A method and system for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing

By establishing a unified coordinate system during the assembly of hull sections, and combining global pose perception and local visual servoing, a continuous assembly error field is constructed. This solves the problems of insufficient accuracy and automation in traditional methods, and achieves high-precision, real-time assembly adjustment, thereby improving the construction quality and efficiency of hull sections.

CN123104875APending Publication Date: 2026-11-17QILU UNIVERSITY OF TECHNOLOGY (SHANDONG ACADEMY OF SCIENCES) +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611602621.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-16
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to achieve high precision, real-time performance, and automation during the assembly of ship hull sections, especially in the assembly of large ship hull sections. Traditional discrete measurement point precision control and single vision measurement are insufficient to meet the demands for high efficiency and high precision, while local flexible deformation and global pose information are difficult to acquire stably.

Method used

By establishing a unified coordinate system for the merging area, combining global pose perception and local visual servoing, a continuous merging error field is constructed, rigid-flexible coupling pose fusion is performed, a weldable window is generated, and staged visual servo closed-loop adjustment is carried out to achieve coordinated control of overall pose and local seam quality.

Benefits of technology

This improved the precision and stability of hull section assembly, reduced manual adjustments based on experience and repeated testing, ensured welding quality, and shortened the construction cycle.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123104875A_ABST
    Figure CN123104875A_ABST
Patent Text Reader

Abstract

This invention relates to the field of marine vessel assembly technology, and in particular to a method and system for hull section assembly based on rigid-flexible coupling pose fusion and visual servoing. The method includes establishing a reference coordinate system for the assembly area and extracting theoretical assembly information based on a three-dimensional design model; acquiring global pose observations and calculating the initial six-degree-of-freedom pose; acquiring local visual observations and extracting assembly features; performing rigid-flexible coupling pose fusion based on the acquired global pose observations and local visual observations; constructing a continuous assembly error field and generating a weldable window based on the fused pose and theoretical assembly information; and generating adjustment amounts based on the constructed assembly error field and weldable window. This invention solves the problem that traditional discrete control point measurements cannot continuously reflect the quality of the assembly seam by combining global pose perception with local visual servoing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of marine vessel assembly technology, and in particular to a method and system for assembling hull sections based on rigid-flexible coupling pose fusion and visual servoing. Background Technology

[0002] The assembly of hull sections is one of the key processes in shipbuilding, and its accuracy directly affects the structural strength, welding quality, assembly efficiency, and the smooth progress of subsequent outfitting operations. During the construction, hoisting, transportation, and assembly of ship sections, the hull sections are susceptible to overall positional deviations and localized structural deformations due to factors such as welding shrinkage, jig errors, hoisting deformation, gravitational deflection, and temperature changes. Particularly in areas such as the closure joint, weld edges, longitudinal girder ends, and rib joints, uneven gaps, misalignments, height differences, or end-face angle deviations can lead to forced assembly, increased residual welding stress, excessive post-weld deformation, and even rework and structural quality issues. Therefore, achieving high-precision measurement, real-time assessment, and closed-loop adjustment during the hull section assembly process is of great significance for improving shipbuilding quality and shortening the construction cycle.

[0003] Currently, the precision control of hull section assembly mainly relies on measuring equipment such as total stations, laser trackers, and 3D scanners to measure key control points, combined with manual experience, crane assistance, jacks, adjustable blocks, or hydraulic lifting devices to adjust the section's attitude. While this method can obtain overall position and attitude information of the hull sections to a certain extent, it typically relies on a small number of discrete control points, making it difficult to continuously reflect local assembly conditions such as gaps, misalignments, height differences, and normal angles along the assemblies. Furthermore, the measurement and adjustment processes are often separate, requiring multiple "measurement-judgment-adjustment-remeasurement" steps, resulting in low real-time performance and automation, which fails to meet the demands for high-efficiency, high-precision assembly of large hull sections.

[0004] On the other hand, visual measurement and visual servoing technologies can acquire local structural features near the closure joint, and have good perception capabilities for details such as bevel edges, plate seams, rib endpoints, process holes, and welding boundaries. However, the hull sections are large in scale and complex in structure, and there are problems such as occlusion, reflection, smoke and dust, feature sparsity, and coordinate system drift on site. It is difficult to obtain stable global pose information by relying solely on visual measurement. In addition, the hull sections are not ideal rigid bodies, and local flexible deformation will occur during support, hoisting, and welding. If adjustments are made only based on the rigid body pose, it is easy to have problems such as the overall posture seemingly meeting the requirements but the local closure joint still exceeding the tolerance, or the local joint being improved but the overall stress state deteriorating.

[0005] Therefore, how to break through the limitations of traditional discrete measurement point precision control and single vision measurement, integrate the global pose of the hull section, the visual error of the local closure seam, the feedback of the actuator, the support status and the local flexible deformation information, and establish a continuous closure error field oriented to the welding process requirements, so as to realize the real-time, high-precision, closed-loop adjustment of the hull section closure process, is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a method and system for hull section assembly based on rigid-flexible coupling pose fusion and visual servoing. By establishing a unified coordinate system for the assembly area, global pose perception, local visual servoing, actuator feedback, support status, and designed assembly surface data are fused to construct a continuous assembly error field. Furthermore, based on the weldable window, the assembly state is determined and the adjustment amount is solved in a closed loop, thereby achieving coordinated control of the overall pose of the hull section and the local seam quality.

[0007] In a first aspect, the present invention provides a method for assembling ship hull segments based on rigid-flexible coupling pose fusion and visual servoing, which adopts the following technical solution:

[0008] A method for assembling ship hull segments based on rigid-flexible coupling pose fusion and visual servoing includes:

[0009] Establish a reference coordinate system for the closure area and extract theoretical closure information based on the three-dimensional design model;

[0010] Collect global pose observations and calculate the initial six-DOF pose;

[0011] Collect local visual observations and extract merging features;

[0012] Rigid-flexible coupled pose fusion is performed based on the acquired global pose observations and local visual observations.

[0013] A continuous closing error field is constructed based on fused pose and theoretical closing information, and a weldable window is generated.

[0014] Adjustments are generated based on the constructed closing error field and weldable window;

[0015] Perform phased visual servo closed-loop adjustments based on the adjustment amount and generate records.

[0016] Secondly, a hull section assembly system based on rigid-flexible coupling pose fusion and visual servoing includes: a data acquisition module configured to establish a reference coordinate system for the assembly area and extract theoretical assembly information based on a three-dimensional design model;

[0017] The global module is configured to acquire global pose observations and calculate the initial six-degree-of-freedom pose.

[0018] The local module is configured to acquire local visual observations and extract merging features;

[0019] The fusion module is configured to perform rigid-flexible coupled pose fusion based on the acquired global pose observations and local visual observations;

[0020] The window module is configured to construct a continuous closing error field and generate a weldable window based on fused pose and theoretical closing information;

[0021] The adjustment module is configured to generate adjustment amounts based on the constructed closure error field and weldable window;

[0022] The execution module is configured to perform phased visual servo closed-loop adjustments based on the adjustment amount and generate records.

[0023] Thirdly, the present invention provides a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the aforementioned method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing.

[0024] Fourthly, the present invention provides a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide a method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing.

[0025] In summary, the present invention has the following beneficial technical effects:

[0026] (1) This invention combines global pose perception with local visual servoing, which can obtain the overall six-degree-of-freedom pose of the segment to be joined relative to the reference segment, and can also obtain local gap, misalignment, height difference and end face angle information near the joining seam, thus solving the problem that traditional discrete control point measurement is difficult to continuously reflect the quality of the joining seam.

[0027] (2) This invention uses rigid-flexible coupling pose fusion to simultaneously consider the overall rigid body pose and local flexible deformation of the segments to be joined. It can distinguish between overall posture deviation and local structural deformation, avoid incorrect adjustment of the entire segment posture due to local errors, and improve the accuracy and stability of the joining adjustment.

[0028] (3) By constructing a continuous closure error field and generating a weldable window based on plate thickness, bevel form, welding sequence, shrinkage allowance and local structural stiffness, the present invention expands the closure judgment from the traditional single-point out-of-tolerance judgment to the process window judgment distributed along the closure line, thereby improving the reliability of the pre-welding assembly state judgment.

[0029] (4) This invention solves the problem by adjusting the closed loop, which takes into account the continuous closing error field, weldable window, actuator stroke, execution speed, support reaction force and anti-collision constraint in a unified manner. It can generate a combination of adjustment amounts such as lifting, lateral movement, longitudinal movement, rotation, traction or pressing, reducing manual experience adjustment and repeated retesting.

[0030] (5) This invention forms a quality data chain from measurement, adjustment, locking to welding process monitoring by pre-welding locking record and in-welding offset monitoring, which helps to reduce post-weld rework and provides data support for quality traceability and process optimization of hull section assembly. Attached Figure Description

[0031] Figure 1 This is a flowchart of the high-precision assembly method for ship hull sections provided in Embodiment 1 of the present invention;

[0032] Figure 2 This is a diagram showing the transformation relationship of the multi-coordinate system coordinate chain provided in Embodiment 1 of the present invention;

[0033] Figure 3 This is a block diagram of rigid-flexible coupling pose fusion provided in Embodiment 1 of the present invention;

[0034] Figure 4 This is a schematic diagram of the construction of the continuous closing error field provided in Embodiment 1 of the present invention;

[0035] Figure 5 This is a diagram of the high-precision assembly system architecture for hull sections provided in Embodiment 2 of the present invention.

[0036] Figure 6 This is a schematic diagram of the experimental verification simulation scenario provided in Embodiment 1 of the present invention;

[0037] Figure 7 This is a comparison chart of the closing error index before and after adjustment, provided in Embodiment 1 of the present invention. Detailed Implementation

[0038] The present invention will be further described in detail below with reference to the accompanying drawings.

[0039] Example 1

[0040] Reference Figure 1 This embodiment presents a method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing. Specifically, the method provided in this embodiment will be described in detail below according to S1 to S8.

[0041] As shown in S1 to S8, the method provided in this embodiment is as follows: First, a unified coordinate reference is established in the closure area, incorporating the design model, measuring equipment, vision equipment, the segment to be closure, the reference segment, and the actuator into the same coordinate chain. Then, the overall pose of the segment to be closure is obtained using global measuring equipment such as a total station, laser tracker, and 3D scanner, while local features of the closure seam are obtained using a binocular camera, line laser camera, or depth camera. Next, the global pose, local visual error, actuator feedback, and support state are fused using rigid-flexible coupling to estimate both the overall six-degree-of-freedom pose of the segment and the local flexible deformation of the closure joint. Afterward, a continuous closure error field is generated along the closure line, and the weldable window is used as the closure state criterion. Finally, if the error field does not enter the weldable window, the actuator adjustment amount is solved and closed-loop adjustment is performed until the pre-welding locking condition is met.

[0042] The above method, through the architecture of "global pose perception - local visual servoing - rigid-flexible coupling fusion - error field judgment - closed-loop solution of adjustment amount", realizes the overall attitude control and local joint quality control during the hull section assembly process. It solves the problems that traditional discrete control point measurement is difficult to reflect the continuous quality of the assemblies, the insufficiency of single visual measurement stability, and the low efficiency of manual experience adjustment.

[0043] S1. Establish the reference coordinate system for the merging area and complete the coordinate calibration.

[0044] like Figure 2 As shown, the multi-coordinate system coordinate chain transformation relationship in this step uses the merging area reference coordinate system as a unified intermediate coordinate system. Merging area reference points, reflective targets, visual calibration plates, or fixed reference stakes are arranged within the slipway, dock, assembly platform, or jig area to establish the merging area reference coordinate system, denoted as [reference needed]. The coordinate system of the design model is denoted as... The reference segmented coordinate system is denoted as Let the coordinate system of the segments to be merged be denoted as The camera coordinate system is denoted as The coordinate system of the actuator is denoted as .

[0045] The homogeneous transformation matrices of each coordinate system relative to the reference coordinate system of the merging area are obtained through total station measurement, camera calibration, target recognition, or point cloud registration. .in, Indicates the design coordinate system To the reference coordinate system of the closure area The transformation relationship, Representing the reference piecewise coordinate system To the reference coordinate system of the closure area The transformation relationship, Represents the segmented coordinate system to be joined To the reference coordinate system of the closure area The transformation relationship. Representing the camera coordinate system To the reference coordinate system of the closure area The transformation relationship, Indicates the coordinate system of the actuator To the reference coordinate system of the closure area The transformation relationship.

[0046] Among them, in this step , , and Used to establish the design model, reference segments, and the reference coordinate system of the camera and actuators relative to the closure area. The static or periodic update calibration relationship; obtained in this step. S3 is used only as the initial pose of the segment to be joined, and is updated online based on the current on-site measurement results of the wheel. This avoids the confusion between coordinate calibration and online pose measurement.

[0047] To avoid ambiguity in the direction of coordinate transformation, this embodiment adopts the following consistent convention: coordinate system Homogeneous coordinate transformation to the reference coordinate system of the merging region ,Right now:

[0048] ,

[0049] in, express Any coordinate system in the coordinate system. The homogeneous transformation matrix is: A matrix, in its specific form, is:

[0050] ,

[0051] in, coordinate system Relative to the reference coordinate system of the merging area The rotation matrix satisfies and ; coordinate system The origin is in the reference coordinate system of the merging area. The translation vector is given below, with length measured in millimeters.

[0052] To obtain by total station measurement For example, select no fewer than three non-collinear calibration control points on the segment to be joined, preferably four or more spatially evenly distributed and non-coplanar calibration control points. Let the first... Each calibration control point is located in the coordinate system of the segment to be merged. The known coordinates in are The total station measured the control point in the reference coordinate system of the closure area. The coordinates in are The rotation matrix is ​​then obtained by solving the following equation. Translation vector :

[0053] ,

[0054] The constraints are and ,in, The weights are set according to the measurement accuracy of the control points, and are used for measurements of equal accuracy. In the specific solution, first calculate the weighted centroids of the two sets of control points:

[0055] ,

[0056] Reconstruct the decentralized covariance matrix:

[0057] ,

[0058] right Perform singular value decomposition ,get:

[0059] ,

[0060] ,

[0061] It is a correction matrix. For the design model coordinate system... and reference piecewise coordinate system The correspondence between design control points and actual measured control points is obtained in the manner described above. and For the camera coordinate system After obtaining the corresponding 3D points from the depth camera, the rigid body registration method described above is used to directly obtain... For the coordinate system of the actuator The zero point of the measuring actuator, the point on the guide axis, and the lateral reference point in the coordinate system With coordinate system The corresponding coordinates are obtained using the rigid body registration method described above. .

[0062] After calibration, calculate the residuals for each calibration control point:

[0063] ,

[0064] And the root mean square residual:

[0065] ,

[0066] when Not greater than the preset calibration tolerance The transformation matrix is ​​accepted if the condition is met; otherwise, outliers are removed and the measurement or registration is repeated. The preset calibration tolerance is determined based on the measurement accuracy of the total station, camera, or actuator, as well as the requirements of the alignment process. During the alignment process, fixed check points can be periodically remeasured; when the check residual exceeds the drift threshold, the corresponding homogeneous transformation matrix is ​​updated.

[0067] The relative pose of the segment to be joined with respect to the reference segment is represented as:

[0068] ,

[0069] in, This indicates that the coordinates in the reference coordinate system A of the merging area will be transformed inversely to the reference segment coordinate system B.

[0070] This step enables coordinate unification among the design model, measurement equipment, vision equipment, actuators, and hull sections, providing a benchmark for subsequent pose fusion and error field calculation.

[0071] S2, import the 3D design model and extract the theoretical fusion information.

[0072] Import the 3D design models of the segments to be joined and the reference segments, and extract information such as the theoretical joining line, theoretical joining surface, key control points, plate thickness, bevel type, welding sequence, welding shrinkage allowance and process tolerance.

[0073] Specifically, this embodiment uses the assembly models of the segments to be joined and the reference segments created in SolidWorks as an example. The models are imported in SLDASM, SLDPRT, or neutral formats such as STEP AP242 and Parasolid. After the models are imported, the solids, surfaces, topological edges, assembly relationships, and attribute information in the models are obtained through the data reading interface of the CAD software. The theoretical joining surfaces of the segments to be joined and the reference segments are determined according to the joining joint number, weld number, or assembly correspondence, and the boundary curves of the theoretical joining surfaces are read. The boundary curves of each segment are connected according to the topological connection relationship to form the theoretical joining line.

[0074] The endpoints, bends, and intersections with ribs and longitudinal bones of the theoretical closure line are used as key control points; the plate thickness is read from the plate thickness attribute of the model or calculated based on the distance between the inner and outer surfaces of the plate; the bevel form is determined based on the included angle and cross-sectional profile of the adjacent bevel surfaces of the closure edge; the welding sequence, welding shrinkage allowance, and process tolerance are read from the model attributes, PMI manufacturing information, or external welding process files, and associated with the corresponding theoretical closure line through the weld number.

[0075] The theoretical convergence line is represented by the arc length parameter. Defined curve:

[0076] ,

[0077] in, It is the length and position along the closure line. This represents the total length of the closure line. When the theoretical closure line is mapped to the site, the benchmark segment or the construction benchmark of the closure area is used as the anchoring object; the mapped theoretical closure line is used as the target benchmark for adjusting the closure segment. The closure line is discretized into... One sampling point:

[0078] ,

[0079] For each sampling point Extract the theoretical closure surface normal. Tangential direction of the closure line Local height direction , plate thickness and bevel parameters .

[0080] S3, acquire global pose observations and calculate the initial six-degree-of-freedom pose.

[0081] Using a total station, laser tracker, 3D laser scanner, or structured light scanning device, control points, targets, or local point clouds on the segment to be joined and the reference segment are measured. Let the segment to be joined in the design model be the first... The control points are The control point was measured on-site in the reference coordinate system of the closure area. The coordinates below are Then the initial pose of the segments to be joined can be obtained through least squares registration.

[0082] The global pose measurement component includes a total station or laser tracker, a set of reference targets mounted on a reference segment, a set of moving targets mounted on the segment to be merged, and a fusion control cabinet communicatively connected to the total station or laser tracker. Both the reference target set and the moving target set include at least four spatially distributed and non-coplanar targets, each fixed to its corresponding segment by a rigid mounting bracket; the reference target set is used to establish the coordinate system of the reference segment. The moving target group is used to establish the coordinate system of the segments to be joined. .

[0083] It should be noted that at the current measurement moment, the segment to be joined is... The control points are in the fixed coordinate system of the segment to be joined. The coordinates below are known reference coordinates, obtained through pre-calibration or from a 3D design model; the coordinates obtained from on-site measurements are the reference coordinates of the same control point in the closure area. The coordinates below are the coordinates of the same entity control point in different coordinate systems. Therefore, based on the first segment to be merged... Each control point is located in the reference coordinate system of the closure area. Measured coordinates below Using the same control point in the coordinate system of the segments to be joined The known coordinates below Based on this, the coordinate correspondence of the control points in the two coordinate systems is established, and the coordinate system of the segments to be joined is obtained by solving the weighted rigid body registration model. Transform to the reference coordinate system of the merging area rotation matrix Translation vector , represented as: in, To control the number of points, For the solution to be found Rotation matrix, Let be the three-dimensional translation vector to be solved. Let the measurement weight of the i-th control point be obtained; and The homogeneous transformation matrix is ​​then constructed. This avoids The dimensional ambiguity caused by directly multiplying a homogeneous matrix with a 3D point that has not been expanded to homogeneous coordinates.

[0084] To further obtain the pose change required for the segment to be joined from its current state to its designed joining state, let the target relative pose of the segment to be joined relative to the reference segment, determined by the design assembly relationship, be: The pose deviation between the current relative pose and the target relative pose is:

[0085] ,

[0086] in, This represents the pose change required for the segment to be joined to reach the target relative pose from its current relative pose. Through... By performing translation and rotation decomposition, the three-dimensional displacement deviation and three-dimensional attitude angle deviation required for subsequent adjustment of the segments to be joined can be obtained.

[0087] Specifically, It can be represented as a homogeneous transformation matrix that includes rotation and translation:

[0088] ,

[0089] in, For the segmented coordinate system to be merged Relative to the reference piecewise coordinate system The rotation matrix, For the segmented coordinate system to be merged The origin is in the reference piecewise coordinate system The translation vector is given below. The translation vector can be expressed as:

[0090] ,

[0091] but Represent the coordinate systems of the segments to be joined. The origin is in the reference piecewise coordinate system The vertical, lateral, and transverse relative position components below. For rotation matrices It can be converted into attitude angles according to a preset Euler angle decomposition order. This embodiment adopts... Euler angle decomposition convention, i.e. .like The element is represented as Then it can be calculated that:

[0092] ,

[0093] ,

[0094] ,

[0095] Represents the arctangent function with quadrant determination. These represent the roll, pitch, and bow-stern rotation angles, respectively. Therefore, the relative pose transformation matrix... Convert to a six-DOF pose vector:

[0096] ,

[0097] in, These represent the longitudinal, lateral, and vertical relative position components of the segment to be joined relative to the reference segment, respectively. These represent the roll, pitch, and bow-stern rotation angles, respectively.

[0098] S4: Collect local visual observations and extract merging features.

[0099] Once the segments to be joined enter the visual capture range, images, depth maps, or laser line cross-sectional data near the joining seam are acquired using a fixed camera, a moving camera, a binocular camera, a depth camera, a line laser camera, or a camera-laser-line combination sensor.

[0100] In this embodiment, the local visual measurement component preferably includes an industrial camera, a line laser, a rigid mounting bracket, an auxiliary lighting device, and a calibration plate. The industrial camera and the line laser are fixed to the rigid mounting bracket at a preset angle, and the laser plane formed by the line laser intersects simultaneously with the reference segment bevel and the segment bevel to be joined. Multiple local visual measurement components are arranged at intervals along the joining seam, or mounted on a scanning guide rail that can move along the joining seam.

[0101] The collected visual data is subjected to image denoising, distortion correction, brightness equalization, edge enhancement and outlier removal. Then, the closing end face, bevel edge, plate seam line, rib end point, longitudinal bone end point, process hole, temporary visual target or natural structural corner point are identified.

[0102] Specifically, the above preprocessing includes: assuming the acquired original image is... The following steps are performed sequentially:

[0103] ,

[0104] ,

[0105] ,

[0106] ,

[0107] ,

[0108] in, It is a bilateral filtering operator that weights neighboring pixels based on pixel spatial distance and gray level difference, and is used to remove image noise and preserve bevel edges; and These are the spatial distance parameter and the grayscale difference parameter, respectively. For distortion correction operators, The camera intrinsic parameter matrix, It is a vector of distortion coefficients that includes radial distortion parameters and tangential distortion parameters; To limit the contrast-adaptive histogram equalization operator, The histogram clipping threshold; For Gaussian smoothing kernel, This represents the convolution operation. This is the edge enhancement coefficient; This includes an edge extraction operator that incorporates gradient computation, nonmaximum suppression, and double thresholding. and These are the low threshold and the high threshold, respectively. This is the set of candidate edges obtained after processing.

[0109] After candidate edge extraction, the laser stripe feature points are converted from image pixel coordinates to three-dimensional spatial coordinates. Let the first... The pixel coordinates of the laser stripe feature points are: Its position in the camera coordinate system The corresponding unit line of sight is:

[0110] ,

[0111] Assume the line laser is in the camera coordinate system The laser plane formed below satisfies Then, the scale and three-dimensional coordinates of the intersection point between the camera's line of sight and the laser plane are as follows:

[0112] ,

[0113] ,

[0114] Then, based on the camera coordinate system obtained from S1 To the reference coordinate system of the closure area The calibration relationship converts 3D points into:

[0115] ,

[0116] in, This is the intrinsic parameter matrix of an industrial camera. and For line laser plane parameters, For laser stripe feature points in the camera coordinate system The three-dimensional coordinates below and camera coordinate system To the reference coordinate system of the closure area The rotation matrix and translation vector, For the transformed reference coordinate system in the merging area The three-dimensional coordinates below.

[0117] For the original 3D point set formed by the depth map or laser line cross section Calculate each point and its... Average distance between neighboring points:

[0118] ,

[0119] And retain the satisfaction Three-dimensional points, where, and These are the mean and standard deviation of the average neighborhood distance, respectively. This represents the outlier removal coefficient. The system identifies the merging feature based on the candidate edge set and the 3D point set after outlier removal.

[0120] In the Within each laser cross section, the intersection points of the laser stripes with the edge of the reference segment bevel and the edge of the segment to be joined are extracted respectively, thus obtaining the reference side points located within the same laser cross section. and the side point to be joined and treat both as the same sampling location. The corresponding measurement point pairs at the location are used to avoid incorrect pairing between the reference side edge points and the edge points to be joined.

[0121] Set at sampling point At this point, the coordinates of the reference segment edge point in the reference coordinate system of the merging area are: The coordinates of the edge points of the segments to be joined in the reference coordinate system of the joining area are: Then the local relative geometric difference is:

[0122] ,

[0123] in, This represents the local spatial difference between the two, which is used in subsequent decomposition to obtain local merging errors such as gaps, misalignments, and height differences.

[0124] S5 performs rigid-flexible coupling pose fusion of global pose observation, local visual observation, actuator feedback, support status and design closing surface data to obtain fused pose and local deformation parameters.

[0125] like Figure 3 As shown, since the hull sections are not ideal rigid bodies, local warping, deflection and port deformation may occur during the support, hoisting, lifting, lateral movement and welding processes. Therefore, this embodiment estimates the local flexible deformation near the closure joint while estimating the overall six-degree-of-freedom pose of the sections to be joined.

[0126] Furthermore, to avoid confusion between the overall rigid body pose and local flexible deformation in the fusion estimation, this embodiment represents the local flexible deformation near the closure joint as a combination of finite-dimensional deformation basis functions distributed along the theoretical closure line. That is, the local deformation field is represented as a linear combination of several preset bending, torsion, port warping, or support settlement modes, with the corresponding modal coefficients used as local flexible deformation parameters. The deformation basis functions are determined by the finite element model of the structural segment to be closed, historical measurement data, or calibration loading tests. During calculation, constraints orthogonal to the overall translation and rotation are set, or zero-mean constraints are set along the theoretical closure line, so that the overall six-degree-of-freedom pose mainly reflects the segmented rigid body motion, and the local flexible deformation parameters mainly reflect the relative deformation near the port.

[0127] The rigid-flexible coupling pose fusion component is housed within a fusion control cabinet, which includes an industrial computer, a synchronous trigger, a data acquisition card, an analog input interface, a communication interface, and an actuator control interface. The global pose measurement component and the local vision measurement component are connected to the industrial computer via an Ethernet interface. The support reaction force sensor and the actuator displacement sensor are connected to the industrial computer via the data acquisition card. The actuator control interface is connected to the lifting unit, the traversing unit, the traction unit, and the clamping unit via a CAN bus, RS485 bus, or EtherCAT bus.

[0128] To improve the ability to express the nonlinear relationship between multi-source observation data and the pose of the segments to be joined and the local flexible deformation, this embodiment sets up a deep regression network based on residual connections. The input of the deep regression network includes the global six-degree-of-freedom pose obtained in S3, the local relative geometric difference obtained in S4, the validity mask, the observation confidence, the adjustment amount of the actuator in the previous round, and the current support reaction force.

[0129] Let the first The network input feature vector collected in each round is:

[0130] ,

[0131] in, Indicates the first Six-DOF pose obtained from global pose observation; Indicates the first The set of local relative geometric differences obtained from local visual observation of the wheel; The mask representing the validity of the nth round of sampling; This indicates the observation reliability of each sampling point in the nth round; This indicates the amount of adjustment made by the implementing agency in the previous round; This represents the current support reaction force at each support point or actuator. During the initial fusion calculation, [the force will be...]. Set it to the zero vector.

[0132] Global pose observation, local visual observation, actuator feedback, and support reaction force data are assigned a unified timestamp by a synchronous trigger. Local visual data are resampled into a fixed number of N three-dimensional spatial differences according to the sampling positions of the theoretical closure line determined by S2. When some sampling positions have no valid observations, they are marked by a validity mask and filled by interpolation of adjacent valid sampling points or preset values, so that the input dimension of the depth regression network remains fixed.

[0133] Furthermore, the validity mask is not only used to fill in missing sampling points, but also serves as an explicit input to the deep regression network. Specifically, the validity mask is used to fill in missing sampling points. The validity mask for round sampling is denoted as The observation confidence level of each sampling point is denoted as . The network input feature vector includes the global six-degree-of-freedom pose, the set of local relative geometric differences, and the validity mask. Observation credibility The previous round of actuator adjustment and the current support reaction force are considered. Therefore, when a sampling point is filled with interpolated or preset values, the network can distinguish between the actual observed value and the filled value, avoiding misjudging missing observations as actual closure errors.

[0134] The set of local relative geometric differences is represented as follows:

[0135] ,

[0136] in, Indicates the first Theoretical closure line in wheel measurement Local relative geometric difference at each sampling point This indicates the number of sampling points on the theoretical closure line.

[0137] The deep regression network includes an input layer, a feature extraction layer, a residual coding module, and an output layer.

[0138] The input layer is used to receive input feature vectors. The feature extraction layer maps the input feature vector to a 256-dimensional feature space and sequentially performs fully connected, batch normalization, and ReLU activation operations. The residual encoding module consists of four structurally identical residual blocks connected in series. Each residual block includes two fully connected layers, and a skip connection is set between the input and output of the residual block to extract nonlinear correlation features between multi-source observation data. The output layer does not set an activation function and directly outputs the regression results.

[0139] The output of the deep regression network is represented as follows:

[0140] ,

[0141] in, Indicates network parameters as Deep regression networks; This represents the six-degree pose correction amount predicted by the neural network; This represents the local flexible deformation parameters predicted by the neural network.

[0142] Furthermore, to ensure the fusion result is identifiable, this embodiment simultaneously introduces global control point residuals, local visual residuals, and execution feedback residuals during training or online calibration. Error components that can be explained solely by global control point residuals are preferentially assigned to the overall six-DOF pose correction; if the error component appears continuously in the local area of ​​the merging line and cannot be eliminated by overall pose correction, it is assigned to the local flexible deformation parameters. Through the above residual decomposition rules, the fusion result can simultaneously output both the overall pose and local deformation, reducing redundant compensation between them.

[0143] The deep regression network can be pre-trained using hull segment deformation data generated by finite element simulation, historical assembly process data, or retest data before and after on-site adjustments. After training, the network parameters are fixed or updated based on on-site retest data to predict the nonlinear correspondence between multi-source observations and pose corrections, as well as local flexible deformation parameters.

[0144] The six-DOF pose correction labels in the training samples are obtained from the difference between the known target relative pose and the measured relative pose under each training state, while the local flexible deformation labels are obtained from finite element calculations, displacement sensor measurements, or calibration loading tests. The network training uses a weighted sum of pose correction errors and local flexible deformation errors as the loss function.

[0145] ,

[0146] in, For deep regression network parameters, and These are the six-degree-of-freedom pose correction and the local flexible deformation prediction output by the network, respectively. and These are the corresponding training labels. and These are the training weights for pose correction error and local flexible deformation error, respectively.

[0147] The six-degree-of-freedom pose correction and local flexible deformation parameters output by the neural network are not directly used as the final control result, but rather as prior inputs to the rigid-flexible coupled pose fusion model. Define the first... The round fusion state vector is:

[0148] ,

[0149] in, This represents the overall six-degree-of-freedom pose of the segment to be joined relative to the reference segment; Indicates the local flexible deformation parameters near the closure joint; Indicates sensor bias; This indicates the compliance parameters of the actuator or support system.

[0150] During the initialization of the fusion state, the initial value of the overall six-DOF pose is obtained by S3, the initial value of the local flexible deformation parameter is determined by the local spatial difference obtained by S4 and the prediction result of the depth regression network, the initial value of the sensor bias is obtained by the sensor static calibration, and the initial value of the compliance parameter of the actuator or support system is obtained by the loading calibration test or the rated parameters of the equipment. At the same time, allowable value ranges are set for the sensor bias and compliance parameters to avoid indistinguishable estimation results among the overall pose, local deformation, sensor bias and compliance parameters.

[0151] Based on global pose observation, local visual observation, actuator feedback, and support status, global pose residuals are constructed respectively. Local visual residuals Feedback from the implementing agency on residuals and support state residuals The fusion state is solved using the following optimization model:

[0152] ,

[0153] in, , , , These are the weight matrices corresponding to global pose observation, local visual observation, actuator feedback, and support status, respectively. These are the constraint coefficients for the neural network pose prediction results; These are the constraint coefficients for the neural network's prediction of flexible deformation.

[0154] The first four terms in the aforementioned optimization model ensure consistency between the fusion result and the actual measurement data, while the last two terms introduce pose correction and local flexible deformation parameters from the deep regression network prediction. Thus, the neural network learns the nonlinear relationship between multi-source observations, and the rigid-flexible coupling optimization model corrects the network prediction results based on the actual measurement residuals.

[0155] When local visual observation is affected by occlusion, reflection, smoke, or missing features, the system reduces the weight matrix of local visual observation. When the global control point or point cloud registration residual increases, reduce the global pose observation weight matrix. When the feedback from the actuator is inconsistent with the actual pose change, the abnormal penalty corresponding to the actuator feedback residual is increased; when the support reaction force is close to the boundary of the allowable range, the weight of the support state residual is increased.

[0156] Through the above deep regression network prediction and rigid-flexible coupling optimization solution, the first... Wheel fusion pose and local flexible deformation parameters The fused pose and local flexible deformation parameters serve as inputs for S6 to construct a continuous closing error field, used to calculate and correct gaps, misalignments, height differences, and end-face angles at each sampling point.

[0157] S6, construct a continuous closure error field along the theoretical closure line or theoretical closure surface, and generate a weldable window based on the welding process parameters.

[0158] like Figure 4 As shown, this step constructs a continuous merging error field distributed along the theoretical merging line. Based on the direction of the theoretical merging line, the local visual observations obtained in S4, and the fusion pose and local flexible deformation results obtained in S5, the local relative geometric differences are decomposed into gaps, misalignments, height differences, and end-face angle errors.

[0159] Specifically, the system first maps the theoretical closure line in the design model to the closure area reference coordinate system. And set according to the preset spacing and the endpoints of the closure line, the corner points, the positions of the ribs and the longitudinal bones. One sampling point:

[0160] ,

[0161] For each sampling point, the system establishes a local orthogonal coordinate base, including the unit normal of the merging surface. Tangential direction of the closure line unit and local height direction .in:

[0162] ,

[0163] The local coordinate base is used to represent the local spatial difference obtained in S4 and corrected by the fusion state in S5.

[0164] Decompose the error into error components with different engineering implications:

[0165] ,

[0166] ,

[0167] ,

[0168] ,

[0169] in, Indicates the gap between the two halves. This indicates a misalignment along the tangential direction of the closure line or the direction of the plate edge. This indicates the difference in height between the two segmented plate surfaces in the local height direction. This represents the normal angle between the two bevel faces; This indicates the unit normal of the local end face of the segment to be joined. Indicates the unit normal of the local end face of the reference segment; This indicates that the calculated value is restricted to a range. Inside.

[0170] Before calculating the included angle of the end face normals, the end face normals of the segments to be joined and the reference segments are unified to the same reference direction; when the two use opposite external normal definitions, one of the end face normals is reversed. Furthermore, the first... The discrete convergence error of each sampling point is expressed as:

[0171] ,

[0172] Considering the issues of measurement discontinuities, local obstructions, abnormal observations, and inconsistent stiffness at different structural locations among the sampling points, this embodiment does not directly connect the discrete error points sequentially. Instead, it uses a cubic weighted average of observation reliability and local structural stiffness. Spline function reconstruction of continuous closing error field:

[0173] ,

[0174] in, For the first Three times spline basis functions These are the corresponding error field coefficients. for Number of spline basis functions This represents the total theoretical closure line length. The error field coefficients are solved using the following optimization model:

[0175] ,

[0176] in, For the first The reliability of observations at each sampling point; This is the error normalization matrix, used to eliminate dimensional differences between gaps, misalignments, height differences, and end face angles; For position Local structural stiffness at the location; This represents the structural continuity constraint coefficient.

[0177] The first term of the aforementioned optimization model is used to make the continuous closing error field approximate the local visual observation results, and the second term is used to suppress unreasonable error abrupt changes along the closing line direction based on the continuity of the hull structure and the local structural stiffness. For areas with high local stiffness, such as ribs and longitudinal bones, the corresponding continuity constraints are increased; for areas with high local flexibility or anti-deformation requirements, the corresponding continuity constraints are decreased.

[0178] When the reliability of visual observation decreases at a sampling point due to occlusion, reflection, smoke, or missing features, the system reduces the accuracy of that point's sampling rate. This allows the error at that location to be interpolated or predicted primarily based on adjacent reliable sampling points, the theoretical closure line shape, and structural continuity constraints. When the length of the low-confidence continuous region exceeds a preset threshold... This area is marked as the retest area. The final continuous merging error field is:

[0179] ,

[0180] in, , , and These represent the gap field, misalignment field, height difference field, and end face angle error field, which are continuously distributed along the theoretical closure line, respectively.

[0181] Furthermore, based on plate thickness, bevel type, welding sequence, welding shrinkage allowance, local structural stiffness, and the current closure stage, weldable windows at various locations along the theoretical closure line are generated. (Location defined) The welding process parameter vector at the location is:

[0182] ,

[0183] in, Indicates plate thickness. Indicates the bevel form, Indicates the welding sequence. This indicates the allowance for welding shrinkage. Indicates the local structural stiffness. This indicates the current closing stage.

[0184] Establish window parameter mapping relationships based on welding process databases or welding process specifications:

[0185] ,

[0186] in, This represents a mapping function that determines the allowable error range based on welding process parameters. In one optional implementation, the window parameter mapping relationship is established using welding process specifications, shipyard assembly standards, or historical assembly quality data. The system determines the upper and lower limits or target ranges for clearance, misalignment, height difference, and end-face normal angle based on plate thickness, bevel type, welding sequence, shrinkage allowance, and local structural stiffness, and uses these as the weldable window parameters for the current position. For different ship types, plate thickness ranges, welding methods, or enterprise process standards, the above allowable ranges can be retrieved from the process database or pre-configured by process engineers, but a defined window boundary should be formed within the same control cycle for subsequent window distance calculations and closed-loop adjustment judgments. Position The weldable window at the location is represented as follows:

[0187] ,

[0188] in, This represents the local error vector to be determined. and These are the lower and upper limits of the allowable gap, respectively. To allow for misaligned edge thresholds, To allow for a high-low difference threshold, To allow the threshold of the included angle of the end face normal.

[0189] The weldable window is not a fixed, single threshold, but rather an allowable error range that varies depending on the sampling location, local structural form, and welding process requirements. For areas with larger plate thickness or higher local structural stiffness, stricter thresholds are set for misalignment and height differences; for areas with welding shrinkage allowance requirements, the allowable gap range is adjusted accordingly based on the welding sequence and heat input direction.

[0190] To avoid directly mixing and comparing millimeter and angular quantities, a normalized distance from the continuous closing error field to the weldable window is defined:

[0191] ,

[0192] in, This is a diagonal normalized matrix composed of the reciprocals of the allowable dimensions of gaps, misalignments, height differences, and end-face angles. When lie in When the error is within the window, the distance is 0; when one or more error components exceed the window boundary, the distance represents the degree of normalized out-of-tolerance.

[0193] The system calculates the maximum distance from the current error field of the entire closure line to the weldable window:

[0194] ,

[0195] In actual calculations, the continuous closing error field can be finely sampled along the theoretical closing line, and the maximum normalized deviation among all finely sampled points can be taken. If the following conditions are met:

[0196] ,

[0197] And the length of the low-confidence continuous region does not exceed If the current closing state is found to be within the weldable window, pre-welding locking can be performed; otherwise, if the current closing state does not meet the pre-welding locking conditions, further adjustments are required.

[0198] when When the threshold requirement is met but there are continuous low-confidence areas, the system requires retesting or supplementary observations of that area; when When the threshold is exceeded, the system outputs the out-of-tolerance sampling location, out-of-tolerance error type, and normalized out-of-tolerance degree, and passes them to the adjustment calculation step in S7.

[0199] S7 determines whether the continuous closing error field enters the weldable window, and if it does not, solves for the adjustment amount with confidence-adaptive safety constraints.

[0200] When S6 calculates the window distance index Less than or equal to the preset threshold And the length of the low-confidence continuous region does not exceed Enter the weldable window based on the current closing state and output a pre-weld lock command. When the length of the low-confidence continuous region exceeds... The system first outputs instructions for retesting or supplementing observations, and does not use the prediction results for the low-confidence area as the sole basis for automatic adjustment of the actuator.

[0201] When the window distance indicator Greater than the preset threshold If the local visual observation meets the minimum confidence requirement for solving the adjustment amount, it is determined that the current closing state has not entered the weldable window. The system solves for the next adjustment amount based on the continuous closing error field, actuator stroke, speed limit, support reaction force, anti-collision boundary, and welding process constraints.

[0202] Before calculating the adjustment amount, the system determines the main source of error based on the out-of-tolerance location, error type, and the overall six-degree-of-freedom pose and local flexible deformation parameters obtained from S5. If multiple sampling positions show errors with consistent directions and similar trends, the system prioritizes overall pose deviation and activates overall lifting, lateral movement, longitudinal movement, or traction units. If the error is mainly concentrated in a local area and corresponds to changes in local flexible deformation or support state, the system activates local lifting, local support, or clamping units.

[0203] Assume the implementing agency has a total of There are one controllable degree of freedom, and the adjustment amount is expressed as:

[0204] ,

[0205] in, , The total number of controllable degrees of freedom of the implementing agency. For the implementing agency number, Indicates the first The adjustment amount of each lifting unit, lateral movement unit, longitudinal movement unit, traction unit, clamping unit, or local support unit within the current control cycle.

[0206] The relationship between the actuator adjustment and the pose increment of the six-DOF segment to be closed is expressed as follows:

[0207] ,

[0208] in, To execute the adjusted six-DOF pose increments, including three translation increments and three rotation increments; For the execution agency Jacobian matrix, its first... Column represents the first The effect of a single actuator generating a unit adjustment on the pose of the six degrees of freedom of the closing segment.

[0209] The Jacobian matrix of the execution mechanism Based on the actuator layout, support point location, and calibration test results, and corrected according to the actuator or support system compliance parameters obtained from S5, the actual adjustment effect is reflected by the actuator clearance, support elasticity, and structural flexibility.

[0210] Furthermore, implement adjustment amounts The subsequent prediction continuous closing error field is expressed as:

[0211] ,

[0212] in, The position of the arc length on the theoretical closure line. This is the theoretical total length of the closure line; The pre-adjustment continuous closing error vector constructed for S6 includes gap, misalignment, height difference and end face angle error in sequence; To implement the adjustment amount The prediction error vector after; Let be the error field sensitivity matrix.

[0213] The The Column represents the first When an implementing agency generates a unit adjustment, the position of the theoretical closure line is... The influence of gaps, misalignment, height differences, and end face angles. According to the Jacobi matrix of the implementing agencies The local coordinate base of the theoretical closure line and the local flexible deformation parameters obtained from S5 were calculated.

[0214] The adjustment amount is solved using the following optimization model:

[0215] ,

[0216] in, The theoretical closure line position The error weights at each location are as follows: critical structural locations, locations with severe deviations, or locations with priority in the welding sequence correspond to larger weights. The weldable window generated for S6; This is the normalized distance from the adjusted prediction error vector to the weldable window; To adjust the penalty coefficient; To support the penalty coefficient for changes in reaction force; Represents the Euclidean norm; Indicates the amount of adjustment to be performed. The resulting vector of reaction force changes at each support point.

[0217] The first term of the above optimization model is used to push the adjusted continuous closing error field into the weldable window; the second term is used to limit the single adjustment amount to avoid excessive adjustment and excessive movement of the mechanism; and the third term is used to suppress abrupt changes in support reaction force to avoid abnormal stress on local structures.

[0218] The optimization model satisfies the following constraints on actuator stroke, speed, and support reaction force:

[0219] ,

[0220] ,

[0221] ,

[0222] in, These are the lower and upper limit vectors of the allowable adjustment amount for each actuator, and the vector inequality holds for each corresponding component. For the first The permissible speed of each implementing agency; To control the cycle.

[0223] For the present The reaction force vector of each support point; These are the lower and upper limits of the allowable reaction force at each support point, respectively, and the reaction force constraint is established for each support point separately.

[0224] Furthermore, to avoid adjusting according to a fixed safety distance when the reliability of visual observation decreases, this embodiment calculates the current minimum effective reliability based on the continuous observation reliability obtained in S6:

[0225] ,

[0226] For position The higher the value, the more reliable the local visual observation. This represents the lowest valid observation confidence level within the entire theoretical convergence line.

[0227] The collision avoidance safety distance is dynamically adjusted based on the minimum effective confidence level.

[0228] ,

[0229] The dynamic safety distance used in the current control cycle; This is the basic safe distance when visual observation is normal; This is a low-confidence safety distance magnification factor. It is applied when visual observation is affected by occlusion, reflection, smoke, or missing features. The system will increase the collision avoidance safety distance accordingly.

[0230] Define the execution adjustment amount The subsequent collision avoidance control barrier function:

[0231] ,

[0232] in, To implement the adjustment amount Then, the predicted minimum distance between the segment to be joined and the benchmark segment, surrounding equipment, or temporary support structure; This is the adjusted collision avoidance safety margin.

[0233] The predicted minimum distance is obtained based on the coordinate transformation relationship established in S1, the three-dimensional design model in S2, and the adjusted predicted pose. When, it indicates that the collision avoidance requirements are met; when When the predicted minimum distance is less than the dynamic safe distance, it indicates that the predicted minimum distance is less than the dynamic safe distance.

[0234] Add the following control barrier constraints to the optimization model:

[0235] ,

[0236] in, The original collision avoidance safety margin. This is the barrier adjustment coefficient. This constraint not only requires that the adjusted state not exceed the collision safety boundary, but also limits the speed at which the segments to be joined approach the collision boundary.

[0237] When applying the control barrier constraint, the current state should satisfy:

[0238] ,

[0239] If the current state no longer meets the safety distance requirements, the system stops automatic adjustment and outputs a reversal or manual safety handling command. Therefore, the complete constraints of the optimization model are expressed as:

[0240] ,

[0241] Solving the objective function under the above constraints yields the optimal adjustment for the current control cycle:

[0242] ,

[0243] in, This represents the adjustment amount that minimizes the objective function under all constraints. This represents the optimal actuator adjustment amount output during the current control cycle.

[0244] The optimal adjustment amount The data is transmitted to S8, where it is executed by the corresponding lifting, lateral, longitudinal, traction, clamping, or local support units. After execution, the system re-acquires the global pose, local vision, and support status, and updates the fused pose, local flexible deformation parameters, continuous closing error field, and window distance index. Then, reassess whether to enter the weldable window.

[0245] If the optimization model has no feasible solution under the current constraints, the system first keeps the actuator inactive and outputs constraint conflict types according to preset priorities. These preset priorities include, in order: anti-collision safety constraints, support reaction force constraints, actuator travel constraints, actuator speed constraints, and weldable window convergence targets. When an anti-collision safety constraint or support reaction force constraint is violated, the system stops automatic adjustment and outputs a manual confirmation or rollback command. When infeasibility is caused solely by travel or speed constraints, the system uses slack variables to resolve the phased adjustment amount and breaks down a single adjustment into multiple rounds of small-step adjustments, thereby avoiding the continued output of infeasible adjustment commands in a solutionless state.

[0246] S8 performs phased visual servo closed-loop adjustment based on the adjustment amount and generates pre-welding lock records and quality traceability data.

[0247] This step is used to convert the adjustment amount obtained by S7 into field execution instructions, and after each execution, global pose and local visual data are re-acquired to determine whether the continuous closing error field has entered the weldable window.

[0248] Let the first At the start of the adjustment cycle, the fused pose obtained by S5 is: The continuous closing error field obtained by S6 is The window distance index is The theoretical adjustment obtained by S7 is .

[0249] The system determines the adjustment step size coefficient based on the current closing stage. The actual adjustment amount implemented in this round is obtained as follows:

[0250] ,

[0251] in, For the first The actual adjustment amount is executed in the round; in the coarse positioning stage, Take the larger value; during the fine-fit seam stage and the pre-weld locking stage, Take the smaller value. Before execution, predict the adjusted continuous closing error field based on the error response matrix:

[0252] ,

[0253] in, For the first The wheel error response matrix is ​​used to describe the effect of the adjustment amount of each actuator on the position of the closing line. The influence relationship between gaps, misalignments, height differences, and end face angle errors; the error response matrix is ​​determined based on the actuator layout parameters, structural model, calibration test, and compliance parameters obtained in S5.

[0254] The predicted adjusted window distance index is:

[0255] ,

[0256] in, This represents the maximum normalized distance from the adjusted continuous error field to the weldable window.

[0257] The implementing agency shall After completing the lifting, lateral movement, longitudinal movement, traction, or clamping operations, the system re-acquires actuator feedback, support reaction force, global pose, and local visual data, and re-executes rigid-flexible coupling pose fusion and continuous error field calculation to obtain the measured fused pose. Continuous closing error field And window distance metrics:

[0258] ,

[0259] To compare the predicted adjustment effects with the actual adjustment effects, a consistency coefficient is calculated:

[0260] ,

[0261] in, This indicates the degree of agreement between the actual decrease in error and the predicted decrease in error. To prevent positive numbers with a denominator of zero. When When the value is small, it indicates that the actual adjustment effect is lower than the predicted effect, and the system reduces the adjustment step size in the next round; when When the value is close to 1, it indicates that the actual adjustment effect is basically consistent with the predicted effect, and the system maintains the current adjustment step size; when Larger and If necessary, the system can appropriately increase the adjustment step size for the next round. The adjustment step size is updated as follows:

[0262] ,

[0263] in, Adjust the step size coefficient. and The threshold for consistency judgment, The step size decay coefficient is less than 1. This is a step size increment factor greater than 1. The system determines that the segment to be joined has entered the pre-welding locking zone when the following conditions are met:

[0264] ,

[0265] in, The length of a continuous low-confidence region. This is the current supporting reaction force. If the above conditions are not met, then... , The feedback and support status from the implementing agency serve as input for the next round, and S5 to S8 are re-executed.

[0266] To complete the step size update logic, when the decrease in prediction error is less than a preset positive number or the adjusted window distance is not lower than the current window distance, the consistency coefficient is adjusted to low confidence, and the step size for the next round of adjustment is reduced. When the consistency coefficient is greater than the second threshold but the actual window distance does not decrease, the step size is not increased; instead, it is maintained or decreased, and the error response matrix is ​​re-estimated. If, for several consecutive rounds, the actual window distance does not decrease, the observation confidence is lower than the threshold, or the deviation between the actuator feedback and the command exceeds the threshold, the system suspends automatic closed-loop adjustment and is marked as a retest or manual intervention state.

[0267] Once the continuous closing error field enters the weldable window, the system generates a pre-welding locking command and records the final relative pose, final continuous error field, window distance index, cumulative adjustment of each actuator, low confidence area, support reaction force, fixture locking status, operation time, and controller number, forming quality traceability data.

[0268] Experimental verification

[0269] To illustrate the feasibility of the method described in this embodiment, a simulation verification environment for hull section assembly can be constructed. Preset translational deviations, attitude deviations, and local port deformations are set for the sections to be assembled, and global measurement equipment, local vision equipment, actuator feedback, and support status data are simulated. The verification process is not limited to actual ship production data, but rather used to illustrate the working principles of coordinate calibration, rigid-flexible coupling pose fusion, continuous error field construction, weldable window determination, and closed-loop adjustment in this method.

[0270] In the specific verification, firstly, the geometric models of the reference segment and the segment to be joined are established in the simulation model, and multiple sampling points are set along the theoretical joining line; then, the initial six-degree-of-freedom pose deviation and local flexible deformation are applied to the segment to be joined to generate the joining error field before adjustment; then, according to S5 to S8 of this embodiment, the fusion estimation, window determination and adjustment amount solution are performed, and the gap, misalignment, height difference, end face angle and window distance index after each round of adjustment are recorded.

[0271] In an exemplary simulation verification, 20 sampling points are set along the theoretical closure line. The initial longitudinal deviation of the segment to be closed is 8mm, the lateral deviation is 6mm, the vertical deviation is 5mm, and the attitude angle deviation is 0.30° to 0.50°. A warping deformation of up to 3mm is superimposed locally at the port. After using the fusion estimation, weldable window determination, and staged closed-loop adjustment of this embodiment, after 4 rounds of adjustment, the maximum closure gap decreased from 5.8mm to 2.6mm, the maximum misalignment decreased from 3.2mm to 0.9mm, the maximum height difference decreased from 2.7mm to 0.8mm, the end face normal angle decreased from 4.5° to 1.6°, and the window distance index decreased from 1.42 to 0.38. The above example is used to illustrate the executable logic between the steps and does not limit the equipment model, plate thickness range, or final values ​​in actual engineering.

[0272] like Figure 6 As shown, the experimental verification simulation scenario includes a reference segment, a segment to be joined, a joining seam area, a global measurement device, a local vision device, and an actuator, which is used to simulate the acquisition of multi-source observation data and the closed-loop adjustment process during the on-site joining process.

[0273] like Figure 7As shown, the joining error indicators before and after adjustment can be compared using bar charts and line graphs. The bar chart represents the changes in gap, misalignment, height difference, end face angle, and window distance before and after adjustment, while the line graph represents the convergence trend of the error after adjustment. This graph illustrates that after closed-loop adjustment, the degree of deviation of the continuous joining error field relative to the weldable window is reduced.

[0274] The simulation verification and image processing results described above demonstrate that the method described in this embodiment can link design closure information, global pose observation, local visual features, actuator feedback, and support status to form a verification link from error identification and window determination to closed-loop adjustment.

[0275] Example 2

[0276] Reference Figure 5 This embodiment provides a hull segment assembly system based on rigid-flexible coupling pose fusion and visual servoing.

[0277] A computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing.

[0278] A terminal device includes a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor as described in the "Ship Section Assembly Method Based on Rigid-Flexible Coupling Pose Fusion and Visual Servo".

[0279] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for assembling ship hull sections based on rigid-flexible coupling pose fusion and visual servoing, characterized in that, include: Establish a reference coordinate system for the closure area and extract theoretical closure information based on the three-dimensional design model; Collect global pose observations and calculate the initial six-DOF pose; Collect local visual observations and extract merging features; Rigid-flexible coupled pose fusion is performed based on the acquired global pose observations and local visual observations. A continuous closing error field is constructed based on fused pose and theoretical closing information, and a weldable window is generated. Adjustments are generated based on the constructed closing error field and weldable window; Perform phased visual servo closed-loop adjustments based on the adjustment amount and generate records.

2. The method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing according to claim 1, characterized in that, The process of establishing a reference coordinate system for the closure area and extracting theoretical closure information based on a three-dimensional design model includes arranging closure area reference points, reflective targets, visual calibration plates, or fixed reference piles within the slipway, dock, assembly platform, or jig area to establish a closure area reference coordinate system, denoted as […]. The coordinate system of the design model is denoted as The reference segmented coordinate system is denoted as Let the coordinate system of the segments to be merged be denoted as The camera coordinate system is denoted as The coordinate system of the actuator is denoted as The homogeneous transformation matrices of each coordinate system relative to the reference coordinate system of the merging area are obtained through total station measurement, camera calibration, target recognition, or point cloud registration. To avoid ambiguity in the direction of coordinate transformation, a unified convention is adopted. coordinate system Homogeneous coordinate transformation to the reference coordinate system of the merging region ,Right now: in, express For any coordinate system in the given coordinate system, the homogeneous transformation matrix is ​​in the form of: in, coordinate system Relative to the reference coordinate system of the merging area The rotation matrix satisfies and ; coordinate system The origin is in the reference coordinate system of the merging area. The translation vector is then used; then, at least three non-collinear calibration control points are selected on the segment to be merged, and let the first... Each calibration control point is located in the coordinate system of the segment to be merged. The known coordinates in are The total station measured the control point in the reference coordinate system of the closure area. The coordinates in are The rotation matrix is ​​then obtained by solving the expression. Translation vector The expression is: The constraints are and ,in, The weights are set according to the measurement accuracy of the control points, and are used for measurements of equal accuracy. In the specific solution, first calculate the weighted centroids of the two sets of control points: Reconstruct the decentralized covariance matrix: ,right Perform singular value decomposition ,get: , , This is the correction matrix; after calibration, the residuals and root mean square residuals of each calibration control point are calculated, and the final relative pose of the segment to be merged relative to the reference segment is expressed as: in, This indicates that the coordinates in the reference coordinate system A of the merging area will be transformed inversely to the reference segment coordinate system B.

3. The method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing according to claim 2, characterized in that, The acquisition of global pose observations and calculation of the initial six-degree-of-freedom pose include, at the current measurement moment, based on the segment to be joined, the first... Each control point is located in the reference coordinate system of the closure area. Measured coordinates below Using the same control point in the coordinate system of the segments to be joined Known coordinates below Based on this, the coordinate correspondence of the control points in the two coordinate systems is established, and the coordinate system of the segments to be joined is obtained by solving the weighted rigid body registration model. Transform to the reference coordinate system of the merging area rotation matrix Translation vector , represented as: in, To control the number of points, For the solution to be found Rotation matrix, Let be the three-dimensional translation vector to be solved. Let the measurement weight of the i-th control point be obtained; and The homogeneous transformation matrix is ​​then constructed. To obtain the pose change required for the segment to be joined from its current state to its designed joining state, let the target relative pose of the segment to be joined relative to the reference segment, determined by the design assembly relationship, be: The pose deviation between the current relative pose and the target relative pose is: in, This represents the pose change required for the segment to be joined to reach the target relative pose from its current relative pose. This is achieved by... By performing translation and rotation decomposition, the three-dimensional displacement deviation and three-dimensional attitude angle deviation required for subsequent adjustments of the segments to be joined are obtained. Represented as a homogeneous transformation matrix including rotation and translation: , For the segmented coordinate system to be merged Relative to the reference piecewise coordinate system The rotation matrix, For the segmented coordinate system to be merged The origin is in the reference piecewise coordinate system The translation vector below is expressed as: ,but These are respectively taken as the longitudinal, lateral, and vertical positional deviations of the segment to be joined relative to the reference segment; for the rotation matrix... Converted into attitude angles according to the preset Euler angle decomposition order, when using the ZYX Euler angle decomposition method, if The element is represented as Then the calculation yields: , , ,in Represents the arctangent function with quadrant determination. These represent the roll, pitch, and bow-stern rotation angles, respectively, thus forming the relative pose transformation matrix. Convert to a six-DOF pose vector: in, These represent the longitudinal, lateral, and vertical relative position components of the segment to be joined relative to the reference segment.

4. The method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing according to claim 3, characterized in that, The process of acquiring local visual observations and extracting closure features includes: when the segment to be closed enters the visual capture range, using a fixed camera, a moving camera, a binocular camera, a depth camera, a line laser camera, or a camera-laser line combination sensor to acquire images, depth maps, or laser line cross-sectional data near the closure seam; and performing image denoising, distortion correction, brightness equalization, edge enhancement, and outlier removal on the acquired visual data; subsequently identifying the closure end face, bevel edge, plate seam line, rib endpoints, longitudinal bone endpoints, process holes, temporary visual targets, or natural structural corner points. Specifically, let the acquired original image be... Image processing is performed sequentially to achieve candidate edge extraction: , , , in, It is a bilateral filtering operator that weights neighboring pixels based on pixel spatial distance and gray level difference, and is used to remove image noise and preserve bevel edges; and These are the spatial distance parameter and the grayscale difference parameter, respectively. For distortion correction operators, The camera intrinsic parameter matrix, It is a vector of distortion coefficients that includes radial distortion parameters and tangential distortion parameters; To limit the contrast-adaptive histogram equalization operator, The histogram clipping threshold; For Gaussian smoothing kernel, This represents the convolution operation. This is the edge enhancement coefficient; This includes an edge extraction operator that incorporates gradient computation, nonmaximum suppression, and double thresholding. and These are the low threshold and the high threshold, respectively. This is the set of candidate edges obtained after processing; after completing the candidate edge extraction, the laser stripe feature points are converted from image pixel coordinates to three-dimensional spatial coordinates, let the first... The pixel coordinates of the laser stripe feature points are: In the camera coordinate system The corresponding unit line of sight is: Assume the line laser is in the camera coordinate system The laser plane formed below satisfies Then, the scale and three-dimensional coordinates of the intersection point between the camera's line of sight and the laser plane are as follows: , Then, based on the camera coordinate system To the reference coordinate system of the closure area The calibration relationship converts 3D points into: ,in, This is the intrinsic parameter matrix of an industrial camera. and For line laser plane parameters, For laser stripe feature points in the camera coordinate system The three-dimensional coordinates below and camera coordinate system To the reference coordinate system of the closure area The rotation matrix and translation vector, For the reference coordinate system in the merging area after transformation The three-dimensional coordinates below; for the original three-dimensional point set formed by the depth map or laser line cross section. Calculate each point and its... Average distance between neighboring points: and retain the satisfaction Three-dimensional points, where, and These are the mean and standard deviation of the average neighborhood distance, respectively. is the outlier removal coefficient; in the Within each laser cross section, the intersection points of the laser stripes with the edge of the reference segment bevel and the edge of the segment to be joined are extracted respectively, thus obtaining the reference side points located within the same laser cross section. and the side point to be joined and treat both as the same sampling location. The corresponding measurement point pairs at the location are used to avoid incorrect pairing between the reference side edge points and the edge points to be joined.

5. The method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing according to claim 4, characterized in that, The rigid-flexible coupled pose fusion based on acquired global pose observations and local visual observations includes setting the first... The network input feature vector collected in the first round is: in, Indicates the first Six-DOF pose obtained from global pose observation; Indicates the first The set of local relative geometric differences obtained from local visual observation of the wheel; The mask representing the validity of the nth round of sampling; This indicates the observation confidence level of each sampling point in the nth round. This indicates the amount of adjustment made by the implementing agency in the previous round; This represents the support reaction force of each support point or actuator at present. During the initial fusion calculation, it will... Set as the zero vector, global pose observation, local visual observation, actuator feedback, and support reaction force data are assigned a unified timestamp by a synchronous trigger. Local visual data is resampled into a fixed number of N three-dimensional spatial differences according to the determined theoretical convergence line sampling positions. When some sampling positions have no valid observations, they are marked by a validity mask and filled by interpolation of adjacent valid sampling points or preset values, thereby keeping the input dimension of the depth regression network fixed. The validity mask for round sampling is denoted as The observation confidence level of each sampling point is denoted as . The network input feature vector includes the global six-degree-of-freedom pose, the set of local relative geometric differences, and the validity mask. Observation credibility The previous round of actuator adjustment and the current support reaction force, represented by the local relative geometric difference set, are as follows: ,in, Indicates the first Theoretical closure line in wheel measurement Local relative geometric difference at each sampling point This represents the number of sampling points on the theoretical convergence line; the input layer of a deep regression network receives the input feature vector. The feature extraction layer maps the input feature vector to a 256-dimensional feature space and sequentially performs fully connected, batch normalization, and ReLU activation operations; the output of the deep regression network is represented as: in, Indicates network parameters as Deep regression networks; This represents the six-degree pose correction amount predicted by the neural network; This represents the local flexible deformation parameters predicted by the neural network; the network training uses a weighted sum of pose correction error and local flexible deformation error as the loss function: in, For deep regression network parameters, and These are the six-degree-of-freedom pose correction and the local flexible deformation prediction output by the network, respectively. and These are the corresponding training labels. and Let be the training weights for pose correction error and local flexible deformation error, respectively, and define the _th_ weight. The round fusion state vector is: in, This represents the overall six-degree-of-freedom pose of the segment to be joined relative to the reference segment; Indicates the local flexible deformation parameters near the closure joint; Indicates sensor bias; This represents the compliance parameters of the actuator or support system; finally, based on global pose observation, local visual observation, actuator feedback, and support status, global pose residuals are constructed respectively. Local visual residuals Feedback from the implementing agency on residuals and support state residuals The fusion state is solved through an optimization model and is expressed as: in, , , , These are the weight matrices corresponding to global pose observation, local visual observation, actuator feedback, and support status, respectively. These are the constraint coefficients for the neural network pose prediction results; These are the constraint coefficients for the neural network's prediction of flexible deformation.

6. The method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing according to claim 5, characterized in that, The process of constructing a continuous merging error field and generating a weldable window based on fused pose and theoretical merging information includes decomposing local relative geometric differences into gaps, misalignments, height differences, and end-face angle errors based on the theoretical merging line direction, local visual observations, and fused pose and local flexible deformation results. First, the theoretical merging line is mapped to the merging area reference coordinate system. And set according to the preset spacing and the endpoints of the closure line, the corner points, the positions of the ribs and the longitudinal bones. One sampling point: For each sampling point, a local orthogonal coordinate base is established, including the unit normal of the merging surface. Tangential direction of the closure line unit and local height direction ,in: The local coordinate base is used to calculate the local spatial difference after the fusion state is corrected. Decompose the error into error components with different engineering implications: , , , ,in, Indicates the gap between the two halves. This indicates a misalignment along the tangential direction of the closure line or the direction of the plate edge. This indicates the difference in height between the two segmented plate surfaces in the local height direction. This represents the normal angle between the two bevel faces; This indicates the unit normal of the local end face of the segment to be joined. Indicates the unit normal of the local end face of the reference segment; This indicates that the calculated value is restricted to a range. Inside; and will the first The discrete convergence error of each sampling point is expressed as: Finally, a third-order weighted average of observation confidence and local structural stiffness was used. Spline function reconstruction of continuous closing error field: in, For the first Three times spline basis functions These are the corresponding error field coefficients. for Number of spline basis functions Given the theoretical total length of the closure line, the error field coefficients are solved using the following optimization model, expressed as: , in, For the first The reliability of observations at each sampling point; This is the error normalization matrix, used to eliminate dimensional differences between gaps, misalignments, height differences, and end face angles; For position Local structural stiffness at the location; This represents the structural continuity constraint coefficient.

7. The method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing according to claim 6, characterized in that, The process of constructing a continuous closure error field and generating weldable windows based on fused pose and theoretical closure information also includes generating weldable windows at various positions along the theoretical closure line based on plate thickness, bevel type, welding sequence, welding shrinkage allowance, local structural stiffness, and the current closure stage, and defining the positions. The welding process parameter vector at the location is: ,in, Indicates plate thickness. Indicates the bevel form, Indicates the welding sequence. This indicates the allowance for welding shrinkage. Indicates the local structural stiffness. This indicates the current closure stage; a window parameter mapping relationship is established based on the welding process database or welding process specification: , in, This represents a mapping function that determines the allowable error range based on welding process parameters, and its position. The weldable window at the location is represented as follows: in, This represents the local error vector to be determined. and These are the lower and upper limits of the allowable gap, respectively. To allow for misaligned edge thresholds, To allow for a high-low difference threshold, To allow a threshold for the included angle of the end face normal; to avoid direct comparison of millimeter and angular quantities, a normalized distance from the continuous closing error field to the weldable window is defined: in, The matrix is ​​a diagonal normalized matrix composed of the reciprocals of the allowable dimensions of gaps, misalignments, height differences, and end face angles. lie in When the error is within the window, this distance is 0; when one or more error components exceed the window boundary, this distance represents the normalized out-of-tolerance level; finally, the maximum distance from the current error field of the entire closure line to the weldable window is calculated: Simultaneously, the continuous closing error field is sampled more densely along the theoretical closing line, and the maximum normalized deviation among all the densely sampled points is taken. If the following conditions are met: And the length of the low-confidence continuous region does not exceed If the current closing state is met, the welding window is entered; otherwise, the welding window is not met and further adjustments are needed.

8. The method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing according to claim 7, characterized in that, The adjustment amount generated based on the constructed closing error field and weldable window includes determining the current error source according to the out-of-tolerance location, error type, and six-degree-of-freedom pose and local flexible deformation parameters. Assume the actuator has a total of... There are one controllable degree of freedom, and the adjustment amount is expressed as: in, , The total number of controllable degrees of freedom of the implementing agency. For the implementing agency number, Indicates the first The adjustment amounts of each lifting unit, lateral movement unit, longitudinal movement unit, traction unit, clamping unit, or local support unit within the current control cycle; the relationship between the actuator adjustment amounts and the six-degree-of-freedom pose increments of the segment to be closed is expressed as follows: in, To execute the adjusted six-degree-of-freedom pose increment; For the execution agency Jacobian matrix, its first... Column represents the first The effect of each actuator generating a unit adjustment on the pose of the six-DOF segment to be closed; and thus the execution adjustment. The subsequent prediction continuous closing error field is expressed as: in, The position of the arc length on the theoretical closure line. This represents the total theoretical length of the closure line. The pre-adjustment continuous closing error vector constructed for S6; To implement the adjustment amount The prediction error vector after; The error field sensitivity matrix, with the adjustment amount obtained through an optimization model, is expressed as: , in, The theoretical closure line position Error weights at each location; For weldable windows; This is the normalized distance from the adjusted prediction error vector to the weldable window; To adjust the penalty coefficient; To support the penalty coefficient for changes in reaction force; Represents the Euclidean norm; Indicates the amount of adjustment to be performed. The resulting vector of reaction force changes at each support point; simultaneously, the optimized model satisfies the constraints of actuator stroke, speed, and support reaction force; to avoid adjusting according to a fixed safety distance when the reliability of visual observation decreases, the current minimum effective reliability is calculated based on the reliability of continuous observation: , For position The reliability of the observations at that location; The minimum valid observation confidence level along the entire theoretical convergence line; and the collision avoidance safety distance is dynamically adjusted based on the minimum valid confidence level: , The dynamic safety distance used in the current control cycle; This is the basic safe distance when visual observation is normal; This is a low-confidence safety distance magnification factor, used when visual observation is affected by occlusion, reflection, smoke, or missing features. The system correspondingly increases the collision avoidance safety distance and defines the adjustment amount to be executed. The subsequent collision avoidance control barrier function: in, To implement the adjustment amount Then, the predicted minimum distance between the segment to be joined and the benchmark segment, surrounding equipment, or temporary support structure; To adjust the collision avoidance safety margin, control barrier constraints are finally added to the optimization model: in, The original collision avoidance safety margin. Here is the barrier adjustment coefficient; the complete constraint of the optimization model is expressed as: Solve the objective function under constraints to obtain the optimal adjustment for the current control cycle: in, This represents the adjustment amount that minimizes the objective function under all constraints. This is the optimal actuator adjustment amount output in the current control cycle.

9. The method for hull segment assembly based on rigid-flexible coupling pose fusion and visual servoing according to claim 8, characterized in that, The step of performing phased visual servo closed-loop adjustment based on the adjustment amount includes setting the first... At the start of the adjustment cycle, the obtained fused pose is: The obtained continuous closing error field is The window distance index is The theoretical adjustment amount obtained is Then, determine the adjustment step size coefficient based on the current closing stage. The actual adjustment amount implemented in this round is obtained as follows: in, For the first The actual adjustment amount is executed in the round; in the coarse positioning stage, Take the larger value; during the fine-fit seam stage and the pre-weld locking stage, Take the smaller value, and before execution, predict the adjusted continuous closing error field based on the error response matrix: ,in, For the first Round error response matrix; the predicted adjusted window distance index is: in, This represents the maximum normalized distance from the predicted adjusted continuous error field to the weldable window; the actuator follows... After completing the lifting, lateral movement, longitudinal movement, traction, or clamping operations, the actuator feedback, support reaction force, global pose, and local visual data are re-acquired. The rigid-flexible coupling pose fusion and continuous error field calculation are then re-performed to obtain the measured fused pose. Continuous closing error field And window distance metrics: To compare the predicted adjustment effects with the actual adjustment effects, a consistency coefficient is calculated: ,in, This indicates the degree of agreement between the actual decrease in error and the predicted decrease in error. To prevent positive numbers with a denominator of zero; adjust the step size to update as follows: in, Adjust the step size coefficient. and The threshold for consistency judgment, The step size decay coefficient is less than 1. The step size increment is greater than 1. When the judgment condition is met, the segment to be joined is determined to enter the pre-welding locking zone. in, The length of a continuous low-confidence region. This serves as the current supporting force.

10. A hull segment assembly system based on rigid-flexible coupling pose fusion and visual servoing, executing the hull segment assembly method based on rigid-flexible coupling pose fusion and visual servoing as described in claim 1, characterized in that... include: The data acquisition module is configured to establish a reference coordinate system for the merging area and extract theoretical merging information based on the three-dimensional design model; The global module is configured to acquire global pose observations and calculate the initial six-degree-of-freedom pose. The local module is configured to acquire local visual observations and extract merging features; The fusion module is configured to perform rigid-flexible coupled pose fusion based on the acquired global pose observations and local visual observations; The window module is configured to construct a continuous closing error field and generate a weldable window based on fused pose and theoretical closing information; The adjustment module is configured to generate adjustment amounts based on the constructed closure error field and weldable window; The execution module is configured to perform phased visual servo closed-loop adjustments based on the adjustment amount and generate records.