Large-scale segmented assembly simulation method based on process optimization and knowledge decision

Abstract

A large segmented assembly simulation method based on process optimization and knowledge decision includes process data acquisition and classification, assembly relation analysis, assembly model design, process algorithm optimization and digital twinning, during construction, a worker can hold a terminal device to perform image recognition, then perform dynamic demonstration of procedures and guide segmented assembly work, and the process optimization and the knowledge decision are integrated. And automation, digitization and systematization from segmented assembly design to production are realized. According to the method, a digital twinborn technology and a decision tree algorithm are utilized, and various assembly characteristics and constraint conditions are deeply analyzed, so that a self-reasonable and efficient assembly process flow is automatically generated; meanwhile, information of the unit modules being assembled in the assembly site and information of the transfer equipment are obtained in real time through the Internet of Things technology, the outfield assembly positions of the to-be-assembled unit modules are reasonably arranged, and the utilization rate of the transfer equipment is increased.

Description

Technical Field

[0001] The present invention relates to the technical field of stern assembly, and particularly to a large-section assembly simulation method based on process optimization and knowledge decision-making. Background Art

[0002] Stern assembly is an important part of the sectional assembly operation of bulk carriers. Usually, the ship's power system and engine room system are distributed in the stern of the ship. Therefore, the assembly of stern unit modules is the most complex. At present, workers mainly rely on the part coordinates marked on paper drawings for assembly, which requires very high requirements for drawing design and part processing, and the optimal installation path is not reasonably planned, resulting in low overall efficiency and high error costs. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a large-section assembly simulation method based on process optimization and knowledge decision-making in view of the deficiencies of the prior art. By using digital twin technology and decision tree algorithm, through in-depth analysis of various assembly features and constraint conditions, an automatic and reasonable and efficient assembly process flow is generated; at the same time, the Internet of Things technology is used to obtain the information of the unit modules being assembled on the assembly site and the information of transfer equipment in real time, and the outfield assembly positions of the unit modules to be assembled are reasonably arranged to improve the utilization rate of transfer equipment.

[0004] The technical problem to be solved by the present invention is realized through the following technical solutions. The present invention is a large-section assembly simulation method based on process optimization and knowledge decision-making, and the steps of this method are as follows:

[0005] (1) Process data collection and classification

[0006] Collect the assembly historical data of each system for the stern section assembly of bulk carriers, including part information, assembly process parameters, environmental constraints, assembly quality records, assembly targets. At the same time, collect relevant process standard documents to clarify various process requirements and specifications, use OCR technology to quickly extract data from paper and picture documents, and obtain data from data sources such as the production management system, quality inspection system, and process document library of ship manufacturing enterprises. Then, perform data cleaning and data numbering on the large amount of collected data, and finally save the processed data to the database.

[0007] (2) Assembly relationship analysis

[0008] Based on the existing process documents, establish an assembly process library, derive, call, and inherit corresponding process functions from this library, extract eigenvalue, and establish a process data set.

[0009] (3) Assembly model design

[0010] Train a decision tree model based on existing assembly processes and datasets, and adopt a multi-variable mode where each variable represents a process. Use the componentized feature entities of the bulk carrier section as the basis for pruning and boundary division. After the model test passes, input the new shipbuilding data and quickly generate the new ship assembly process through the model;

[0011] (4) Process algorithm optimization

[0012] Extract the knowledge from the construction process instruction book of the new ship, establish the knowledge expression of the construction rules, and weight them according to the importance of the processes. Based on the rules and weights, use the process optimization algorithm to adjust the assembly process;

[0013] (5) Digital twin

[0014] Develop a digital twin simulation platform, simulate based on the process plan after algorithm optimization in the platform, conduct spatial verification of human factors and interference on the automatically generated assembly plan, and return the error data for process adjustment to ensure the feasibility of the process plan;

[0015] During construction, after the workers hold the terminal device for image recognition, conduct dynamic demonstration of the processes to guide the assembly operation of the section, and realize the automation, digitization, and systematization from the sectional assembly design to production.

[0016] The technical problem to be solved by the present invention can also be further realized by the following technical solution. For the above-mentioned large-section assembly simulation method based on process optimization and knowledge decision-making, the specific implementation path of step (2) is as follows:

[0017] The first step: Structural processing and knowledge modeling of process documents. Use natural language processing technology to parse unstructured documents such as CAD drawings, process cards, and operation manuals, extract key information such as process names, steps, tools, and quality standards, use XML and JSON format standards to establish a process metadata model, realize standardized data storage, and then establish a process association model based on ontology, define the logical relationship between processes, and use a graph database to store the process knowledge graph for supporting complex path queries and inferences;

[0018] The second step: Intelligent derivation and invocation of the process library. Use parametric process template design to encapsulate typical processes into configurable templates, drive process adjustment through variable parameters, and use process derivation rules to realize automatic process expansion through decision tree algorithms;

[0019] The third step: Feature value extraction and process dataset construction. First, parameterize the process feature dimensions of geometric features, process features, resource features, and quality features, and then establish a "process execution feedback closed-loop" to input the actual production data back into the process library, and optimize the feature weights and process parameters through machine learning algorithms.

[0020] The technical problem to be solved by the present invention can also be further realized by the following technical solutions. For the above-mentioned large-scale sectional assembly simulation method based on process optimization and knowledge decision-making, the algorithm training process of step (3) is as follows:

[0021] The first step: Collect data from four dimensions of the physical attributes, process constraints, environmental variables, and target results of the assembly scene to form a structured training set;

[0022] The second step: Perform data preprocessing on the sorted training set data, and perform secondary processing on the data using methods such as missing value filling, standardization / normalization, and categorical encoding to solve problems of missing values, noise, and heterogeneity, and provide a standardized format for model input;

[0023] The third step: Extract the key features of the stern assembly scene, and enhance the model's modeling ability for spatial, temporal, and process dependencies by means of key feature screening, temporal feature extraction, and spatial feature construction;

[0024] The fourth step: Encode the expert rules and process constraints for the assembly of the stern system unit module established previously into the model to prevent the algorithm output from making decisions that violate engineering common sense;

[0025] The fifth step: Construct a decision tree based on the Gini coefficient to achieve classification of assembly steps and regression prediction of time-consuming / quality;

[0026] The sixth step: Verify the effectiveness of the model in the stern assembly scene and continuously iterate and update for improvement.

[0027] The technical problem to be solved by the present invention can also be further realized by the following technical solutions. For the above-mentioned large-scale sectional assembly simulation method based on process optimization and knowledge decision-making, the specific operation of step (4) is as follows:

[0028] First, perform structured parsing on the process guidance data for the stern section assembly of newly built ships, construct a knowledge graph for newly built ships by sorting out the relationship types and node types, and complete the extraction and expression of the knowledge of the construction rules for newly built ships;

[0029] Secondly, optimize the process importance weighting system, and quantify the evaluation dimensions from four aspects of safety criticality, quality sensitivity, timing rigidity, and resource scarcity in combination with the characteristics of new ship construction. Among them, for safety criticality, use the AHP (Analytic Hierarchy Process) to divide the severity of the consequences caused by safety accidents into levels; for quality sensitivity, use the inverse proportional weighting method, especially for the welding and small-probability high-risk assembly work scenarios in the block assembly link; for timing rigidity, use 1 - elasticity coefficient to adjust the weights of each process according to the implementation progress of the assembly plan; for resource scarcity, use logarithmic normalization, which is obtained by the occupation duration of special equipment / total construction period. According to the above classification method, use the dynamic weight calculation formula:

[0030]

[0031] Where:

[0032] S i : Safety criticality weight;

[0033] Q i : Quality sensitivity weight;

[0034] T i : Timing rigidity weight;

[0035] R i : Resource scarcity weight;

[0036] K ship : Ship type correction coefficient, benchmark ship type = 1.0, bulk carrier = 0.9;

[0037] Thirdly, use the new shipbuilding knowledge graph to optimize the rule constraint splitting, and then improve the CART algorithm in combination with the process dynamic weight formula. The traditional Gini coefficient is improved to:

[0038]

[0039] Where, W k is the importance weight of process k, and its weighting logic is to assign higher weights to critical processes, making the impact of their splitting on the Gini coefficient greater, so as to guide the algorithm to optimize the critical path first;

[0040] D represents the complete training data set of the current node, including all samples to be segmented;

[0041] D k represents the kth sub-node after splitting, satisfying specific feature conditions;

[0042] |D| and |D k | respectively represent the number of samples in the data set D and the D k subset;

[0043] Finally, by introducing the corrected Gini coefficient, a comprehensive optimization of the CART algorithm in terms of splitting rules, tree structure bias, anti-noise ability, and multi-objective balance is carried out in a specific segmented assembly scenario.

[0044] The technical problem to be solved by the present invention can also be further realized by the following technical solutions. For the above-mentioned large-scale segmented assembly simulation method based on process optimization and knowledge decision-making, this method is implemented by developing a segmented assembly simulation platform. The segmented assembly simulation platform includes a process data collection and classification module, an assembly relationship analysis module, an assembly model design module, a process optimization algorithm module, and a digital twin and dynamic simulation module.

[0045] The process data collection and classification module is used to collect the historical assembly data of the stern section of a bulk carrier, including process data such as part information, assembly process parameters, process standard documents, environmental constraints, and assembly quality records.

[0046] The assembly relationship analysis module is used to establish an assembly process library based on the existing process documents, and derive, call, and inherit the corresponding process functions from this library, extract eigenvalue, and establish a process data set.

[0047] The assembly model design module is used to train a decision tree model according to the existing assembly processes and data sets.

[0048] The process optimization algorithm module is used to extract knowledge using the construction process instruction book of a new ship, establish the knowledge expression of construction rules, and weight them according to the importance of the processes. Then, based on the rules and weights, use the process optimization algorithm to adjust the assembly processes, so as to fully consider the characteristics of the new ship under construction and the characteristics of each process itself, and realize the optimal assembly process plan.

[0049] The digital twin and dynamic simulation module is used to develop a digital twin simulation platform, conduct human factor and interference spatial verification on the automatically generated assembly plan, and return error data for process adjustment to ensure that the process plan is feasible.

[0050] Compared with the prior art, the present invention proposes a large-scale segmented assembly simulation method based on process optimization and knowledge decision-making. Using the decision tree algorithm, based on a large amount of historical assembly data of ship stern section unit modules and related process data of ships, in-depth analysis is carried out on various assembly features and constraint conditions, and the most reasonable and efficient assembly process flow is automatically sorted out, realizing specific optimization of the decision tree algorithm.

[0051] For the unit module to be assembled, in the digital twin simulation assembly simulation platform of the present invention, the material information of the unit module is imported and the simulation button is clicked. The system combines the latest process documents of the new ship and uses the trained algorithm model to perform simulation assembly tests on the unit to be assembled and output the simulation results;

[0052] Secondly, in combination with digital twin technology, the operator of the present invention can, with the help of the twin model of the module unit, clearly view the specific operation steps of each assembly process, the installation positions and sequences of components, and other detailed information from an intuitive three-dimensional perspective. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is the assembly simulation system diagram of the present invention;

[0054] Figure 2 is the decision tree algorithm training and testing flowchart of the present invention;

[0055] Figure 3 is the digital twin platform assembly simulation flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0057] Refer to Figures 1-3 , a large-section assembly simulation method based on process optimization and knowledge decision-making, and the specific implementation is as follows:

[0058] The first stage: process data collection and classification;

[0059] Collect the assembly historical data of each system for the stern section assembly of the bulk carrier, including component information such as dimensions, materials, and weights; assembly process parameters such as welding current, voltage, and assembly sequence; environmental constraints such as the load capacity of lifting equipment, environmental temperature and humidity, and vibration interference; assembly quality records such as defect types and positions; and assembly objectives such as assembly time, quality score, cost consumption, and safety.

[0060] Meanwhile, collect relevant process standard documents, clarify various process requirements and specifications, and use OCR technology to quickly extract data from paper and picture documents. Secondly, obtain data from data sources such as the production management system, quality inspection system, and process document library of shipbuilding enterprises, clean and number the large amount of collected data, and finally save the processed data to the database.

[0061] The second stage: Assembly relationship analysis;

[0062] Based on the existing process documents, establish an assembly process library from which corresponding process functions can be derived, called, and inherited, extract eigenvalue, and establish a process data set. The specific implementation path is as follows:

[0063] The first step: Structured processing and knowledge modeling of process documents. Use natural language processing (NLP) technology to parse unstructured documents such as CAD drawings, process cards, and operation manuals, extract key information such as process names, steps, tools, and quality standards, establish a process metadata model using XML and JSON format standards to achieve standardized data storage, and then establish a process association model based on ontology, define the logical relationship between processes, and use a graph database to store the process knowledge graph to support complex path queries and inferences.

[0064] The second step: Intelligent derivation and call of the process library. Use parametric process template design to encapsulate typical processes such as section welding and pipeline installation as configurable templates, drive process adjustment through variable parameters such as plate thickness and weld length, and use process derivation rules to achieve automatic process expansion through decision tree algorithms.

[0065] The third step: Eigenvalue extraction and process data set construction. First, perform parametric processing on process feature dimensions such as geometric features, process features, resource features, and quality features, and then establish a "closed-loop feedback of process execution" to reverse input actual production data into the process library, and optimize feature weights and process parameters through machine learning algorithms.

[0066] The third stage: Assembly model design;

[0067] According to the existing assembly processes and data sets, train a decision tree model and adopt a multi-variable mode, where each variable is a process. Use the componentized feature entities of the bulk carrier section as the basis for pruning and boundary division. After the model test is passed, input new shipbuilding data, and new ship assembly processes can be quickly generated through the model. This process generation method fully combines big data and algorithms, improves the original ship design idea, greatly improves the design cycle of assembly processes, reduces manual workload, improves assembly automation, and reduces dependence on the personal abilities of engineers; The algorithm training and testing process is as Figure 2 shown;

[0068] The algorithm training process is as follows:

[0069] Step 1: Collect data from four dimensions: the physical properties, process constraints, environmental variables, and target results of the assembly scenario to form a structured training set;

[0070] Step 2: Perform data preprocessing on the organized training set data. Use processing methods such as missing value filling, standardization / normalization, and categorical encoding to reprocess the data, solve problems of missing values, noise, and heterogeneity, and provide a standardized format for model input;

[0071] Step 3: Extract the key features of the stern assembly scenario. Use methods such as key feature screening, temporal feature extraction, and spatial feature construction to enhance the model's modeling ability for spatial, temporal, and process dependencies;

[0072] Step 4: Encode the expert rules and process constraints for the assembly of the stern system unit module established previously into the model to prevent the algorithm from outputting decisions that violate engineering common sense;

[0073] Step 5: Build a decision tree based on the Gini coefficient to achieve classification of assembly steps and regression prediction of time-consuming / quality;

[0074] Step 6: Verify the effectiveness of the model in the stern assembly scenario and continuously iterate, update, and improve.

[0075] Fourth stage: Process algorithm optimization;

[0076] The process algorithm established using the decision tree model is obtained by simulating the algorithm according to the existing process data, but the importance of the processes and the construction characteristics and guidelines of newly built ships are not considered. Therefore, based on this, the present invention further optimizes; extracts the knowledge of the construction process guidebook of the newly built ship, establishes the knowledge expression of the construction rules, and weights according to the importance of the processes. Based on the rules and weights, uses the process optimization algorithm to adjust the assembly processes, making it fully consider the characteristics of the newly built ship and the characteristics of each process itself, and realizing the optimal solution of the assembly process. The specific implementation is as follows:

[0077] First, perform structured parsing on the assembly process guide data of the stern section of the newly built ship, and construct a knowledge graph of the newly built ship by sorting out the relationship types and node types. Thus, the extraction and expression of the construction rules knowledge of the newly built ship are completed;

[0078] Secondly, optimize the importance weighting system of the process. Combine the characteristics of new shipbuilding to quantify the evaluation dimensions from four aspects: safety criticality, quality sensitivity, timing rigidity, and resource scarcity. Among them, for safety criticality, use the AHP (Analytic Hierarchy Process) to divide the severity level of the consequences caused by safety accidents into grades (1-5); for quality sensitivity, use the inverse proportion weighting method, especially for the welding and small-probability high-risk assembly work scenarios in the block assembly link; for timing rigidity, use 1 - elasticity coefficient to adjust the weights of each process according to the implementation progress of the assembly plan; for resource scarcity, use logarithmic normalization, obtained by the occupation duration of special equipment / total construction period. According to the above classification method, use the dynamic weight calculation formula:

[0079]

[0080] Where:

[0081] S i : Weight of safety criticality;

[0082] Q i : Weight of quality sensitivity;

[0083] T i : Weight of timing rigidity;

[0084] R i : Weight of resource scarcity;

[0085] K ship : Ship type correction coefficient (benchmark ship type = 1.0, bulk carrier = 0.9);

[0086] The weights are not fixed, but are updated in real time with the progress of block construction and environmental changes.

[0087] Thirdly, use the knowledge graph of new shipbuilding to optimize the rule constraint splitting, and then improve the CART algorithm in combination with the process dynamic weight formula. The traditional Gini coefficient is improved to:

[0088]

[0089] Among them, W k is the importance weight of process k, and its weighting logic is to assign higher weights to critical processes, making the impact of their splitting on the Gini coefficient greater, so as to guide the algorithm to optimize the critical path first.

[0090] D represents the complete training data set of the current node (before branching), including all samples to be segmented (i.e., assembly process instances);

[0091] D k represents the kth child node (sub-data set) after splitting, satisfying specific feature conditions;

[0092] |D| and |Dk represent the sample numbers of the dataset D and its subset D k respectively;

[0093] Finally, by introducing the modified Gini coefficient, a comprehensive optimization of the CART algorithm in terms of splitting rules, tree structure bias, anti-noise ability, and multi-objective balance is achieved in a specific segmented assembly scenario.

[0094] Phase 5: Digital Twin;

[0095] Develop a digital twin simulation platform, simulate the optimized process plan based on the algorithm in the platform, conduct spatial verification of human factors, interference, etc. on the automatically generated assembly plan, and return error data for process adjustment to ensure the feasibility of the process plan.

[0096] During construction, workers can hold terminal devices for image recognition and then conduct dynamic demonstrations of the processes to guide the segmented assembly operations, achieving automation, digitization, and systematization from segmented assembly design to production.

[0097] Before conducting assembly simulation, technicians need to collect the historical process data of the stern assembly of the bulk carrier and train and test the simulation algorithm. Before assembling the module units of the stern section of the bulk carrier, the staff import the trained simulation assembly algorithm model into the digital twin platform, and import the part information and model data of the unit module to be assembled. By clicking the simulation button, conduct simulation on the unit module to be assembled, obtain the optimal assembly process, and generate the corresponding instruction manual to guide the workers in assembly.

[0098] The selection of the simulation algorithm for training and testing includes process data collection and classification, assembly relationship analysis, and assembly model design. Process data collection and classification mainly complete the collection of historical assembly data of the stern section of the bulk carrier, including part information, assembly process parameters, process standard documents, environmental constraints, assembly quality records, and other process data; assembly relationship analysis completes the establishment of an assembly process library based on the existing process documents, from which corresponding process functions can be derived, called, and inherited, extract feature values, and establish a process dataset; assembly model design completes the training of a decision tree model based on the existing assembly processes and datasets.

[0099] The digital twin platform includes a data import module, a model library management module, a simulation module, an algorithm management module, a data management module, and a twin simulation module. The data import module is used to import all relevant data of the unit to be assembled and classify and manage it. The model library management module is used to import the module part models of the unit to be assembled and manage them. The simulation module conducts assembly simulation on the unit to be assembled and generates corresponding assembly guidance documents. The algorithm management module is used to manage the simulation algorithms imported into the platform and collect various information during the algorithm execution process for subsequent algorithm optimization. The data management model is used to manage the data imported by users into the platform. The twin simulation module conducts spatial verification of human factors, interference, etc. on the automatically generated assembly plan, and returns error data for process adjustment to ensure that the process plan is feasible. At the same time, three-dimensional simulation technology is used to achieve visual guidance for assembly.

Claims

1. A large-scale sectional assembly simulation method based on process optimization and knowledge decision-making, characterized in that: The method steps are as follows: (1) Process data collection and classification Collect the assembly historical data of each system for the stern section assembly of bulk carriers, including component information, assembly process parameters, environmental constraints, assembly quality records, and assembly targets. At the same time, collect relevant process standard documents to clarify various process requirements and specifications. Use OCR technology to quickly extract data from paper and picture documents, and obtain data from the data sources of the production management system, quality inspection system, and process document library of the shipbuilding enterprise. Then, perform data cleaning and data numbering on the large amount of collected data, and finally save the processed data to the database; (2) Assembly relationship analysis Based on the existing process documents, establish an assembly process library, derive, call, and inherit the corresponding process functions from this library, extract eigenvalue, and establish a process data set; (3) Assembly model design According to the existing assembly processes and data sets, train a decision tree model and adopt a multi-variable mode, where each variable is a process. Use the componentized feature entities of the bulk carrier section as the basis for pruning and boundary division. After the model test passes, input the new shipbuilding data, and quickly generate the new ship assembly process through the model; (4) Process algorithm optimization Extract the knowledge from the construction process instruction book of the new ship, establish the knowledge expression of the construction rules, and weight them according to the importance of the processes. Based on the rules and weights, use the process optimization algorithm to adjust the assembly processes; (5) Digital twin Develop a digital twin simulation platform, simulate based on the process plan after algorithm optimization in the platform, conduct human factor and interference spatial verification on the automatically generated assembly plan, and return the error data for process adjustment to ensure that the process plan is feasible; During construction, the workers hold terminal devices to perform image recognition and then conduct dynamic demonstrations of the processes to guide the assembly operation of the section, realizing the automation, digitization, and systematization from section assembly design to production.

2. The large-scale sectional assembly simulation method based on process optimization and knowledge decision-making according to claim 1, characterized in that: The specific implementation path of step (2) is as follows: The first step: Process document structuring and knowledge modeling. Use natural language processing technology to parse unstructured documents such as CAD drawings, process cards, and operation manuals, extract key information such as process names, steps, tools, and quality standards, establish a process metadata model using XML and JSON format standards to achieve standardized data storage, and then establish a process association model based on ontology, define the logical relationships between processes, and store the process knowledge graph using a graph database to support complex path queries and inferences; The second step: Intelligent derivation and call of the process library. Use parametric process template design to encapsulate typical processes into configurable templates, drive process adjustment through variable parameters, and use process derivation rules to automatically expand processes through decision tree algorithms; The third step: Eigenvalue extraction and process data set construction. First, perform parametric processing on the process feature dimensions of geometric features, process features, resource features, and quality features, and then establish a "process execution feedback loop" to reverse input the actual production data into the process library, and optimize the feature weights and process parameters through machine learning algorithms.

3. The large-scale segmented assembly simulation method based on process optimization and knowledge decision-making according to claim 1, characterized in that: The algorithm training process of step (3) is as follows: Step 1: Collect data from four dimensions: the physical properties, process constraints, environmental variables, and target results of the assembly scenario to form a structured training set; Step 2: Perform data preprocessing on the organized training set data. Use methods such as missing value filling, standardization / normalization, and categorical encoding to reprocess the data, addressing missing values, noise, and heterogeneity issues, and providing a standardized format for model input; Step 3: Extract the key features of the stern assembly scenario. Adopt methods such as key feature screening, temporal feature extraction, and spatial feature construction to enhance the model's modeling ability for spatial, temporal, and process dependencies; Step 4: Encode the expert rules and process constraints for the stern system unit module assembly established previously into the model to prevent the algorithm from outputting decisions that violate engineering common sense; Step 5: Build a decision tree based on the Gini coefficient to achieve classification of assembly steps and regression prediction of time-consuming / quality; Step 6: Verify the effectiveness of the model in the stern assembly scenario and continuously iterate, update, and improve; 4. The large-scale segmented assembly simulation method based on process optimization and knowledge decision-making according to claim 1, characterized in that: The specific operations of step (4) are as follows: First, perform structured parsing on the process guidance data for the stern section assembly of newly built ships. Construct a knowledge graph for newly built ships by sorting out the relationship types and node types, and complete the extraction and expression of the construction rules knowledge for newly built ships; Second, optimize the importance weighting system for processes. Combine the characteristics of new ship construction to quantify the evaluation dimensions from four aspects: safety criticality, quality sensitivity, temporal rigidity, and resource scarcity. For safety criticality, use the AHP (Analytic Hierarchy Process) to divide the severity of the consequences caused by safety accidents into levels; For quality sensitivity, use the inverse proportional weighting method, especially for welding and small-probability high-risk assembly work scenarios in the section assembly link; For temporal rigidity, use 1 - elastic coefficient to adjust the weights of each process according to the implementation progress of the assembly plan; for resource scarcity, use logarithmic normalization, obtained by the occupation duration of special equipment / total construction period. According to the above division method, use the dynamic weight calculation formula: Where: S i : Safety-critical weight; Q i : mass sensitivity weight; T i : Temporal rigidity weight; R i : Resource scarcity weight; K ship : Ship form correction coefficient, reference ship form = 1.0, bulk carrier = 0.9; Third, use the knowledge graph of newly built ships to optimize the rule constraint splitting, and then improve the CART algorithm in combination with the process dynamic weight formula. The traditional Gini coefficient is improved to: Among them, W k is the importance weight of process k. Its weighting logic is to assign higher weights to critical processes, making the impact of their splitting on the Gini coefficient greater, thereby guiding the algorithm to preferentially optimize the critical path; D represents the complete training data set of the current node, including all samples to be split; D k represents the k-th child node after splitting, satisfying specific characteristic conditions; |D| and |D k represent the number of samples in the dataset D and D k subset, respectively; Finally, through the introduction of the modified Gini coefficient, comprehensively optimize the CART algorithm in terms of splitting rules, tree structure bias, anti-noise ability, and multi-objective balance in a specific section assembly scenario.

5. The large-scale segment assembly simulation method based on process optimization and knowledge decision-making according to claim 1, wherein: This method is implemented by developing a section assembly simulation platform, which includes a process data collection and classification module, an assembly relationship analysis module, an assembly model design module, a process optimization algorithm module, and a digital twin and dynamic simulation module; The process data collection and classification module is used to collect the historical assembly data of the stern section of bulk carriers, including process data such as part information, assembly process parameters, process standard documents, environmental constraints, and assembly quality records; The assembly relationship analysis module is used to establish an assembly process library based on the existing process documents, and derive, call, and inherit the corresponding process functions from this library, extract feature values, and establish a process data set; The assembly model design module is used to train a decision tree model based on existing assembly processes and datasets; The process optimization algorithm module is used to extract knowledge using the construction process instruction book of the new ship, establish the knowledge expression of construction rules, weight them according to the importance of processes, and then use the process optimization algorithm based on the rules and weights to adjust the assembly processes, so as to fully consider the characteristics of the newly built ship and the characteristics of each process itself, and achieve the optimal assembly process plan; The digital twin and dynamic simulation module is used to develop a digital twin simulation platform, conduct spatial verification of human factors and interference on the automatically generated assembly plan, and return error data for process adjustment to ensure that the process plan is feasible.

Images