A laser three-dimensional scanning assisted double stern line type CSOV sectional assembly precision control method

Abstract

The application relates to the technical field of shipbuilding precision control, and discloses a laser three-dimensional scanning auxiliary double-stern line type CSOV sectional assembly precision control method, which realizes precision budget allocation through double-stern super-sectional division, coordinate system building and error transmission modeling, optimizes the assembly sequence in combination with a genetic algorithm, globally optimizes the precision, builds a multi-source fusion scanning system, restores sectional real geometric deviation in combination with a YOLOv8 blind area identification and strain and deformation decoupling algorithm, pre-processes point clouds in a parallel architecture, realizes high-precision alignment through two-stage strategies of FPFH coarse registration and ICP fine registration, completes three-dimensional comparison and deviation visualization, identifies three types of core errors, calculates compensation and generates a six-degree-of-freedom positioning scheme, and builds a BP neural network welding shrinkage prediction model to realize welding deformation closed-loop control in combination with a PID algorithm; the application can control the total section assembly comprehensive deviation within + / - 2 mm, improves assembly precision and construction efficiency, and is suitable for high-precision modular construction of high-end ships.

Description

Technical Field

[0001] This invention relates to the field of shipbuilding precision control technology, and in particular to a laser three-dimensional scanning-assisted method for precision control of segmented assembly of twin-stern CSOVs. Background Technology

[0002] As offshore wind power continues to develop towards deeper and more distant waters, CSOVs (Offshore Engineering Service Vessels), as the core equipment for the entire life cycle operation and maintenance of wind power, are facing increasingly higher requirements in terms of construction quality and performance. The Ulstein TwinX-Stern twin-stern hull form, with its excellent seakeeping, fuel economy, and operational stability, has become the mainstream hull form for high-end CSOVs. The complex curved structure of this hull form means that the assembly precision of 6-8 super sections directly determines the ship's hydrodynamic performance, propulsion efficiency, and operational safety, making it a core control point in CSOV construction.

[0003] Currently, the construction of twin-stern linear CSOVs in China presents a situation of "leading design but weak manufacturing," and existing laser scanning-assisted assembly technology still has many shortcomings in practical applications:

[0004] Firstly, the segmentation and assembly sequence rely heavily on engineering experience and lack a quantitative model for error propagation and a reverse allocation mechanism for precision budget. The cumulative error of the overall assembly is prone to exceed the standard, which cannot meet the precision requirements of high-end products.

[0005] Secondly, the lack of a unified fusion mechanism for multi-source scanning data makes it difficult to dynamically adjust the scanning path according to the curvature of the surface, easily leading to data blind spots. Furthermore, the elastic deformation during hoisting and the segmented manufacturing deviations are coupled and difficult to decouple accurately, resulting in distorted scanning results. Thirdly, the large volume of point cloud data in the dual-stern segment makes traditional serial processing inefficient, and the single registration algorithm has slow convergence speed and insufficient docking accuracy. Fourthly, the docking error identification dimension is singular, and the compensation method is fixed, making it difficult to achieve high-precision positioning. Fifthly, the lack of welding shrinkage prediction and dynamic anti-deformation control during welding results in a high rate of deviations after welding, making it difficult to control rework costs and construction cycle. Sixthly, the data of the entire assembly process is not systematically archived, and the process iteration lacks data support, making it difficult to form a replicable high-precision assembly process system.

[0006] While leading international shipbuilding companies have mastered mature twin-stern section assembly technology, their core algorithms and process solutions remain confidential. Furthermore, the technology is highly dependent on high-end specialized equipment, resulting in high construction costs and making it difficult to adapt to the existing equipment conditions of domestic shipyards. Therefore, there is an urgent need to develop a laser 3D scanning-assisted twin-stern linear CSOV section assembly precision control method that is suitable for the current domestic construction situation, in order to fill the domestic technological gap and support the independent upgrading of the CSOV industry. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to provide a laser 3D scanning-assisted method for precision control of twin-stern linear CSOV segment assembly. This method integrates global optimization, multi-source scanning, GPU parallel processing, intelligent prediction, and dynamic control technologies to construct a full-process, high-precision, and intelligent twin-stern segment assembly precision control system. It incorporates proprietary control algorithms and formulas to achieve precise control of key aspects such as error propagation, deformation decoupling, and welding shrinkage. The method is adaptable to the existing equipment levels of domestic shipyards, forming a low-cost and replicable process solution.

[0008] A laser 3D scanning-assisted method for controlling the assembly accuracy of a segmented double-stern linear CSOV includes the following steps:

[0009] S1: Based on the curved structural characteristics of the twin-stern hull and the constraints of the assembly process, the super segmentation is completed, a global coordinate system and a segment local coordinate system are constructed, the local coordinate system and the global coordinate system are accurately connected, and it serves as a unified measurement benchmark for precision control.

[0010] Based on this benchmark, an assembly error propagation model is established. The accuracy budget of each segment is allocated in reverse according to the total assembly accuracy index, and a fitness function is constructed to optimize and determine the segment assembly sequence with the goal of minimizing the cumulative assembly error.

[0011] S2: Build a multi-source fusion 3D scanning system to scan the segments to be assembled, calculate the surface curvature value based on the surface complexity, classify the surface complexity level based on the preset curvature threshold, plan the corresponding scanning path, and identify and fill in the data blind spots.

[0012] Synchronously collect segmented hoisting strain data to decouple the interference of hoisting deformation on the scanning results and restore the true geometric deviation of the segments;

[0013] S3: A parallel computing architecture is used to preprocess the point cloud data. A combined registration algorithm of feature pre-alignment and fine adjustment is used to uniformly register the point cloud data in each segment's local coordinate system to the global coordinate system by combining coarse registration and fine registration.

[0014] S4: Compare the preprocessed point cloud data with the theoretical point cloud data of the design, calculate the deviation values ​​of key areas, and generate deviation visualization results and deviation analysis reports.

[0015] S5: Based on the deviation analysis results, identify the multi-dimensional errors of the segmented docking end faces, calculate the corresponding compensation amount, and generate a six-degree-of-freedom adjustment scheme by combining the assembly process requirements and the adjustment equipment parameters.

[0016] S6: Complete the segmented precise positioning and temporary fixation according to the positioning scheme, and establish a neural network welding shrinkage prediction model. With welding process parameters as input and welding shrinkage as output, the model is trained through welding test data to predict the welding shrinkage deformation.

[0017] During the welding process, deformation data is monitored in real time and compared with the predicted shrinkage. A PID dynamic adjustment algorithm is used to dynamically adjust the welding parameters to implement anti-deformation control. After welding, the out-of-tolerance areas are repaired with precision, and the repair is verified by scanning again.

[0018] S7: Perform a full-domain 3D scan inspection on the assembled segment, collect overall point cloud data, and compare it with the designed theoretical model to determine whether the accuracy index is qualified. If the inspection is qualified, proceed to the next segment assembly process; if it is not qualified, return to step S5 to recalculate the compensation amount, and repeat steps S6-S7 for adjustment.

[0019] Furthermore: In step S1, the CSOV hull is divided into 6-8 super sections, and the docking surfaces of the sections avoid areas of abrupt curvature changes. Each super section contains a complete double-stern curved surface unit and a propeller base mounting area. A global coordinate system is established with the hull keel centerline as the X-axis, the width direction as the Y-axis, and the draft direction as the Z-axis. A local section coordinate system is established with the geometric center of each super section as the origin. The coordinate system is precisely docked through coordinate transformation.

[0020] Furthermore: In step S1, the cumulative assembly error of the whole section is quantified by the assembly error propagation model. The error propagation coefficient is determined according to the stiffness of the segment structure. The greater the stiffness, the closer the error propagation coefficient is to 1.0. Based on the requirement that the assembly accuracy of the whole section does not exceed ±2mm, the error budget of the core segment of the double stern curved surface is allocated to not exceed ±0.8mm, and the error budget of the ordinary segment does not exceed ±1.2mm.

[0021] A genetic algorithm is used to optimize the assembly sequence. A fitness function is constructed with the goal of minimizing the total assembly error and assembly cycle. The optimal assembly sequence is obtained through iterative optimization by selection, crossover and mutation operations, with ≥100 iterations.

[0022] Furthermore: In step S2, the multi-source fusion 3D scanning system consists of a laser scanning unit, a structured light scanning unit, and a photogrammetry unit; wherein the laser scanning accuracy is ≤ ±0.1mm, the structured light scanning resolution is ≤ 0.05mm, and the photogrammetry accuracy is ≤ ±0.15mm;

[0023] In step S2, the high-complexity surface region and the low-complexity surface region are divided according to the curvature threshold ≥0.05m⁻¹. The high-complexity region adopts dense scanning with a scanning spacing ≤5mm, and the low-complexity region adopts sparse scanning with a scanning spacing ≤10mm. The YOLOv8AI algorithm is used to identify the scanning blind zone in real time. The blind zone determination criterion is point cloud density <10 points / mm².

[0024] In step S2, strain sensors are arranged at the edge of the segmented double stern curved surface and near the propeller base to collect hoisting strain data; a strain-deformation conversion matrix is ​​established based on the finite element model, and the hoisting deformation deviation is deducted from the scanning measurement deviation through the deformation decoupling algorithm to restore the actual geometric deviation of the segment.

[0025] Furthermore: In step S3, the CUDA parallel computing architecture is used to perform point cloud denoising, filtering, and simplification preprocessing operations based on GPU acceleration; a parallel Gaussian filtering algorithm is used to remove point cloud noise, with a filtering window radius of 4; and a voxel grid method is used to simplify the point cloud, with the simplification rate controlled at 40%~50%.

[0026] In step S3, the fast point feature histogram features of the point cloud are extracted first and the feature distance is calculated to achieve coarse alignment with a coarse alignment error of ≤ ±0.5mm. Then, the ICP algorithm is used for fine adjustment, and the rotation matrix and translation vector are iteratively optimized. The iteration termination condition is that the deviation between two adjacent iterations is ≤ 0.02mm, the number of iterations does not exceed 50, and the final total registration error is ≤ ±0.15mm.

[0027] Furthermore: In step S4, the visualization results are a color deviation cloud map and a deviation distribution histogram; the deviation distribution is presented using pseudo-color mapping, where a deviation ≤ ±0.5mm is a qualified area, a deviation in the range of ±0.5~±1.5mm is a warning area, and a deviation > ±1.5mm is an out-of-tolerance area.

[0028] Furthermore: In step S5, the identified multi-dimensional errors of the mating end face include mating gap, misalignment deviation, and angle deviation; wherein the identification accuracy of mating gap and misalignment deviation is ≤ ±0.05mm, and the identification accuracy of angle deviation is ≤ ±0.1°; marking the out-of-tolerance areas where the gap is > 0.5mm, misalignment is > ±0.8mm, and angle is > ±0.5°.

[0029] In step S5, a linear interpolation compensation algorithm is used to calculate the compensation amount for each deviation point. The compensation amount and the deviation value are linearly related, and the calculation error of the compensation amount is ≤ ±0.1mm. A symmetrical compensation strategy is adopted for the symmetrical parts of the two sterns to ensure the symmetry of the two stern lines.

[0030] Furthermore: In step S6, a six-degree-of-freedom positioning device is used to perform segmented positioning, and a laser tracking device is used to collect three-dimensional coordinate data in real time during the positioning process, with a positioning accuracy of ≤ ±0.1mm; after the positioning is completed, a positioning fixture is used to temporarily fix the segments, and the positioning accuracy is verified by scanning again;

[0031] In step S6, a BP neural network welding shrinkage prediction model is established, with welding current, voltage, speed, and weld thickness as input parameters and welding shrinkage as output parameter. The model training error is ≤5%. During the welding process, a PID dynamic adjustment algorithm is used to adjust the welding parameters in real time and implement anti-deformation control to control the welding deformation within ±0.2mm.

[0032] In step S6, after welding is completed, the segments are laser scanned and retested. For areas with out-of-tolerance, local grinding and welding are used to repair the accuracy. After repair, the accuracy is scanned again to verify whether it meets the standard.

[0033] Furthermore: In step S7, the qualified indicators for assembly accuracy inspection include: assembly deviation of double stern curved outer plate ≤ ±2mm, gap between segmented mating surfaces ≤ 0.5mm, misalignment deviation ≤ ±0.8mm, and angle deviation ≤ ±0.5°; if the inspection fails, the compensation amount is recalculated, and the adjustment, welding, and inspection process is repeated.

[0034] Furthermore: A laser three-dimensional scanning-assisted double-stern linear CSOV segmented assembly accuracy control system, comprising:

[0035] The module includes a global accuracy control module, a 3D scanning and deformation decoupling module, a point cloud processing and unified registration module, a model comparison and deviation analysis module, a docking error compensation module, a welding process control module, and an accuracy inspection and process optimization module.

[0036] The global precision control module is used to divide super segments based on the curved structural characteristics of the twin-stern hull and assembly process constraints, construct a two-level coordinate reference system of global and segment, establish an assembly error control model, and inversely allocate the precision index of each segment to optimize and determine the segment assembly sequence.

[0037] The 3D scanning and deformation decoupling module is used to collect 3D point cloud data of the segments to be assembled, plan differentiated scanning paths based on surface features and automatically fill in scanning blind spots, synchronously collect segment hoisting deformation data and decouple hoisting deformation interference, and restore the true geometric deviation of the segments.

[0038] The point cloud processing and unified registration module is used to preprocess the point cloud data using a parallel processing architecture, and to uniformly match each segment of point cloud to the global coordinate system through a coarse-fine registration method; the model comparison and deviation analysis module is used to perform a three-dimensional comparison between the restored segmented actual point cloud data and the theoretical model value data, calculate the deviation of key areas and generate deviation analysis results.

[0039] The docking error compensation module is used to identify multi-dimensional errors of the segmented docking end faces based on the deviation analysis results, calculate the corresponding compensation amount, and generate a segmented adjustment scheme.

[0040] The welding process control module is used to complete segmented precise positioning and temporary fixation according to the positioning scheme, predict the amount of welding shrinkage deformation, dynamically adjust welding parameters to implement anti-deformation control during the welding process, and perform precision repair on out-of-tolerance areas after welding.

[0041] The precision inspection and process optimization module is used to perform full-domain three-dimensional precision inspection on the assembled segments, determine whether the precision indicators are qualified, archive the assembly data of the whole process, and iteratively optimize the assembly process parameters based on historical data.

[0042] The present invention has the following beneficial effects:

[0043] (1) Solved the problems of unreasonable segmentation and assembly sequence and excessive error propagation accumulation: Through global optimization and genetic algorithm optimization, the scientific design of segmentation and assembly sequence is realized. The error propagation model and reverse error budget allocation ensure that the total assembly cumulative error is ≤ ±2mm, which meets the accuracy requirements of high-end CSOV and reduces the risk of error accumulation.

[0044] (2) It solves the problems of low accuracy and deformation interference in multi-source scanning data fusion: multi-source data fusion and intelligent scanning path planning eliminate data blind spots and redundancy, and improve data integrity; the deformation decoupling algorithm accurately separates hoisting deformation from actual geometric deviation, improves the authenticity of scanning data, and provides a reliable basis for accuracy judgment.

[0045] (3) It solves the problems of low efficiency and insufficient registration accuracy in point cloud processing: The CUDA parallel computing architecture improves the efficiency of point cloud processing, and the FPFH and ICP combined registration algorithm controls the registration error within ±0.15mm, realizing the accurate docking of the local coordinate system and the global coordinate system, which greatly improves the efficiency and accuracy of point cloud processing.

[0046] (4) It solves the problems of inaccurate identification of docking error and lack of targeted compensation scheme: the multi-dimensional error identification algorithm can accurately identify various docking errors, the linear interpolation compensation algorithm realizes the dynamic calculation of compensation amount, the adjustment accuracy is ≤ ±0.1mm, and the deviation rate is reduced after adjustment.

[0047] (5) It solves the problem of difficult control of welding shrinkage deformation: the prediction accuracy of the BP neural network welding shrinkage prediction model is ≥95%, and the PID dynamic anti-deformation control controls the welding deformation within ±0.2mm, reducing the post-weld deviation rate and reducing rework costs and assembly cycle.

[0048] (6) Solved the problem of assembly data not being systematically archived: The segmented assembly quality database realizes full-process data archiving and traceability, and optimizes the process through historical data analysis and iteration to form a replicable high-precision assembly scheme that is compatible with the existing equipment level of domestic shipyards, reducing manufacturing costs, shortening the assembly cycle, and reducing rework rate.

[0049] (7) Filling the domestic technology gap: For the segmented assembly of twin-stern linear CSOV, a full-process high-precision control system was constructed, and exclusive algorithms and formulas were integrated to fill the domestic patent gap in high-precision manufacturing process of twin-stern linear CSOV, promote the independent upgrading of CSOV industry, enhance the market competitiveness of my country's high-end CSOV, and provide technical support for the development of offshore wind power equipment industry. Attached Figure Description

[0050] Figure 1 This is a flowchart of a laser 3D scanning-assisted method for controlling the precision of segmented assembly of a twin-stern linear CSOV. Detailed Implementation

[0051] The present invention will be further described in detail below with reference to specific embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0052] Please see the appendix Figure 1 The present invention provides an embodiment 1: a laser three-dimensional scanning-assisted method for controlling the assembly accuracy of a segmented double-stern linear CSOV, comprising the following specific implementation steps:

[0053] S1: Based on the curved structural characteristics of the twin-stern hull and the constraints of the assembly process, the super segmentation is completed, a global coordinate system and a segment local coordinate system are constructed, the local coordinate system and the global coordinate system are accurately connected, and it serves as a unified measurement benchmark for precision control.

[0054] Based on this benchmark, an assembly error propagation model is established. The accuracy budget of each segment is allocated in reverse according to the total assembly accuracy index, and a fitness function is constructed to optimize and determine the segment assembly sequence with the goal of minimizing the cumulative assembly error.

[0055] S2: Build a multi-source fusion 3D scanning system to scan the segments to be assembled, calculate the surface curvature value based on the surface complexity, classify the surface complexity level based on the preset curvature threshold, plan the corresponding scanning path, and identify and fill in the data blind spots.

[0056] Synchronously collect segmented hoisting strain data to decouple the interference of hoisting deformation on the scanning results and restore the true geometric deviation of the segments;

[0057] S3: A parallel computing architecture is used to preprocess the point cloud data. A combined registration algorithm of feature pre-alignment and fine adjustment is used to uniformly register the point cloud data in each segment's local coordinate system to the global coordinate system by combining coarse registration and fine registration.

[0058] S4: Compare the preprocessed point cloud data with the theoretical point cloud data of the design, calculate the deviation values ​​of key areas, and generate deviation visualization results and deviation analysis reports.

[0059] S5: Based on the deviation analysis results, identify the multi-dimensional errors of the segmented docking end faces, calculate the corresponding compensation amount, and generate a six-degree-of-freedom adjustment scheme by combining the assembly process requirements and the adjustment equipment parameters.

[0060] S6: Complete the segmented precise positioning and temporary fixation according to the positioning scheme, and establish a neural network welding shrinkage prediction model. With welding process parameters as input and welding shrinkage as output, the model is trained through welding test data to predict the welding shrinkage deformation.

[0061] During the welding process, deformation data is monitored in real time and compared with the predicted shrinkage. A PID dynamic adjustment algorithm is used to dynamically adjust the welding parameters to implement anti-deformation control. After welding, the out-of-tolerance areas are repaired with precision, and the repair is verified by scanning again.

[0062] S7: Perform a full-domain 3D scan inspection on the assembled segment, collect overall point cloud data, and compare it with the designed theoretical model to determine whether the accuracy index is qualified. If the inspection is qualified, proceed to the next segment assembly process; if it is not qualified, return to step S5 to recalculate the compensation amount, and repeat steps S6-S7 for adjustment.

[0063] Furthermore, in another embodiment of the present invention, in step S1, the CSOV hull is divided into 6-8 super segments, and the segment docking surfaces avoid areas of abrupt changes in surface curvature. Each super segment contains a complete double-stern curved surface unit and a propeller base mounting area. A global coordinate system is established with the hull keel centerline as the X-axis, the beam direction as the Y-axis, and the draft direction as the Z-axis. A local segment coordinate system is established with the geometric center of each super segment as the origin. Precise docking of the coordinate systems is achieved through coordinate transformation.

[0064] Furthermore, in another embodiment of the present invention, in step S1, the cumulative error of the overall assembly is quantified by the assembly error propagation model. The error propagation coefficient is determined according to the stiffness of the segment structure. The greater the stiffness, the closer the error propagation coefficient is to 1.0. Based on the requirement that the overall assembly accuracy does not exceed ±2mm, the error budget of the double stern curved core segment is allocated to not exceed ±0.8mm, and the error budget of the ordinary segment does not exceed ±1.2mm.

[0065] A genetic algorithm is used to optimize the assembly sequence. A fitness function is constructed with the goal of minimizing the total assembly error and assembly cycle. The optimal assembly sequence is obtained through iterative optimization by selection, crossover and mutation operations, with ≥100 iterations.

[0066] Furthermore, in another embodiment of the present invention, in step S2, the multi-source fusion three-dimensional scanning system consists of a laser scanning unit, a structured light scanning unit, and a photogrammetry unit; wherein the laser scanning accuracy is ≤ ±0.1 mm, the structured light scanning resolution is ≤ 0.05 mm, and the photogrammetry accuracy is ≤ ±0.15 mm.

[0067] In step S2, the curvature threshold is ≥0.05m. -1 To standardize the division of surface regions into high-complexity and low-complexity regions, high-complexity regions are characterized by dense scanning with a scan spacing ≤ 5 mm, while low-complexity regions are characterized by sparse scanning with a scan spacing ≤ 10 mm. The YOLOv8AI algorithm is used to identify scanning blind zones in real time, with the blind zone determination criterion being a point cloud density < 10 points / mm². 2 ;

[0068] In step S2, strain sensors are arranged at the edge of the segmented double stern curved surface and near the propeller base to collect hoisting strain data; a strain-deformation transformation matrix is ​​established based on the finite element model, and the hoisting deformation deviation is deducted from the scanning measurement deviation through the deformation decoupling algorithm to restore the actual geometric deviation of the segment;

[0069] Furthermore, in another embodiment of the present invention, in step S3, a CUDA parallel computing architecture is used to perform preprocessing operations of point cloud denoising, filtering, and simplification based on GPU acceleration; a parallel Gaussian filtering algorithm is used to remove point cloud noise, with a filtering window radius of 4; and a voxel grid method is used to simplify the point cloud, with the simplification rate controlled at 40%~50%.

[0070] In step S3, the fast point feature histogram features of the point cloud are extracted first and the feature distance is calculated to achieve coarse alignment with a coarse alignment error of ≤ ±0.5mm. Then, the ICP algorithm is used for fine adjustment, and the rotation matrix and translation vector are iteratively optimized. The iteration termination condition is that the deviation between two adjacent iterations is ≤ 0.02mm, the number of iterations does not exceed 50, and the final total registration error is ≤ ±0.15mm.

[0071] Furthermore, in another embodiment of the present invention, in step S4, the visualization result is to generate a color deviation cloud map and a deviation distribution histogram; the deviation distribution is presented using pseudo-color mapping, wherein a deviation ≤ ±0.5mm is a qualified area, a deviation in the range of ±0.5~±1.5mm is a warning area, and a deviation > ±1.5mm is an out-of-tolerance area.

[0072] Furthermore, in another embodiment of the present invention, in step S5, the identified multi-dimensional errors of the mating end face include mating gap, misalignment deviation and angle deviation; wherein the identification accuracy of mating gap and misalignment deviation is ≤ ±0.05mm, and the identification accuracy of angle deviation is ≤ ±0.1°; and the out-of-tolerance areas are marked with gap > 0.5mm, misalignment > ±0.8mm, and angle > ±0.5°.

[0073] In step S5, a linear interpolation compensation algorithm is used to calculate the compensation amount for each deviation point. The compensation amount and the deviation value have a linear correspondence, and the calculation error of the compensation amount is ≤ ±0.1mm. A symmetrical compensation strategy is adopted for the symmetrical parts of the twin sterns to ensure the symmetry of the twin stern profile.

[0074] Furthermore, in another embodiment of the present invention, in step S6, a six-degree-of-freedom positioning device is used to perform segmented positioning, and a laser tracking device is used to collect three-dimensional coordinate data in real time during the positioning process, with a positioning accuracy of ≤ ±0.1mm; after the positioning is completed, a positioning fixture is used to temporarily fix the segments, and the positioning accuracy is verified by scanning again.

[0075] In step S6, a BP neural network welding shrinkage prediction model is established, with welding current, voltage, speed, and weld thickness as input parameters and welding shrinkage as output parameter. The model training error is ≤5%. During the welding process, a PID dynamic adjustment algorithm is used to adjust the welding parameters in real time and implement anti-deformation control to keep the welding deformation within ±0.2mm.

[0076] In step S6, after welding is completed, the segments are laser scanned and retested. For areas with out-of-tolerance, local grinding and welding are used to repair the accuracy. After repair, the accuracy is scanned again to verify whether it meets the standard.

[0077] Furthermore, in another embodiment of the present invention, in step S7, the qualified indicators for assembly accuracy inspection include: assembly deviation of double stern curved outer plate ≤ ±2mm, gap between segmented mating surfaces ≤ 0.5mm, misalignment deviation ≤ ±0.8mm, and angle deviation ≤ ±0.5°; if the inspection fails, the compensation amount is recalculated, and the adjustment, welding, and inspection process is repeated.

[0078] An embodiment 2 of the present invention provides a laser three-dimensional scanning-assisted segmented assembly accuracy control system for twin-stern linear CSOVs, comprising:

[0079] The module includes a global accuracy control module, a 3D scanning and deformation decoupling module, a point cloud processing and unified registration module, a model comparison and deviation analysis module, a docking error compensation module, a welding process control module, and an accuracy inspection and process optimization module.

[0080] The global precision control module is used to divide super segments based on the curved structural characteristics of the twin-stern hull and assembly process constraints, construct a two-level coordinate reference system of global and segment, establish an assembly error control model, and inversely allocate the precision index of each segment to optimize and determine the segment assembly sequence.

[0081] The 3D scanning and deformation decoupling module is used to collect 3D point cloud data of the segments to be assembled, plan differentiated scanning paths based on surface features and automatically fill in scanning blind spots, synchronously collect segment hoisting deformation data and decouple hoisting deformation interference, and restore the true geometric deviation of the segments.

[0082] The point cloud processing and unified registration module is used to perform preprocessing on point cloud data using a parallel processing architecture, and to uniformly match each segment of point cloud to the global coordinate system through a combination of coarse and fine registration. The model comparison and deviation analysis module is used to perform three-dimensional comparison between the restored segmented actual point cloud data and the theoretical model value data, calculate the deviation of key areas and generate deviation analysis results.

[0083] The docking error compensation module is used to identify multi-dimensional errors of the segmented docking end faces based on the deviation analysis results, calculate the corresponding compensation amount, and generate a segmented adjustment scheme.

[0084] The welding process control module is used to complete segmented precise positioning and temporary fixation according to the positioning plan, predict the amount of welding shrinkage deformation, dynamically adjust welding parameters during the welding process to implement anti-deformation control, and perform precision repair on out-of-tolerance areas after welding.

[0085] The Precision Inspection and Process Optimization module is used to perform full-domain three-dimensional precision inspection on the assembled segments, determine whether the precision indicators are qualified, archive the assembly data of the whole process, and iteratively optimize the assembly process parameters based on historical data.

[0086] An application example provided by this invention:

[0087] Step S1: Twin-stern line segmentation and global optimization

[0088] 1.1 Segmentation: Based on the surface features, structural strength distribution, and assembly process requirements of the Ulstein TwinX-Stern twin stern hull, the CSOV hull is divided into 6-8 super segments. The segmentation of the symmetrical twin stern parts is optimized to ensure that the segment docking surfaces avoid areas of abrupt curvature changes, thus reducing docking difficulty. Each super segment contains a complete twin stern curved surface unit and a propeller base mounting area, which facilitates modular assembly.

[0089] 1.2 Coordinate System Establishment: A global coordinate system is established with the keel centerline of the CSOV hull as the X-axis, the beam direction as the Y-axis, and the draft direction as the Z-axis. A segmental local coordinate system is established with the geometric center of each super segment as the origin. The local coordinate system and the global coordinate system are accurately connected through coordinate transformation formulas, with a connection error ≤ ±0.1mm. The theoretical value point cloud data of the segments is extracted from the design model, including the three-dimensional coordinates of key parts such as the twin stern surfaces, docking surfaces, and propeller base mounting surfaces, forming a theoretical value database.

[0090] 1.3 Error Propagation Modeling and Budget Allocation: Establish a segmented assembly error propagation model, using formulas... Quantify the cumulative error of the overall assembly, including the error propagation coefficient. Determined based on the stiffness of the segmented structure (the greater the stiffness, the better). (The closer to 1.0) The cumulative overall error for the assembly of the main section is expressed in mm. This represents the total number of segments in the assembly sequence. For the first The manufacturing and assembly errors of each segment are expressed in units of... ;

[0091] Based on the overall assembly accuracy requirement (≤±2mm), the error budget for each segment is allocated in reverse. The error budget for the core segment (double stern curved surface segment) is ≤±0.8mm, and the error budget for ordinary segments is ≤±1.2mm.

[0092] 1.4 Assembly Sequence Optimization: A genetic algorithm is used to optimize the assembly sequence and construct a fitness function. The weighting coefficient Take 0.4, Encode individuals for segmented assembly sequences. For sequence The corresponding total assembly cycle, in units of The assembly sequence optimization objective is to minimize the total assembly error and assembly cycle by using integer encoding. The assembly sequence is iteratively optimized through selection, crossover and mutation operations (iteration times ≥ 100 times) to finally determine the optimal assembly sequence, ensuring that the cumulative error propagation is minimized and the assembly cycle is minimized.

[0093] Step S2: Multi-source fusion scanning and deformation decoupling

[0094] 2.1 Multi-source scanning system setup: Build a multi-source fusion scanning system integrating laser scanning, structured light scanning, and photogrammetry. The laser scanner's scanning accuracy should be ≤ ±0.1mm, the structured light scanning resolution ≤ 0.05mm, and the photogrammetry accuracy ≤ ±0.15mm. Fix the scanning equipment in a stable position on the assembly site to ensure that the scanning range covers the entire section to be assembled without significant obstruction.

[0095] 2.2 Intelligent Scanning Path Planning: Based on the complexity analysis of the double stern surface, the surface curvature value is calculated, with a curvature threshold ≥ 0.05m. -1 As a standard, the curved surface is divided into high-complexity regions (the transition between the two stern curved surfaces) and low-complexity regions; the scanning path is intelligently planned, with dense scanning (scanning spacing ≤ 5mm) used in high-complexity regions and sparse scanning (scanning spacing ≤ 10mm) used in low-complexity regions, thereby improving scanning efficiency and data accuracy.

[0096] 2.3 Blind Spot Identification and Supplementary Scanning: The YOLOv8AI algorithm is used to analyze the scanned data in real time to identify data blind spots (scanned point cloud density < 10 points / mm). 2 The recognition accuracy is ≥99%; when a blind spot is detected, a supplementary scan is automatically triggered with a response time of ≤10s to ensure complete scan data without blind spots.

[0097] 2.4 Decoupling of Lifting Deformation: Strain sensors were placed at key locations in the segment (edges of the double stern curved surfaces and near the propeller base) to collect strain data during the segment lifting process; a strain-deformation transformation matrix was established based on the finite element model. Through the deformation decoupling algorithm formula From the scanning measurement deviation Subtract hoisting deformation deviation Restore the actual geometric deviation of the segments. , The deviation component caused by elastic deformation during hoisting, in units of This ensures that the scanned data accurately reflects the actual state of the segments.

[0098] Step S3: GPU Parallel Point Cloud Processing

[0099] 3.1 Parallel point cloud preprocessing: The CUDA parallel computing architecture is adopted, and GPU-accelerated point cloud denoising, filtering, and simplification preprocessing are performed.

[0100] A parallel Gaussian filtering algorithm is used, through the formula Remove point cloud noise, where the window radius k=4 and the Gaussian weighting coefficient is... It decreases as the distance from the center of the window increases; For point clouds in coordinates The depth value at that location, in units of , The standard deviation of the Gaussian kernel. The effective number of point clouds within the window is used for normalization. The voxel grid method is used to simplify the point cloud, with the simplification rate controlled at 40%~50%. While ensuring data accuracy, the point cloud processing efficiency is improved by ≥60%.

[0101] 3.2 Point Cloud Registration: A combined registration algorithm of FPFH feature pre-alignment and ICP fine-tuning is adopted. First, the FPFH features of the point cloud are extracted, and then the registration is performed using the formula... Calculate feature distances to achieve coarse alignment of point clouds, with a coarse alignment error ≤ ±0.5mm. For point FPFH feature number Dimensional components; For point FPFH feature number Dimensional components; This represents the feature dimension of FPFH, with a value of 33.

[0102] The ICP algorithm is then used for fine-tuning, and the rotation matrix and translation vector are iteratively optimized. The iteration termination condition is that the deviation between two adjacent iterations is ≤0.02mm, and the number of iterations does not exceed 50. Finally, the accurate registration of the segmented local coordinate system point cloud to the global coordinate system is achieved, with a total registration error of ≤±0.15mm.

[0103] Step S4: Comparison and analysis of actual state and theoretical model in segments

[0104] 4.1 Three-dimensional comparison: Import the segmented actual geometric model (preprocessed point cloud data) restored in step S2 and the theoretical value point cloud data into the comparison software, and use a point-to-point comparison method to conduct a comprehensive comparison of key parts such as the double stern curved surface, docking surface, and propeller base mounting surface.

[0105] 4.2 Deviation Calculation and Visualization: A deviation calculation algorithm is adopted, using formulas... Calculate the deviation value for each measuring point, where the total number of measuring points M ≥ 10000 to ensure the comprehensiveness of the deviation calculation. For the actual point cloud The three-dimensional coordinate vector of each measurement point; For the corresponding point cloud in the theory The three-dimensional coordinate vector of each measurement point; The three-dimensional spatial deviation value of the measuring point, in units of ;

[0106] A color deviation cloud map is generated, and a pseudo-color mapping is used to present the deviation distribution. Green (≤±0.5mm) is the qualified area, yellow (±0.5~±1.5mm) is the warning area, and red (>±1.5mm) is the out-of-tolerance area. At the same time, a deviation distribution histogram is generated to count the number of measuring points in different deviation intervals, intuitively present the deviation distribution characteristics, and generate a deviation analysis report.

[0107] Step S5: Identification and Compensation of Segmented Docking Errors

[0108] 5.1 Docking Error Identification: Based on the deviation analysis report, a multi-dimensional error identification algorithm is used to accurately identify three types of core errors in the segmented docking end faces: docking gap (identification accuracy ≤ ±0.05mm), misalignment deviation (identification accuracy ≤ ±0.05mm), and angle deviation (identification accuracy ≤ ±0.1°). Out-of-tolerance areas are marked (gap > 0.5mm, misalignment > ±0.8mm, angle > ±0.5°), and the location and magnitude of the error are clearly defined.

[0109] 5.2 Compensation Calculation: A linear interpolation compensation algorithm is adopted, using the formula... Calculate the compensation amount for each deviation point, where the maximum compensation amount is... Take 2mm, For the first The measured deviation value of each measuring point, in units of ; The maximum deviation value within the compensation area, in units of ; The minimum deviation value within the compensation area, in units of ; The compensation amount for the measuring point, in units of ;

[0110] Compensation calculation error Ensure that the compensation amount corresponds linearly to the deviation value, and the calculation error of the compensation amount is ≤ ±0.1mm; for the deviation of the symmetrical parts of the two sterns, a symmetrical compensation strategy is adopted to ensure the symmetry of the double stern profile.

[0111] 5.3 Adjustment Plan Formulation: Based on the compensation calculation results, segmented assembly process requirements and the adjustment range of the six-degree-of-freedom adjustment equipment, a personalized segmented adjustment plan is formulated, clarifying the adjustment sequence, adjustment direction, adjustment amount and adjustment accuracy requirements. Special adjustment measures are formulated for areas with excessive tolerances to ensure that the deviation meets the standards after adjustment.

[0112] Step S6: Precise segment positioning, assembly, and welding control

[0113] 6.1 Precise Positioning: The segments to be assembled are hoisted to the assembly station, and a six-degree-of-freedom positioning device is used to adjust the segment posture according to the positioning plan; the laser tracking device is activated to collect three-dimensional coordinate data in real time during the segment positioning process, monitor the positioning accuracy, and ensure that the segment posture is stable during the positioning process with a positioning accuracy of ≤±0.1mm; after the positioning is completed, the segments are temporarily fixed using positioning fixtures, and laser scanning is performed again to verify that the deviation after positioning meets the preset requirements.

[0114] 6.2 Welding Shrinkage Prediction: A BP neural network welding shrinkage prediction model was established, using welding current, voltage, speed, and weld thickness as input parameters, and welding shrinkage as the output parameter. The model was trained using a large amount of welding test data, with a training error ≤5%. A prediction formula was adopted. Predicting welding shrinkage provides a basis for anti-deformation control. For the first One input parameter; For the first The connection weights corresponding to each input; For neuron bias terms; This represents the number of neurons in the input layer. The Sigmoid activation function is expressed as follows: ;

[0115] 6.3 Dynamic Control During Welding: A laser scanning device is used to monitor deformation data in real time during the welding process. The measured deformation is compared with the predicted shrinkage to calculate the deviation. A PID dynamic adjustment algorithm is employed, using the formula... Dynamically adjust welding parameters (current, voltage, speed), among which , It is a comprehensive adjustment parameter for welding, used to simultaneously adjust welding current, voltage and travel speed, implement anti-deformation control, and control welding deformation within ±0.2mm.

[0116] 6.4 Post-weld repair: After welding is completed, the segments are laser scanned and re-measured to identify areas that exceed the tolerance after welding. For areas that exceed the tolerance, local grinding, repair welding and other methods are used to repair the accuracy. After repair, the segments are scanned again to verify that the accuracy after repair meets the preset standards.

[0117] Step S7: Assembly quality inspection and data archiving

[0118] 7.1 Assembly Quality Inspection: Perform full-range laser 3D scanning on the assembled sections to collect overall point cloud data. Compare this data with the theoretical model to verify whether the assembly deviation of the double stern curved outer plate, the section joint gap, the misalignment deviation, and the angle deviation meet the preset standards (double stern curved outer plate assembly deviation ≤ ±2mm, section joint gap ≤ 0.5mm, misalignment deviation ≤ ±0.8mm, angle deviation ≤ ±0.5°). If the inspection is qualified, proceed to the next section assembly process; if not, return to step S5 to recalculate the compensation amount and repeat steps S6-S7 for adjustment.

[0119] 7.2 Data Archiving and Process Optimization: Organize and archive the 3D scanning data, deviation analysis reports, compensation records, positioning parameters, welding parameters, and precision inspection reports of the entire assembly process to establish a segmented assembly quality database; analyze historical data through the database, summarize deviation patterns and process optimization space, iteratively optimize the segmentation scheme, assembly sequence, welding parameters, etc., to form a replicable high-precision assembly process scheme, providing technical support for subsequent mass production.

[0120] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for controlling the assembly accuracy of a segmented double-stern linear CSOV assisted by laser three-dimensional scanning, characterized in that, Includes the following steps: S1: Based on the curved structural characteristics of the twin-stern hull and the constraints of the assembly process, the super segmentation is completed, a global coordinate system and a segment local coordinate system are constructed, the local coordinate system and the global coordinate system are accurately connected, and it serves as a unified measurement benchmark for precision control. Based on this benchmark, an assembly error propagation model is established. The accuracy budget of each segment is allocated in reverse according to the total assembly accuracy index, and a fitness function is constructed to optimize and determine the segment assembly sequence with the goal of minimizing the cumulative assembly error. S2: Build a multi-source fusion 3D scanning system to scan the segments to be assembled, calculate the surface curvature value based on the surface complexity, classify the surface complexity level based on the preset curvature threshold, plan the corresponding scanning path, and identify and fill in the data blind spots. Synchronously collect segmented hoisting strain data to decouple the interference of hoisting deformation on the scanning results and restore the true geometric deviation of the segments; S3: A parallel computing architecture is used to preprocess the point cloud data. A combined registration algorithm of feature pre-alignment and fine adjustment is used to uniformly register the point cloud data in each segment's local coordinate system to the global coordinate system by combining coarse registration and fine registration. S4: Compare the preprocessed point cloud data with the theoretical point cloud data of the design, calculate the deviation values ​​of key areas, and generate deviation visualization results and deviation analysis reports. S5: Based on the deviation analysis results, identify the multi-dimensional errors of the segmented docking end faces, calculate the corresponding compensation amount, and generate a six-degree-of-freedom adjustment scheme by combining the assembly process requirements and the adjustment equipment parameters. S6: Complete the segmented precise positioning and temporary fixation according to the positioning scheme, and establish a neural network welding shrinkage prediction model. With welding process parameters as input and welding shrinkage as output, the model is trained through welding test data to predict the welding shrinkage deformation. During the welding process, deformation data is monitored in real time and compared with the predicted shrinkage. A PID dynamic adjustment algorithm is used to dynamically adjust the welding parameters to implement anti-deformation control. After welding, the out-of-tolerance areas are repaired with precision, and the repair is verified by scanning again. S7: Perform a full-domain 3D scan inspection on the assembled segments, collect overall point cloud data, compare it with the designed theoretical model, and determine whether the accuracy indicators are qualified; after the inspection is qualified, proceed to the next segment assembly process. If the result is unsatisfactory, return to step S5 to recalculate the compensation amount, and repeat steps S6-S7 for adjustment.

2. The laser three-dimensional scanning-assisted segmented assembly accuracy control method for twin-stern linear CSOVs according to claim 1, characterized in that, In step S1, the CSOV hull is divided into 6-8 super sections. The docking surfaces of the sections avoid areas of abrupt curvature changes. Each super section contains a complete double-stern curved surface unit and a propeller base mounting area. A global coordinate system is established with the hull keel centerline as the X-axis, the width direction as the Y-axis, and the draft direction as the Z-axis. A local section coordinate system is established with the geometric center of each super section as the origin. The coordinate system is precisely docked through coordinate transformation.

3. The laser three-dimensional scanning-assisted segmented assembly accuracy control method for twin-stern linear CSOVs according to claim 1, characterized in that, In step S1, the cumulative assembly error of the whole section is quantified by the assembly error propagation model. The error propagation coefficient is determined according to the stiffness of the segment structure. The greater the stiffness, the closer the error propagation coefficient is to 1.

0. Based on the requirement that the assembly accuracy of the whole section does not exceed ±2mm, the error budget of the double stern curved core segment is allocated to not exceed ±0.8mm, and the error budget of the ordinary segment does not exceed ±1.2mm. A genetic algorithm is used to optimize the assembly sequence. A fitness function is constructed with the goal of minimizing the total assembly error and assembly cycle. The optimal assembly sequence is obtained through iterative optimization by selection, crossover and mutation operations, with ≥100 iterations.

4. The laser three-dimensional scanning-assisted segmented assembly accuracy control method for twin-stern linear CSOVs according to claim 1, characterized in that, In step S2, the multi-source fusion 3D scanning system consists of a laser scanning unit, a structured light scanning unit, and a photogrammetry unit; wherein the laser scanning accuracy is ≤ ±0.1mm, the structured light scanning resolution is ≤ 0.05mm, and the photogrammetry accuracy is ≤ ±0.15mm. In step S2, the curvature threshold is ≥0.05m. -1 To standardize the division of surface regions into high-complexity and low-complexity regions, high-complexity regions are characterized by dense scanning with a scan spacing ≤ 5 mm, while low-complexity regions are characterized by sparse scanning with a scan spacing ≤ 10 mm. The YOLOv8AI algorithm is used to identify scanning blind zones in real time, with the blind zone determination criterion being a point cloud density < 10 points / mm². 2 ; In step S2, strain sensors are arranged at the edge of the segmented double stern curved surface and near the propeller base to collect hoisting strain data; Based on the finite element model, a strain-deformation transformation matrix is ​​established. The hoisting deformation deviation is deducted from the scanning measurement deviation through a deformation decoupling algorithm to restore the actual geometric deviation of the segments.

5. The laser three-dimensional scanning-assisted segmented assembly accuracy control method for twin-stern linear CSOVs according to claim 1, characterized in that, In step S3, the CUDA parallel computing architecture is used to perform point cloud denoising, filtering, and simplification preprocessing operations based on GPU acceleration; a parallel Gaussian filtering algorithm is used to remove point cloud noise, with a filtering window radius of 4. The point cloud was simplified using a voxel grid method, with a simplification rate controlled at 40%~50%. In step S3, the fast point feature histogram features of the point cloud are extracted first and the feature distance is calculated to achieve coarse alignment with a coarse alignment error of ≤ ±0.5mm. Then, the ICP algorithm is used for fine adjustment, and the rotation matrix and translation vector are iteratively optimized. The iteration termination condition is that the deviation between two adjacent iterations is ≤ 0.02mm, the number of iterations does not exceed 50, and the final total registration error is ≤ ±0.15mm.

6. The laser three-dimensional scanning-assisted segmented assembly accuracy control method for twin-stern linear CSOVs according to claim 1, characterized in that, In step S4, the visualization results are a color deviation cloud map and a deviation distribution histogram; the deviation distribution is presented using pseudo-color mapping, where a deviation ≤ ±0.5mm is a qualified area, a deviation in the range of ±0.5~±1.5mm is a warning area, and a deviation > ±1.5mm is an out-of-tolerance area.

7. The laser three-dimensional scanning-assisted segmented assembly accuracy control method for twin-stern linear CSOVs according to claim 1, characterized in that, In step S5, the identified multi-dimensional errors of the mating end face include mating gap, misalignment deviation, and angle deviation; wherein the identification accuracy of mating gap and misalignment deviation is ≤ ±0.05mm, and the identification accuracy of angle deviation is ≤ ±0.1°; marking the out-of-tolerance areas where the gap is > 0.5mm, misalignment is > ±0.8mm, and angle is > ±0.5°. In step S5, a linear interpolation compensation algorithm is used to calculate the compensation amount for each deviation point. The compensation amount and the deviation value are linearly related, and the calculation error of the compensation amount is ≤ ±0.1mm. A symmetrical compensation strategy is adopted for the symmetrical parts of the two sterns to ensure the symmetry of the two stern lines.

8. The laser three-dimensional scanning-assisted segmented assembly accuracy control method for twin-stern linear CSOVs according to claim 1, characterized in that, In step S6, a six-degree-of-freedom positioning device is used to perform segmented positioning. The three-dimensional coordinate data during the positioning process is collected in real time by a laser tracking device, and the positioning accuracy is ≤ ±0.1mm. After the positioning is completed, a positioning fixture is used to temporarily fix the segments, and the positioning accuracy is verified by scanning again. In step S6, a BP neural network welding shrinkage prediction model is established, with welding current, voltage, speed, and weld thickness as input parameters and welding shrinkage as output parameter. The model training error is ≤5%. During the welding process, a PID dynamic adjustment algorithm is used to adjust the welding parameters in real time and implement anti-deformation control to control the welding deformation within ±0.2mm. In step S6, after welding is completed, the segments are laser scanned and retested. For areas with out-of-tolerance, local grinding and welding are used to repair the accuracy. After repair, the accuracy is scanned again to verify whether it meets the standard.

9. The laser three-dimensional scanning-assisted segmented assembly accuracy control method for twin-stern linear CSOVs according to claim 1, characterized in that, In step S7, the qualified indicators for assembly accuracy inspection include: assembly deviation of double stern curved outer plate ≤ ±2mm, gap between segmented mating surfaces ≤ 0.5mm, misalignment deviation ≤ ±0.8mm, and angle deviation ≤ ±0.5°; if the inspection fails, the compensation amount is recalculated, and the adjustment, welding, and inspection process is repeated.

10. A laser three-dimensional scanning-assisted segmented assembly accuracy control system for a double-stern-shaped CSOV, applied to the laser three-dimensional scanning-assisted segmented assembly accuracy control method for a double-stern-shaped CSOV as described in any one of claims 1-9, characterized in that, include: The module includes a global accuracy control module, a 3D scanning and deformation decoupling module, a point cloud processing and unified registration module, a model comparison and deviation analysis module, a docking error compensation module, a welding process control module, and an accuracy inspection and process optimization module. The global precision control module is used to divide super segments based on the curved structural characteristics of the twin-stern hull and assembly process constraints, construct a two-level coordinate reference system of global and segment, establish an assembly error control model, and inversely allocate the precision index of each segment to optimize and determine the segment assembly sequence. The 3D scanning and deformation decoupling module is used to collect 3D point cloud data of the segments to be assembled, plan differentiated scanning paths based on surface features and automatically fill in scanning blind spots, synchronously collect segment hoisting deformation data and decouple hoisting deformation interference, and restore the true geometric deviation of the segments. The point cloud processing and unified registration module is used to preprocess the point cloud data using a parallel processing architecture, and to uniformly match each segment of point cloud to the global coordinate system through a coarse-fine registration method; the model comparison and deviation analysis module is used to perform a three-dimensional comparison between the restored segmented actual point cloud data and the theoretical model value data, calculate the deviation of key areas and generate deviation analysis results. The docking error compensation module is used to identify multi-dimensional errors of the segmented docking end faces based on the deviation analysis results, calculate the corresponding compensation amount, and generate a segmented adjustment scheme. The welding process control module is used to complete segmented precise positioning and temporary fixation according to the positioning scheme, predict the amount of welding shrinkage deformation, dynamically adjust welding parameters to implement anti-deformation control during the welding process, and perform precision repair on out-of-tolerance areas after welding. The precision inspection and process optimization module is used to perform full-domain three-dimensional precision inspection on the assembled segments, determine whether the precision indicators are qualified, archive the assembly data of the whole process, and iteratively optimize the assembly process parameters based on historical data.

Images