Radar system troubleshooting process generation method and system based on large model

Abstract

The invention relates to the field of fault detection, in particular to a radar system troubleshooting process generation method and system based on a large model. The method comprises the following steps: S1, picture extraction; s2, fault tree picture recognition and structured conversion; s3, troubleshooting guide picture recognition and structured conversion; s4, obtaining a mapping relation between the fault tree and the troubleshooting process; s5, constructing a fault knowledge database; and S6, performing fault retrieval and troubleshooting guidance. The system comprises a fault knowledge database, a picture extraction module, a fault tree picture extraction module, a troubleshooting guide picture extraction module, a mapping module and a fault retrieval and troubleshooting guide module. According to the method, the fault knowledge database of the radar system is constructed, so that the matched troubleshooting process can be quickly and accurately positioned; a natural language processing and semantic understanding technology is realized by means of a large model; and the next troubleshooting measure is automatically selected according to the detection result in the troubleshooting process, so that intelligent promotion and optimization of the troubleshooting process are realized.

Description

Technical Field

[0001] This invention relates to the field of fault detection, and specifically to a method and system for generating troubleshooting procedures for radar systems based on large models. Background Technology

[0002] As a crucial component of modern combat equipment, radar systems require robust operation and troubleshooting capabilities. These capabilities include ensuring continuous and stable operation, minimizing downtime, and reducing maintenance costs. However, current radar system troubleshooting primarily relies on technicians consulting paper or electronic fault trees and specialized troubleshooting guides, combined with their accumulated experience, to diagnose and resolve faults. This traditional approach has significant limitations in practice: radar systems have complex internal circuitry, dense signal links, and numerous interconnected fault factors. Technicians must manually sift through massive amounts of fault data and guide documents. Especially when dealing with cross-module faults, repeatedly comparing fault tree nodes with troubleshooting steps in the processing unit not only leads to low efficiency but also risks misjudgment due to missed critical fault link information during manual retrieval, thus prolonging the fault handling cycle.

[0003] To address the aforementioned issues, the industry has attempted to construct a digital fault database for radar systems and combine it with simple search functions to guide troubleshooting. However, many key problems remain unresolved in practical applications. For example, fault data and troubleshooting guidelines are stored as separate documents, lacking a unified mapping mechanism; the correspondence between specific fault nodes in the fault tree and the corresponding troubleshooting steps in the troubleshooting guidelines is scattered and unclear; the correspondence between faults described in natural language and fault tree nodes and troubleshooting guidelines cannot be accurately identified; and radar system troubleshooting requires a series of steps, including circuit testing, signal acquisition, and module linkage testing, but existing systems cannot adjust subsequent guidance paths in real time based on intermediate results during the troubleshooting process, requiring technicians to manually backtrack the fault tree and re-query the corresponding troubleshooting steps.

[0004] Therefore, a new radar system troubleshooting process generation technology is needed to solve problems such as unclear correlation between faults and troubleshooting steps, low accuracy of natural language retrieval, and poor dynamic adaptability of troubleshooting paths in the current radar system troubleshooting guidance process, so as to improve the efficiency and accuracy of radar system troubleshooting. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention provides a method and system for generating troubleshooting procedures for radar systems based on a large model. By clarifying the correspondence between fault tree nodes and troubleshooting steps in the radar system, a rapid location and matching troubleshooting procedure is achieved. The large model enables dynamic adaptation during the retrieval and troubleshooting guidance process by describing fault phenomena in natural language.

[0006] Specifically, on the one hand, the present invention provides a method for generating a troubleshooting process for a radar system based on a large model, which includes the following steps: S1, Image Extraction: Obtain the fault manual file of the radar system, convert it into processable fault manual data, identify the fault modes and corresponding fault tree images and troubleshooting guide images according to the document structure, and save the identified images. S2, Fault Tree Image Recognition and Structure Conversion; S21, Use a multimodal large model to extract text and logical relationships from fault tree image files; S22, Generate a structured text description of the fault tree; S3, troubleshooting guide image recognition and structure conversion; S31, Use a multimodal large model to extract the text and logical relationships in the troubleshooting guide image file; S32, Generate a descriptive troubleshooting logic flow; S33, Generate step-by-step troubleshooting text; S4, obtain the mapping relationship between the fault tree and the troubleshooting process; Align the fault causes in the structured text description of the fault tree in step S22 with the step-by-step troubleshooting text generated in step S3. Using a large model matching algorithm, locate the specific troubleshooting step in the step-by-step troubleshooting text for each fault cause in the fault tree, and establish an accurate "fault cause - step-by-step troubleshooting text" mapping relationship. S5, build a fault knowledge database; The fault tree structured text description, step-by-step troubleshooting text, and the mapping relationship between the fault tree and the troubleshooting process obtained from steps S2, S3, and S4 are stored according to a predefined database pattern to obtain a fault knowledge database. S6, Fault retrieval and troubleshooting guidance; S61, Fault Mode Recognition: Identify fault phenomenon descriptions or fault modes from user questions to obtain standardized user descriptions of fault modes. S62, Match database-related fault modes: Match with related fault modes stored in the database; S63, Human-computer interactive troubleshooting guidance based on a large model: Enables dynamic guidance during the troubleshooting process based on the fault mode and on-site feedback.

[0007] Preferably, S1, image extraction, the specific steps are as follows: S11, Read the fault manual; The acquired radar system-related fault manuals are converted into processable fault manual data. The first-level chapter title of the radar system-related fault manual is the equipment name, the second-level chapter title is the fault mode, the third-level chapter title is the fault tree, and the text includes fault tree images. The third-level chapter title is the troubleshooting guide, and the text includes troubleshooting guide images. S12, identify fault modes; Based on the processable fault manual data, and considering the document structure, fault modes are automatically identified from the second-level chapter headings. S13, Extract image; Images extracted from the text of a third-level chapter titled "Fault Tree" under a second-level chapter title are considered to be fault tree images for the corresponding fault mode; images extracted from the text of a third-level chapter titled "Troubleshooting Guide" under a second-level chapter title are considered to be troubleshooting guide images for the corresponding fault mode. S14: Image naming and saving; For the extracted images, the chapter title is added as a prefix to the image file name to distinguish different sources and ensure that each image file name is unique.

[0008] Preferably, in step S22, generating a fault tree structured text description involves the following steps: Based on the text content and logical relationships of the fault tree extracted by S21, a structured description of the fault tree is generated by a large language model to store the logical relationships between the names of the fault tree nodes. Each fault tree node is determined as a fault mode or fault cause. The fault tree node located in the middle layer of the fault tree is the fault cause of the fault tree node in the upper layer and also the fault mode of the fault tree node in the lower layer.

[0009] Preferably, the fault knowledge database in S5 includes three tables: a fault mode table, a troubleshooting guide table, and a mapping table; The fault mode table contains fields for: fault mode ID, fault mode, and fault cause; the troubleshooting guide table contains fields for: step-by-step troubleshooting text ID and step-by-step troubleshooting text; the mapping table contains fields for: fault mode ID, troubleshooting operation step mapping relationship, and step-by-step troubleshooting text ID; an inverted index and a semantic vector index are built for the fault knowledge database.

[0010] Preferably, step S61, fault mode identification, involves identifying fault phenomenon descriptions or fault modes from user queries to obtain standardized user descriptions of fault modes; this includes the following steps: S611 directly obtains fault inquiries in text form input by users in natural language through the interactive interface, and saves the fault inquiries as user problem data to be processed; S612: The received user question data to be processed is preprocessed to obtain standardized user question text. Preprocessing removes irrelevant symbols, redundant expressions, and ambiguous words through a cleaning operation, thereby restating the user questions. S613, based on standardized user problem text, uses named entity recognition and keyword extraction algorithms to locate fault phenomenon descriptions and candidate fault mode keywords, thereby extracting key information; S614 compares the extracted key information with a preset fault terminology dictionary, converts the key information into standardized fault terms in the fault terminology dictionary, and obtains a standardized fault mode user description.

[0011] Preferably, step S62, matching database-related fault modes, involves matching with relevant fault modes stored in the database; this includes the following steps: S621 uses standardized fault mode user representations to query the fault knowledge database in S5 using a keyword matching algorithm to obtain a preliminary set of matching fault mode candidates. S622, For the initially matched fault mode candidate set, the semantic vector of the standardized fault mode user expression is used to calculate the cosine similarity with the semantic vector of each fault mode in the fault mode candidate set to obtain a fault mode candidate set with similarity score. S623, sort the fault modes in the fault mode candidate set in descending order according to the similarity score. If the fault mode with the highest score is higher than the preset threshold, it is taken as the final matching result. If the highest score is lower than the preset threshold, it is considered that no matching result was found, and the user is informed that there is no relevant knowledge in the database. S624, obtain the fault mode structured data; For the fault modes retrieved in S623, the structured data of the fault modes is obtained from the database. The structured data of the fault modes includes at least the fault cause, the troubleshooting operation steps mapping steps, and the troubleshooting guide text.

[0012] Preferably, step S63, based on a large model, provides interactive troubleshooting guidance: enabling dynamic guidance during the troubleshooting process based on fault modes and on-site feedback; including the following steps: S631, based on fault mode structured data, calls the large model to generate initial troubleshooting guidance steps, and returns information such as the operation or inspection items in the troubleshooting guidance steps to the user; S632 receives feedback information obtained after on-site operation or inspection is performed according to the troubleshooting guidance steps; S633: The feedback information is correlated and matched with the structured data of the fault mode in the current step. The large model determines whether to perform further troubleshooting operations, supplement feedback information, or end the troubleshooting operation based on the feedback information. If it is determined that further troubleshooting operations should be performed, S634 is executed; if supplementary feedback information is required, return to S632 and give a prompt message; if the troubleshooting operation is ended, S635 is executed. S634, call the large model to generate targeted troubleshooting steps: update the generated troubleshooting guidance text to the visualization interface, push it to the user terminal in real time, and return to S632; S635 records all troubleshooting steps and feedback information in the log for future optimization of the troubleshooting guide.

[0013] This application also discloses a troubleshooting process generation system for radar systems based on a large model, which includes: a fault knowledge database, an image extraction module, a fault tree image extraction module, a troubleshooting guide image extraction module, a mapping module, and a fault retrieval and troubleshooting guidance module. The fault knowledge database includes a fault mode table, a troubleshooting guide table, and a mapping table; the fault knowledge database establishes an inverted index and a semantic vector index. The image extraction module converts the fault manual into processable fault manual data, and retrieves and saves fault tree images and corresponding troubleshooting guide images based on chapter titles. The fault tree image extraction module uses a multimodal large model to extract text and logical relationships from fault tree image files to generate a structured text description of the fault tree; The troubleshooting guide image extraction module uses a multimodal large model to extract text and logical relationships from the troubleshooting guide image files to generate step-by-step troubleshooting text; The mapping module aligns the fault causes in the structured text description of the fault tree with the step-by-step troubleshooting text. Through a large model matching algorithm, it locates the specific troubleshooting steps in the step-by-step troubleshooting text for each fault cause in the fault tree, establishing a precise "fault cause - step-by-step troubleshooting text" mapping relationship. The fault retrieval and troubleshooting guidance module is used to provide dynamically guided troubleshooting steps based on the fault inquiries input by the user in natural language, the troubleshooting guide, and feedback information during the troubleshooting process, provided by a large model.

[0014] Preferably, the fault retrieval and troubleshooting guidance module includes a fault mode identification submodule, a fault mode matching submodule, and a troubleshooting guidance submodule; The fault mode recognition submodule is used to convert fault inquiries input by users in natural language into standard fault modes; The fault mode matching submodule is used to query the fault knowledge database to obtain structured fault mode data based on standard fault modes. The troubleshooting guidance submodule provides corresponding troubleshooting steps based on the human-computer interaction of the large model and the structured data and feedback information of the fault modes.

[0015] Preferably, the fault mode table fields in the fault knowledge database include: fault mode id, fault mode, and fault cause; the troubleshooting guide table fields include: step-by-step troubleshooting text id and step-by-step troubleshooting text; the mapping table fields include: fault mode id, troubleshooting operation step mapping relationship, and step-by-step troubleshooting text id.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By constructing a structured association mapping database between radar system fault tree nodes and troubleshooting guide steps, the matching troubleshooting process can be quickly and accurately located, solving the problem of scattered storage of fault data and troubleshooting guides and the lack of a unified association mechanism.

[0017] 2. By leveraging large models to implement natural language processing and semantic understanding technologies, it supports users to search for fault phenomena using natural language descriptions. This breaks through the limitations of traditional keyword-based precise retrieval, solves the problem of poor adaptability of traditional keyword retrieval to natural language descriptions, and improves the recognition accuracy of fault questions.

[0018] 3. By using a large model, the troubleshooting guidance process is dynamically adapted. The next troubleshooting measure is automatically selected based on the detection results during the troubleshooting process. This effectively solves the problem of poor dynamic adaptability of the current troubleshooting guidance, realizes the intelligent advancement and optimization of the troubleshooting process, and reduces operation and maintenance costs. Attached Figure Description

[0019] Figure 1 This is a technical framework diagram of the troubleshooting process for radar systems based on a large model, as described in this invention. Figure 2 This invention relates to a method for generating troubleshooting procedures for radar systems based on large models. Figure 3 This invention relates to a troubleshooting process generation system for radar systems based on a large model. Figure 4 This is a schematic diagram of the fault tree for the inconsistency between radar data and external loading in this invention. Detailed Implementation

[0020] To fully explain the technical content, objectives, and effects of this invention, the embodiments of this invention will be described in detail using a radar system processing unit as an example, in conjunction with the accompanying drawings.

[0021] This invention aims to construct a fault knowledge database for radar systems using large-scale models. During troubleshooting, it can provide corresponding troubleshooting steps based on the given fault description and automatically provide the next troubleshooting steps by receiving feedback information, achieving interactive troubleshooting guidance. Figure 1 The technology shown.

[0022] This invention discloses a method for generating troubleshooting procedures for radar systems based on large models, such as... Figure 2 As shown, it includes the following steps: S1, Image Extraction; A fault knowledge database is constructed, comprising structured fault trees, troubleshooting guide images, and their relationships. The structured fault trees and troubleshooting guide images are extracted from fault manuals. Fault tree identification and conversion, troubleshooting guide image identification and conversion, and relationship construction are automated using multimodal information processing and natural language understanding technologies. The specific steps are as follows: S11, Read the fault manual; Obtain the relevant electronic fault manuals and convert them into processable fault manual data. The fault manual content should include at least: the system name as the first-level chapter title, the fault mode name as the second-level chapter title, the fault tree and troubleshooting guide as the third-level chapter titles. The main text of the fault tree at the third-level chapter title should include a fault tree image, and the main text of the troubleshooting guide at the third-level chapter title should include a troubleshooting guide image.

[0023] In this embodiment, a Word file of the radar system's fault manual is obtained. A Python script is used to read the fault manual Word file and convert it into processable fault manual data. The preferred format for this processable fault manual data is JSON. The first-level chapter title of a chapter in the radar system's fault manual is "Processing Unit," the second-level chapter title is "Radar Data and External Loading Inconsistency," the third-level chapter title is "Fault Tree" (including fault tree images), and the third-level chapter title is "Troubleshooting Guide" (including troubleshooting guide images).

[0024] S12, identify fault modes; Based on the available fault manual data, and considering the document structure, fault modes are automatically identified from the second-level chapter headings. S13, Extract image; Images extracted from the text of a section with the tertiary chapter title "Fault Tree" under a secondary chapter title are considered fault tree images for the corresponding fault mode; images extracted from the text of a section with the tertiary chapter title "Troubleshooting Guide" under a secondary chapter title are considered troubleshooting guide images for the corresponding fault mode.

[0025] S14: Image naming and saving; For the extracted images, the chapter title is added as a prefix to the image filename to distinguish different sources and ensure that each image filename is unique. Each image is then saved as an independent image file with the prefixed filename; images obtained from the fault tree section are called fault tree image files, and images obtained from the troubleshooting guide section are called troubleshooting guide image files.

[0026] For example, in this embodiment, the second-level chapter titles are unique. Therefore, the image obtained when the first-level chapter title is "Processing Unit", the second-level chapter title is "Radar Data and External Loading Inconsistent", and the third-level chapter title is "Fault Tree" is named "Radar Data and External Loading Inconsistent_gzs.jpg". The image obtained when the third-level chapter title is "Troubleshooting Guide" is named "Radar Data and External Loading Inconsistent_pglct.jpg".

[0027] S2, Fault Tree Image Recognition and Structure Conversion; S21, Use a multimodal large model to extract text and logical relationships from fault tree image files; The fault tree image file obtained in S14 is used as input, and the text content and logical relationships in the fault tree image are extracted using a multimodal large model.

[0028] The text content is the name of the fault tree node within the fault tree node in the fault tree image.

[0029] For example, based on the fault tree image named "Radar data and add-on loading mismatch_gzs.jpg", Figure 4 As shown, the following fault tree node names can be extracted: [Radar data and external loading are inconsistent, processing unit cable fault, data processing module voltage abnormality, transformer fault, data processing module fault].

[0030] Logical relationships include the logical gate relationships between fault tree nodes.

[0031] S22, Generate a structured text description of the fault tree; Based on the fault tree text content and logical relationships extracted by S21, a large language model generates a structured description of the fault tree to store the logical relationships between fault tree node names. Each fault tree node is identified as a fault mode or fault cause. Fault tree nodes located in the middle layer of the fault tree are the fault causes of the fault tree nodes in the layer above them and also the fault modes of the fault tree nodes in the layer below them.

[0032] For example, for a fault tree image file named "Radar data and external loading inconsistency_gzs.jpg", the structured text description of the fault tree can be generated based on the text content and logical relationship obtained in S21. The description is as follows: [Data processing module failure - leading to - abnormal data processing module voltage, transformer failure - leading to - abnormal data processing module voltage, abnormal data processing module voltage - radar data and external loading inconsistency, processing unit cable failure - leading to - radar data and external loading inconsistency]. That is, the fault mode is [Radar data and external loading inconsistency], the fault cause is [processing unit cable failure, abnormal data processing module voltage], the fault mode is [abnormal data processing module voltage], and the fault cause is [transformer failure, data processing module failure].

[0033] S3, Troubleshooting guide image recognition and structure conversion; S31, Extracting text and logical relationships from troubleshooting guide image files using a multimodal large model; The troubleshooting guide image file obtained in S14 is used as input, and the text content and logical relationships in the troubleshooting guide image are extracted using a multimodal large model.

[0034] The text content is the troubleshooting measures within the step boxes in the troubleshooting guide image. For example, based on the troubleshooting guide image file named "Radar Data and External Loading Inconsistent_pglct.jpg", the following troubleshooting measures can be extracted: [Whether the fault disappears after reloading the external data, check whether the data processing module voltage is abnormal, replace the transformer, replace the data processing module, replace the bus cable, and end the troubleshooting].

[0035] The logical relationships include the arrow flow of the troubleshooting guide image step boxes.

[0036] S32, Generate a descriptive troubleshooting logic flow; Based on the troubleshooting guide image text content and logical relationships extracted from S31, a descriptive troubleshooting logic flow is generated by a large language model.

[0037] Based on the troubleshooting guide image file named "Radar Data and External Loading Inconsistency_pglct.jpg", the following descriptive troubleshooting logic flow can be obtained: [Does the fault disappear after reloading the external data? - Yes - End troubleshooting; Does the fault disappear after reloading the external data? - No - Check if the data processing module voltage is abnormal; Check if the data processing module voltage is abnormal - Yes - Replace the transformer; Does the fault disappear after replacing the transformer? - Yes - End troubleshooting; Does the fault disappear after replacing the transformer? - No - Replace the data processing module; Does the fault disappear after replacing the data processing module? - Yes - End troubleshooting; Check if the data processing module voltage is abnormal - No - Replace the bus cable - End troubleshooting].

[0038] S33, Generate step-by-step troubleshooting text; The descriptive troubleshooting logic flow generated based on S32 is transformed into step-by-step troubleshooting text by a large language model. Step-by-step troubleshooting text refers to describing the troubleshooting process in a step-by-step format.

[0039] For example, a troubleshooting guide image file named "Radar Data and External Loading Inconsistency_pglct.jpg" provides the following troubleshooting steps: Step 1: Does the fault disappear after reloading the external data? If yes, end the troubleshooting; otherwise, proceed to Step 2. Step 2: Check if the data processing module voltage is abnormal. If yes, proceed to Step 3; otherwise, proceed to Step 5. Step 3: Does the fault disappear after replacing the transformer? If yes, end the troubleshooting; otherwise, proceed to Step 4. Step 4: Does the fault disappear after replacing the data processing module? If yes, end the troubleshooting; otherwise, proceed to Step 2. Step 5: Replace the bus cable and end the troubleshooting.

[0040] S4, obtain the mapping relationship between the fault tree and the troubleshooting process; Align the fault causes in step S22 with the step-by-step troubleshooting text generated in step S3. Using a large model matching algorithm, locate the specific troubleshooting step in the step-by-step troubleshooting text for each fault cause in the fault tree, establishing a precise "fault cause - step-by-step troubleshooting text" mapping relationship. The large model matching algorithm combines semantic understanding and string matching through training.

[0041] For example, the fault cause obtained in S22 is [processing unit cable fault, data processing module voltage abnormality, transformer fault, data processing module fault]. The step-by-step troubleshooting text extracted in S33 is: Step 1, does the fault disappear after reloading the external data? If yes, end the troubleshooting; if not, proceed to Step 2. Step 2, check if the data processing module voltage is abnormal. If yes, proceed to Step 3; if not, proceed to Step 5. Step 3, does the fault disappear after replacing the transformer? If yes, end the troubleshooting. If not, proceed to Step 4. Step 4, does the fault disappear after replacing the data processing module? If yes, end the troubleshooting. If not, proceed to Step 2. Step 5, replace the bus cable and end the troubleshooting. Through the large model matching algorithm, the following mapping relationship is obtained: [data processing module voltage abnormality - Step 2, transformer fault - Step 3, data processing module fault - Step 4, processing unit cable fault - Step 5].

[0042] S5, build a fault knowledge database; The fault tree structured text description, step-by-step troubleshooting text, and the mapping relationship between the fault tree and the troubleshooting process obtained from steps S2, S3, and S4 are stored according to a predefined database schema to obtain a fault knowledge database. This database includes three tables: a fault mode table with fields containing [fault mode id, fault mode, fault cause]. A fault tree can contain multiple fault modes, and each fault mode corresponds to a fault mode id and multiple fault causes. The multiple fault mode IDs of a fault tree are hierarchically encoded according to the logical relationship of the fault tree, with the top fault node being called the main fault mode; a troubleshooting guide table with fields containing [step-by-step troubleshooting text id, step-by-step troubleshooting text]; and a mapping table with fields containing [fault mode id, troubleshooting operation step mapping relationship, step-by-step troubleshooting text id]. The troubleshooting operation step mapping relationship is the mapping relationship between a fault cause determined in S4 and the step-by-step troubleshooting text. An inverted index and a semantic vector index are built into the fault knowledge database, ultimately forming a highly structured fault knowledge database that can be queried and reasoned about by the troubleshooting guidance system.

[0043] S6, Fault retrieval and troubleshooting guidance; This paper addresses the need to accurately identify fault descriptions or fault patterns from user issues and precisely match them with relevant fault patterns stored in a database. The accurate identification of fault descriptions or fault patterns from user issues is based on natural language processing, while the precise matching with relevant fault patterns stored in the database is achieved using knowledge matching technology. The specific steps are as follows: S61, Fault Mode Recognition; S611 directly obtains fault inquiries in text form input by users in natural language through the interactive interface, and saves the fault inquiries as user problem data to be processed.

[0044] S612: The received user question data to be processed is preprocessed to obtain standardized user question text. Preprocessing removes irrelevant symbols, redundant expressions, and ambiguous words through a cleaning operation, thereby restating the user questions.

[0045] S613, based on standardized user problem text, uses named entity recognition and keyword extraction algorithms to locate fault phenomenon descriptions and candidate fault mode keywords, thereby extracting key information.

[0046] S614 compares the extracted key information with a preset fault terminology dictionary, converts the key information into standardized fault terms in the fault terminology dictionary, and obtains a standardized fault mode user description.

[0047] For example, a user's fault inquiry is: the machine is making a strange "clicking" sound. After preprocessing, the key information is "the machine is making a strange clicking sound". The user then standardizes the fault mode and describes it as "abnormal noise from the equipment".

[0048] S62, matches database-related fault modes; S621, using standardized fault mode user representations, the fault knowledge database in S5 is queried using a keyword matching algorithm to obtain a preliminary set of matching fault mode candidates. In this embodiment, only the primary fault mode is queried.

[0049] S622, for the initially matched fault mode candidate set, the semantic vector of the standardized fault mode user expression is used to calculate the cosine similarity with the semantic vector of each fault mode in the fault mode candidate set to obtain a fault mode candidate set with similarity score.

[0050] S623: Sort the fault modes in the fault mode candidate set in descending order according to the similarity score. If the fault mode with the highest score is higher than a preset threshold, it is taken as the final matching result. If the highest score is lower than the preset threshold, it is considered that no matching result was found, and the user is informed that there is no relevant knowledge in the database.

[0051] S624, obtains the fault mode structured data.

[0052] For the fault modes retrieved in S623, structured fault mode data is obtained from the database. The structured fault mode data includes at least the fault cause, troubleshooting guide text, and fault cause-step troubleshooting text mapping relationship.

[0053] For example, after a user submits a question, the structured fault mode data might be: { Fault Cause: [Radar data and external loading are inconsistent, processing unit cable fault, data processing module voltage abnormality, transformer fault, data processing module fault]; Operation Steps: [Radar data and external loading are inconsistent - Step 1, data processing module voltage abnormality - Step 2, transformer fault - Step 3, data processing module fault - Step 4, processing unit cable fault - Step 5]; Troubleshooting Guide Text: "Step 1, Does the fault disappear after reloading the external data? If yes, end the troubleshooting process; if not, proceed to Step 2; Step 2, Check if the data processing module voltage is abnormal. If yes, proceed to Step 3; if not, proceed to Step 5; Step 3, Does the fault disappear after replacing the transformer? If yes, end the troubleshooting process. If not, proceed to Step 4; Step 4, Does the fault disappear after replacing the data processing module? If yes, end the troubleshooting process. If not, proceed to Step 2; Step 5, Replace the bus cable, end the troubleshooting process.]}

[0054] S63, a human-computer interactive troubleshooting guide based on a large model; During troubleshooting, users follow the retrieved troubleshooting guidelines and report the observed phenomena or instrument information to the large model. The large model then understands the troubleshooting process and, based on the observed phenomena or instrument information from this step, determines whether further information is needed, or generates the next targeted troubleshooting steps and dynamically adjusts the guidance process. This part, based on natural language generation and contextual understanding technology, implements initial guidance for fault modes and dynamic guidance based on on-site feedback. The specific steps are as follows: S631, based on fault mode structured data, calls the large model to generate initial troubleshooting guidance steps, and returns information such as operations or inspection items in the troubleshooting guidance steps to the user.

[0055] S632 receives feedback information obtained after on-site operations or inspections are performed according to the troubleshooting guidance steps.

[0056] S633: The feedback information is correlated and matched with the structured data of the fault mode in the current step. The large model determines whether to perform further troubleshooting operations, supplement feedback information, or end the troubleshooting operation based on the feedback information. If it is determined that further troubleshooting operations should be performed, S634 is executed. If supplementary feedback information is required, return to S632 and give a prompt message. When the troubleshooting operation is ended, S635 is executed.

[0057] S634, call the large model to generate targeted troubleshooting steps: update the generated troubleshooting guidance text to the visualization interface, push it to the user terminal in real time, and return to S632. S635 records all troubleshooting steps and feedback information in the log for future optimization of the troubleshooting guide.

[0058] This invention also discloses a troubleshooting process generation system for radar systems based on large models, such as... Figure 3 As shown, it includes: a fault knowledge database 1, an image extraction module 2, a fault tree image extraction module 3, a troubleshooting guide image extraction module 4, a mapping module 5, and a fault retrieval and troubleshooting guidance module 6.

[0059] The fault knowledge database 1 includes a fault mode table, a troubleshooting guide table, and a mapping table. The fault mode table fields include: [fault mode id, fault mode, fault cause]. A fault tree can contain multiple fault modes, with each fault mode corresponding to one id and multiple fault causes. Multiple fault mode IDs within a fault tree are hierarchically encoded according to the logical relationship of the fault tree. The fault node at the top of the fault tree is called the main fault mode. The troubleshooting guide table fields include: [step-based troubleshooting text id, step-based troubleshooting text]. The mapping table fields include: [fault mode id, troubleshooting operation step mapping relationship, step-based troubleshooting text id]. An inverted index and a semantic vector index are built into the fault knowledge database, ultimately forming a highly structured fault knowledge database that can be queried and reasoned about by the troubleshooting guidance system.

[0060] Image extraction module 2 converts the fault manual into processable fault manual data, and obtains and saves fault tree images and corresponding troubleshooting guide images based on chapter titles.

[0061] Fault tree image extraction module 3 uses a multimodal large model to extract text and logical relationships from fault tree image files to generate a structured text description of the fault tree.

[0062] Troubleshooting Guide Image Extraction Module 4 uses a multimodal large model to extract text and logical relationships from the troubleshooting guide image files to generate step-by-step troubleshooting text.

[0063] Mapping module 5 aligns the fault causes in the structured text description of the fault tree with the step-by-step troubleshooting text. Through a large model matching algorithm, it locates the specific troubleshooting steps in the step-by-step troubleshooting text for each fault cause in the fault tree, establishing a precise "fault cause - step-by-step troubleshooting text" mapping relationship.

[0064] The Fault Retrieval and Troubleshooting Guidance Module 6 is used to dynamically guide troubleshooting steps based on user-inputted fault information in natural language, troubleshooting guidelines, and feedback information during the troubleshooting process, using a large-scale model. This module includes a fault mode recognition submodule, a fault mode matching submodule, and a troubleshooting guidance submodule. The fault mode recognition submodule converts user-inputted fault information in natural language into standard fault modes. The fault mode matching submodule retrieves structured fault mode data from the fault knowledge database based on the standard fault modes. The troubleshooting guidance submodule, based on the large-scale model's interactive interface, provides corresponding troubleshooting steps according to the structured fault mode data and feedback information.

[0065] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for generating troubleshooting procedures for radar systems based on large models, characterized in that: It includes the following steps: S1, Image Extraction: Obtain the fault manual file of the radar system, convert it into processable fault manual data, identify the fault modes and corresponding fault tree images and troubleshooting guide images according to the document structure, and save the identified images. S2, Fault Tree Image Recognition and Structure Conversion; S21, Use a multimodal large model to extract text and logical relationships from fault tree image files; S22, Generate a structured text description of the fault tree; S3, troubleshooting guide image recognition and structure conversion; S31, Use a multimodal large model to extract text and logical relationships from the troubleshooting guide image file; S32, Generate a descriptive troubleshooting logic flow; S33, Generate step-by-step troubleshooting text; S4, obtain the mapping relationship between the fault tree and the troubleshooting process; Align the fault causes in the structured text description of the fault tree in step S22 with the step-by-step troubleshooting text generated in step S3. Using a large model matching algorithm, locate the specific troubleshooting step for each fault cause in the fault tree in the step-by-step troubleshooting text, and establish an accurate mapping relationship between fault causes and step-by-step troubleshooting text. S5, build a fault knowledge database; The fault tree structured text description, step-by-step troubleshooting text, and the mapping relationship between the fault tree and the troubleshooting process obtained from steps S2, S3, and S4 are stored according to a predefined database pattern to obtain a fault knowledge database. S6, Fault retrieval and troubleshooting guidance; S61, Fault Mode Recognition: Identify fault phenomenon descriptions or fault modes from user questions to obtain standardized user descriptions of fault modes. S62, Match database-related fault modes: Match with related fault modes stored in the database; S63, Human-computer interactive troubleshooting guidance based on a large model: Enables dynamic guidance during the troubleshooting process based on the fault mode and on-site feedback.

2. The method for generating troubleshooting procedures for radar systems based on large models according to claim 1, characterized in that: S1, Image Extraction, the specific steps are as follows: S11, Read the fault manual; The acquired radar system-related fault manuals are converted into processable fault manual data. The first-level chapter title of the radar system-related fault manual is the equipment name, the second-level chapter title is the fault mode, the third-level chapter title is the fault tree, and the text includes fault tree images. The third-level chapter title is the troubleshooting guide, and the text includes troubleshooting guide images. S12, identify fault modes; Based on the processable fault manual data, and considering the document structure, fault modes are automatically identified from the second-level chapter headings. S13, Extract image; Images extracted from the text of a third-level chapter titled "Fault Tree" under a second-level chapter title are considered to be fault tree images for the corresponding fault mode; images extracted from the text of a third-level chapter titled "Troubleshooting Guide" under a second-level chapter title are considered to be troubleshooting guide images for the corresponding fault mode. S14: Image naming and saving; For the extracted images, the chapter title is added as a prefix to the image file name to distinguish different sources and ensure that each image file name is unique.

3. The method for generating troubleshooting procedures for radar systems based on large models according to claim 1, characterized in that: S22, Generate a structured text description of the fault tree. The specific steps are as follows: Based on the text content and logical relationships of the fault tree extracted by S21, a structured description of the fault tree is generated by a large language model to store the logical relationships between the names of the fault tree nodes. Each fault tree node is determined as a fault mode or fault cause. The fault tree node located in the middle layer of the fault tree is the fault cause of the fault tree node in the upper layer and also the fault mode of the fault tree node in the lower layer.

4. The method for generating troubleshooting procedures for radar systems based on large models according to claim 1, characterized in that: The fault knowledge database in S5 includes three tables: a fault mode table, a troubleshooting guide table, and a mapping table. The fault mode table contains fields for: fault mode id, fault mode, and fault cause; the troubleshooting guide table contains fields for: step-by-step troubleshooting text id and step-by-step troubleshooting text. The mapping table fields include: fault mode ID, troubleshooting operation step mapping relationship, and step-based troubleshooting text ID; an inverted index and a semantic vector index are built for the fault knowledge database.

5. The method for generating troubleshooting procedures for radar systems based on large models according to claim 1, characterized in that: S61, Fault Mode Identification: Identify fault phenomenon descriptions or fault modes from user queries to obtain standardized user descriptions of fault modes; including the following steps: S611 directly obtains fault inquiries in text form input by users in natural language through the interactive interface, and saves the fault inquiries as user problem data to be processed; S612, the received user question data to be processed is preprocessed to obtain a standardized user question text; the preprocessing removes irrelevant symbols, redundant expressions and ambiguous words through a cleaning operation to realize the reformulation of the user question; S613, based on standardized user problem text, uses named entity recognition and keyword extraction algorithms to locate fault phenomenon descriptions and candidate fault mode keywords, thereby extracting key information; S614 compares the extracted key information with the preset fault terminology dictionary, and converts the key information into standardized fault terms in the fault terminology dictionary to obtain a standardized fault mode user description.

6. The method for generating troubleshooting procedures for radar systems based on large models according to claim 1, characterized in that: S62, Matching database-related fault modes: Matching with relevant fault modes stored in the database; including the following steps: S621 uses standardized fault mode user representations to query the fault knowledge database in S5 using a keyword matching algorithm to obtain a preliminary set of matching fault mode candidates. S622, For the initially matched fault mode candidate set, the semantic vector of the standardized fault mode user expression is used to calculate the cosine similarity with the semantic vector of each fault mode in the fault mode candidate set to obtain a fault mode candidate set with similarity score. S623, sort the fault modes in the fault mode candidate set in descending order according to the similarity score. If the fault mode with the highest score is higher than the preset threshold, it is taken as the final matching result. If the highest score is lower than the preset threshold, it is considered that no matching result was found, and the user is informed that there is no relevant knowledge in the database. S624, obtain the fault mode structured data; For the fault modes retrieved in S623, structured fault mode data is obtained from the database. The structured fault mode data includes at least the fault cause, the mapping relationship of troubleshooting operation steps, and the troubleshooting guide text.

7. The method for generating troubleshooting procedures for radar systems based on large models according to claim 1, characterized in that: S63, based on a large model, provides interactive troubleshooting guidance: This enables dynamic guidance during the troubleshooting process based on fault modes and on-site feedback; it includes the following steps: S631, based on fault mode structured data, calls the large model to generate initial troubleshooting guidance steps, and returns information such as the operation or inspection items in the troubleshooting guidance steps to the user; S632 receives feedback information obtained after on-site operation or inspection is performed according to the troubleshooting guidance steps; S633: The feedback information is correlated and matched with the structured data of the fault mode in the current step. The large model determines whether to perform further troubleshooting operations, supplement feedback information, or end the troubleshooting operation based on the feedback information. If it is determined that further troubleshooting operations should be performed, S634 is executed; if supplementary feedback information is required, return to S632 and give a prompt message; if the troubleshooting operation is ended, S635 is executed. S634, call the large model to generate targeted troubleshooting steps: update the generated troubleshooting guidance text to the visualization interface, push it to the user terminal in real time, and return to S632; S635 records all troubleshooting steps and feedback information in the log for future optimization of the troubleshooting guide.

8. A troubleshooting process generation system for radar systems based on large models, characterized in that: It includes: The system includes a fault knowledge database, an image extraction module, a fault tree image extraction module, a troubleshooting guide image extraction module, a mapping module, and a fault retrieval and troubleshooting guidance module. The fault knowledge database includes a fault mode table, a troubleshooting guide table, and a mapping table; an inverted index and a semantic vector index are established for the fault knowledge database. The image extraction module converts the fault manual into processable fault manual data, and retrieves and saves fault tree images and corresponding troubleshooting guide images based on chapter titles. The fault tree image extraction module uses a multimodal large model to extract text and logical relationships from fault tree image files to generate a structured text description of the fault tree; The troubleshooting guide image extraction module uses a multimodal large model to extract text and logical relationships from the troubleshooting guide image files to generate step-by-step troubleshooting text; The mapping module aligns the fault causes in the structured text description of the fault tree with the step-by-step troubleshooting text. Through a large model matching algorithm, it locates the specific troubleshooting steps for each fault cause in the fault tree within the step-by-step troubleshooting text, establishing a precise fault cause-step-by-step troubleshooting text mapping relationship. The fault retrieval and troubleshooting guidance module is used to provide dynamically guided troubleshooting steps based on the fault inquiries input by the user in natural language, the troubleshooting guide, and feedback information during the troubleshooting process, provided by a large model.

9. The radar system troubleshooting process generation system based on a large model according to claim 8, characterized in that: The fault retrieval and troubleshooting guidance module includes a fault mode identification submodule, a fault mode matching submodule, and a troubleshooting guidance submodule. The fault mode recognition submodule is used to convert fault inquiries input by users in natural language into standard fault modes; The fault mode matching submodule is used to query the fault knowledge database to obtain structured fault mode data based on standard fault modes. The troubleshooting guidance submodule provides corresponding troubleshooting steps based on the human-computer interaction of the large model and the structured data and feedback information of the fault modes.

10. The radar system troubleshooting process generation system based on a large model according to claim 8, characterized in that: The fault knowledge database includes the following fields in the fault mode table: fault mode id, fault mode, and fault cause; and the troubleshooting guide table includes the following fields: step-by-step troubleshooting text id and step-by-step troubleshooting text. The mapping table fields include: fault mode ID, troubleshooting operation step mapping relationship, and step-by-step troubleshooting text ID.

Images