Ship welding quality problem knowledge graph construction and analysis method

Abstract

The invention relates to the technical field of ship welding quality management, and discloses a ship welding quality problem knowledge graph construction method, which comprises the following steps of S1, defining event types and event relationships related to ship welding quality problems; s2, based on dependency syntactic analysis, pattern matching, similarity calculation and transition probability calculation methods, events and relationships between the events are extracted from the ship welding quality problem case report; and S3, utilizing Neo4j graph database software to construct a ship welding quality problem knowledge graph according to the events and relationships extracted in the step S2, and visually displaying the whole process of the welding quality problem from cause to result through the constructed knowledge graph, such as a typical evolution conduction path. By means of knowledge graph analysis, quality management personnel can quickly position key factors influencing the welding quality, the quality problem analysis efficiency is remarkably improved, and the defect that a traditional analysis method depends on subjective experiences is effectively overcome.

Description

Technical Field

[0001] This invention relates to the field of ship welding quality management technology, specifically to a method for constructing and analyzing a knowledge graph of ship welding quality problems. Background Technology

[0002] In the shipbuilding industry, welding quality is directly related to the safety, reliability, and service life of ships, and is one of the core quality control points in the shipbuilding process. As ships develop towards larger and higher value-added models, the amount of welding work has increased significantly, welding processes have become increasingly complex, and the difficulty of controlling welding quality problems has also increased accordingly.

[0003] Currently, leading shipbuilding companies both domestically and internationally have adopted a series of measures to improve welding quality and efficiency. These include vigorously promoting advanced welding methods such as robotic welding, automated welding, and intelligent welding; actively researching and applying new welding materials to improve welding performance; and accelerating the training of modern, highly skilled welding personnel to enhance operational skills. However, these improvement methods have significant limitations. On the one hand, the methods are too simplistic, relying excessively on the investment in new technologies, processes, and equipment. There is a serious lack of research on the systematic management and improvement of all aspects affecting welding quality, including personnel, machinery, materials, methods, and environment (i.e., people, machines, materials, methods, and environment), resulting in a failure to form a comprehensive, multi-layered quality control system.

[0004] Traditional methods primarily rely on traditional quality management tools such as cause-and-effect diagrams (fishbone diagrams), system diagrams, correlation diagrams, and Pareto charts. While these tools can perform preliminary qualitative analysis and data statistics on quality issues—for example, statistically analyzing the frequency and distribution of different types of welding defects (such as porosity, cracks, and lack of fusion)—they only summarize surface quality data and make simple judgments about quality. They cannot delve into the complete process of welding quality problems, from the emergence and development of potential risk factors to the final manifestation of quality defects. The formation of welding quality problems is often the result of the interaction and mutual influence of multiple factors. There are complex and multi-dimensional intrinsic relationships between different influencing factors, but traditional analytical tools lack the ability to quantitatively analyze and model these complex relationships, making it difficult to accurately depict the dynamic evolution of welding quality problems over time and under changing conditions. This directly leads to a lack of scientific basis and specificity in the quality control strategies formulated by enterprises, failing to fundamentally block the evolution path of quality problems, and hindering the proactive prevention and systematic control of welding quality problems. This severely restricts the further improvement of ship welding quality management.

[0005] To effectively address the aforementioned technical challenges, achieve in-depth analysis of the evolutionary logic of welding quality problems, and conduct comprehensive, systematic, and precise analysis of these problems, thereby constructing a scientific and effective welding quality control system, this invention proposes a method for constructing and analyzing a knowledge graph of ship welding quality problems based on knowledge graph technology. This method integrates and correlates various types of welding quality-related information by constructing a knowledge graph, analyzes the transmission paths and evolutionary patterns of quality problems, provides strong technical support for the systematic control of ship welding quality, and prevents or suppresses welding quality problems at their root. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing ship welding quality control methods, such as their simplistic approach and insufficient analytical depth, and to provide a method for constructing and analyzing a knowledge graph of ship welding quality issues, so as to achieve systematic analysis and effective control of welding quality problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for constructing a knowledge graph of ship welding quality issues includes the following steps: S1: Define the types of events and relationships related to ship welding quality issues; S2: Based on dependency parsing, pattern matching, similarity calculation, and transition probability calculation methods, extract events and relationships between events from case reports of ship welding quality problems; S3: Using Neo4j graph database software, construct a knowledge graph of ship welding quality issues based on the events and relationships extracted in S2.

[0008] The event types described in S1 include personnel operation-related events, equipment status-related events, material property-related events, process parameter-related events, environmental condition-related events, and quality problem result events.

[0009] As a further explanation of the present invention, the event relationships described in S1 include causal relationships and sequential relationships.

[0010] As a further explanation of the present invention, the processing procedure for the case report in S2 is as follows: dependency parsing is used to parse the text syntax structure, pattern matching is used to identify text fragments of preset event types, similarity calculation is used to normalize the fuzzy event description, and transition probability calculation is used to determine the correlation strength between events.

[0011] As a further explanation of the present invention, the knowledge graph constructed in S3 includes nodes and relationships. The node labels include reason_result and subservient. The relationship type corresponds to the event relationship defined in S1. The attribute information of the nodes and relationships includes the attribute keys from, relation, and to.

[0012] As a further explanation of the present invention, the personnel operation-related events include insufficient sense of responsibility of construction workers, inadequate communication of management requirements, and construction workers' violation of drawing requirements.

[0013] As a further explanation of the present invention, the causal relationship is a relationship in which one event directly leads to the occurrence of another event, and the sequential relationship is a relationship in which one event occurs and another event occurs in sequence.

[0014] A method for analyzing a knowledge graph of ship welding quality issues includes the following steps: P1: Based on the knowledge graph of ship welding quality problems constructed by the method described in claim 1, calculate the evolution and transmission path of ship welding quality problems; P2: Based on the analysis results of the evolution and transmission path, a causal model of ship welding quality problems is formed, and targeted welding quality control measures are formulated.

[0015] As a further illustration of the present invention, in P1, a path query algorithm is executed in the Neo4j graph database to obtain the complete evolutionary transmission path from cause to effect for different quality problems.

[0016] As a further explanation of the present invention, the causal model in P2 is constructed based on the correlation and evolutionary transmission path of events in the knowledge graph, and is used to characterize the influence logic of human, machine, material, method and environment factors on welding quality problems.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention is applied to welding quality management in shipbuilding enterprises. Through the construction of a knowledge graph, it visually displays the entire process of welding quality problems from cause to effect, as shown in the typical evolutionary transmission path. With the help of knowledge graph analysis, quality management personnel can quickly locate key factors affecting welding quality, significantly improving the efficiency of quality problem analysis and effectively avoiding the drawbacks of traditional analysis methods that rely on subjective empiricism. The resulting knowledge base and knowledge association paths for welding quality issues provide solid data support for formulating control measures, making countermeasures more targeted and effective. After applying this method, the incidence of ship welding quality problems significantly decreased, and the level of welding quality management was effectively improved, verifying the practicality and effectiveness of the invention. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the construction of a knowledge graph for welding quality issues in this invention. Figure 2 This is a detailed flowchart of the data preprocessing process of the present invention; Figure 3This is a detailed flowchart of the extraction of atomic events for welding quality problems in this invention; Figure 4 This is a detailed flowchart of the extraction of welding quality problem event relationships in this invention; Figure 5 This is a detailed flowchart illustrating the generalization of atomic events in welding quality issues according to the present invention. Figure 6 This is a detailed flowchart of the event transition probability calculation in this invention; Figure 7 A detailed flowchart illustrating the construction of the welding quality problem knowledge graph for N Shipbuilding Company in this invention; Figure 8 This is a schematic diagram illustrating the creation of knowledge nodes in this invention. Figure 9 This is a schematic diagram of the welding quality problem knowledge graph of the present invention; Figure 10 This is a schematic diagram illustrating the evolution and transmission path of the "management personnel's requests not being properly communicated" issue in this invention. Figure 11 This is a schematic diagram illustrating the evolution and transmission path of "inadequate pre-job training in companies" in this invention. Figure 12 This is a schematic diagram illustrating the evolution and transmission path of "improper storage of welding materials" in this invention; Figure 13 This is a schematic diagram illustrating the evolution and transmission path of the "lack of responsibility among construction workers" in this invention. Figure 14 This is a schematic diagram illustrating the causal model of ship welding quality problems constructed based on the knowledge graph analysis of welding quality problems in this invention. Detailed Implementation

[0019] In the following description, various embodiments of the invention will be described with reference to the accompanying drawings.

[0020] Example 1: Please refer to Figures 1-14 The present invention provides a technical solution: Step 1: Event Type and Relationship Definition Based on a comprehensive survey of factors affecting ship welding quality, the event types are defined as follows: 1. Personnel-related incidents: These include specific scenarios such as insufficient responsibility of construction personnel (e.g., failure to operate according to procedures), inadequate communication of management requirements (e.g., omission of technical briefings), construction personnel violating drawing requirements (e.g., unauthorized changes to welding sequence), and deviations in the execution of inspection standards by quality inspectors. 2. Equipment status related events: including abnormal equipment conditions such as excessive fluctuations in welding equipment current and voltage, unstable arc caused by welding torch wear, uneven wire feeding caused by wire feeding motor failure, and water leakage in the cooling system affecting equipment operation; 3. Material characteristic related events: These involve material quality issues such as excessive humidity in the welding material storage environment leading to moisture absorption, welding material model not matching the process documents, flux particle size not meeting requirements affecting slag removal performance, and shielding gas containing excessive impurities. 4. Process parameter related events: including parameter setting issues such as welding current exceeding the process range, welding voltage and current mismatch, welding speed being too fast leading to incomplete fusion, and preheating temperature not reaching the specified process value; 5. Environmental Conditions Related Events: These include environmental factors such as excessive humidity in the work area (e.g., failure to control humidity during the rainy season), changes in welding material performance due to high temperature, damage to gas protection due to excessive wind speed, and contamination of the molten pool by excessive dust concentration. 6. Quality problem outcome events: These include typical welding defects such as surface porosity of welds, incomplete root penetration, longitudinal cracks in welds, excessive undercut depth, excessive weld beads, and excessive joint deformation.

[0021] Event relationships are defined as two core types: Causal relationship: refers to the direct cause of the preceding event to the occurrence of the following event, such as "welding materials getting damp" directly leading to "porosity in the weld"; Sequential relationship: refers to the relationship in which events occur sequentially in time, such as "lack of pre-job training" followed by "insufficient operator skills".

[0022] S2: Collected over 500 case reports on ship welding quality issues from the past three years, covering quality problems in different welded parts such as hull structure, compartments, and pipelines. Utilized the LTP (Language Technology Platform) dependency parsing tool to perform grammatical analysis on the report text, identifying subject-verb-object structures and modifiers, and extracting core information such as "event subject + action + result".

[0023] Relationships are extracted based on a pre-defined pattern library: causal relationship patterns include "caused by…", "caused by…", "…is the cause of…", etc.; sequential relationship patterns include "first…then…", "after…", "following…", etc. Pattern matching directly identifies causal relationship pairs such as "improper storage of welding materials leads to insufficient weld strength" and sequential relationship pairs such as "inadequate training followed by non-standard operation" from the text.

[0024] For ambiguous descriptive text (such as "improper storage of welding materials"), cosine similarity calculation is used for normalization, and the formula is: ,in , Let and represent the i-th component of the two text vectors respectively. By calculating the cosine similarity between the fuzzy event description text to be processed and the standard event type text, the fuzzy description with a similarity higher than a preset threshold (the threshold is set to 0.7 in this embodiment) is normalized to the corresponding standard event type.

[0025] The strength of the association between events is determined by calculating the transition probability. The formula for calculating the transition probability is: ,in This represents the transition probability of event B occurring after event A has occurred. This indicates the number of times events A and B occur consecutively. This represents the total number of times event A occurs. When the transition probability is higher than a preset threshold (set to 0.5 in this embodiment), it is determined that event A and event B have a sequential relationship.

[0026] For example, extract the causal relationship between "inadequate pre-job training" and "insufficient operational skills of construction workers" from case reports, as well as the sequential relationship between "construction workers violating drawing requirements" and "management personnel's oversight in inspection".

[0027] Step 3: Input the 120 extracted events as nodes, 85 causal relationships, and 62 sequential relationships as edges into Neo4j. Node labels are set to reason_result (80 causal and result nodes) and subservient (40 auxiliary explanatory nodes); relationship types are labeled as "causal relationship" and "sequential relationship"; node attributes include event ID, description, occurrence time, and involved processes; relationship attributes include correlation strength (e.g., P=0.6), first discovery time, etc. Nodes are linked using the from, relation, and to attribute keys to form a complete knowledge graph containing node and relationship attributes, which can intuitively display the correlation chain such as "improper welding material storage → welding material getting damp → porosity formation".

[0028] P1: Evolutionary Transmission Path Analysis Execute path lookup algorithms in the Neo4j graph database, such as path lookup statements like "MATCHp=()-[r*1..5]->()RETURNp", to obtain the evolution and transmission paths of typical quality problems. For example, the transmission path of "improper storage of welding materials" is: excessive humidity in the welding material storage warehouse (A) → causal relationship → damp welding materials (B) → causal relationship → poor molten pool protection during welding (C) → causal relationship → porosity in the weld (D) → causal relationship → leakage during hydrostatic testing (E), clearly showing the complete transmission process from storage problems to the final leakage.

[0029] P2: Causation Model Construction and Control Measures Formulation Based on path analysis, a multi-factor causal model of "human-machine-material-method-environment" was constructed. The model shows that "material storage," "personnel operation," and "process execution" are the three key causal modules. For the path of "inadequate communication of management requirements," the following measures were implemented: ① Establish a digital briefing platform to achieve electronic signature; ② Conduct weekly spot checks on the effectiveness of communication, and include the records in performance evaluations; ③ Adopt a "double briefing" system (technical + quality) for key processes. For the path of "improper storage of welding materials," the following measures were implemented: ① Install an automatic temperature and humidity control system in the warehouse for real-time early warning; ② Store welding materials in designated areas, labeling the entry / expiration time; ③ Establish a drying record log for welding materials before use.

[0030] Working Principle: During use, a unified standard is established for subsequent information extraction by clearly defining the types and relationships of events related to ship welding quality issues. Natural language processing technology is used to deeply analyze unstructured case reports, transforming textual information into structured event and relationship data, solving the problem that traditional text data is difficult to use directly for in-depth analysis. Leveraging the graph structure characteristics of the Neo4j graph database, events and relationships are visualized and stored as nodes and edges, achieving the integration of quality information and breaking the limitations of information silos in traditional data storage. The graph database's path query function is used to uncover the evolution paths of quality problems. Based on path analysis, causal models are built and control measures are formulated, forming a closed-loop quality management process from data collection to decision support, enabling precise traceability and effective control of welding quality issues.

[0031] Example 2 Please see Figures 1-14 The present invention provides a technical solution: Step 1: Definition of event types and relationships Based on a comprehensive survey of factors affecting ship welding quality, the event types are defined as follows: 1. Personnel-related incidents: These include specific scenarios such as insufficient responsibility of construction personnel (e.g., failure to operate according to procedures), inadequate communication of management requirements (e.g., omission of technical briefings), construction personnel violating drawing requirements (e.g., unauthorized changes to welding sequence), and deviations in the execution of inspection standards by quality inspectors. 2. Equipment status related events: including abnormal equipment conditions such as excessive fluctuations in welding equipment current and voltage, unstable arc caused by welding torch wear, uneven wire feeding caused by wire feeding motor failure, and water leakage in the cooling system affecting equipment operation; 3. Material characteristic related events: These involve material quality issues such as excessive humidity in the welding material storage environment leading to moisture absorption, welding material model not matching the process documents, flux particle size not meeting requirements affecting slag removal performance, and shielding gas containing excessive impurities. 4. Process parameter related events: including parameter setting issues such as welding current exceeding the process range, welding voltage and current mismatch, welding speed being too fast leading to incomplete fusion, and preheating temperature not reaching the specified process value; 5. Environmental Conditions Related Events: These include environmental factors such as excessive humidity in the work area (e.g., failure to control humidity during the rainy season), changes in welding material performance due to high temperature, damage to gas protection due to excessive wind speed, and contamination of the molten pool by excessive dust concentration. 6. Quality problem outcome events: These include typical welding defects such as surface porosity of welds, incomplete root penetration, longitudinal cracks in welds, excessive undercut depth, excessive weld beads, and excessive joint deformation.

[0032] Event relationships are defined as two core types: Causal relationship: refers to the direct cause of the preceding event to the occurrence of the following event, such as "welding materials getting damp" directly leading to "porosity in the weld"; Sequential relationship: refers to the relationship in which events occur sequentially in time, such as "lack of pre-job training" followed by "insufficient operator skills".

[0033] S2: Collected over 500 case reports on ship welding quality issues from the past three years, covering quality problems in different welded parts such as hull structure, compartments, and pipelines. Utilized the LTP (Language Technology Platform) dependency parsing tool to perform grammatical analysis on the report text, identifying subject-verb-object structures and modifiers, and extracting core information such as "event subject + action + result".

[0034] Relationships are extracted based on a pre-defined pattern library: causal relationship patterns include "caused by…", "caused by…", "…is the cause of…", etc.; sequential relationship patterns include "first…then…", "after…", "following…", etc. Pattern matching directly identifies causal relationship pairs such as "improper storage of welding materials leads to insufficient weld strength" and sequential relationship pairs such as "inadequate training followed by non-standard operation" from the text.

[0035] For ambiguous descriptive text (such as "improper storage of welding materials"), cosine similarity calculation is used for normalization, and the formula is: ,in , Let and represent the i-th component of the two text vectors respectively. By calculating the cosine similarity between the fuzzy event description text to be processed and the standard event type text, the fuzzy description with a similarity higher than a preset threshold (the threshold is set to 0.7 in this embodiment) is normalized to the corresponding standard event type.

[0036] The strength of the association between events is determined by calculating the transition probability. The formula for calculating the transition probability is: ,in This represents the transition probability of event B occurring after event A has occurred. This indicates the number of times events A and B occur consecutively. This represents the total number of times event A occurs. When the transition probability is higher than a preset threshold (set to 0.5 in this embodiment), it is determined that event A and event B have a sequential relationship.

[0037] For example, extract the causal relationship between "inadequate pre-job training" and "insufficient operational skills of construction workers" from case reports, as well as the sequential relationship between "construction workers violating drawing requirements" and "management personnel's oversight in inspection".

[0038] Step 3: Input the 120 extracted events as nodes, 85 causal relationships, and 62 sequential relationships as edges into Neo4j. Node labels are set to reason_result (80 causal and result nodes) and subservient (40 auxiliary explanatory nodes); relationship types are labeled "causal relationship" and "sequential relationship"; node attributes include event ID, description, occurrence time, and involved processes; relationship attributes include correlation strength (e.g., P=0.6), first discovery time, etc. Nodes are linked using the from, relation, and to attribute keys to form a complete knowledge graph containing node and relationship attributes, which can intuitively display the correlation chain such as "improper welding material storage → welding material getting damp → porosity formation". P1: Evolutionary Transmission Path Analysis Adjust the transition probability threshold to 0.4 to identify more weak sequential relationships. Execute the query "MATCHp=(n:reason_result)-[r*1..6]->(m:reason_result)RETURNp" to obtain paths of length 1-6. For example, the path for "construction personnel lacking responsibility" is: construction personnel did not preheat according to regulations (A) → causal relationship → poor weld fusion (B) → causal relationship → non-destructive testing failure (C) → sequential relationship → rework arrangement (D) → sequential relationship → project delay (E) → sequential relationship → ship delivery delay (F), fully presenting the impact chain of quality problems on delivery.

[0039] P2: Causation Model Construction and Control Measures Formulation An extended causation model incorporating the "quality-schedule-cost" relationship was constructed. The model showed that quality issues impacted production plans through rework. To address the "insufficient responsibility among construction workers" pathway, the following measures were implemented: ① Conducting specialized training on "quality responsibility and cost" to strengthen awareness of responsibility; ② Setting quality control points in key processes and increasing the frequency of process inspections; ③ Establishing a rework cost accounting mechanism, linking costs to team performance; ④ Optimizing production plans and reserving a reasonable rework buffer period. These multi-dimensional measures effectively prevented the transmission of quality issues to the schedule and cost domains.

[0040] Working Principle: In use, by lowering the transition probability threshold, the scope of sequential relationship identification is expanded, capturing more potentially weakly correlated event pairs and compensating for indirect correlations that might have been missed in Example 1. It increases the depth of path querying, tracking the extended impact of quality issues on production schedules, costs, etc., breaking through the limitations of traditional quality analysis that only focuses on the defect itself. Based on more comprehensive path information, an extended causation model is constructed, incorporating the direct impact and indirect consequences of quality issues into the analysis. This ensures that control measures not only target the quality defect itself but also cover all aspects of risk transmission, achieving an upgrade from defect control to full-process risk prevention, and enhancing the foresight and comprehensiveness of quality management.

[0041] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

Claims

1. A method for constructing a knowledge graph of ship welding quality issues, characterized in that, Includes the following steps: S1: Define the types of events and relationships related to ship welding quality issues; S2: Based on dependency parsing, pattern matching, similarity calculation, and transition probability calculation methods, extract events and relationships between events from case reports of ship welding quality problems; S3: Using Neo4j graph database software, construct a knowledge graph of ship welding quality issues based on the events and relationships extracted in S2.

2. The method for constructing a knowledge graph of ship welding quality issues according to claim 1, characterized in that: The event types described in S1 include events related to personnel operation, events related to equipment status, events related to material properties, events related to process parameters, events related to environmental conditions, and events resulting from quality problems.

3. The method for constructing a knowledge graph of ship welding quality issues according to claim 1, characterized in that: The event relationships described in S1 include causal relationships and sequential relationships.

4. The method for constructing a knowledge graph of ship welding quality issues according to claim 1, characterized in that: The process of processing case reports in S2 is as follows: dependency parsing is used to parse the grammatical structure of the text, pattern matching is used to identify text fragments of preset event types, similarity calculation is used to normalize fuzzy event descriptions, and transition probability calculation is used to determine the strength of association between events.

5. The method for constructing a knowledge graph of ship welding quality issues according to claim 1, characterized in that: The knowledge graph constructed in S3 contains nodes and relationships. Node labels include reason_result and subservient. The relationship type corresponds to the event relationship defined in S1. The attribute information of nodes and relationships includes the attribute keys from, relation, and to.

6. The method for constructing a knowledge graph of ship welding quality issues according to claim 2, characterized in that: The personnel-related incidents mentioned include insufficient sense of responsibility among construction workers, inadequate communication of management requirements, and violations of drawing requirements by construction workers.

7. The method for constructing a knowledge graph of ship welding quality issues according to claim 3, characterized in that: The causal relationship is a relationship in which one event directly leads to the occurrence of another event, and the sequential relationship is a relationship in which one event occurs and another event occurs in sequence.

8. A method for analyzing a knowledge graph of ship welding quality problems, characterized in that, Includes the following steps: P1: Based on the knowledge graph of ship welding quality problems constructed by the method described in claim 1, calculate the evolution and transmission path of ship welding quality problems; P2: Based on the analysis results of the evolution and transmission path, a causal model of ship welding quality problems is formed, and targeted welding quality control measures are formulated.

9. The method for analyzing a knowledge graph of ship welding quality problems according to claim 8, characterized in that: In P1, a path query algorithm is executed in the Neo4j graph database to obtain the complete evolutionary transmission path from cause to effect for different quality problems.

10. The method for analyzing a knowledge graph of ship welding quality problems according to claim 8, characterized in that: The P2 causal model is constructed based on the relationship and evolutionary transmission path of events in the knowledge graph, and is used to characterize the influence logic of human, machine, material, method and environment factors on welding quality problems.

Images