally, the coarse registration steps are as follows: first, calculate the geometric center of the two point clouds, then translate and align the centers of the two point clouds, then obtain the normal vector of the plane containing the two point clouds, and align the normal vectors of the two point clouds, and finally search for the rotation angle in the range of 0-360° with the aligned normal vector as the rotation axis to minimize the contour alignment error of the two point clouds in the plane.

[0016] Optionally, the fine registration step is as follows: using a multi-stage iterative nearest point algorithm to gradually reduce the matching distance threshold until the registration result reaches the set target value.

[0017] Optionally, the steps for calculating the grasping posture are as follows: obtain the direction vector of the grasping part and the normal vector of the point cloud on the upper surface of the actual part, and cross-multiply the two to obtain a new direction vector. Then, cross-multiply the new direction vector with the direction vector of the grasping part to obtain a new normal vector of the point cloud on the upper surface of the actual part. Process the direction vector of the grasping part, the new direction vector, and the new normal vector of the point cloud on the upper surface of the actual part and combine them column by column to obtain a rotation matrix. Calculate the robot's rotation posture according to the rotation matrix and the definition of Euler angles of the robot, and finally obtain the robot's grasping posture coordinates.

[0018] Optionally, the step of obtaining the direction vector of the gripping part is as follows: define the model gripping point and the midpoint of the model assembly edge and convert them into the actual gripping point and the actual assembly edge midpoint in the robot coordinate system, and obtain the direction vector based on the actual gripping point and the actual assembly edge midpoint.

[0019] Optionally, the step of obtaining the normal vector of the point cloud on the upper surface of the actual part is as follows: calculate the geometric center of the point cloud of the actual part, and construct a covariance matrix based on the center point of the point cloud and other points. Use the eigenvector corresponding to the smallest eigenvalue of the matrix as the global normal vector of the plane to be obtained. Considering that the normal vector has directional ambiguity, define the normal vector with an acute angle to the positive Z-axis of the robot coordinate system as the normal vector of the point cloud on the upper surface of the actual part.

[0020] As described above, the automated assembly and welding method for ship subgroups based on global point cloud matching of the present invention has the following beneficial effects: The present invention requires no manual intervention or predefined rules to select key feature points. From contour point cloud generation and data acquisition to registration calculation, the entire process is automatically completed by the algorithm, greatly improving the level of intelligence. This not only reduces the technical dependence and experience requirements of operators but also avoids subjective errors introduced by manual point selection, ensuring the consistency and repeatability of the process. Attached Figure Description

[0021] Figure 1 The diagram shows a flowchart of the automatic assembly and welding method for ship crews based on global point cloud matching in an embodiment of the present invention. Detailed Implementation

[0022] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0023] In the detailed description of embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged and not to scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0024] For ease of description, spatial relation terms such as “below,” “under,” “lower than,” “below,” “above,” and “upper” may be used herein to describe the relationship between one element or feature shown in the accompanying drawings and other elements or features. It will be understood that these spatial relation terms are intended to include directions other than those depicted in the drawings for the device in use or operation. Furthermore, when a layer is referred to as being “between” two layers, it can be the only layer between the two layers, or there may be one or more layers in between. The term “between” as used herein includes both endpoint values.

[0025] In the context of this application, the structure described above the first feature may include embodiments in which the first and second features are formed in direct contact, or embodiments in which additional features are formed between the first and second features, such that the first and second features may not be in direct contact.

[0026] It should be noted that the illustrations provided in this embodiment are only schematic representations of the basic concept of the present invention. Therefore, the illustrations only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0027] like Figure 1 As shown, this embodiment provides an automatic assembly and welding method for ship subgroups based on global point cloud matching. The specific steps include:

[0028] Theoretical point cloud data is generated based on the theoretical model of the part.

[0029] Specifically, the Gen file of the assembled parts is obtained from the CNC cutting system. Dedicated analysis software is used to extract the part contour information and generate point cloud data for the theoretical model. The Gen file includes the name of each part, plate thickness, X and Y coordinates of the part vertices, and geometric descriptions of the vertices. Vertices are the key inflection points that form the boundary lines of the part's external or internal contour, including the start and end points of straight lines, and the start and end points of curves (arcs, splines) and key control points. Taking the base plate part as an example, key vertices refer to the start and end points of all straight lines and all arcs. The geometric description of the vertices refers to the information that helps the vertices determine the unique contour, including the center coordinates, radius, and sweep angle. In the part contour described in the Gen file, all line segments (straight lines and curves) are stored in arc format, including the start point, end point, center, radius, and sweep angle. When the radius is 0, the center is (0, 0), and the sweep angle is 0, this arc segment appears as a straight line.

[0030] The physical part is scanned and the scanned data is preprocessed to obtain the measured point cloud data.

[0031] Specifically, a 3D camera is used to acquire measured point cloud data of parts (base plate, stiffening plate) located on the assembly platform; the acquired raw point cloud is preprocessed, including downsampling and noise reduction, to improve the efficiency and accuracy of subsequent processing. The 3D camera is located on the robot, and the assembly platform has a cuboid structure with its upper and lower surfaces parallel to the robot's XY plane, and any side surface of the assembly platform is always parallel or perpendicular to the robot's XZ plane.

[0032] After obtaining the measured point cloud data of the part, the point cloud is downsampled. Downsampling refers to reducing the number of points in the point cloud data through algorithms to reduce the computational load. Various downsampling algorithms can be used, including but not limited to voxel downsampling, random downsampling, uniform downsampling, and curvature downsampling. Taking voxel downsampling as an example, a three-dimensional grid (voxel grid) is created in the entire point cloud space, with each voxel being a tiny cube. For each voxel, all points contained within it are approximated by a single point. This representative point is typically calculated by taking the average of the coordinates of all points within the voxel (centroid), or by directly taking the center point of the voxel.

[0033] Because point cloud acquisition simultaneously captures point clouds from the assembly platform surface, parts on the assembly platform, and other invalid point clouds, it is necessary to denoise the point cloud data after downsampling. Noise denoising includes two parts: background removal and contour extraction.

[0034] Background removal refers to removing point clouds outside the part surface point cloud. First, coordinate filtering is used. The downsampled point cloud's XY coordinates are filtered based on the maximum and minimum X, maximum and minimum Y coordinates of points along the assembly platform's edge contour within the point cloud space. The filtered point cloud is entirely within the assembly platform contour; at this point, the point cloud data only includes the point cloud of the assembly platform's upper surface and the part point cloud on the platform. The height of the assembly platform's upper surface is set to the minimum Z value, and the thickness of the part allowed to be assembled on the platform plus the minimum Z value is set to the maximum Z value. Coordinate filtering continues to remove point clouds outside the range of the minimum to maximum Z values. The filtered point cloud now only contains the part point cloud.

[0035] Normal filtering is used to remove point clouds from the side edges of a part's contour. The process involves first calculating the normal vector of each point in the point cloud using PCA (Principal Component Analysis). Then, a normal vector is specified, typically along the X, Y, or Z axis. The angle between the normal vector of each point and the specified normal vector is calculated, and a threshold range for this angle is set. This filters out points that meet the requirements. Using the Z-axis as the specified normal vector, the angle between the normal vector of each point and the Z-axis is calculated. Points with angle values ​​between 10° and 170° are filtered out, resulting in the point cloud of the part's upper surface.

[0036] Then, an edge estimation algorithm is applied to the point cloud of the upper surface of the part to obtain the part's contour. The steps of the edge estimation algorithm are as follows: First, the normal vector of each point in the point cloud is calculated using the PCA method. Then, the number of neighboring points is specified, such as 30; an angle threshold is set, such as 90 degrees; and a scale value is set, such as 0.8. The angle between the normal vector of each point and the normal vector of all its neighboring points is calculated. The proportion of points with angles greater than the threshold is counted. If the calculated scale value is greater than the set scale value, the point is considered an edge point. Retaining all edge points yields the edge contour point cloud of the upper surface of the part.

[0037] The theoretical point cloud data and the measured point cloud data are registered in multiple stages.

[0038] Specifically, multi-stage registration includes two stages: coarse registration and fine registration. The rotation transformation matrix obtained at each registration step and the final rotation transformation matrix after registration are both 4×4 homogeneous transformation matrices, with the general form being: Where R is a 3x3 rotation matrix representing the rotation of the coordinate system; t is a 3x1 translation matrix representing the offset of the origin of the coordinate system. In the following calculation description, the theoretical model point cloud is referred to as M_pointcloud, and the measured point cloud is referred to as P_pointcloud. A point cloud is a set containing n three-dimensional coordinates (X, Y, Z). To transform the point cloud, it needs to be written as an nx3 matrix, where each row corresponds to a point coordinate, and the three columns represent the XYZ coordinates of each point. For the subsequent transformation, a fourth column needs to be added, which is all 1s. The point cloud transformation is the multiplication of the transpose of the N×4 point cloud matrix transformation matrix: The first three columns of the transformed matrix are taken as the transformed point cloud.

[0039] The steps for coarse registration are as follows: First, calculate the geometric centers of the two point clouds. This yields the theoretical point cloud center M_center(x1, y1, z1) and the measured point cloud center P_center(x2, y2, z2). Then, perform translation alignment. The translation transformation matrix T1 = ... E represents a 3×3 identity matrix. .

[0040] The least squares method is used to fit the plane containing the model contour point cloud and the actual contour point cloud, respectively. The equation of the theoretical model point cloud plane is Max + Mby + Mcz + Md = 0, and its normal vector Nm_vector(Ma, Mb, Mc) is obtained. The equation of the plane containing the actual contour point cloud is Pax + Pby + Pcz + Pd = 0, and its normal vector Np_vector(Na, Nb, Nc) is obtained. Normalization is then performed on both Nm_vector and Np_vector. For example:

[0041] Nm1_vector=(Ma1, Mb1, Mc1)=(Ma / Lm, Mb / Lm, Mc / Lm),

[0042] Np1_vector=(Na1, Nb1, Nc1)=(Na / Lp, Nb / Lp, Nc / Lp),

[0043] Then, a rotation matrix is ​​constructed to align the normal vectors of the theoretical point cloud plane with those of the measured point cloud plane. First, the rotation axis vector K(Ka, Kb, Kc) is found, i.e., The rotation axis vector is normalized to obtain K1 (Ka1, Kb1, Kc1), and the rotation axis vector is perpendicular to both the theoretical model point cloud plane normal vector and the actual point cloud plane normal vector.

[0044] Next, the angle between the plane normal vector of the theoretical model point cloud and the plane normal vector of the measured point cloud is calculated. According to K1 and Construct rotation matrix And perform normal vector alignment to obtain the transformation matrix T2= This transformation involves only rotation and not translation; therefore, the translation matrix t is a zero matrix. , .

[0045] After normal alignment, in-plane rotation alignment is performed: using the aligned normal vector as the rotation axis, the optimal rotation angle is searched within the range of 0-360° to minimize the contour alignment error of the two point clouds in the plane, resulting in the transformation matrix T3. Details are as follows:

[0046] First, the normalized normal vector of the measured point cloud is used as the rotation axis when using the T2 transformation matrix. As a variable, it gradually increases from 0° to 360°, with a step size of 5°. The angles are 5°, 10°, 15°, 20°...360°, for each... Construct a rotation transformation matrix to transform the theoretical model point cloud and calculate the nearest point distance between the transformed theoretical model point cloud and the measured point cloud.

[0047] Select the angle corresponding to the minimum distance between the nearest points, such as 20°. Using 20° as the median angle, set a step size range both upwards and downwards, such as 15° to 25°. Use the normalized normal vector of the actual point cloud when calculating the T2 transformation matrix as the rotation axis, and the rotation angle is... The rotation transformation matrix is ​​constructed by gradually increasing the angle from 15° to 25° with a step size of 1°. This matrix is ​​used to transform the theoretical model point cloud, and the nearest point distance between the transformed theoretical model point cloud and the actual point cloud is calculated. The angle with the smallest nearest point distance is selected as the optimal rotation angle. The normalized normal vector of the measured point cloud used when calculating the T2 transformation matrix is ​​taken as the rotation axis, and the optimal rotation angle is used... Construct a rotation transformation matrix for the rotation angle to obtain T3.

[0048] The theoretical model point cloud is transformed using transformation matrix T3, and the transformed point cloud is used as the input point cloud for subsequent fine registration.

[0049] Based on coarse registration, a multi-stage iterative nearest point (ICP) algorithm is used for fine registration, gradually reducing the matching distance threshold, and finally obtaining a high-precision total transformation matrix T_total.

[0050] During ICP registration, the transformation estimation method is selected as point-to-point, the convergence condition is selected as the maximum number of iterations such as 50, and the initial distance threshold is set to 1% of the maximum distance in a suitable direction of the source point cloud for registration.

[0051] If registration fails, the distance threshold is increased to twice the initial threshold for re-registration. If that fails, the threshold is increased again for registration until success is achieved.

[0052] After successful registration, record the obtained registration transformation matrix T4. Use the obtained transformation matrix T4 to transform the source point cloud as the new source point cloud. Reduce the distance threshold of the previous successful registration and perform registration again, such as reducing the distance threshold to 0.8 times the original value. Repeat the registration process, recording the transformation matrix after each registration, and continuously reducing the distance threshold until the registration fit parameter reaches the set target value, such as 0.5. In this way, a high-precision registration result can be obtained.

[0053] Starting from the coarse registration, take the transpose of all rotation transformation matrices T1, T2, T3, T4, ... and multiply them to obtain the transpose of the high-precision registration matrix T_total. Taking its transpose will give you T_total.

[0054] After the theoretical model and the actual model are registered, the grasping posture and assembly posture are calculated.

[0055] Specifically, in the theoretical model, based on the working method and physical dimensions of the gripping tool, a point offset by a fixed distance from the midpoint of the part's assembly edge to the interior of the part is defined as the model gripping point. For example, when calculating the gripping point of the actual rib using the above method, the model gripping point M1 and the midpoint of the model assembly edge M2 are defined. The model gripping point M1 and the midpoint of the model assembly edge M2 are transformed into the robot coordinate system to obtain the actual gripping point coordinates P1 and the actual assembly edge midpoint coordinates P2. The actual gripping point coordinates are obtained by subtracting the actual assembly edge midpoint coordinates from the actual gripping point coordinates. This direction vector (denoted as D_vector) is used to calculate the robot's gripping posture. In addition, the robot's gripping posture also needs to be based on the normal vector of the point cloud on the upper surface of the actual part (denoted as N_vector).

[0056] The normal vector of the point cloud on the actual part's upper surface is calculated using principal component analysis. The entire point cloud dataset is treated as a whole, and principal component analysis is used to find the direction in which the points are most concentrated; this direction is defined as the normal vector direction of the plane in question. The three-dimensional geometric center of all points in this point cloud is then calculated. Based on center point P and other points Construct a 3×3 covariance matrix C( This matrix represents the distribution of the entire point cloud around its center.

[0057] Eigenvalue decomposition of the covariance matrix C yields three eigenvalues ​​λ0, λ1, and λ2 (usually satisfying λ0 ≥ λ1 ≥ λ2 ≥ 0) and their corresponding mutually orthogonal unit eigenvectors v0, v1, and v2. The eigenvector v2 corresponding to the smallest eigenvalue λ2 represents the direction in which the point cloud distribution is most concentrated. For an ideal plane, all points are distributed on a two-dimensional plane, therefore the variance of the distribution is minimized in the direction perpendicular to the plane. Thus, this vector v2 is defined as the global normal vector N_vector of the plane in question. The normal vector calculated through principal component analysis has directional ambiguity (i.e., N_vector and -N_vector). The normal vector that makes an acute angle with the positive Z-axis of the robot coordinate system is defined as the normal vector of the point cloud on the upper surface of the actual part.

[0058] To calculate the grasping posture, according to the right-hand rule, a new direction vector, denoted as Y_vector, is obtained by cross product of N_vector and D_vector. To ensure orthogonality, a new N_vector is obtained by cross product of D_vector and Y_vector. This N_vector is then denoted as Z_vector, and D_vector as X_vector. X_vector, Y_vector, and Z_vector are normalized and combined column-wise to obtain the rotation matrix R. Then, the robot's rotation posture is calculated based on the rotation matrix R and the definition of Euler angles for the robot, thus finally obtaining the robot's grasping posture coordinates.

[0059] When calculating the assembly posture, the model assembly points are defined according to the working mode of the gripping tool and the physical dimensions. For example, the model assembly point is defined as the point at which the midpoint of the base plate assembly line is offset by a specific distance along the plate thickness direction.

[0060] Using the total transformation matrix obtained from the above steps, the model assembly point and the midpoint of the model assembly line are transformed into the robot coordinate system to obtain the actual assembly point coordinates and the actual assembly line midpoint coordinates. The actual assembly point coordinates are obtained by subtracting the actual assembly line midpoint coordinates from the actual assembly point coordinates, which is used to calculate the robot assembly posture.

[0061] The normal vector of the point cloud on the upper surface of the bottom plate is calculated using the method for calculating the normal vector of the upper surface of the rib plate in the above steps. Then, the posture of the bottom plate assembly robot is calculated according to the method for calculating the gripping posture of the rib plate. The posture of the bottom plate assembly robot can be obtained.

[0062] Finally, based on the calculated gripping posture, the robot controls the gripping fixture to grip the stiffener and rotate it 90° to the assembly posture, ready to perform the welding action. For example, the robot controls the stiffener to move to the calculated actual assembly posture position for assembly and welding.

[0063] This embodiment also discloses a system for performing the above method, the system including a data processing unit, a 3D vision acquisition unit, a point cloud registration unit, a robot control unit, and an execution unit.

[0064] The data processing unit is used to parse Gen file data, extract part contour vertex information, and generate theoretical model point cloud.

[0065] The 3D vision acquisition unit includes a line laser scanner and a corresponding optical positioning system installed at the end of the robot, used to acquire high-precision point cloud data of the measured parts.

[0066] Point cloud registration unit: An industrial computer or industrial control computer equipped with a high-performance processor is configured to run point cloud registration algorithms, execute multi-stage registration processes, and calculate high-precision coordinate transformation matrices.

[0067] The robot control unit is used to control a six-axis industrial robot, receiving the coordinates and attitude information of the gripping point and assembly point sent by the point cloud registration unit, and planning the motion trajectory.

[0068] The execution unit includes an electromagnetic gripping device and welding equipment driven by a robot control unit, used to perform gripping, handling, assembly and final welding operations of the stiffeners.

[0069] This invention completely abandons the traditional approach of relying on a limited number of key vertices, innovatively employing full-contour point cloud data for high-precision registration. Through a multi-segment coarse registration method, it rapidly calculates a good initial pose of the point cloud using feature data such as the point cloud center, point cloud plane normal vector, and the nearest point distance between point clouds, eliminating the problem of initial pose causing fine registration calculation failures and improving robustness. Through the ICP (Iterative Closest Point) algorithm, it calculates the transformation matrix using the complete geometric contour information of the part, fundamentally eliminating catastrophic assembly failures caused by individual vertex recognition errors (such as those caused by sheet metal deformation, paint reflection, or sensor noise). This method is insensitive to local defects and noise, and its accuracy and stability are far superior to three-point-based calculation methods, providing a reliable guarantee for achieving high-precision automated assembly of ship groups.

[0070] This method is based on general point cloud data processing technology and does not rely on specific geometric features (such as the requirement of three easily identifiable vertices). Therefore, it is not only applicable to parts with regular shapes, but also effective for parts with complex contours, curves, or irregular shapes, greatly expanding the application boundaries of automated assembly and welding technology and having broad applicability and promotional value.

[0071] This invention scientifically defines gripping and assembly points (e.g., located inside the midpoint of the assembly edge) on the model based on the physical dimensions of the gripping and assembly tools, and maps them to the actual coordinate system through a high-precision transformation matrix. This method ensures the physical accessibility and safety of the gripping and assembly process, avoids operational failures caused by gripping points located outside the contour or interfering with the tooling, and improves the reliability and efficiency of the entire automation system.

[0072] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. An automated assembly and welding method for ship sub-groups based on global point cloud matching, characterized in that, The specific steps include: Theoretical point cloud data is generated based on the theoretical model of the parts; Scan the physical part and preprocess the scanned data to obtain measured point cloud data; The theoretical point cloud data and the measured point cloud data are registered in multiple stages. After the theoretical model and the actual model are registered, the grasping posture and assembly posture are calculated. The part is grasped and assembled based on computational gripping and assembly postures.

2. The automatic assembly and welding method for ship sub-groups based on global point cloud matching according to claim 1, characterized in that: The steps for generating theoretical point cloud data based on the theoretical model of the part are as follows: obtain the Gen file of the part, extract the part contour information, and generate theoretical model point cloud data.

3. The automatic assembly and welding method for ship sub-groups based on global point cloud matching according to claim 1, characterized in that: Based on the dimensions of the assembly tool, define the model assembly points, perform coordinate transformation on the model assembly points to obtain the actual assembly points, and calculate the assembly posture based on the actual assembly points.

4. The automatic assembly and welding method for ship sub-groups based on global point cloud matching according to claim 1, characterized in that: The steps for preprocessing the scanned data are as follows: after scanning the physical part, obtain the actual point cloud data of the part, first downsample the actual point cloud, then denoise the background of the actual point cloud and extract the contour.

5. The automatic assembly and welding method for ship sub-groups based on global point cloud matching according to claim 1, characterized in that: The steps for multi-stage registration of theoretical point cloud data and measured point cloud data include coarse registration and fine registration.

6. The automatic assembly and welding method for ship sub-groups based on global point cloud matching according to claim 5, characterized in that: The coarse registration steps are as follows: first, calculate the geometric center of the two point clouds, then translate and align the centers of the two point clouds, then obtain the normal vector of the plane containing the two point clouds, and align the normal vectors of the two point clouds, and finally search for the rotation angle in the range of 0-360° with the aligned normal vector as the rotation axis to minimize the contour alignment error of the two point clouds in the plane.

7. The automatic assembly and welding method for ship sub-groups based on global point cloud matching according to claim 5, characterized in that: The fine registration process involves using a multi-stage iterative nearest-point algorithm to gradually reduce the matching distance threshold until the registration result reaches the set target value.

8. The automatic assembly and welding method for ship sub-groups based on global point cloud matching according to claim 1, characterized in that: The steps for calculating the grasping posture are as follows: obtain the direction vector of the grasping part and the normal vector of the point cloud on the upper surface of the actual part, and cross-multiply the two to obtain a new direction vector. Then, cross-multiply the new direction vector with the direction vector of the grasping part to obtain a new normal vector of the point cloud on the upper surface of the actual part. Process the direction vector of the grasping part, the new direction vector, and the new normal vector of the point cloud on the upper surface of the actual part and combine them column by column to obtain a rotation matrix. Calculate the robot's rotation posture according to the rotation matrix and the definition of Euler angles of the robot, and finally obtain the robot's grasping posture coordinates.

9. The automatic assembly and welding method for ship sub-groups based on global point cloud matching according to claim 8, characterized in that: The steps for obtaining the direction vector of the gripping part are as follows: define the model gripping point and the midpoint of the model assembly edge and convert them into the actual gripping point and the actual assembly edge midpoint in the robot coordinate system, and obtain the direction vector based on the actual gripping point and the actual assembly edge midpoint.

10. The automatic assembly and welding method for ship sub-groups based on global point cloud matching according to claim 8, characterized in that: The steps for obtaining the normal vector of the point cloud on the upper surface of the actual part are as follows: calculate the geometric center of the point cloud of the actual part, and construct a covariance matrix based on the center point and other points of the point cloud. Use the eigenvector corresponding to the smallest eigenvalue of the matrix as the global normal vector of the plane to be obtained. Considering the ambiguity of the direction of the normal vector, define the normal vector with an acute angle to the positive Z-axis of the robot coordinate system as the normal vector of the point cloud on the upper surface of the actual part.

Images