Pressurized water reactor system fault tree building method based on large language model

Abstract

The invention discloses a pressurized water reactor system fault tree building method based on a large language model, and relates to the field of fault tree analysis. In order to solve the defects that in the prior art, a fault tree construction process depends on manpower, efficiency is low and logic consistency is insufficient, the technical scheme provided by the invention comprises the following steps: outputting structured module failure reason items according to a system success criterion, a system hierarchical structure, an equipment list, a connection relationship and an equipment failure mode of a pressurized water reactor system; on the basis of the module failure reason items, the overall failure of the system is set as a top event, and a subsystem failure event is output; on the basis of the subsystem failure event, continuing to expand and map to a module failure intermediate event causing subsystem failure, and outputting a module failure event; based on module failure events, tracing to specific equipment and failure forms thereof one by one, generating equipment failure bottom events, and constructing and outputting a preliminary fault tree; and generating a compact final fault tree.

Description

Technical Field

[0001] This involves the field of fault tree analysis, specifically the construction of fault trees for pressurized water reactor systems based on large language models. Background Technology

[0002] Fault tree analysis, a typical system reliability analysis method, has been widely used in high-safety industries such as nuclear energy, aerospace, chemical engineering, and aviation since its introduction in the 1960s. Its basic idea is to decompose top-level failure events of a system into lower-level events at the subsystem, module, and equipment levels through logical reasoning, thereby revealing the mechanism and path of system failure. In the nuclear power field, especially in pressurized water reactor systems, fault tree analysis is one of the important foundational tools of probabilistic safety analysis (PSA), providing decision-making support for system design, operation and maintenance, and risk assessment.

[0003] Existing fault tree construction methods primarily rely on manual methods. Researchers or engineering experts typically deduce the fault tree step by step based on their understanding of the system structure and failure mechanisms, combined with experience. While this method is intuitive, it heavily depends on expert knowledge, resulting in a time-consuming and inefficient tree-building process, and the results are easily influenced by subjective factors. For example, in the analysis of pressurized water reactor cooling systems, engineers need to manually identify the failure modes of each device and their impact on system functionality. When the system is large and the failure paths are complex, omissions or logical inconsistencies are highly likely to occur.

[0004] With the development of artificial intelligence, scholars have attempted to use knowledge graphs, machine learning, and other methods to assist in fault tree construction. For example, some studies have proposed automated tree construction methods based on knowledge bases, which semi-automatically generate fault tree structures by pre-establishing a failure mode database; other literature explores the use of natural language processing methods based on deep learning to extract failure information from equipment documentation and map it into the fault tree model. However, these methods still have certain limitations when dealing with highly complex engineering objects such as pressurized water reactor systems with multiple coupling effects. On the one hand, traditional knowledge graph methods struggle to cover all potential equipment interaction relationships, easily overlooking collaborative failure paths; on the other hand, while deep learning methods can automatically extract some failure information, their capabilities in logical reasoning and hierarchical modeling are insufficient, leading to redundant branches or fault trees that do not conform to engineering realities.

[0005] In summary, existing technologies suffer from drawbacks such as reliance on manual fault tree construction, low efficiency, insufficient logical consistency, and difficulty in addressing multi-level failure paths and synergistic effects of multiple faults in complex systems such as pressurized water reactors. Summary of the Invention

[0006] To address the shortcomings of existing technologies, such as reliance on manual fault tree construction, low efficiency, insufficient logical consistency, and difficulty in handling multi-level failure paths and multiple fault synergistic effects in complex systems like pressurized water reactors, the technical solution provided by this invention is as follows: A fault tree construction method for pressurized water reactor systems based on a large language model includes: The system success criteria, system hierarchy, equipment list, connection relationships and equipment failure modes of the pressurized water reactor system are taken as input information. The large language model parses and infers the causes of module failures, and outputs structured module failure cause entries as subsequent input steps. Based on the module failure cause entries, the overall system failure is set as the top event and mapped to the subsystem failure intermediate event that caused the system failure. The subsystem failure event is then output as the subsequent input. Based on the subsystem failure event, continue to expand the mapping to the intermediate module failure event that caused the subsystem failure, and output the module failure event as the subsequent input. Based on module failure events, trace each failure event back to the specific device and its failure mode, generate device failure base events, construct a preliminary fault tree and output the steps as subsequent inputs; The steps involve performing a cleaning operation based on the initial fault tree, automatically eliminating duplicate events, equivalent branches, and redundant paths to generate a compact final fault tree.

[0007] Furthermore, a preferred implementation is provided, wherein the system information includes system success criteria, system hierarchy, device list, connection relationships, and device failure modes.

[0008] Furthermore, a preferred implementation is provided, where the top event is a system-wide failure event and is mapped to a subsystem failure intermediate event. If a specific failure cause is found, an entry is directly generated; otherwise, the subsystem is uniformly marked as failed.

[0009] Furthermore, a preferred implementation is provided: when a subsystem failure event is expanded down to a module failure intermediate event, if a unique failure cause is found, an entry is directly generated; if no cause is found, the module is uniformly marked as failed.

[0010] Furthermore, a preferred implementation is provided: when tracing module failure events one by one to the specific device failure event, the device failure and its failure mode are generated by combining the module failure cause entries output by the large language model.

[0011] Furthermore, a preferred implementation is provided, in which the final fault tree has a compact structure and clear logic, making it suitable for reliability analysis and probabilistic safety assessment of pressurized water reactor systems.

[0012] A fault tree construction device for pressurized water reactor systems based on a large language model is also provided, including: The system success criteria, system hierarchy, equipment list, connection relationships and equipment failure modes of the pressurized water reactor system are taken as input information. The large language model parses and infers the module failure causes and outputs structured module failure cause entries as subsequent module inputs. Based on the module failure cause entries, the overall system failure is set as the top event and mapped to the intermediate events of the subsystem failure that caused the system failure. The subsystem failure events are then output as the modules for subsequent input. Based on the subsystem failure event, the mapping continues to the intermediate failure event of the module that caused the subsystem failure, and the output module failure event is used as the module for subsequent input. Based on module failure events, each event is traced back to the specific device and its failure mode to generate device failure base events, construct a preliminary fault tree, and output the module as subsequent input. Based on the initial fault tree, a cleaning operation is performed to automatically eliminate duplicate events, equivalent branches, and redundant paths, generating a compact final fault tree module.

[0013] A computer storage medium is also provided for storing a computer program, which, when read by the computer, executes the method.

[0014] A computer is also provided, including a processor and a storage medium, wherein the computer executes the method when the processor reads a computer program stored in the storage medium.

[0015] A computer program product is also provided, which, when executed, implements the method described.

[0016] Compared with the prior art, the advantages of the technical solution provided by the present invention are as follows: This solution automatically parses the hierarchical structure, equipment connections, and failure modes of a pressurized water reactor system using a large language model, generating structured entries for module failure causes, significantly reducing reliance on expert experience. Compared to traditional tree-building methods that rely on manual analysis, this approach can cover the entire system in a short time, improving the efficiency and consistency of fault tree construction and avoiding omissions or ambiguities that are prone to occur in manual reasoning.

[0017] This solution employs a top-down hierarchical mapping mechanism, progressively decomposing the overall system failure into failure events at the subsystem, module, and device levels, ensuring a clear hierarchical structure of the fault tree. Existing knowledge base-assisted methods often only achieve local failure mode mapping and lack complete hierarchical expansion capabilities, while this mechanism can fully cover the complex multi-level failure paths of pressurized water reactor systems.

[0018] This approach leverages the logical reasoning and multi-fault collaborative analysis capabilities of large language models to automatically identify associated faults between devices and generate precise propagation paths. In traditional research, knowledge graph methods struggle to capture dynamic multi-fault coupling effects, while deep learning methods often remain at the level of single-event extraction. This approach, however, better reveals complex interactions and enhances the ability to characterize the true risks of a system.

[0019] This proposal suggests a fault tree cleaning method based on a combination of rules and models. It automatically scans and optimizes the initially generated fault tree, removing duplicate events, equivalent branches, and redundant paths, ultimately yielding a compact result that meets engineering analysis requirements. Compared to fault trees directly generated by existing deep learning methods, the cleaning process in this proposal makes the results more practical in terms of logical correctness and structural simplicity.

[0020] This solution establishes a scalable and standardized tree-building process that can be applied to complex nuclear energy scenarios such as pressurized water reactor systems. Existing automation research typically focuses on specific subsystems or conducts experimental verification on a limited scale, lacking universality and generalizability. In contrast, the standardized process of this solution provides high-quality input for subsequent reliability analysis and probabilistic safety assessment, offering advantages for engineering implementation.

[0021] It is applicable to the automated construction of fault trees and reliability analysis of pressurized water reactor systems in the nuclear energy field. Attached Figure Description

[0022] Figure 1 A system framework diagram representing a fault tree construction method for pressurized water reactor systems based on a large language model; Figure 2 This represents the PCS system flowchart. Figure 3 Partial results indicating the cause of the inference module failure; Figure 4 This represents the result of constructing the fault tree. Detailed Implementation

[0023] To make the advantages and benefits of the technical solution provided by the present invention clearer, the technical solution provided by the present invention will now be described in further detail with reference to the accompanying drawings, specifically: Implementation Method 1: This implementation method provides a fault tree construction method for pressurized water reactor systems based on a large language model, including: The system success criteria, system hierarchy, equipment list, connection relationships and equipment failure modes of the pressurized water reactor system are taken as input information. The large language model parses and infers the causes of module failures, and outputs structured module failure cause entries as subsequent input steps. Based on the module failure cause entries, the overall system failure is set as the top event and mapped to the subsystem failure intermediate event that caused the system failure. The subsystem failure event is then output as the subsequent input. Based on the subsystem failure event, continue to expand the mapping to the intermediate module failure event that caused the subsystem failure, and output the module failure event as the subsequent input. Based on module failure events, trace each failure event back to the specific device and its failure mode, generate device failure base events, construct a preliminary fault tree and output the steps as subsequent inputs; The steps involve performing a cleaning operation based on the initial fault tree, automatically eliminating duplicate events, equivalent branches, and redundant paths to generate a compact final fault tree.

[0024] System information includes system success criteria, system hierarchy, equipment list, connectivity relationships, and equipment failure modes.

[0025] When the top event is a system-wide failure event and is mapped to a subsystem failure intermediate event, if a specific failure cause is found, an entry is directly generated; otherwise, the subsystem is uniformly marked as failed.

[0026] When a subsystem failure event is expanded down to a module failure intermediate event, if a unique failure cause is found, an entry is generated directly; otherwise, the module is uniformly marked as failed.

[0027] When tracing module failure events one by one to the specific device failure event, the device failure and its failure mode are generated by combining the module failure cause entries output by the large language model.

[0028] The final fault tree has a compact structure and clear logic, making it suitable for reliability analysis and probabilistic safety assessment of pressurized water reactor systems.

[0029] Implementation Method Two: This implementation method is a further detailed description of the technical solution provided in Implementation Method One, specifically: By combining large language models with rules, a complete process from system information input to final fault tree generation is achieved, specifically including the following steps: Step 1: System Information Input and Failure Cause Reasoning First, the basic information of the pressurized water reactor system is input into the large language model. This information includes system success criteria, system hierarchy, equipment list, connections between equipment, and possible failure modes. The large language model performs semantic parsing and logical reasoning on the input information to determine the causes of failure in each module and establishes collaborative propagation paths for multiple faults based on the interactions between equipment. The output is a structured list of module failure cause entries, which provides the foundation for subsequent hierarchical fault tree mapping.

[0030] Step 2: System top event setting and subsystem failure mapping After receiving the module failure cause entries obtained in Step 1, the overall system failure event is set as the top event. Using the inference results of the large language model, the top event is mapped level by level to intermediate events in subsystems that may lead to overall system failure. If a system-specific failure cause is found during the retrieval process, an entry is directly generated; otherwise, all subsystems associated with this system are collected, and the word "failure" is appended to their names as subsystem-level failure causes. The output of this step is a set of failure events at the subsystem level.

[0031] Step 3: Subsystem failure decomposition to module failure event Using the subsystem failure events from the previous step as input, the process continues to map them to the corresponding module failure intermediate events. During this process, if a specific failure cause for the module can be retrieved from the database or model inference results, it is directly extracted and an entry is generated; otherwise, each module under the subsystem is marked as "failed," forming a subsystem-level set of failure events. The output of this step is the module failure event, laying the foundation for further tracing down to the underlying events.

[0032] Step 4: Tracing module failure back to the device failure event. Using the aforementioned module failure events as input, the process continues to trace back to the device level, mapping specific device failure events. During this process, combined with the module failure cause entries generated in step one, the module failures are mapped level by level to specific devices and their possible failure modes until they cannot be further divided, forming the base events of the fault tree. The output of this step is a preliminary fault tree, which contains the logical relationships between the system, subsystem, module, and device levels.

[0033] Step 5: Preliminary Fault Tree Cleaning and Optimization The preliminary fault tree generated in step four is used as input to perform a cleaning operation. During cleaning, duplicate events, equivalent branches, and redundant paths in the fault tree are automatically scanned. Unnecessary information is removed through merging, deletion, or pruning, thus ensuring the clarity and compactness of the fault tree's logical structure. The final output is an optimized complete fault tree with a reasonable structure and consistent logic, which can be directly applied to reliability analysis and probabilistic safety assessment.

[0034] Through the above steps, this implementation method realizes a complete link from system information input to final fault tree generation, and the output of each step serves as the input for the next step, ensuring the coherence and logic of the method. Compared with existing technologies, this scheme not only improves the automation and logical rigor of fault tree construction, but also effectively addresses the problem of multiple fault synergistic effects in pressurized water reactor systems, ensuring that the fault tree has higher reliability and application value in practical engineering analysis.

[0035] Implementation method three, combined with Figure 1-4 This embodiment further describes the technical solution provided above in detail through specific examples, specifically: The technical problem to be solved by this embodiment is to provide a fault tree construction method for pressurized water reactor systems based on a large language model. By utilizing the natural language processing and logical reasoning capabilities of the large language model, the automatic construction of fault tree structure from system information can be achieved, solving the problems of low efficiency, poor consistency, and difficulty in covering the multi-level associations of complex systems in traditional manual tree construction methods.

[0036] Compared with existing technologies, the fault tree construction method for pressurized water reactor systems based on large language models in this embodiment has the following advantages: (1) This implementation method automatically parses the system hierarchy, device connection relationship and failure mode through a large language model, and generates structured entries for the reasons for module failure, which significantly improves the automation level of fault tree construction and reduces the reliance on expert experience; (2) A top-down hierarchical mapping mechanism is adopted to decompose system failures into subsystem, module and equipment failure events step by step, so as to ensure that the fault tree logic hierarchy is clear and fully covers the multi-level failure paths of the pressurized water reactor system. (3) By leveraging the multi-fault synergy analysis capabilities of the large language model, the correlation between equipment faults can be automatically identified, generating accurate propagation paths containing complex interactions, thus avoiding the common logical omission problem in traditional methods. (4) A fault tree cleaning method based on the combination of rules and models is proposed to automatically eliminate duplicate events, redundant branches and equivalent paths, ensuring that the final fault tree structure is compact and meets the requirements of engineering analysis. (5) This implementation provides a scalable standardized tree building process that is applicable to complex industrial scenarios such as nuclear energy systems, providing high-quality input for subsequent reliability analysis and probabilistic safety assessment (PSA).

[0037] like Figure 1 As shown, it includes the following steps: Step 1: Input system information into the large language model to determine the cause of the inference module failure; The system information includes system success criteria, system hierarchy, equipment list, connection relationships, and equipment failure modes.

[0038] The failure cause of the inference module is determined by analyzing the impact of equipment failure on module function, evaluating the synergistic effect of multiple failures, and establishing a precise propagation path.

[0039] The output is a structured list of module failure cause entries converted from module failure causes.

[0040] Step 2: Select the overall system failure as the top event and map it to the intermediate events of the subsystem failure that caused the system failure; The intermediate events of the mapping subsystem failure are processed by searching for specific failure reasons for each system. If a specific failure reason exists, an entry is generated; otherwise, all associated subsystems are collected and each is appended with the word "failure" as a system-level failure reason.

[0041] Step 3: Continue to expand downwards, mapping to the intermediate events of module failure that caused the subsystem to fail; The intermediate events of the mapping module failure are handled by searching for specific failure reasons for each subsystem. If a specific failure reason exists, an entry is generated; otherwise, all associated modules are collected and each module is appended with the word "failure" as a subsystem-level failure reason.

[0042] Step 4: Continue to expand downwards, mapping to the device failure underlying event that caused the module failure, and build a preliminary fault tree; In step 4, the mapping device failure base events are traced back to specific devices and their failure modes one by one according to the failure causes generated in step 1, forming indivisible base events, thereby constructing a preliminary fault tree.

[0043] Step 5: Clean the initial fault tree.

[0044] The cleaning process involves automatically scanning the initial fault tree for duplicate events, equivalent branches, and redundant paths, and performing merging or pruning operations to ensure that the final fault tree structure is compact and logically clear.

[0045] In a specific embodiment: Example 1 Taking the passive containment cooling system (PCS) as an example, the fault tree construction method for pressurized water reactor systems based on large language models includes the following steps: Step 1: Include Figure 2 The PCS system information, including the PCS system flowchart, is input into the large language model. The reasons for the inference module failure are analyzed, and some results are shown below. Figure 3 As shown; Step 2: Select "failure of the first column of the passive containment cooling system" as the top event and map it to the intermediate event of the subsystem failure that caused the system failure; Step 3: Continue to expand downwards, mapping to the intermediate events of module failure that caused the subsystem to fail; Step 4: Continue to expand downwards, mapping to the device failure underlying event that caused the module failure, and build a preliminary fault tree; Step 5: Clean the initial fault tree; the results are as follows. Figure 4 As shown.

[0046] The above description of several specific embodiments further details the technical solution provided by the present invention in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the above-described specific embodiments are not intended to limit the present invention. Any reasonable modifications and improvements to the present invention, combinations of embodiments, and equivalent substitutions based on the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fault tree construction method for pressurized water reactor systems based on a large language model, characterized in that, include: The system success criteria, system hierarchy, equipment list, connection relationships and equipment failure modes of the pressurized water reactor system are taken as input information. The large language model parses and infers the causes of module failures, and outputs structured module failure cause entries as subsequent input steps. Based on the module failure cause entries, the overall system failure is set as the top event and mapped to the subsystem failure intermediate event that caused the system failure. The subsystem failure event is then output as the subsequent input. Based on the subsystem failure event, continue to expand the mapping to the intermediate module failure event that caused the subsystem failure, and output the module failure event as the subsequent input. Based on module failure events, trace each failure event back to the specific device and its failure mode, generate device failure base events, construct a preliminary fault tree and output the steps as subsequent inputs; The steps involve performing a cleaning operation based on the initial fault tree, automatically eliminating duplicate events, equivalent branches, and redundant paths to generate a compact final fault tree.

2. The method for constructing a fault tree for a pressurized water reactor system based on a large language model according to claim 1, characterized in that, System information includes system success criteria, system hierarchy, equipment list, connectivity relationships, and equipment failure modes.

3. The method for constructing a fault tree for a pressurized water reactor system based on a large language model according to claim 1, characterized in that, When the top event is a system-wide failure event and is mapped to a subsystem failure intermediate event, if a specific failure cause is found, an entry is directly generated; otherwise, the subsystem is uniformly marked as failed.

4. The method for constructing a fault tree for a pressurized water reactor system based on a large language model according to claim 1, characterized in that, When a subsystem failure event is expanded down to a module failure intermediate event, if a unique failure cause is found, an entry is generated directly; otherwise, the module is uniformly marked as failed.

5. The method for constructing a fault tree for a pressurized water reactor system based on a large language model according to claim 1, characterized in that, When tracing module failure events one by one to the specific device failure event, the device failure and its failure mode are generated by combining the module failure cause entries output by the large language model.

6. The method for constructing a fault tree for a pressurized water reactor system based on a large language model according to claim 1, characterized in that, The final fault tree has a compact structure and clear logic, making it suitable for reliability analysis and probabilistic safety assessment of pressurized water reactor systems.

7. A fault tree construction device for a pressurized water reactor system based on a large language model, characterized in that, include: The system success criteria, system hierarchy, equipment list, connection relationships and equipment failure modes of the pressurized water reactor system are taken as input information. The large language model parses and infers the module failure causes and outputs structured module failure cause entries as subsequent module inputs. Based on the module failure cause entries, the overall system failure is set as the top event and mapped to the intermediate events of the subsystem failure that caused the system failure. The subsystem failure events are then output as the modules for subsequent input. Based on the subsystem failure event, the mapping continues to the intermediate failure event of the module that caused the subsystem failure, and the output module failure event is used as the module for subsequent input. Based on module failure events, each event is traced back to the specific device and its failure mode to generate device failure base events, construct a preliminary fault tree, and output the module as subsequent input. Based on the initial fault tree, a cleaning operation is performed to automatically eliminate duplicate events, equivalent branches, and redundant paths, generating a compact final fault tree module.

8. A computer storage medium for storing computer programs, characterized in that, When the computer program is read by the computer, the computer executes the method of claim 1.

9. A computer, comprising a processor and a storage medium, characterized in that, When the processor reads the computer program stored in the storage medium, the computer executes the method of claim 1.

10. A computer program product, as a computer program, is characterized by: When the computer program is executed, it implements the method of claim 1.

Images