CN-121996698-A - Radar system troubleshooting flow generating method and system based on large model

CN121996698ACN 121996698 ACN121996698 ACN 121996698ACN-121996698-A

Abstract

The invention relates to the field of fault detection, in particular to a method and a system for generating a radar system troubleshooting flow based on a large model. The method comprises the steps of S1, picture extraction, S2, fault tree picture identification and structural conversion, S3, fault elimination guide picture identification and structural conversion, S4, obtaining a mapping relation between a fault tree and a fault elimination flow, S5, constructing a fault knowledge database, and S6, fault retrieval and fault elimination guide. The system comprises a fault knowledge database, a picture extraction module, a fault tree picture extraction module, a fault troubleshooting guide picture extraction module, a mapping module and a fault retrieval and troubleshooting guide module. The fault knowledge database of the radar system is constructed, the matched fault elimination flow can be positioned quickly and accurately, natural language processing and semantic understanding technology is realized by means of a large model, next fault elimination measures are selected automatically according to detection results in the fault elimination process, and intelligent promotion and optimization of the fault elimination flow are realized.

Inventors

WU DUO
GAO LONG
ZENG JIANGHUI
AN RAN
HE LIU
WANG XIAOTIAN

Assignees

中国航空综合技术研究所

Dates

Publication Date: 20260508
Application Date: 20251230

Claims (10)

1. A radar system troubleshooting flow generating method based on a large model is characterized by comprising the following steps: S1, extracting pictures, namely acquiring a file of a fault manual of a radar system, converting the file into processable fault manual data, identifying a fault mode, a fault tree picture and an troubleshooting guide picture corresponding to the fault mode according to a document structure, and storing the identified pictures; s2, fault tree picture identification and structural conversion; S21, extracting texts and logic relations in the fault tree picture file by using the multi-mode large model; S22, generating a fault tree structured text description; S3, identifying and structuring conversion of the troubleshooting guide picture; S31, extracting text and logic relation in the troubleshooting guide picture file by using the multi-mode large model; s32, generating a descriptive troubleshooting logic flow; s33, generating a step-by-step troubleshooting text; s4, obtaining a mapping relation between the fault tree and the troubleshooting flow; Aligning the fault reasons in the structured text description of the fault tree in the step S22 with the step troubleshooting text generated in the step S3, positioning each fault reason in the fault tree to a specific troubleshooting step in the step troubleshooting text through a large model matching algorithm, and establishing an accurate mapping relation between the fault reasons and the step troubleshooting text; s5, constructing a fault knowledge database; Storing the structural text description of the fault tree, the step troubleshooting text and the mapping relation between the fault tree and the troubleshooting flow obtained in the steps S2, S3 and S4 according to a predefined database mode to obtain a fault knowledge database; S6, fault searching and troubleshooting guiding; s61, fault mode identification, namely identifying fault phenomenon expression or fault mode from a user question to obtain standardized fault mode user expression; s62, matching the related fault modes of the database, namely matching the related fault modes stored in the database; S63, man-machine interaction type troubleshooting guiding based on the large model is realized, and dynamic guiding is carried out according to a fault mode and site feedback in the troubleshooting process.
2. The method for generating a fault rejection flow of a radar system based on a large model as in claim 1, wherein S1, the picture extraction comprises the following specific steps: S11, reading a fault manual; converting the acquired radar system related fault manual into processable fault manual data, wherein a first-stage section title of the radar system related fault manual is a device name, a second-stage section title of the radar system related fault manual is a fault mode, a third-stage section title of the radar system related fault manual is a fault tree, a text comprises fault tree pictures, the third-stage section title of the radar system related fault manual is a troubleshooting guide, and the text comprises troubleshooting guide pictures; s12, identifying a fault mode; according to the processable fault manual data, aiming at the document structure, automatically identifying and obtaining a fault mode from the second-level chapter title; s13, extracting pictures; The picture extracted from the text of which the three-level chapter title under the two-level chapter title is a fault tree picture of the corresponding fault mode is regarded as the fault tree picture of the corresponding fault mode; s14, naming and storing the pictures; And adding chapter titles as prefixes into the picture file names of the extracted pictures according to the difference of chapters to distinguish different sources, so as to ensure that each picture file name has uniqueness.
3. The method for generating a fault flow of a large model based radar system according to claim 1, wherein S22, generating a fault tree structured text description comprises the following specific steps: Based on the text content and the logic relation of the fault tree extracted in the S21, generating a fault tree structural description by a large language model, wherein the fault tree structural description is used for storing the logic relation among the names of the fault tree nodes, each fault tree node is determined to be a fault mode or a fault reason, and the fault tree node in the middle layer of the fault tree is the fault reason of the fault tree node in the upper layer of the fault tree node and is also the fault mode of the fault tree node in the lower layer of the fault tree node.
4. The method for generating a fault flow of a large model based radar system as claimed in claim 1, wherein the fault knowledge database in S5 comprises three tables including a fault mode table, a fault rejection guide table and a mapping table; The fault mode table field comprises a fault mode id, a fault mode and a fault reason, the troubleshooting guide table field comprises a step troubleshooting text id and a step troubleshooting text, the mapping table field comprises a fault mode id, a troubleshooting operation step mapping relation and a step troubleshooting text id, and an inverted index and a semantic vector index are established for a fault knowledge database.
5. The method for generating a fault flow of a large model based radar system according to claim 1, wherein S61, the fault pattern recognition comprises the steps of recognizing a fault phenomenon expression or a fault pattern from a user question to obtain a standardized fault pattern user expression, and the method comprises the following steps: s611, directly acquiring fault consultation in a text form input by a user through an interactive interface, and storing the fault consultation as user problem data to be processed; s612, preprocessing the received user problem data to be processed to obtain normalized user problem text, wherein the preprocessing removes irrelevant symbols, redundant expressions and ambiguous words through cleaning operation to realize re-expression of the user problem; S613, based on normalized user problem text, using named entity recognition and keyword extraction algorithm to locate fault phenomenon descriptions and candidate fault mode keywords in the text, and extracting key information; S614, the extracted key information is compared with a preset fault term dictionary, and the key information is converted into standardized fault words in the fault term dictionary, so that standardized fault mode user expression is obtained.
6. The method for generating a fault flow for a large model based radar system of claim 1, wherein S62, matching the database-related failure modes, and storing the database-related failure modes, comprises the steps of: s621, inquiring a fault knowledge database in S5 through a keyword matching algorithm to obtain a preliminarily matched fault mode candidate set by using a standardized fault mode user expression; S622, performing cosine similarity calculation on the semantic vectors expressed by the standardized fault mode users and the semantic vectors of all fault modes in the fault mode candidate set to obtain a fault mode candidate set with similarity scores; S623, sorting the fault modes in the fault mode candidate set in a descending order according to the similarity score, if the highest score is higher than the fault mode with the preset threshold value, taking the fault mode as a final matching result, and if the highest score is lower than the preset threshold value, considering that the matching result is not found, and informing a user that no related knowledge exists in the database; s624, obtaining fault mode structured data; And (3) acquiring fault mode structured data from a database aiming at the fault mode retrieved in the step S623, wherein the fault mode structured data at least comprises fault reasons, troubleshooting operation step mapping relations and troubleshooting guide texts.
7. The method for generating a fault rejection flow of a large model based radar system of claim 1, wherein the step S63 of conducting a man-machine interactive fault rejection based on the large model comprises the steps of conducting a dynamic boot according to a fault mode in combination with a site feedback during the fault rejection, and the method comprises the following steps: s631, calling a large model to generate an initial troubleshooting guide step based on the fault mode structured data, and returning information such as operation or check items in the troubleshooting guide step to a user; S632, receiving feedback information obtained after the operation or the inspection of the site according to the troubleshooting guiding step; S633, carrying out association matching on feedback information and fault mode structured data matched with the current step, judging whether the fault operation is further executed, the feedback information is supplemented or the fault operation is ended according to the feedback information by the large model, executing S634 when the fault operation is further executed, returning to S632 and giving prompt information when the feedback information is needed, and executing S635 when the fault operation is ended; S634, calling the large model to generate a targeted next troubleshooting step, namely updating the generated next troubleshooting guide text to a visual interface, pushing the updated next troubleshooting guide text to the user terminal in real time, and returning to S632; And S635, recording all executed troubleshooting steps and feedback information into a log for subsequent optimization of the troubleshooting guide.
8. A radar system troubleshooting flow generating system based on a large model is characterized by comprising a fault knowledge database, a picture extraction module, a fault tree picture extraction module, an troubleshooting guide picture extraction module, a mapping module and a fault retrieval and troubleshooting guide module; the fault knowledge database comprises a fault mode table, a fault rejection guide table and a mapping table, wherein an inverted index and a semantic vector index are established for the fault knowledge database; the picture extraction module is used for converting the fault manual into processable fault manual data, acquiring fault tree pictures and corresponding troubleshooting guide pictures according to chapter titles and storing the fault tree pictures and the corresponding troubleshooting guide pictures; The fault tree picture extraction module is used for extracting texts and logic relations in the fault tree picture file by using the multi-mode large model to generate fault tree structured text description; the troubleshooting guide picture extraction module is used for extracting texts and logical relations in the troubleshooting guide picture file by using the multi-mode large model to generate a step troubleshooting text; The mapping module aligns fault reasons in the fault tree structured text description with the step troubleshooting text, positions each fault reason in the fault tree to a specific troubleshooting step in the step troubleshooting text through a large model matching algorithm, and establishes an accurate fault reason-step troubleshooting text mapping relation; the fault searching and troubleshooting guiding module is used for providing a dynamically guided troubleshooting step by a large model according to the troubleshooting guide and feedback information in the troubleshooting process according to the fault consultation input by a user by adopting natural language.
9. The system for generating a fault flow for a large model based radar system of claim 8, wherein said fault retrieving and troubleshooting guidance module comprises a fault pattern recognition sub-module, a fault pattern matching sub-module, and a troubleshooting guidance sub-module; The fault mode recognition submodule is used for converting fault consultation input by a user through natural language into a standard fault mode; The fault mode matching sub-module is used for inquiring in a fault knowledge database according to a standard fault mode to obtain fault mode structured data; the troubleshooting guiding submodule gives out corresponding troubleshooting steps based on man-machine interaction of the large model according to the fault mode structural data and feedback information.
10. The system for generating a fault flow for a large model based radar system of claim 8, wherein the fault knowledge database comprises a fault pattern table field comprising a fault pattern id, a fault pattern and a fault cause, a fault guide table field comprising a step fault elimination text id and a step fault elimination text, and a mapping table field comprising a fault pattern id, a fault elimination operation step mapping relation and a step fault elimination text id.

Description

Radar system troubleshooting flow generating method and system based on large model Technical Field The invention relates to the field of fault detection, in particular to a method and a system for generating a radar system troubleshooting flow based on a large model. Background In the operation and maintenance and fault handling process of the radar system serving as an important device of modern combat equipment, the core requirements of ensuring continuous and stable operation of the radar system, shortening the time of fault shutdown, reducing operation and maintenance guarantee cost and the like are required to be met. However, the current troubleshooting work of the radar system mainly relies on technical staff to consult data such as fault trees in paper or electronic version, special troubleshooting guidelines and the like, and the troubleshooting experience accumulated by the technical staff is combined to diagnose and troubleshoot the fault phenomenon. However, the traditional troubleshooting mode has obvious limitations in practical application, namely, the radar system has complex internal circuit, dense signal links and a plurality of fault association factors, technicians need to manually search corresponding information in massive fault data and guideline documents, and particularly when the inter-module association faults are faced, the association relations between fault tree nodes and the troubleshooting steps of a processing unit need to be repeatedly compared, so that the troubleshooting efficiency is low, misjudgment is easily caused by missing key fault link information due to manual search, and the fault handling period is further prolonged. Aiming at the problems, in the industry, an attempt is made to realize troubleshooting guidance by constructing a digital fault database of a radar system and combining a simple search function, but a plurality of key problems still exist in practical application, such as that fault data and troubleshooting guidance are stored in a form of independent documents, a unified association mapping mechanism is lacking, the corresponding relation between specific fault nodes of a processing unit in a fault tree and corresponding troubleshooting steps in the troubleshooting guidance is scattered and undefined, the corresponding relation between faults described by natural language and the fault tree nodes and the troubleshooting guidance cannot be accurately recognized, the radar system troubleshooting needs to gradually complete series of steps such as circuit detection, signal acquisition, module linkage test and the like, and the prior system cannot adjust a subsequent guide path in real time according to an intermediate result in the troubleshooting process, and a technician is required to manually trace back the fault tree and inquire the corresponding troubleshooting steps. Therefore, a new technology for generating a fault rejection flow of a radar system is needed at present to solve the problems of undefined association of faults and fault rejection steps, low accuracy of natural language retrieval, poor dynamic adaptability of a fault rejection path and the like in the fault rejection guiding process of the radar system, and improve the fault rejection efficiency and accuracy of the radar system. Disclosure of Invention In order to solve the defects of the prior art, the invention provides a large-model-based radar system troubleshooting flow generating method and system, which are used for realizing a quick positioning and matching troubleshooting flow by defining the corresponding relation between fault tree nodes and troubleshooting steps of a radar system and realizing dynamic adaptation in the process of searching and troubleshooting guidance by describing a fault phenomenon in natural language by means of a large model. Specifically, in one aspect, the present invention provides a method for generating a fault rejection flow of a radar system based on a large model, which includes the following steps: S1, extracting pictures, namely acquiring a file of a fault manual of a radar system, converting the file into processable fault manual data, identifying a fault mode, a fault tree picture and an troubleshooting guide picture corresponding to the fault mode according to a document structure, and storing the identified pictures; s2, fault tree picture identification and structural conversion; S21, extracting texts and logic relations in the fault tree picture file by using the multi-mode large model; S22, generating a fault tree structured text description; S3, identifying and structuring conversion of the troubleshooting guide picture; S31, extracting text and logic relation in the troubleshooting guide picture file by using the multi-mode large model; s32, generating a descriptive troubleshooting logic flow; s33, generating a step-by-step troubleshooting text; s4, obtaining a mapping relation between the fault tree and the troubleshooting fl