A CSOV ship twin stern line modular assembly process and segment butt joint error compensation method

Abstract

This invention relates to the field of marine engineering service and maintenance vessel manufacturing technology, and discloses a modular assembly process for the twin stern hull of a CSOV vessel and a method for compensating for segmented docking errors. This method constructs a twin-stern symmetrical control and integrated collaborative design system by dividing the vessel into super segments through multi-objective optimization; it achieves real-time accuracy monitoring using laser scanning and digital twins, optimizes the assembly sequence using genetic algorithms, and identifies interference through virtual assembly; it utilizes SVM for intelligent classification of docking errors and matches mechanical, thermal, and process-related multi-strategy collaborative compensation; and it combines LSTM lifting attitude prediction and PID welding synchronous control, achieving closed-loop process optimization through parallel accuracy detection and big data analysis. This invention can control the twin-stern symmetrical deviation within ±0.3mm, significantly improving assembly accuracy and efficiency, adapting to existing equipment in domestic shipyards, and promoting the independent upgrading of the domestic high-end CSOV industry.

Description

Technical Field

[0001] This invention relates to the field of CSOV (Construction Service Vessel) manufacturing technology, and in particular to a modular assembly process for a twin-stern hull of a CSOV and a method for compensating for segment docking errors. Background Technology

[0002] As the global offshore wind power industry expands into deep and open waters, the CSOV (Conditional Sea Container Vehicle) is the core equipment for the entire lifecycle operation and maintenance of offshore wind power. Its manufacturing precision and efficiency directly determine the safety, efficiency, and market competitiveness of operation and maintenance. The Ulstein TwinX-Stern twin-stern styling has become the mainstream styling for high-end CSOVs due to its excellent seakeeping, propulsion efficiency, and operational stability. However, this styling has a complex curved structure, high symmetry requirements, and extremely difficult modular assembly and segmented docking error compensation.

[0003] Currently, domestic CSOV manufacturing faces the dilemma of "strong design, weak manufacturing, and slow optimization," with existing technologies exhibiting the following core bottlenecks:

[0004] Lack of collaborative design mechanism: The segmentation only considers structural strength and does not combine pre-outfitting with green power system integration for collaborative optimization, resulting in frequent interference during final assembly; the symmetry control of the two sterns adopts passive adjustment and lacks active control algorithm, resulting in excessive accumulation of symmetry deviation; the digitalization level of pre-outfitting interfaces is low and relies on manual experience; the green power system is disconnected from the final assembly process, and pipeline layout conflicts with segment docking.

[0005] The manufacturing process is not sufficiently digitalized: the laser point cloud noise filtering effect is poor, and the digital twin model cannot synchronize the manufacturing status in real time; the monitoring of the symmetrical deviation of the two sterns lacks a scientific early warning mechanism, and the accumulated deviation is transmitted to the final assembly stage; the optimization of the final assembly sequence is prone to getting trapped in local optima, and the virtual assembly only checks hard interference while ignoring soft interference (pipeline spacing, operating space, etc.), and the feasibility is not verified in combination with the symmetrical characteristics of the two sterns.

[0006] Weaknesses in error compensation and intelligent control: The types of docking errors are complex, and the existing identification relies on human experience and lacks self-learning ability; the compensation strategy is fixed and singular, and does not adaptively select according to the type of error, resulting in poor coordination between mechanical, thermal, and process compensation; the accuracy of hoisting posture prediction is low, the welding process lacks intelligent synchronous control, and the inconsistency of parameters on both sides of the stern leads to asymmetrical deformation.

[0007] Quality inspection and process optimization are lagging behind: the sequential inspection method is inefficient, there is a lack of big data analysis to uncover process patterns, and there is no closed-loop mechanism for process optimization, making it difficult to achieve continuous improvement.

[0008] While leading international companies (such as Ulstein and VARD) have mastered the relevant core technologies, their processes rely on high-end specialized equipment, which is costly and cannot be adapted to the existing equipment level of domestic shipyards; they have not been optimized for the production conditions and CSOV operation and maintenance needs of domestic shipyards, resulting in poor applicability.

[0009] Existing domestic patents mostly focus on single links, lacking integrated design and intelligent control of the entire process. They suffer from problems of "emphasizing design but neglecting process" and "emphasizing individual technical points but neglecting system integration," making it difficult to form an effective technical defense. Against this background, the development of a modular assembly process for the twin stern hull of CSOV ships that integrates integrated collaborative design, intelligent precision control, and data-driven optimization, as well as a method for compensating for segment docking errors, has become an urgent need to promote the independent upgrading of my country's CSOV industry. Summary of the Invention

[0010] To address the shortcomings of existing technologies, the present invention aims to provide a modular assembly process for CSOV twin-stern hull lines and a method for compensating for segment docking errors. By combining the manufacturing characteristics of Ulstein TwinX-Stern twin-stern hull line CSOVs with the equipment level of domestic shipyards, the invention achieves a triple improvement in modular assembly efficiency, precision, and intelligence, forming a low-cost and replicable process solution.

[0011] A modular assembly process for a CSOV (Conductor-Shipboard) with twin stern lines and a method for compensating for section docking errors includes the following steps:

[0012] S1: Based on the geometric features of the twin-stern linear curved surface, structural strength requirements, shipyard hoisting equipment load-bearing capacity, and the integration requirements of pre-outfitting and green power, a multi-objective collaborative optimization algorithm is used to divide the CSOV hull into multiple super sections;

[0013] Based on the super segmentation results, a dual-stern symmetric active control design, digital interface management of pre-outfitting units, and integrated collaborative design of green power and final assembly are established.

[0014] S2: Perform a full-range scan of the super segment during the manufacturing process, collect segment point cloud data, and perform noise reduction and filtering preprocessing on the point cloud data;

[0015] Based on the preprocessed point cloud data, a segmented digital twin model is constructed, which includes segmented geometric features, pre-outfitting status and manufacturing precision information, and the segmented digital twin model is updated in real time.

[0016] Based on the active control design of dual stern symmetry in S1, the dual stern symmetry reference surface is selected, the edges of the dual stern curved surfaces and key symmetry points of the propeller base are selected, the dual stern symmetry deviation is monitored and an early warning threshold is set;

[0017] S3: An adaptive genetic algorithm is used to construct a multi-objective fitness function, and the optimal segmented assembly order is obtained through iterative optimization.

[0018] A virtual assembly environment is built based on the segmented digital twin model. The optimal assembly sequence, segmented geometric data, pre-outfitting data, and green power equipment data are imported. A multi-dimensional interference inspection algorithm is used to inspect the virtual assembly process for interference, detect hard interference and soft interference, and verify the symmetry and feasibility of the segmented docking of the two sterns in conjunction with the symmetrical reference plane of the two sterns.

[0019] S4: Hoist the segments to the final assembly station and preliminarily position them according to the optimal assembly sequence, collect point cloud data of the segment docking end faces, extract the actual geometric data of the docking end faces, compare the geometric data in the segment digital twin model, and calculate the docking error;

[0020] The SVM machine learning algorithm is used to identify and classify the docking errors, extract the feature parameters of systematic errors, random errors and deformation errors, and establish an error classification database with self-learning function;

[0021] S5: Based on the error classification results, an adaptive selection algorithm for compensation strategies is used to automatically match the optimal compensation strategy. Systematic errors are preferentially compensated by process compensation, random errors are preferentially compensated by mechanical compensation, and deformation errors are compensated by a combination of thermal and mechanical compensation.

[0022] Calculate the optimal compensation amount for various types of errors, implement multi-strategy collaborative compensation according to the optimal compensation amount, monitor the docking accuracy in real time during the compensation process and iterate the compensation until the accuracy requirements are met;

[0023] S6: An attention-based LSTM prediction model is used to collect historical attitude data during the hoisting process to predict the hoisting attitude, and the hoisting equipment parameters are actively adjusted based on the prediction results;

[0024] By synchronously controlling the welding parameters on both sides of the stern using a fuzzy adaptive PID algorithm, and predicting and suppressing welding deformation using a welding deformation prediction formula, the welding operation is completed using a symmetrical welding process, and the modular integration and installation of each functional module is completed simultaneously.

[0025] S7: Use a parallel accuracy detection algorithm to perform a global accuracy inspection on the assembled twin-stern hull. If the inspection fails, return to step S5 for readjustment.

[0026] A big data analysis platform for final assembly accuracy was established. By mining the correlation patterns between process parameters and assembly accuracy through association rules, a process iterative optimization model was constructed to achieve closed-loop optimization of process parameters.

[0027] Furthermore: In step S1, dividing the CSOV hull into multiple super sections specifically includes:

[0028] Based on the curvature distribution of the Ulstein TwinX-Stern twin-stern linear surface, structural strength requirements, shipyard lifting equipment load-bearing capacity, and pre-outfitting and green power integration needs, the CSOV hull is divided into 6-8 super sections using the aforementioned multi-objective collaborative optimization algorithm. During the division process, abrupt curvature changes in the twin-stern surface are avoided, with the curvature of these abrupt curvature changes not less than 0.05 per meter. Each super section integrates complete twin-stern surface units, pre-outfitting units, and green power equipment installation areas. The docking surfaces of each super section adopt standardized designs, and the section weight is controlled within the load-bearing capacity of the shipyard's existing lifting equipment.

[0029] Further: In step S1, the dual-stern symmetry active control design includes:

[0030] A global coordinate system for the CSOV hull is established, with the keel centerline as the X-axis, the beam direction as the Y-axis, and the draft direction as the Z-axis. A global symmetry reference plane is established with the stern symmetry center as the reference. A PID algorithm for active control of stern symmetry is adopted, and the symmetry adjustment is calculated based on the proportional, integral, and derivative terms of the real-time symmetry deviation value, where the proportional coefficient is 1.4, the integral coefficient is 0.2, and the derivative coefficient is 0.3. The segment design parameters are adjusted in real time to ensure that the stern symmetry deviation is no greater than ±0.3 mm.

[0031] The digital interface management of the pre-outfitting unit includes:

[0032] A digital interface management system for pre-outfitting units is constructed, and the interface dimensions, coordinates, and connection methods of pre-outfitting units are standardized. An interface matching degree evaluation model is established, which evaluates the interface matching degree by calculating the average Euclidean distance between the actual coordinates and theoretical coordinates of each interface measuring point in the three-dimensional coordinate system. The number of interface measuring points is not less than 50, and the interface is judged to be qualified when the interface matching degree is not greater than ±0.1 mm.

[0033] The integrated design of green power and final assembly includes:

[0034] Conduct integrated collaborative design of green power systems and final assembly processes, establish a collaborative coupling degree calculation model, and use the weighted average of the coupling degree of each collaborative link as the overall collaborative coupling degree. The number of collaborative links shall not be less than 4, and the weight coefficient of each link shall be in the range of 0.2 to 0.3. When the collaborative coupling degree is not less than 0.85, it shall be judged as qualified.

[0035] Furthermore: In step S2, a laser three-dimensional scanning device is used to perform a full-range scan of the super segment during the manufacturing process. The scanning accuracy is no greater than ±0.1 mm, the scanning data acquisition density is no less than 10 points per square millimeter, and the scanning data covers the key parts of the double stern curved surface, the docking surface, and the pre-outfitting interface.

[0036] The denoising and filtering preprocessing adopts an improved laser point cloud filtering algorithm, namely tensor voting combined with random sampling consensus algorithm. The point cloud data is filtered by tensor voting value and voting weight, where the voting weight is in the range of 0.1 to 0.3, and the point cloud error after filtering is no greater than ±0.05 mm.

[0037] The point cloud data includes:

[0038] Twin-stern structure surface data: Three-dimensional coordinate point data of the entire twin-stern surface contour, used to determine the twin-stern symmetry deviation and surface assembly accuracy;

[0039] Segmented docking end face data: Three-dimensional geometric point data of each super segment docking surface, used to calculate assembly errors such as segmented docking gap, misalignment, and angle;

[0040] Key interface data for pre-outfitting: Three-dimensional coordinate data of interfaces of pre-outfitting units such as pipelines and equipment, used to verify the interface matching accuracy;

[0041] Core base installation data: Three-dimensional point data of the mounting surfaces of key equipment such as the thruster base, ensuring the positioning accuracy of the core equipment;

[0042] The segmented digital twin model is updated in real time using a digital twin real-time update mechanism. Based on the model data at the current moment, the update coefficient and the difference between the actual scan data and the model data at the current moment are used for correction and update. The update coefficient is 0.7, which realizes real-time synchronization between the model and the segmented manufacturing process.

[0043] The monitoring of stern symmetry deviation includes:

[0044] Based on the aforementioned stern-to-stern symmetry reference plane, select at least 20 pairs of key symmetry points on the edges of the stern curved surfaces and the propeller base. Employ a stern-to-stern symmetry deviation monitoring algorithm to obtain the real-time symmetry deviation by calculating the spatial distance of the corresponding key symmetry points in the three-dimensional coordinate system. Set the warning threshold to ±0.3 mm. When the absolute value of the real-time symmetry deviation exceeds the warning threshold, a deviation warning is triggered.

[0045] Furthermore: In step S3, the multi-objective fitness function constructed by the adaptive genetic algorithm uses the reciprocal of the weighted sum of the assembly cycle, the cumulative error, and the interference risk as the fitness value;

[0046] The weights for assembly cycle, error accumulation, and interference risk are 0.4, 0.3, and 0.3 respectively. The optimization objectives are to shorten the assembly cycle, reduce error accumulation, and reduce interference risk. The crossover probability decreases linearly with the number of iterations, while the mutation probability increases linearly with the number of iterations. The maximum number of iterations is 120.

[0047] The multi-dimensional interference inspection algorithm performs hard interference detection and soft interference detection on the virtual assembly process. The hard interference includes structural collisions and equipment collisions, and the soft interference includes insufficient pipeline spacing, insufficient equipment installation space, and insufficient operating space. If interference problems exist, return to S1 to adjust the segment division or pre-outfitting layout until all interference problems are resolved.

[0048] Furthermore, in step S4, the SVM machine learning algorithm is used to identify and classify the docking error, with the optimization objective of maximizing the classification interval and minimizing the classification error. A penalty coefficient is introduced to control the tolerance for classification errors, wherein the penalty coefficient is 10, to ensure that the error classification accuracy is not less than 98%.

[0049] The error classification database with self-learning function adopts a self-learning update mechanism. Based on the original classification model, it updates the model by combining the learning rate and the difference between the new error data and the output of the original model. The learning rate is 0.15, and the classification model is automatically updated when new error data is added.

[0050] Further: In step S5, the adaptive selection algorithm for compensation strategy selects the optimal compensation strategy by calculating the weighted fit of three strategies: mechanical compensation, thermal compensation, and process compensation. The mechanical compensation fit weight is 0.4, the thermal compensation fit weight is 0.3, and the process compensation fit weight is 0.3. The optimal compensation strategy is automatically matched based on the maximum weighted fit.

[0051] The optimal compensation amount for various types of errors is calculated using a multi-strategy collaborative compensation algorithm. The mechanical compensation amount, thermal compensation amount, and process compensation amount are weighted and summed according to their compensation weights to obtain the total compensation amount. The compensation weights of the three compensation methods are adjusted according to the error type and the sum is 1.

[0052] The thermal compensation adopts fuzzy PID control, and the control algorithm is the same as the symmetric active control PID algorithm in step S1.2. The parameters are adjusted to a proportional coefficient of 1.2, an integral coefficient of 0.15, and a derivative coefficient of 0.25 to ensure that the thermal compensation temperature control accuracy is no greater than ±5 degrees Celsius.

[0053] The implementation of multi-strategy collaborative compensation includes:

[0054] Mechanical compensation is achieved using a six-degree-of-freedom positioning device, thermal compensation using a controllable heating device, and process compensation using welding sequence adjustment and positioning tooling optimization. During the compensation process, laser tracking equipment is used to monitor the docking accuracy in real time. After each compensation is completed, point cloud data of the docking end face is collected and residual error is calculated until the residual error is no greater than ±0.1 mm, the docking gap is no greater than 0.5 mm, the misalignment deviation is no greater than ±0.8 mm, and the angle deviation is no greater than ±0.5 degrees.

[0055] Further: In step S6, the attention-based LSTM prediction model uses the attention mechanism to weightedly fuse the hidden state information of historical hoisting posture data, and obtains the hoisting posture prediction value through output layer mapping. The root mean square error between the predicted posture and the true posture is used as the loss function, where the observation time is 10 time steps and the prediction time is 5 time steps. The hoisting equipment parameters are actively adjusted according to the prediction results to ensure that the segmented positioning deviation is no greater than ±0.1 mm.

[0056] The fuzzy adaptive PID algorithm synchronously controls the welding parameters on both sides of the stern, collects welding current, voltage, speed, and temperature parameters on both sides in real time and adjusts them dynamically; combined with the welding deformation prediction model, the welding deformation is predicted based on the linear combination of welding current, welding speed, and welding temperature, where the welding current correction coefficient is 0.002, the welding speed correction coefficient is 0.001, and the welding temperature correction coefficient is 0.0005, and the welding parameters are adjusted in advance to ensure symmetrical welding deformation of the stern.

[0057] The symmetrical welding process adopts a butt welding followed by fillet welding and a centrally symmetrical welding process. Welding is carried out simultaneously from the center of symmetry of the two sterns to both sides. The welding speed is controlled at 2 to 3 millimeters per second, the welding current is 200 to 250 amperes, and the welding voltage is 22 to 25 volts. During the welding process, a laser scanning device is used to monitor the welding deformation in real time. If the deformation exceeds ±0.2 millimeters, the welding parameters are adjusted immediately to ensure that the welding deformation is controlled to be no greater than ±0.2 millimeters.

[0058] Further: In step S7, the parallel precision detection algorithm obtains the global average deviation by calculating the average spatial distance between each detection point of the hull after final assembly and the theoretical coordinates, wherein the total number of detection points is not less than 10,000; the focus is on inspecting the fit of the twin stern curved surfaces, the segment docking accuracy, and the positioning accuracy of the propeller base, ensuring that the assembly deviation of the twin stern curved outer plate is not greater than ±2 mm, the gap between the segment docking surfaces is not greater than 0.5 mm, the positioning deviation of the propeller base is not greater than ±1 mm, and the symmetry deviation of the twin sterns is not greater than ±0.3 mm;

[0059] The assembly precision big data analysis platform integrates data from the entire assembly process and employs an association rule mining algorithm to mine the correlation between process parameters and assembly precision by calculating the support and confidence of the simultaneous occurrence of process parameters and precision indicators. The process iterative optimization model, based on the current process parameters, iteratively updates the process parameters along the gradient direction of the precision evaluation indicators, with an iteration step size of 0.05. Iteratively optimizes segmented parameters, welding parameters, and compensation strategies, and feeds the optimized process parameters back to subsequent assembly stages, establishing a closed-loop mechanism for process optimization.

[0060] Furthermore: A modular assembly process for a CSOV (Conductor-Shipboard) with twin stern lines and a section docking error compensation system, comprising:

[0061] The super-segment division and collaborative design module is used to divide the CSOV hull into multiple super segments based on the geometric features of the twin stern linear curved surfaces, structural strength requirements, shipyard hoisting equipment load-bearing capacity, and pre-outfitting and green power integration requirements. It also establishes twin stern symmetry active control design, pre-outfitting unit digital interface management, and green power and final assembly integrated collaborative design.

[0062] The segmented digital twin and symmetry monitoring module is used to perform full-range laser scanning on the super segments in the manufacturing process, collect segmented point cloud data and perform noise reduction and filtering preprocessing, extract actual segmented geometric data, construct segmented digital twin models containing segmented geometric features, pre-outfitting status and manufacturing accuracy information and update them in real time, and select the edges of the double stern curved surfaces and key symmetry points of the propeller base based on the double stern symmetry reference plane to monitor the double stern symmetry deviation and set early warning thresholds.

[0063] The assembly sequence optimization and virtual assembly module is used to construct a multi-objective fitness function using an adaptive genetic algorithm to iteratively optimize and obtain the optimal segmented assembly sequence. Based on the segmented digital twin model, a virtual assembly environment is built. The optimal assembly sequence, segmented geometric data, pre-outfitting data, and green power equipment data are imported. A multi-dimensional interference inspection algorithm is used to perform hard and soft interference detection on the virtual assembly process. The symmetry and feasibility of the segmented docking of the two sterns are verified in conjunction with the symmetrical reference plane of the two sterns.

[0064] The intelligent identification and classification module for docking errors is used to collect point cloud data of the docking end faces of the segments after the initial positioning of the segmented hoisting, extract the actual geometric data of the docking end faces, compare the geometric data in the segmented digital twin model to calculate the docking error, use the SVM machine learning algorithm to identify and classify the docking error, extract the feature parameters of systematic error, random error and deformation error, and establish an error classification database with self-learning function.

[0065] The adaptive multi-strategy collaborative compensation module is used to automatically match the optimal compensation strategy based on the error classification results using an adaptive selection algorithm. Systematic errors are preferentially compensated by process compensation, random errors are preferentially compensated by mechanical compensation, and deformation errors are compensated by a combination of thermal and mechanical compensation. The module calculates the optimal compensation amount for each type of error and implements multi-strategy collaborative compensation. During the compensation process, the docking accuracy is monitored in real time and the compensation is iteratively performed until the accuracy requirements are met.

[0066] The hoisting attitude prediction and welding synchronization control module is used to collect historical attitude data during the hoisting process using an attention-based LSTM prediction model to predict the hoisting attitude. Based on the prediction results, the hoisting equipment parameters are actively adjusted. The welding parameters on both sides of the stern are synchronously controlled by a fuzzy adaptive PID algorithm. The welding deformation is predicted and suppressed by the welding deformation prediction formula. The welding operation is completed by using a symmetrical welding process. The modular integration and installation of each functional module are completed simultaneously.

[0067] The global accuracy detection and process optimization module is used to perform global accuracy inspection on the assembled twin-stern hull using a parallel accuracy detection algorithm. If the inspection fails, it triggers a readjustment and compensation. It also builds a big data analysis platform for assembly accuracy, mines the correlation between process parameters and assembly accuracy through association rules, constructs a process iterative optimization model, and realizes closed-loop optimization of process parameters.

[0068] The central control unit is communicatively connected to each of the above modules and is used to coordinate and control the operation sequence, data interaction and command issuance of each module.

[0069] The present invention has the following beneficial effects:

[0070] (1) Through multi-objective collaborative optimization and dedicated algorithms, the collaborative linkage of segmentation, pre-outfitting, green power integration and final assembly process is realized, effectively controlling structural symmetry deviation, improving the interface matching and process collaboration of each component, avoiding assembly interference problems, optimizing the overall final assembly operation process, and shortening the final assembly construction cycle.

[0071] (2) By relying on the improved laser point cloud filtering algorithm to optimize the point cloud data acquisition and processing effect, combined with the characteristics of the real-time digital twin model and the synchronous update of the manufacturing process, and with the symmetric deviation early warning mechanism, the problem of deviation accumulation is avoided from the source, effectively improving the overall accuracy of segmented manufacturing and providing a solid foundation for the accuracy of subsequent final assembly operations.

[0072] (3) Optimize the assembly sequence with adaptive genetic algorithm to improve the convergence effect and accuracy of the assembly plan. Check and avoid various soft and hard interference risks in advance through multi-dimensional interference inspection, ensure that the virtual assembly effect is highly consistent with the actual assembly operation, and reduce rework and process adjustment work in the assembly stage.

[0073] (4) The SVM machine learning algorithm is used to optimize the error classification and recognition effect. It is equipped with an error database with self-learning function, which can realize the dynamic iterative optimization of the classification model, accurately adapt to various error scenarios, provide reliable support for the reasonable selection of error compensation strategies, and greatly improve the work efficiency of error identification and judgment.

[0074] (5) By using the adaptive selection algorithm of the compensation strategy to match the optimal compensation combination scheme, the overall compensation effect is improved by relying on the multi-strategy collaborative compensation mode. Combined with the thermal compensation fuzzy PID control technology, the problem of secondary deformation during assembly is effectively avoided, the hull docking error is greatly improved, and the hidden danger of assembly deviation is reduced.

[0075] (6) The LSTM hoisting attitude prediction model is used to optimize the hoisting and positioning effect of the components. Combined with the fuzzy adaptive PID welding synchronous control technology, the symmetry of welding deformation of the double stern structure is ensured, the welding deformation problem is accurately controlled, welding defects and welding rework are effectively reduced, and the welding quality is improved.

[0076] (7) Optimize the inspection operation mode through parallel precision detection algorithm, improve the overall inspection work efficiency, rely on big data analysis and process knowledge mining to extract core optimization rules, and cooperate with process parameter iterative optimization model to realize continuous improvement of manufacturing process, steadily improve assembly accuracy and operation efficiency, and form a standardized, scalable and reusable modular assembly process solution.

[0077] (8) This invention can be directly adapted to existing production lines without the need to add large-scale special production equipment, effectively controlling the overall manufacturing cost; it can comprehensively optimize the core operating performance of CSOV vessels, help product performance meet international advanced standards, promote the independent development of the domestic CSOV industry, and significantly enhance the market core competitiveness of domestic high-end CSOV products. Attached Figure Description

[0078] Figure 1 This is a flowchart of a modular assembly process for a CSOV ship with twin stern lines and a method for compensating for segment docking errors. Detailed Implementation

[0079] 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.

[0080] Please see the appendix Figure 1 The present invention provides an embodiment 1: a modular assembly process for a CSOV ship with a twin-stern hull shape and a method for compensating for segment docking errors, comprising the following specific implementation steps:

[0081] S1: Based on the geometric features of the twin-stern linear curved surface, structural strength requirements, shipyard hoisting equipment load-bearing capacity, and the integration requirements of pre-outfitting and green power, a multi-objective collaborative optimization algorithm is used to divide the CSOV hull into multiple super sections;

[0082] Based on the super segmentation results, a dual-stern symmetric active control design, digital interface management of pre-outfitting units, and integrated collaborative design of green power and final assembly are established.

[0083] S2: Perform a full-range scan of the super segment during the manufacturing process, collect segment point cloud data, and perform noise reduction and filtering preprocessing on the point cloud data;

[0084] Based on the preprocessed point cloud data, a segmented digital twin model is constructed, which includes segmented geometric features, pre-outfitting status and manufacturing precision information, and the segmented digital twin model is updated in real time.

[0085] Based on the active control design of dual stern symmetry in S1, the dual stern symmetry reference surface is selected, the edges of the dual stern curved surfaces and key symmetry points of the propeller base are selected, the dual stern symmetry deviation is monitored and an early warning threshold is set;

[0086] S3: An adaptive genetic algorithm is used to construct a multi-objective fitness function, and the optimal segmented assembly order is obtained through iterative optimization.

[0087] A virtual assembly environment is built based on the segmented digital twin model. The optimal assembly sequence, segmented geometric data, pre-outfitting data, and green power equipment data are imported. A multi-dimensional interference inspection algorithm is used to inspect the virtual assembly process for interference, detect hard interference and soft interference, and verify the symmetry and feasibility of the segmented docking of the two sterns in conjunction with the symmetrical reference plane of the two sterns.

[0088] S4: Hoist the segments to the final assembly station and preliminarily position them according to the optimal assembly sequence, collect point cloud data of the segment docking end faces, extract the actual geometric data of the docking end faces, compare the geometric data in the segment digital twin model, and calculate the docking error;

[0089] The SVM machine learning algorithm is used to identify and classify the docking errors, extract the feature parameters of systematic errors, random errors and deformation errors, and establish an error classification database with self-learning function;

[0090] S5: Based on the error classification results, an adaptive selection algorithm for compensation strategies is used to automatically match the optimal compensation strategy. Systematic errors are preferentially compensated by process compensation, random errors are preferentially compensated by mechanical compensation, and deformation errors are compensated by a combination of thermal and mechanical compensation.

[0091] Calculate the optimal compensation amount for various types of errors, implement multi-strategy collaborative compensation according to the optimal compensation amount, monitor the docking accuracy in real time during the compensation process and iterate the compensation until the accuracy requirements are met;

[0092] S6: An attention-based LSTM prediction model is used to collect historical attitude data during the hoisting process to predict the hoisting attitude, and the hoisting equipment parameters are actively adjusted based on the prediction results;

[0093] By synchronously controlling the welding parameters on both sides of the stern using a fuzzy adaptive PID algorithm, and predicting and suppressing welding deformation using a welding deformation prediction formula, the welding operation is completed using a symmetrical welding process, and the modular integration and installation of each functional module is completed simultaneously.

[0094] S7: Use a parallel accuracy detection algorithm to perform a global accuracy inspection on the assembled twin-stern hull. If the inspection fails, return to step S5 for readjustment.

[0095] A big data analysis platform for final assembly accuracy was established. By mining the correlation patterns between process parameters and assembly accuracy through association rules, a process iterative optimization model was constructed to achieve closed-loop optimization of process parameters.

[0096] Furthermore, in another embodiment of the present invention, step S1, which involves dividing the CSOV hull into multiple supersections, specifically includes:

[0097] Based on the curvature distribution of the Ulstein TwinX-Stern twin-stern linear curved surface, structural strength requirements, shipyard lifting equipment load-bearing capacity, and pre-outfitting and green power integration needs, the aforementioned multi-objective collaborative optimization algorithm is used to divide the CSOV hull into 6-8 super sections. During the division process, abrupt curvature changes in the twin-stern curved surface are avoided, with the curvature of these abrupt curvature changes not less than 0.05 per meter. This ensures that each super section integrates complete twin-stern curved surface units, pre-outfitting units, and green power equipment installation areas. The docking surfaces of each super section adopt standardized designs, and the section weight is controlled within the load-bearing capacity of the shipyard's existing lifting equipment.

[0098] Furthermore, in another embodiment of the present invention, in step S1, the dual-stern symmetry active control design includes:

[0099] A global coordinate system for the CSOV hull is established, with the keel centerline as the X-axis, the beam direction as the Y-axis, and the draft direction as the Z-axis. A global symmetry reference plane is established with the stern symmetry center as the reference. A PID algorithm for active control of stern symmetry is adopted, and the symmetry adjustment is calculated based on the proportional, integral, and derivative terms of the real-time symmetry deviation value, where the proportional coefficient is 1.4, the integral coefficient is 0.2, and the derivative coefficient is 0.3. The segment design parameters are adjusted in real time to ensure that the stern symmetry deviation is no greater than ±0.3 mm.

[0100] The digital interface management of the pre-outfitting unit includes:

[0101] A digital interface management system for pre-outfitting units is constructed, and the interface dimensions, coordinates, and connection methods of pre-outfitting units are standardized. An interface matching degree evaluation model is established, which evaluates the interface matching degree by calculating the average Euclidean distance between the actual coordinates and theoretical coordinates of each interface measuring point in the three-dimensional coordinate system. The number of interface measuring points is not less than 50, and the interface is judged to be qualified when the interface matching degree is not greater than ±0.1 mm.

[0102] The integrated design of green power and final assembly includes:

[0103] Conduct integrated collaborative design of green power systems and final assembly processes, establish a collaborative coupling degree calculation model, and use the weighted average of the coupling degree of each collaborative link as the overall collaborative coupling degree. The number of collaborative links shall not be less than 4, and the weight coefficient of each link shall be in the range of 0.2 to 0.3. When the collaborative coupling degree is not less than 0.85, it shall be judged as qualified.

[0104] Furthermore, in another embodiment of the present invention, in step S2, a laser three-dimensional scanning device is used to perform a full-range scan of the super segment during the manufacturing process. The scanning accuracy is no greater than ±0.1 mm, the scanning data acquisition density is no less than 10 points per square millimeter, and the scanning data covers the key parts of the double stern curved surface, the docking surface, and the pre-outfitting interface.

[0105] The denoising and filtering preprocessing adopts an improved laser point cloud filtering algorithm, namely tensor voting combined with random sampling consensus algorithm. The point cloud data is filtered by tensor voting value and voting weight, where the voting weight is in the range of 0.1 to 0.3, and the point cloud error after filtering is no greater than ±0.05 mm.

[0106] The point cloud data includes:

[0107] Twin-stern structure surface data: Three-dimensional coordinate point data of the entire twin-stern surface contour, used to determine the twin-stern symmetry deviation and surface assembly accuracy;

[0108] Segmented docking end face data: Three-dimensional geometric point data of each super segment docking surface, used to calculate assembly errors such as segmented docking gap, misalignment, and angle;

[0109] Key interface data for pre-outfitting: Three-dimensional coordinate data of interfaces of pre-outfitting units such as pipelines and equipment, used to verify the interface matching accuracy;

[0110] Core base installation data: Three-dimensional point data of the mounting surfaces of key equipment such as the thruster base, ensuring the positioning accuracy of the core equipment;

[0111] The segmented digital twin model is updated in real time using a digital twin real-time update mechanism. Based on the model data at the current moment, the update coefficient and the difference between the actual scan data and the model data at the current moment are used for correction and update. The update coefficient is 0.7, which realizes real-time synchronization between the model and the segmented manufacturing process.

[0112] The monitoring of stern symmetry deviation includes:

[0113] Based on the aforementioned stern symmetry reference plane, select at least 20 pairs of key symmetry points on the edges of the stern curved surfaces and the propeller base. Employ a stern symmetry deviation monitoring algorithm to obtain the real-time symmetry deviation by calculating the spatial distance of the corresponding key symmetry points in the three-dimensional coordinate system. Set the warning threshold to ±0.3 mm. When the absolute value of the real-time symmetry deviation exceeds the warning threshold, a deviation warning is triggered.

[0114] Furthermore, in another embodiment of the present invention, in step S3, the multi-objective fitness function constructed by the adaptive genetic algorithm uses the reciprocal of the weighted sum of the assembly cycle, the cumulative error, and the interference risk as the fitness value.

[0115] The weights for assembly cycle, error accumulation, and interference risk are 0.4, 0.3, and 0.3 respectively. The optimization objectives are to shorten the assembly cycle, reduce error accumulation, and reduce interference risk. The crossover probability decreases linearly with the number of iterations, while the mutation probability increases linearly with the number of iterations. The maximum number of iterations is 120.

[0116] The multi-dimensional interference inspection algorithm performs hard interference detection and soft interference detection on the virtual assembly process. The hard interference includes structural collisions and equipment collisions, and the soft interference includes insufficient pipeline spacing, insufficient equipment installation space, and insufficient operating space. If interference problems exist, return to S1 to adjust the segment division or pre-outfitting layout until all interference problems are resolved.

[0117] Furthermore, in another embodiment of the present invention, in step S4, the SVM machine learning algorithm is used to identify and classify the docking error, with the optimization objective of maximizing the classification interval and minimizing the classification error. A penalty coefficient is introduced to control the tolerance of classification errors, wherein the penalty coefficient is 10, to ensure that the error classification accuracy is not less than 98%.

[0118] The error classification database with self-learning function adopts a self-learning update mechanism. Based on the original classification model, it updates the model by combining the learning rate and the difference between the new error data and the output of the original model. The learning rate is 0.15. When new error data is added, the classification model is automatically updated.

[0119] Furthermore, in another embodiment of the present invention, in step S5, the adaptive selection algorithm for compensation strategy selects the optimal compensation strategy by calculating the weighted fit of three strategies: mechanical compensation, thermal compensation, and process compensation. The mechanical compensation fit weight is 0.4, the thermal compensation fit weight is 0.3, and the process compensation fit weight is 0.3. The optimal compensation strategy is automatically matched based on the maximum weighted fit.

[0120] The optimal compensation amount for various types of errors is calculated using a multi-strategy collaborative compensation algorithm. The mechanical compensation amount, thermal compensation amount, and process compensation amount are weighted and summed according to their compensation weights to obtain the total compensation amount. The compensation weights of the three compensation methods are adjusted according to the error type and the sum is 1.

[0121] The thermal compensation adopts fuzzy PID control, and the control algorithm is the same as the symmetric active control PID algorithm in step S1.2. The parameters are adjusted to a proportional coefficient of 1.2, an integral coefficient of 0.15, and a derivative coefficient of 0.25 to ensure that the thermal compensation temperature control accuracy is no greater than ±5 degrees Celsius.

[0122] The implementation of multi-strategy collaborative compensation includes:

[0123] Mechanical compensation is achieved using a six-degree-of-freedom positioning device, thermal compensation using a controllable heating device, and process compensation using welding sequence adjustment and positioning tooling optimization. During the compensation process, laser tracking equipment is used to monitor the docking accuracy in real time. After each compensation is completed, point cloud data of the docking end face is collected and residual error is calculated until the residual error is no greater than ±0.1 mm, the docking gap is no greater than 0.5 mm, the misalignment deviation is no greater than ±0.8 mm, and the angle deviation is no greater than ±0.5 degrees.

[0124] Furthermore, in another embodiment of the present invention, in step S6, the attention-based LSTM prediction model weightedly fuses the hidden state information of historical hoisting posture data through the attention mechanism, and obtains the hoisting posture prediction value through the output layer mapping. The root mean square error between the predicted posture and the true posture is used as the loss function, wherein the observation time is 10 time steps and the prediction time is 5 time steps; the hoisting equipment parameters are actively adjusted according to the prediction results to ensure that the segmented positioning deviation is no greater than ±0.1 mm.

[0125] The fuzzy adaptive PID algorithm synchronously controls the welding parameters on both sides of the stern, collects welding current, voltage, speed, and temperature parameters on both sides in real time and adjusts them dynamically; combined with the welding deformation prediction model, the welding deformation is predicted based on the linear combination of welding current, welding speed, and welding temperature, where the welding current correction coefficient is 0.002, the welding speed correction coefficient is 0.001, and the welding temperature correction coefficient is 0.0005, and the welding parameters are adjusted in advance to ensure symmetrical welding deformation of the stern.

[0126] The symmetrical welding process employs a butt welding followed by fillet welding, and a centrally symmetrical welding process. Welding is carried out simultaneously from the center of symmetry of the two sterns to both sides, controlling the welding speed at 2 to 3 millimeters per second, the welding current at 200 to 250 amperes, and the welding voltage at 22 to 25 volts. During the welding process, a laser scanning device is used to monitor welding deformation in real time. If the deformation exceeds ±0.2 millimeters, the welding parameters are immediately adjusted to ensure that the welding deformation is controlled to be no greater than ±0.2 millimeters.

[0127] Furthermore, in another embodiment of the present invention, in step S7, the parallel precision detection algorithm obtains the global average deviation by calculating the average spatial distance between each detection point of the hull after final assembly and the theoretical coordinates, wherein the total number of detection points is not less than 10,000; the focus is on inspecting the fit of the twin stern curved surfaces, the segment docking accuracy, and the positioning accuracy of the propeller base, ensuring that the assembly deviation of the twin stern curved outer plate is not greater than ±2 mm, the gap between the segment docking surfaces is not greater than 0.5 mm, the positioning deviation of the propeller base is not greater than ±1 mm, and the symmetry deviation of the twin sterns is not greater than ±0.3 mm;

[0128] The assembly precision big data analysis platform integrates data from the entire assembly process and employs an association rule mining algorithm to mine the correlation between process parameters and assembly precision by calculating the support and confidence of the simultaneous occurrence of process parameters and precision indicators. The process iterative optimization model, based on the current process parameters, iteratively updates the process parameters along the gradient direction of the precision evaluation indicators, with an iteration step size of 0.05. Iteratively optimizes segmented parameters, welding parameters, and compensation strategies, and feeds the optimized process parameters back to subsequent assembly stages, establishing a closed-loop mechanism for process optimization.

[0129] An embodiment 2 of the present invention provides: a modular assembly process for a CSOV ship with twin stern lines and a section docking error compensation system, comprising:

[0130] The super-segment division and collaborative design module is used to divide the CSOV hull into multiple super segments based on the geometric features of the twin stern linear curved surfaces, structural strength requirements, shipyard hoisting equipment load-bearing capacity, and pre-outfitting and green power integration requirements. It also establishes twin stern symmetry active control design, pre-outfitting unit digital interface management, and green power and final assembly integrated collaborative design.

[0131] The segmented digital twin and symmetry monitoring module is used to perform full-range laser scanning on the super segments in the manufacturing process, collect segmented point cloud data and perform noise reduction and filtering preprocessing, extract actual segmented geometric data, construct segmented digital twin models containing segmented geometric features, pre-outfitting status and manufacturing accuracy information and update them in real time, and select the edges of the double stern curved surfaces and key symmetry points of the propeller base based on the double stern symmetry reference plane to monitor the double stern symmetry deviation and set early warning thresholds.

[0132] The assembly sequence optimization and virtual assembly module is used to construct a multi-objective fitness function using an adaptive genetic algorithm to iteratively optimize and obtain the optimal segmented assembly sequence. Based on the segmented digital twin model, a virtual assembly environment is built. The optimal assembly sequence, segmented geometric data, pre-outfitting data, and green power equipment data are imported. A multi-dimensional interference inspection algorithm is used to perform hard and soft interference detection on the virtual assembly process. The symmetry and feasibility of the segmented docking of the two sterns are verified in conjunction with the symmetrical reference plane of the two sterns.

[0133] The intelligent identification and classification module for docking errors is used to collect point cloud data of the docking end faces of the segments after the initial positioning of the segmented hoisting, extract the actual geometric data of the docking end faces, compare the geometric data in the segmented digital twin model to calculate the docking error, use the SVM machine learning algorithm to identify and classify the docking error, extract the feature parameters of systematic error, random error and deformation error, and establish an error classification database with self-learning function.

[0134] The adaptive multi-strategy collaborative compensation module is used to automatically match the optimal compensation strategy based on the error classification results using an adaptive selection algorithm. Systematic errors are preferentially compensated by process compensation, random errors are preferentially compensated by mechanical compensation, and deformation errors are compensated by a combination of thermal and mechanical compensation. The module calculates the optimal compensation amount for each type of error and implements multi-strategy collaborative compensation. During the compensation process, the docking accuracy is monitored in real time and the compensation is iteratively performed until the accuracy requirements are met.

[0135] The hoisting attitude prediction and welding synchronization control module is used to collect historical attitude data during the hoisting process using an attention-based LSTM prediction model to predict the hoisting attitude. Based on the prediction results, the hoisting equipment parameters are actively adjusted. The welding parameters on both sides of the stern are synchronously controlled by a fuzzy adaptive PID algorithm. The welding deformation is predicted and suppressed by the welding deformation prediction formula. The welding operation is completed by using a symmetrical welding process. The modular integration and installation of each functional module are completed simultaneously.

[0136] The global accuracy detection and process optimization module is used to perform global accuracy inspection on the assembled twin-stern hull using a parallel accuracy detection algorithm. If the inspection fails, it triggers a readjustment and compensation. It also builds a big data analysis platform for assembly accuracy, mines the correlation between process parameters and assembly accuracy through association rules, constructs a process iterative optimization model, and realizes closed-loop optimization of process parameters.

[0137] The central control unit is communicatively connected to each of the above modules and is used to coordinate and control the operation sequence, data interaction and command issuance of each module.

[0138] An application example provided by this invention:

[0139] S1: Double-stern-line super-segmentation and integrated collaborative design

[0140] 1.1 Dual-Stern Super-Segmentation Multi-Objective Optimization Partitioning

[0141] Combining the surface curvature distribution characteristics of the CSOV's Ulstein Twin X-Stern twin-stern hull, hull structural strength design specifications, the ultimate load-bearing capacity of shipyard lifting equipment, and the multiple requirements of segmented pre-outfitting construction and integrated green power system, a multi-objective collaborative weighted optimization algorithm is adopted to carry out the super-segment design of the hull. This algorithm takes minimizing structural stress, optimizing lifting load adaptation, maximizing pre-outfitting integration, and minimizing assembly cumulative error as four-dimensional core optimization objectives. A multi-objective quantitative optimization function is constructed to rationally divide the overall hull into 6-8 structurally independent, functionally complete, and modularly adaptable super-segments. The core optimization function is as follows:

[0142]

[0143] The parameters in the formula are defined as follows: To comprehensively optimize the target value, the smaller the value, the higher the optimality of the segmentation scheme; This represents the maximum structural stress in each segment. This refers to the deviation between the segment weight and the rated load for hoisting. The energy loss coefficient for pre-outfitting and green power integration; This represents the theoretical assembly deviation value for segments; The corresponding weighting coefficients have a total weight of 1, focusing on four dimensions: structural strength, hoisting compatibility, integration efficiency, and assembly accuracy.

[0144] The division process strictly avoids areas of abrupt curvature change on the double stern surface. Areas with a curvature ≥0.05m⁻¹ are defined as high-risk areas of curvature change, avoiding segmentation at high curvature locations and eliminating initial deviation problems such as surface structure deformation and docking misalignment from the source.

[0145] Ensure that each super segment is fully integrated with double stern curved surface structural units, standardized pre-outfitting units, and dedicated installation areas for green power equipment; all segment docking surfaces adopt a unified standardized structural design, and the weight of each segment structure is controlled synchronously to make it fully compatible with the load-bearing capacity of existing hoisting equipment.

[0146] 1.2 Design of Active Control for Dual Stern Symmetry

[0147] A three-dimensional global coordinate system for the CSOV hull was established, and the axial reference was defined as follows: the keel centerline was the X-axis, the horizontal transverse width was the Y-axis, the vertical draft of the hull was the Z-axis, and the central symmetry plane of the twin stern structure was the global reference plane, which served as the unified core reference for the accuracy detection, design correction, and construction control of the twin stern symmetry of the entire ship.

[0148] Based on a globally symmetric reference plane, a dual-stern symmetry active control PID algorithm is used to iteratively correct the segmented structure design parameters in real time, dynamically eliminating the asymmetric deviation of the dual-stern structure and ensuring the consistency of the dual-stern linear structure. The core calculation formula of the algorithm is as follows:

[0149] The parameters in the formula are defined as follows: Adjustments were made to ensure symmetry in the twin-stern structure. The real-time symmetry deviation value of the twin-stern structure at time t; This is the proportionality coefficient, with a value of 1.4; This is the integral coefficient, with a value of 0.2; The differential coefficient is 0.3. Through closed-loop iterative control of this algorithm, the overall symmetry deviation of the two sterns of the ship is ultimately guaranteed to be ≤±0.3mm.

[0150] 1.3 Design of Digital Interface Management for Pre-outfitting Units

[0151] A comprehensive digital interface management system for pre-outfitting units is established, standardizing and finalizing the interface dimensions, three-dimensional installation coordinates, and mechanical connection methods of various pre-outfitting units, such as hull pipelines and airborne equipment, to eliminate interface compatibility deviations from the design stage. Simultaneously, a quantitative evaluation model for interface matching is constructed to accurately quantify the assembly qualification of pre-outfitting interfaces. The core calculation formula is as follows:

[0152]

[0153] The parameters in the formula are defined as follows: This is to account for the overall matching error of the pre-outfitting interface; The specification requires a total of ≥50 test points for interface testing. Design the three-dimensional theoretical coordinates for the i-th measurement point; Let be the actual 3D assembly coordinates of the i-th measuring point, when the interface matching error... When the thickness is ≤±0.1mm, the interface is deemed to be compatible and meets the requirements for high-precision modular assembly.

[0154] 1.4 Integrated Collaborative Design of Green Power and Final Assembly

[0155] For green power system configurations such as battery hybrid power and methanol fuel reserved for ships, an integrated collaborative design is carried out for the layout of power equipment, pipeline system arrangement, and modular hull assembly process to avoid spatial conflicts between equipment, pipelines, and hull structure. A quantitative calculation model for the degree of collaborative coupling is constructed to quantify the compatibility between the green power system and each process of hull assembly. The core calculation formula is as follows:

[0156]

[0157] The parameters in the formula are defined as follows: The overall synergistic coupling degree between green power systems and final assembly processes; To ensure the number of core components in the collaborative design, there should be no fewer than four items, covering key aspects such as pipeline layout, equipment installation, operating space reservation, and maintenance access planning. The single coupling degree score is given for the i-th collaborative link; Let be the weight coefficient of the i-th collaborative link, ranging from 0.2 to 0.3, and the sum of the weights of all links is 1. When the overall collaborative coupling degree... When the value is ≥0.85, the integrated collaborative design is deemed qualified, which can effectively avoid structural interference and spatial conflict problems.

[0158] S2: Segmented manufacturing precision control and 3D data acquisition, digital twin modeling

[0159] 2.1 Segmented Full-Dimensional 3D Data Acquisition

[0160] High-precision laser 3D scanning equipment was used to conduct full-area, blind-spot-free 3D scanning and acquisition of each completed super segment, obtaining complete segment point cloud data. The data types mainly include four core types: coordinate data of the twin stern structure curved surface, geometric contour data of the segment docking end face, coordinate data of the pre-outfitting key interface, and point data of the propeller base mounting surface. The scanning data acquisition density is ≥10 points / mm², and the scanning range fully covers the core precision areas such as the twin stern curved surface contour, segment docking end face, pre-outfitting key interface, and propeller base mounting surface, ensuring that the acquired data is complete, blind-spot-free, high-precision, and distortion-free.

[0161] 2.2 Point Cloud Data Preprocessing, Noise Reduction, and Filtering

[0162] To eliminate measurement noise, redundant specks, and anomalous interference data in the original point cloud data while preserving key geometric features of the hull surface, an improved point cloud filtering algorithm combining tensor voting and random sampling consistency is employed for data preprocessing. This scheme provides complete explicit expressions for the tensor voting kernel function T and the weight coefficients ωij, forming a complete filtering calculation system. The core preprocessing formulas are as follows:

[0163]

[0164] The parameters in the formula are defined as follows: The point cloud data value corresponding to the coordinate point after filtering optimization; These are the original scanned point cloud data values; This represents the tensor voting weight coefficient, with a value ranging from 0.1 to 0.3. The tensor voting kernel function value of neighborhood point j with respect to the center point; It is the set of neighborhood points corresponding to the center point.

[0165] Weight explicit expression:

[0166] In the formula: Let Euclidean distance be the spatial distance between the center point and its neighboring points. This is the weighting adjustment coefficient.

[0167] Explicit expression for the tensor voting kernel function:

[0168] In the formula: The angle between the normal vectors of the neighboring points and the center point is used to characterize the continuity and consistency of the hull surface structure. After preprocessing by this algorithm, the overall error of the point cloud data is ≤ ±0.05mm, which can provide high-precision data support for subsequent digital twin modeling, accuracy detection and error analysis.

[0169] 2.3 Construction and Real-time Update of Segmented Digital Twin Model

[0170] Based on preprocessed high-precision point cloud data, a full-coverage, high-fidelity segmented digital twin model is constructed. The model integrates comprehensive information such as the actual geometric shape of each segment, pre-outfitting construction status, and manufacturing precision deviations, achieving a precise one-to-one mapping between physical segments and the virtual model. A dynamic iterative update mechanism is employed to optimize the model's state in real time, eliminating model lag and ensuring real-time synchronization between the virtual model and the physical manufacturing process. The core update formula is as follows:

[0171]

[0172] The parameters in the formula are defined as follows: For the iteratively updated digital twin model data; This is the original model data at the current moment; The model update coefficients are fixed at 0.7. This data represents segmented, measured scanning data at the current moment. Through real-time iterative correction, the real-time manufacturing status of each segment can be accurately reproduced, ensuring consistency between virtual simulation and actual construction.

[0173] 2.4 Real-time monitoring and early warning of symmetrical deviation between the two sterns

[0174] Based on the globally symmetrical reference plane of the twin sterns constructed by S1, core symmetrical feature points such as the edge contours of the twin stern curved surfaces and the mounting points of the propeller base are selected. No fewer than 20 pairs of symmetrical detection points are selected, and a twin stern symmetrical deviation monitoring algorithm is used to quantitatively calculate the real-time symmetrical deviation, realizing dynamic monitoring and early warning of assembly deviation. The core calculation formula is as follows:

[0175] The parameters in the formula are defined as follows: This represents the real-time symmetry deviation value of the symmetrical points on both sterns; The actual three-dimensional coordinates of the corresponding symmetrical detection points on both sides of the stern. Set the deviation warning threshold. When the absolute value of the real-time deviation When the deviation is detected, the system will automatically trigger a deviation warning and adjust the manufacturing and assembly process parameters in a timely manner to avoid the accumulation and amplification of deviations.

[0176] S3: Feasibility Verification of Segmented Assembly Sequence Optimization and Virtual Assembly

[0177] 3.1 Assembly Order Optimization Based on Adaptive Genetic Algorithm

[0178] With the multi-dimensional optimization objectives of shortening the final assembly construction cycle, suppressing the accumulation of assembly errors, and avoiding virtual assembly interference, an adaptive genetic algorithm is used to construct a multi-objective fitness function and iteratively solve for the optimal segmented final assembly sequence. This step fully supplements the chromosome coding rules, the explicit formula of the adaptive mutation operator, and the complete quantitative calculation model of the final assembly cycle T(X), the cumulative error δtotal(X), and the interference risk coefficient I(X), forming a complete algorithm solution system, as detailed below:

[0179]

[0180] The parameters in the formula are defined as follows: The fitness value corresponds to the assembly sequence scheme X. The larger the fitness value, the higher the optimality of the scheme. The total assembly cycle corresponding to scheme X; This represents the cumulative assembly error corresponding to scheme X; The assembly interference risk coefficient corresponding to scheme X; These are weighting coefficients, with values ​​of 0.4, 0.3, and 0.3 respectively.

[0181] 1. Chromosome encoding: Encoding is done using integer sequences; chromosomes... , This is the super segment number, and the sequence order corresponds to the final assembly construction sequence.

[0182] 2. Explicit formula for adaptive mutation operator:

[0183] Two genes on a chromosome are randomly selected for exchange mutation to ensure the legality of the assembled sequence.

[0184] 3. Assembly cycle calculation model:

[0185] In the formula: The standard assembly time for the i-th segment; The time required for segmented switching of hoisting and transportation.

[0186] 4. Error cumulative calculation model:

[0187] In the formula: This represents the single assembly error at the i-th docking point; The error is accumulated with weights, and the weight increases as the time sequence progresses.

[0188] 5. Interference Risk Calculation Model:

[0189] In the formula: The number of hard interferences; This represents the number of soft interferences. For the total number of assembly inspection logs, soft interference is assigned a semi-risk weight.

[0190] The algorithm supports adaptive dynamic adjustment of iteration parameters. The crossover probability decreases linearly with the number of iterations, while the mutation probability increases linearly with the number of iterations, effectively avoiding the algorithm from getting trapped in local optima. The maximum number of iterations... After iterative convergence, the optimal segmented assembly sequence is output, which can effectively shorten the overall assembly cycle by ≥40%.

[0191] 3.2 Setting up a virtual assembly environment

[0192] Based on a segmented digital twin model, a virtual assembly environment is built, and the optimal assembly sequence, segmented geometric data, pre-outfitting data, and green power equipment data are imported to recreate the actual assembly scene and ensure that the virtual environment is consistent with the actual assembly station and equipment parameters.

[0193] 3.3 Multi-dimensional interference inspection and assembly verification

[0194] A multi-dimensional interference inspection algorithm is used to check for interference in the virtual assembly process, detecting both hard interference (structural collision, equipment collision) and soft interference (insufficient pipeline spacing, insufficient equipment installation space, insufficient operating space). Combined with the symmetrical reference plane of the two sterns, the symmetry and feasibility of the segmented docking of the two sterns are verified. If interference problems exist, the process returns to step S1 to adjust the segment division or pre-outfitting layout until all interference problems are resolved, ensuring that the virtual assembly is consistent with the actual final assembly.

[0195] Soft interference is a problem caused by insufficient assembly space, including insufficient pipeline installation spacing, lack of equipment installation space, and insufficient construction operation space. It is detected using a minimum spatial distance threshold judgment formula, the core formula of which is as follows:

[0196] The parameters in the formula are defined as follows: This is the minimum spatial distance between two assembly units; The minimum allowable safe distance thresholds for the corresponding working conditions are: pipeline spacing threshold 20mm, equipment installation reserved threshold 50mm, and construction operation threshold 300mm. The presence of soft interference was determined. No soft interference was determined.

[0197] The overall detection logic of the algorithm is as follows: iterate through the entire virtual assembly process frame by frame, perform hard and soft interference quantitative detection on all assembly units, and simultaneously verify the symmetry and process adaptability of the twin-stern segment docking structure. If any interference problem or symmetry deviation exceeds the standard, immediately return to step S1 to re-optimize the segmentation scheme or pre-outfitting arrangement, iteratively optimizing until all interference problems are completely eliminated and the symmetry accuracy meets the standard, ensuring the fault-free implementation of the actual final assembly operation on site.

[0198] S4: Segmented docking error acquisition and machine learning classification and recognition

[0199] 4.1 Accurate Collection of Docking Error Data

[0200] Following the optimized assembly sequence, each super segment was hoisted into place and preliminarily positioned. Laser 3D scanning equipment was used to collect full-domain point cloud data of the segment docking end face, extract the actual 3D geometric parameters of the docking end face, and compare and analyze them with the theoretical design parameters built into the digital twin model. This accurately solved various original docking errors such as segment docking position deviation, gap deviation, and angle deviation, providing accurate data samples for subsequent error classification, identification, and compensation.

[0201] 4.2 Error Identification and Classification Based on SVM Algorithm

[0202] The Support Vector Machine (SVM) machine learning algorithm is used to intelligently identify and accurately classify the collected multidimensional docking error samples. It can effectively distinguish three typical assembly errors: systematic errors, random errors, and structural deformation errors, and accurately extract the core feature parameters of each type of error. The algorithm's global optimization objective is to maximize the classification margin in high-dimensional space and minimize the sample classification error. This step completes the explicit expressions for the Gaussian kernel function and the SVM decision function, forming a complete error classification algorithm system. The core objective function, constraints, and inference formulas are as follows:

[0203]

[0204]

[0205] The parameters in the formula are defined as follows: The weight vector for the high-dimensional classification hyperplane; This is the error penalty coefficient, fixed at 10, used to balance model complexity and classification accuracy; Let be the classification slack variable for the i-th error sample; Label the sample category; For error feature sample data; This represents the hyperplane offset. This optimized algorithm achieves an assembly error classification and recognition accuracy of ≥98%.

[0206] Explicit expression for the Gaussian kernel function of SVM:

[0207] In the formula: The kernel width parameter controls the scale of feature space mapping.

[0208] SVM final classification decision function:

[0209] In the formula: For Lagrange multipliers, non-zero multipliers correspond to support vector samples. This is the sign discrimination function, which outputs the error category label.

[0210] 4.3 Construction of Self-Learning Error Classification Database

[0211] An error classification database with incremental self-learning iterative capabilities is constructed, storing core data such as characteristic parameters, generation mechanisms, deviation characteristics, and impact range of systematic errors, random errors, and deformation errors. A dynamic self-learning update mechanism is employed to continuously iterate and optimize the error classification model, thereby constantly improving the accuracy of error identification. The core update formula is as follows:

[0212]

[0213] The parameters in the formula are defined as follows: This is the error classification model after iterative updates; The original baseline classification model; The model's self-learning rate is fixed at 0.15. This refers to newly added measured error data collected on-site. Each time new error sample data is added, the model automatically iterates and optimizes, continuously improving error identification accuracy and providing reliable data support for intelligent matching of compensation strategies.

[0214] S5: Multi-strategy adaptive docking error collaborative compensation

[0215] 5.1 Compensation Strategy: Adaptive Matching Selection

[0216] Based on the error intelligent classification results of S4, a multi-strategy adaptive selection algorithm is adopted. Through quantitative weighted fitness calculation, the optimal compensation scheme is accurately matched for different types of errors. The core decision function is as follows:

[0217]

[0218] The parameters in the formula are defined as follows: To assess the overall adaptability of the compensation strategy, a larger value corresponds to the optimal adaptability of the solution. These are the weighting coefficients for the three types of compensation strategies, with mechanical compensation having a weight of 0.4, thermal compensation a weight of 0.3, and process compensation a weight of 0.3. Independent adaptability to a single compensation strategy.

[0219] Based on error characteristics, specific matching rules are formulated: systematic errors are preferentially adapted to process compensation, random errors are preferentially adapted to mechanical compensation, and structural deformation errors are adapted to a combined thermal and mechanical compensation mode. The optimal compensation scheme is automatically output based on the principle of maximum fit. This step supplements the three types of errors with specific fit mapping functions, achieving precise quantitative mapping from error type to compensation strategy fit.

[0220] 1. Systematic error fit mapping:

[0221] 2. Randomness error fit mapping:

[0222] 3. Deformation error adaptation mapping:

[0223] In the formula: This represents the steady-state mean of the system error. The standard deviation of random error; For the measured deformation of the ship's hull structure, the intelligent and precise optimal matching of compensation strategies can be achieved through a precise mapping model.

[0224] 5.2 Accurate Calculation of Multi-Strategy Collaborative Compensation Amount

[0225] Based on the optimal matching compensation strategy, a multi-strategy collaborative compensation algorithm is adopted, which integrates the advantages of mechanical compensation, thermal compensation, and process compensation. The optimal total compensation amount for docking error is calculated by weighting, so as to realize multi-dimensional error collaborative correction. The core calculation formula is as follows:

[0226] The parameters in the formula are defined as follows: The optimal total compensation for docking error; These are three independent compensation quantities: mechanical, thermal, and process. The dynamic weight coefficients corresponding to the compensation method are adaptively adjusted according to the error type, and the sum of the three weights is always 1.

[0227] To achieve accurate quantitative solution of compensation, this step completes the independent calculation models for the three types of compensation and builds a complete fuzzy PID control system adapted to thermal compensation and temperature control, including parameter calculation formulas and a 5×5 fuzzy rule base, to achieve intelligent closed-loop control of the compensation process.

[0228] 1. Independent calculation model for mechanical compensation:

[0229] In the formula: For segmented positioning deviation, This is the mechanical rigidity correction factor.

[0230] 2. Independent calculation model for heat compensation:

[0231] In the formula: The coefficient of linear expansion of steel. To heat the temperature difference, This is the structural compensation length.

[0232] 3. Independent calculation model for process compensation:

[0233] In the formula: For systematic steady-state deviation, This is the process correction factor.

[0234] 4. Complete fuzzy PID control formula:

[0235]

[0236] Reference parameters: ; For temperature deviation, This represents the rate of change of deviation.

[0237] This algorithm constructs a 5×5 fuzzy inference rule base, classifying the input deviation e and the deviation change rate ec into five levels: NB (large negative), NM (medium negative), ZO (zero), PM (medium positive), and PB (large positive). The control logic is as follows: when the deviation is large, the proportional coefficient is increased first to suppress integral saturation and quickly eliminate steady-state deviation; when the deviation change rate is fast, the differential coefficient is increased to dampen system fluctuations; the parameter correction amplitude is weakened in the steady-state range to ensure system stability, achieving thermal compensation temperature control accuracy ≤ ±5℃, no secondary structure deformation, and an overall calculation error of compensation ≤ ±0.1mm.

[0238] The input quantities are divided into {NB, NM, ZO, PM, PB} (negative large, negative medium, zero, positive medium, positive large); when the deviation is large, the proportional coefficient is increased first to suppress integral saturation; when the deviation changes rapidly, the derivative coefficient is increased to dampen fluctuations; the parameter output is weakened in the steady state range to ensure the thermal compensation temperature control accuracy is ≤±5℃, eliminate secondary structure deformation, and the overall error of the compensation calculation is ≤±0.1mm.

[0239] 5.3 Implementation and Iterative Optimization of Precise Compensation

[0240] Based on the optimal collaborative compensation amount, hierarchical and precise compensation is carried out: mechanical rigidity adjustment compensation is achieved through a six-degree-of-freedom positioning device, structural thermal deformation compensation is achieved through a controllable heating and temperature control device, and process flexibility compensation is achieved by adjusting the welding sequence and optimizing the positioning fixtures. Throughout the compensation process, laser tracking equipment is used to monitor the docking accuracy in real time. An iterative convergence closed-loop control algorithm is established, and the compensation termination condition is precisely controlled through a residual error judgment formula to achieve iterative error convergence. The core iterative judgment formula is as follows:

[0241]

[0242] The parameters in the formula are defined as follows: The residual assembly error after compensation in the (k+1)th iteration; These are the measured assembly coordinates after iterative compensation; Assemble coordinates for theoretical standards; The residual error convergence threshold is set. Continuous iterative compensation is performed until the full set of accuracy standards are met: residual error ≤ ±0.1mm, butt gap ≤ 0.5mm, misalignment deviation ≤ ±0.8mm, and angle deviation ≤ 0.5°, thus completing the full closed loop of error compensation.

[0243] S6: Modular Assembly, Intelligent Construction, and Synchronous Integration

[0244] 6.1 Lifting Attitude Prediction and Control Based on Attention LSTM

[0245] To address the issues of large attitude fluctuations and difficulty in controlling positioning accuracy during segmented hoisting, an attention-enhanced LSTM temporal prediction model is employed. This model collects temporal attitude data, including segment angles, spatial positions, and movement speeds, during the hoisting operation. It then predicts subsequent attitude changes in multiple hoisting steps and pre-adjusts the hoisting equipment's operating parameters, achieving proactive attitude prediction and precise control. The model's output formula and loss function are as follows:

[0246]

[0247]

[0248] The parameters in the formula are defined as follows: The predicted hoisting attitude value at time t; It is a non-linear activation function; These are the weights and biases of the model's output layer, respectively. These represent the hidden state and cell state of the LSTM model at time t, respectively. The root mean square error loss function; For historical data observation time steps; Predicting time steps for future posture; The coordinates are the actual attitude coordinates at time t; By predicting the attitude coordinates at time t and dynamically adjusting the hoisting equipment parameters based on the model prediction results, the final positioning deviation of the segments can be precisely controlled within the range of ≤±0.1mm.

[0249] 6.2 Intelligent prediction and parameter control of welding deformation

[0250] To ensure consistent welding processes on both sides of the stern and suppress asymmetrical welding deformation, a fuzzy adaptive PID algorithm is used to synchronously control key process parameters such as welding current, voltage, welding speed, and welding temperature on both sides, achieving bidirectional synchronization between welding progress and process parameters. Simultaneously, a multi-parameter welding deformation prediction model is established to predict welding deformation in advance and correct process parameters, controlling deformation deviation from its source. The core prediction formula is as follows:

[0251] The parameters in the formula are defined as follows: This is the predicted value for welding deformation; For real-time welding current; For real-time welding speed; For real-time welding temperature; These are the deformation correction coefficients corresponding to current, speed, and temperature, respectively, which effectively ensure symmetrical and controllable welding deformation of the twin sterns. This step completes the fuzzy adaptive PID inference system formula:

[0252] Fuzzy Adaptive PID Inference System:

[0253]

[0254] In the formula: To ensure synchronous deviation of welding parameters on both sides, This represents the real-time rate of change of the deviation. As a fuzzy inference mapping function, it can dynamically correct the three parameters of the PID according to the real-time welding conditions, so as to achieve high-precision synchronous control of welding parameters, construction progress and structural deformation on both sides of the stern.

[0255] 6.3 Implementation of Symmetrical Welding Process

[0256] A standardized construction process of "first butt joint shaping, then fillet welding for final form, and symmetrical welding from the center outwards" is adopted, coupled with a closed-loop control algorithm for synchronous and balanced welding on both sides. Welding operations are carried out simultaneously on both sides, using the center of symmetry of the two sterns as the reference starting point. The consistency of welding on both sides is quantitatively controlled through a synchronous deviation constraint formula for welding parameters. The core synchronous control formula is as follows:

[0257]

[0258] The parameters in the formula are defined as follows: This represents the comprehensive synchronous deviation value of the welding parameters on both sides. The welding currents for the left and right sides of the twin sterns are respectively; These are the welding speeds for the left and right sides of the twin sterns, respectively. These are the welding temperatures for the left and right sides of the twin sterns, respectively. The synchronization deviation threshold is fixed at 0.02. Strict control of process parameters is maintained during construction: welding speed 2~3mm / s, welding current 200~250A, and welding voltage 22~25V. Welding deformation is monitored in real-time by laser throughout the process. Parameters are dynamically adjusted when deformation exceeds ±0.2mm, ultimately stabilizing the welding deformation to ≤±0.2mm, ensuring the symmetrical consistency of the twin-stern structure.

[0259] 6.4 Integrated installation of functional modules

[0260] After completing all the procedures of segmented docking, error compensation, and welding finalization, the integrated installation of core functional modules such as the green power system, motion compensation system, and intelligent operation and maintenance system is carried out in a unified manner. The digital interface design standards are strictly matched to ensure that each functional module is accurately adapted to the hull structure. The overall installation error of the module is ≤±1mm, which meets the collaborative operation performance requirements of the ship's DP2+ positioning system and motion compensation system.

[0261] S7: Global accuracy verification and data-driven process closed-loop optimization

[0262] 7.1 Global Precision Detection of Parallel Algorithms

[0263] After the final assembly is completed, a large-scale parallel precision inspection algorithm is used to conduct a full-area high-precision inspection of the twin stern hulls. Through parallel calculation of a massive number of inspection points, the global average assembly deviation of the hull assembly is quickly calculated, balancing inspection efficiency and accuracy. The core calculation formula is as follows:

[0264]

[0265] The parameters in the formula are defined as follows: The average assembly deviation for the entire hull assembly; The total number of monitoring points in the entire area is N, and the standard requires N ≥ 10000. The three-dimensional theoretical coordinates of the detection point; Let be the actual three-dimensional coordinates of the i-th detection point.

[0266] The key accuracy indicators to be verified are: assembly deviation of the double stern curved outer plate ≤ ±2mm, gap between segmented mating surfaces ≤ 0.5mm, positioning deviation of the propeller base ≤ ±1mm, and symmetry deviation of the double stern ≤ ±0.3mm. If the overall accuracy inspection fails, immediately return to step S5 to repeat the error iteration compensation until all accuracy indicators meet the standards.

[0267] 7.2 Data Correlation Mining for Assembly Precision

[0268] A big data analysis platform for final assembly precision was established, collecting process data from the entire process, including segmented manufacturing, error detection, compensation construction, welding operations, and precision inspection. Using the Apriori association rule mining algorithm, quantitative calculations were performed using both support and confidence indicators to deeply explore the intrinsic correlation between process parameters and assembly precision. The core calculation formula is as follows:

[0269]

[0270]

[0271] The parameters in the formula are defined as follows: The correlation support of process parameter A to accuracy index B; The confidence level of the correlation between process parameter A and accuracy index B; The number of samples that simultaneously meet the standards for parameter A and indicator B. The number of samples that meet the standard for parameter A; This represents the total number of statistical samples. Data mining can accurately extract the optimal matching patterns between welding parameters, compensation strategies, segmentation schemes, and assembly accuracy, providing data support for process iteration and optimization.

[0272] 7.3 Iterative Closed-Loop Optimization of Process Parameters

[0273] A data-driven gradient descent process iterative optimization model is constructed, with optimal assembly accuracy as the core objective. Based on the comprehensive accuracy loss function, the optimal correction amount of process parameters is solved in reverse, achieving adaptive iterative optimization of core process parameters for segmentation, welding, and compensation. The core iterative formula is as follows:

[0274]

[0275] The parameters in the formula are defined as follows: These are the optimal process parameters after iterative optimization; These are the original process parameters for the current iteration cycle; The iteration step size is fixed at 0.05. This step provides the comprehensive evaluation loss function for hull assembly accuracy. It completes the analytical expression for the comprehensive assembly accuracy loss function J and its gradient partial derivative:

[0276] 1. Overall assembly accuracy loss function J:

[0277] In the formula: This represents the average deviation of the entire assembly. The symmetrical deviation is due to the double stern structure; The average residual error of segmented docking; It represents the absolute value of average welding deformation; the sum of all weighting coefficients is 1, which allows for comprehensive and precise quantification of assembly precision loss.

[0278] 2. Analytical expression for gradient partial derivative:

[0279]

[0280] Through a precise gradient back-reaming mechanism, core process parameters such as segmentation, welding construction, and error compensation can be optimized in a targeted manner. The optimal process parameters after iterative optimization are fed back to the subsequent hull assembly construction process, forming a closed-loop process iteration system of "construction, testing, analysis, and optimization", which continuously improves the accuracy and efficiency of ship assembly.

[0281] 7.4 Archiving of the entire process data

[0282] The design data, manufacturing data, error data, compensation records, welding parameters, precision inspection data, and process optimization parameters of the entire final assembly process are compiled and archived to form a complete final assembly process archive, providing technical support for subsequent mass production and process optimization.

[0283] 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 modular assembly process for a CSOV (Conductor-Shipboard) with twin stern lines and a method for compensating for section docking errors, characterized in that... Includes the following steps: S1: Based on the geometric features of the twin-stern linear curved surface, structural strength requirements, shipyard hoisting equipment load-bearing capacity, and the integration requirements of pre-outfitting and green power, a multi-objective collaborative optimization algorithm is used to divide the CSOV hull into multiple super sections; Based on the super segmentation results, a dual-stern symmetric active control design, digital interface management of pre-outfitting units, and integrated collaborative design of green power and final assembly are established. S2: Perform a full-range scan of the super segment during the manufacturing process, collect segment point cloud data, and perform noise reduction and filtering preprocessing on the point cloud data; Based on the preprocessed point cloud data, a segmented digital twin model is constructed, which includes segmented geometric features, pre-outfitting status and manufacturing precision information, and the segmented digital twin model is updated in real time. Based on the active control design of dual stern symmetry in S1, the dual stern symmetry reference surface is selected, the edges of the dual stern curved surfaces and key symmetry points of the propeller base are selected, the dual stern symmetry deviation is monitored and an early warning threshold is set; S3: An adaptive genetic algorithm is used to construct a multi-objective fitness function, and the optimal segmented assembly order is obtained through iterative optimization. A virtual assembly environment is built based on the segmented digital twin model. The optimal assembly sequence, segmented geometric data, pre-outfitting data, and green power equipment data are imported. A multi-dimensional interference inspection algorithm is used to inspect the virtual assembly process for interference, detect hard interference and soft interference, and verify the symmetry and feasibility of the segmented docking of the two sterns in conjunction with the symmetrical reference plane of the two sterns. S4: Hoist the segments to the final assembly station and preliminarily position them according to the optimal assembly sequence, collect point cloud data of the segment docking end faces, extract the actual geometric data of the docking end faces, compare the geometric data in the segment digital twin model, and calculate the docking error; The SVM machine learning algorithm is used to identify and classify the docking errors, extract the feature parameters of systematic errors, random errors and deformation errors, and establish an error classification database with self-learning function; S5: Based on the error classification results, an adaptive selection algorithm for compensation strategies is used to automatically match the optimal compensation strategy. Systematic errors are preferentially compensated by process compensation, random errors are preferentially compensated by mechanical compensation, and deformation errors are compensated by a combination of thermal and mechanical compensation. Calculate the optimal compensation amount for various types of errors, implement multi-strategy collaborative compensation according to the optimal compensation amount, monitor the docking accuracy in real time during the compensation process and iterate the compensation until the accuracy requirements are met; S6: An attention-based LSTM prediction model is used to collect historical attitude data during the hoisting process to predict the hoisting attitude, and the hoisting equipment parameters are actively adjusted based on the prediction results; By synchronously controlling the welding parameters on both sides of the stern using a fuzzy adaptive PID algorithm, and predicting and suppressing welding deformation using a welding deformation prediction formula, the welding operation is completed using a symmetrical welding process, and the modular integration and installation of each functional module is completed simultaneously. S7: Use a parallel accuracy detection algorithm to perform a global accuracy inspection on the assembled twin-stern hull. If the inspection fails, return to step S5 for readjustment. A big data analysis platform for final assembly accuracy was established. By mining the correlation patterns between process parameters and assembly accuracy through association rules, a process iterative optimization model was constructed to achieve closed-loop optimization of process parameters.

2. The modular assembly process and segment docking error compensation method for a CSOV twin-stern hull as described in claim 1, characterized in that, In step S1, dividing the CSOV hull into multiple super sections specifically includes: Based on the curvature distribution of the Ulstein TwinX-Stern twin-stern linear surface, structural strength requirements, shipyard lifting equipment load-bearing capacity, and pre-outfitting and green power integration needs, the CSOV hull is divided into 6-8 super sections using the aforementioned multi-objective collaborative optimization algorithm. During the division process, abrupt curvature changes in the twin-stern surface are avoided, with the curvature of these abrupt curvature changes not less than 0.05 per meter. Each super section integrates complete twin-stern surface units, pre-outfitting units, and green power equipment installation areas. The docking surfaces of each super section adopt standardized designs, and the section weight is controlled within the load-bearing capacity of the shipyard's existing lifting equipment.

3. The modular assembly process and segment docking error compensation method for a CSOV twin-stern hull as described in claim 1, characterized in that, In step S1, the dual-stern symmetry active control design includes: A global coordinate system for the CSOV hull is established, with the keel centerline as the X-axis, the beam direction as the Y-axis, and the draft direction as the Z-axis. A global symmetry reference plane is established with the stern symmetry center as the reference. A PID algorithm for active control of stern symmetry is adopted, and the symmetry adjustment is calculated based on the proportional, integral, and derivative terms of the real-time symmetry deviation value, where the proportional coefficient is 1.4, the integral coefficient is 0.2, and the derivative coefficient is 0.

3. The segment design parameters are adjusted in real time to ensure that the stern symmetry deviation is no greater than ±0.3 mm. The digital interface management of the pre-outfitting unit includes: A digital interface management system for pre-outfitting units is constructed, and the interface dimensions, coordinates, and connection methods of pre-outfitting units are standardized. An interface matching degree evaluation model is established, which evaluates the interface matching degree by calculating the average Euclidean distance between the actual coordinates and theoretical coordinates of each interface measuring point in the three-dimensional coordinate system. The number of interface measuring points is not less than 50, and the interface is judged to be qualified when the interface matching degree is not greater than ±0.1 mm. The integrated design of green power and final assembly includes: Conduct integrated collaborative design of green power systems and final assembly processes, establish a collaborative coupling degree calculation model, and use the weighted average of the coupling degree of each collaborative link as the overall collaborative coupling degree. The number of collaborative links shall not be less than 4, and the weight coefficient of each link shall be in the range of 0.2 to 0.

3. When the collaborative coupling degree is not less than 0.85, it shall be judged as qualified.

4. The modular assembly process and segment docking error compensation method for a CSOV twin-stern hull as described in claim 1, characterized in that, In step S2, a laser 3D scanning device is used to perform a full-range scan of the super segment in the manufacturing process. The scanning accuracy is no greater than ±0.1 mm, the scanning data acquisition density is no less than 10 points per square millimeter, and the scanning data covers the key parts of the double stern curved surface, the docking surface, and the pre-outfitting interface. The denoising and filtering preprocessing adopts an improved laser point cloud filtering algorithm, namely tensor voting combined with random sampling consensus algorithm. The point cloud data is filtered by tensor voting value and voting weight, where the voting weight is in the range of 0.1 to 0.3, and the point cloud error after filtering is no greater than ±0.05 mm. The point cloud data includes: Twin-stern structure surface data: Three-dimensional coordinate point data of the entire twin-stern surface contour, used to determine the twin-stern symmetry deviation and surface assembly accuracy; Segmented docking end face data: Three-dimensional geometric point data of each super segment docking surface, used to calculate assembly errors such as segmented docking gap, misalignment, and angle; Key interface data for pre-outfitting: Three-dimensional coordinate data of interfaces of pre-outfitting units such as pipelines and equipment, used to verify the interface matching accuracy; Core base installation data: Three-dimensional point data of the mounting surfaces of key equipment such as the thruster base, ensuring the positioning accuracy of the core equipment; The segmented digital twin model is updated in real time using a digital twin real-time update mechanism. Based on the model data at the current moment, the update coefficient and the difference between the actual scan data and the model data at the current moment are used for correction and update. The update coefficient is 0.7, which realizes real-time synchronization between the model and the segmented manufacturing process. The monitoring of stern symmetry deviation includes: Based on the aforementioned stern-to-stern symmetry reference plane, select at least 20 pairs of key symmetry points on the edges of the stern curved surfaces and the propeller base. Employ a stern-to-stern symmetry deviation monitoring algorithm to obtain the real-time symmetry deviation by calculating the spatial distance of the corresponding key symmetry points in the three-dimensional coordinate system. Set the warning threshold to ±0.3 mm. When the absolute value of the real-time symmetry deviation exceeds the warning threshold, a deviation warning is triggered.

5. The modular assembly process and segment docking error compensation method for a CSOV twin-stern hull as described in claim 1, characterized in that, In step S3, the multi-objective fitness function constructed by the adaptive genetic algorithm uses the reciprocal of the weighted sum of the assembly cycle, the cumulative error, and the interference risk as the fitness value. The weights for assembly cycle, error accumulation, and interference risk are 0.4, 0.3, and 0.3 respectively. The optimization objectives are to shorten the assembly cycle, reduce error accumulation, and reduce interference risk. The crossover probability decreases linearly with the number of iterations, while the mutation probability increases linearly with the number of iterations. The maximum number of iterations is 120. The multi-dimensional interference inspection algorithm performs hard interference detection and soft interference detection on the virtual assembly process. The hard interference includes structural collisions and equipment collisions, and the soft interference includes insufficient pipeline spacing, insufficient equipment installation space, and insufficient operating space. If interference problems exist, return to S1 to adjust the segment division or pre-outfitting layout until all interference problems are resolved.

6. The modular assembly process and segment docking error compensation method for a CSOV twin-stern hull as described in claim 1, characterized in that, In step S4, the SVM machine learning algorithm is used to identify and classify the docking error, with the optimization objective of maximizing the classification margin and minimizing the classification error. A penalty coefficient is introduced to control the tolerance for classification errors, wherein the penalty coefficient is 10, to ensure that the error classification accuracy is not less than 98%. The error classification database with self-learning function adopts a self-learning update mechanism. Based on the original classification model, it updates the model by combining the learning rate and the difference between the new error data and the output of the original model. The learning rate is 0.15, and the classification model is automatically updated when new error data is added.

7. The modular assembly process and segment docking error compensation method for a CSOV twin-stern hull as described in claim 1, characterized in that, In step S5, the adaptive selection algorithm for compensation strategy selects the optimal compensation strategy by calculating the weighted fit of three strategies: mechanical compensation, thermal compensation, and process compensation. The mechanical compensation fit weight is 0.4, the thermal compensation fit weight is 0.3, and the process compensation fit weight is 0.

3. The optimal compensation strategy is automatically matched based on the maximum weighted fit. The optimal compensation amount for various types of errors is calculated using a multi-strategy collaborative compensation algorithm. The mechanical compensation amount, thermal compensation amount, and process compensation amount are weighted and summed according to their compensation weights to obtain the total compensation amount. The compensation weights of the three compensation methods are adjusted according to the error type and the sum is 1. The thermal compensation adopts fuzzy PID control, and the control algorithm is the same as the symmetric active control PID algorithm in step S1.

2. The parameters are adjusted to a proportional coefficient of 1.2, an integral coefficient of 0.15, and a derivative coefficient of 0.25 to ensure that the thermal compensation temperature control accuracy is no greater than ±5 degrees Celsius. The implementation of multi-strategy collaborative compensation includes: Mechanical compensation is achieved using a six-degree-of-freedom positioning device, thermal compensation using a controllable heating device, and process compensation using welding sequence adjustment and positioning tooling optimization. During the compensation process, laser tracking equipment is used to monitor the docking accuracy in real time. After each compensation is completed, point cloud data of the docking end face is collected and residual error is calculated until the residual error is no greater than ±0.1 mm, the docking gap is no greater than 0.5 mm, the misalignment deviation is no greater than ±0.8 mm, and the angle deviation is no greater than ±0.5 degrees.

8. The modular assembly process and segment docking error compensation method for a CSOV twin-stern hull as described in claim 1, characterized in that, In step S6, the attention-based LSTM prediction model uses the attention mechanism to weightedly fuse the hidden state information of historical hoisting posture data, and obtains the hoisting posture prediction value through the output layer mapping. The root mean square error between the predicted posture and the true posture is used as the loss function, where the observation time is 10 time steps and the prediction time is 5 time steps. The hoisting equipment parameters are actively adjusted according to the prediction results to ensure that the segmented positioning deviation is no greater than ±0.1 mm. The fuzzy adaptive PID algorithm synchronously controls the welding parameters on both sides of the stern, collects welding current, voltage, speed, and temperature parameters on both sides in real time and adjusts them dynamically; combined with the welding deformation prediction model, the welding deformation is predicted based on the linear combination of welding current, welding speed, and welding temperature, where the welding current correction coefficient is 0.002, the welding speed correction coefficient is 0.001, and the welding temperature correction coefficient is 0.0005, and the welding parameters are adjusted in advance to ensure symmetrical welding deformation of the stern. The symmetrical welding process adopts a butt welding followed by fillet welding and a centrally symmetrical welding process. Welding is carried out simultaneously from the center of symmetry of the two sterns to both sides. The welding speed is controlled at 2 to 3 millimeters per second, the welding current is 200 to 250 amperes, and the welding voltage is 22 to 25 volts. During the welding process, a laser scanning device is used to monitor the welding deformation in real time. If the deformation exceeds ±0.2 millimeters, the welding parameters are adjusted immediately to ensure that the welding deformation is controlled to be no greater than ±0.2 millimeters.

9. The modular assembly process and segment docking error compensation method for a CSOV twin-stern hull as described in claim 1, characterized in that, In step S7, the parallel precision detection algorithm obtains the global average deviation by calculating the average spatial distance between each detection point of the hull after final assembly and the theoretical coordinates, wherein the total number of detection points is not less than 10,000; the algorithm focuses on inspecting the fit of the twin stern curved surfaces, the segment docking accuracy, and the positioning accuracy of the propeller base, ensuring that the assembly deviation of the twin stern curved outer plate is not greater than ±2 mm, the gap between the segment docking surfaces is not greater than 0.5 mm, the positioning deviation of the propeller base is not greater than ±1 mm, and the symmetry deviation of the twin sterns is not greater than ±0.3 mm; The assembly precision big data analysis platform integrates data from the entire assembly process and uses an association rule mining algorithm to mine the correlation between process parameters and assembly precision by calculating the support and confidence of the simultaneous occurrence of process parameters and precision indicators. The process iterative optimization model is based on the current process parameters and iteratively updates the process parameters along the gradient direction of the accuracy evaluation index. The iteration step size is 0.

05. The model iteratively optimizes the segmentation parameters, welding parameters, and compensation strategies, and feeds the optimized process parameters back to the subsequent assembly stage to establish a closed-loop mechanism for process optimization.

10. A modular assembly process and section docking error compensation system for a CSOV twin-stern hull, applied to the modular assembly process and section docking error compensation method for a CSOV twin-stern hull as described in any one of claims 1-9, characterized in that, include: The super-segment division and collaborative design module is used to divide the CSOV hull into multiple super segments based on the geometric features of the twin stern linear curved surfaces, structural strength requirements, shipyard hoisting equipment load-bearing capacity, and pre-outfitting and green power integration requirements. It also establishes twin stern symmetry active control design, pre-outfitting unit digital interface management, and green power and final assembly integrated collaborative design. The segmented digital twin and symmetry monitoring module is used to perform full-range laser scanning on the super segments in the manufacturing process, collect segmented point cloud data and perform noise reduction and filtering preprocessing, extract actual segmented geometric data, construct segmented digital twin models containing segmented geometric features, pre-outfitting status and manufacturing accuracy information and update them in real time, and select the edges of the double stern curved surfaces and key symmetry points of the propeller base based on the double stern symmetry reference plane to monitor the double stern symmetry deviation and set early warning thresholds. The assembly sequence optimization and virtual assembly module is used to construct a multi-objective fitness function using an adaptive genetic algorithm to iteratively optimize and obtain the optimal segmented assembly sequence. Based on the segmented digital twin model, a virtual assembly environment is built. The optimal assembly sequence, segmented geometric data, pre-outfitting data, and green power equipment data are imported. A multi-dimensional interference inspection algorithm is used to perform hard and soft interference detection on the virtual assembly process. The symmetry and feasibility of the segmented docking of the two sterns are verified in conjunction with the symmetrical reference plane of the two sterns. The intelligent identification and classification module for docking errors is used to collect point cloud data of the docking end faces of the segments after the initial positioning of the segmented hoisting, extract the actual geometric data of the docking end faces, compare the geometric data in the segmented digital twin model to calculate the docking error, use the SVM machine learning algorithm to identify and classify the docking error, extract the feature parameters of systematic error, random error and deformation error, and establish an error classification database with self-learning function. The adaptive multi-strategy collaborative compensation module is used to automatically match the optimal compensation strategy based on the error classification results using an adaptive selection algorithm. Systematic errors are preferentially compensated by process compensation, random errors are preferentially compensated by mechanical compensation, and deformation errors are compensated by a combination of thermal and mechanical compensation. The module calculates the optimal compensation amount for each type of error and implements multi-strategy collaborative compensation. During the compensation process, the docking accuracy is monitored in real time and the compensation is iteratively performed until the accuracy requirements are met. The hoisting attitude prediction and welding synchronization control module is used to collect historical attitude data during the hoisting process using an attention-based LSTM prediction model to predict the hoisting attitude. Based on the prediction results, the hoisting equipment parameters are actively adjusted. The welding parameters on both sides of the stern are synchronously controlled by a fuzzy adaptive PID algorithm. The welding deformation is predicted and suppressed by the welding deformation prediction formula. The welding operation is completed by using a symmetrical welding process. The modular integration and installation of each functional module are completed simultaneously. The global accuracy detection and process optimization module is used to perform global accuracy inspection on the assembled twin-stern hull using a parallel accuracy detection algorithm. If the inspection fails, it triggers a readjustment and compensation. It also builds a big data analysis platform for assembly accuracy, mines the correlation between process parameters and assembly accuracy through association rules, constructs a process iterative optimization model, and realizes closed-loop optimization of process parameters. The central control unit is communicatively connected to each of the above modules and is used to coordinate and control the operation sequence, data interaction and command issuance of each module.

Images