Search

CN-122022763-A - Automobile intelligent diagnosis method based on artificial intelligence

CN122022763ACN 122022763 ACN122022763 ACN 122022763ACN-122022763-A

Abstract

The invention discloses an artificial intelligence automobile intelligent diagnosis method, and belongs to the field of automobile intelligent diagnosis. The method comprises the steps of constructing a system database containing a case base and vehicle operation data, preprocessing the system database before data warehousing, constructing a multi-layer classification system based on the database, classifying vehicle problems by adopting classification models, selecting an adaptive problem processing model according to the problem types, performing scene secondary training and fine tuning in a mode of prompting word template configuration and MCP calling a knowledge base, finally generating a diagnosis result, checking and confirming the diagnosis result, and periodically refluxing the checked and confirmed diagnosis data to the system database. According to the invention, intelligent diagnosis of the automobile is realized through data standard management, accurate classification of problems, efficient training and application of models and a multi-level auditing mechanism, and the coverage and accuracy of diagnostic services are improved.

Inventors

  • WANG LI
  • CHEN FENG
  • Hua Yuanfang

Assignees

  • 奇瑞商用车（安徽）有限公司

Dates

Publication Date
20260512
Application Date
20260113

Claims (10)

  1. 1. An artificial intelligence automobile intelligent diagnosis method is characterized by comprising the following steps: s1, constructing a system database containing a case library and vehicle operation data, and preprocessing the data before warehousing; s2, constructing a multi-layer classification system based on the database, and classifying the vehicle problems by adopting a classification model; S3, selecting an adaptive problem processing model according to the problem category, performing scene secondary training and fine tuning in a manner of prompting word template configuration and MCP calling a knowledge base, and finally generating a diagnosis result; S4, checking and confirming the diagnosis result, wherein the checking and confirming comprises semantic association degree evaluation, compliance and legal risk evaluation, data rationality and matching degree verification and integral confidence degree calculation based on a manual labeling case; And S5, periodically refluxing the diagnosis data confirmed by the verification to the system database.
  2. 2. The intelligent automobile diagnosis method of claim 1, wherein the multi-layer classification system adopts a multi-layer tree structure and comprises a plurality of first-level classes, and a plurality of second-level and third-level subclasses are arranged under each first-level class.
  3. 3. The method for intelligent diagnosis of an artificial intelligence vehicle according to claim 1, wherein the classification model adopts a text classification architecture based on a pre-training language model, and comprises one or a combination of two models: (1) A discriminant classification model for pre-training language model fine tuning, comprising: After a pre-training language model is selected and loaded, one or more layers of fully-connected networks are added to an output layer of the pre-training language model, vector features are mapped to a predefined class space, and probability distribution of each class is output, wherein the pre-training language model comprises BERT, roBERTa, ERNIE; (2) The prompt word driving classification model based on the large language model LLM selects the large language model LLM as a zero sample/few sample classifier, and maps the user problem to the corresponding category by constructing a classification prompt template and category description, wherein the large language model LLM comprises but is not limited to GPT-5, DEEPSEEK R1 and Gemini 2.5 Pro.
  4. 4. The method for intelligent diagnosis of an artificial intelligence vehicle according to claim 1, wherein in step S3, the problem handling model performs the scene secondary training and fine tuning by means of the prompt word template configuration and the MCP invoking knowledge base, comprising: Configuring a prompt word template, namely configuring special system prompt and dialogue templates for each problem processing model to limit the system roles of the system prompt and dialogue templates, and designing a differential prompt template according to different problem categories; The MCP calls a knowledge base, namely, the system configures a unified MCP interface for each problem processing model, in the dialogue process, the problem processing model calls an external tool and a knowledge source defined by the MCP interface through a prompt word, wherein the external tool and the knowledge source comprise a case retrieval tool, a document retrieval tool and a real-time data query tool, and finally, the model performs reasoning and answer generation according to the returned content of the MCP, so that retrieval enhancement diagnosis is realized.
  5. 5. The method for intelligent diagnosis of an automobile according to claim 1, wherein step S3 further comprises constructing an automated model evaluation system for quantitatively evaluating the overall performance of different models on different problem categories, and dynamically selecting models based on the overall performance.
  6. 6. The method for intelligent diagnosis of an artificial intelligence vehicle according to claim 5, wherein said constructing an automated model evaluation system comprises: Constructing an evaluation data set; defining evaluation indexes including accuracy, safety score and diagnostic integrity score; the automatic evaluation flow is established by setting a model evaluation task trigger period, selecting a test sample according to an evaluation data set to test each question processing model and different prompt words and knowledge base configuration versions thereof during task trigger, comparing model output results with standard answers, calculating each index value of each model under different test samples, and storing the index values into a model performance index table: And establishing a dynamic model selection strategy, namely inquiring a model evaluation result corresponding to the problem category from a model performance index table to which the problem category belongs after determining the problem category, and selecting a model according to a preset strategy according to the model evaluation result.
  7. 7. The method for intelligent diagnosis of an artificial intelligence vehicle according to claim 1, wherein the semantic relevance evaluation comprises a text vector similarity evaluation, which uses cosine similarity between a user questioning text and a model diagnosis result as a semantic relevance score, and a key entity and attribute alignment evaluation, which uses a key entity coverage rate in a user questioning text to adjust the semantic relevance score by checking the model diagnosis result.
  8. 8. The intelligent automobile diagnosis method of claim 1, wherein the compliance and legal risk assessment comprises rule base detection and model auxiliary detection, wherein the rule base detection analyzes a model diagnosis result into a structured text and scores the structured text according to a matching result of the structured text and a preset rule base, the model auxiliary detection scores the model diagnosis result through a preset safety inspection model, and finally, the rule base detection score and the model auxiliary detection score are weighted and averaged to obtain the compliance and legal risk score.
  9. 9. The intelligent diagnosis method of artificial intelligence according to claim 1, wherein the data rationality and matching degree scoring comprehensively considers current vehicle operation data and historical cases, and scoring model diagnosis results.
  10. 10. The method for intelligent diagnosis of an artificial intelligence vehicle according to claim 1, wherein the calculating of the overall confidence coefficient based on the artificial labeling case comprises: Selecting a plurality of manual labeling cases as a confidence calibration data set; diagnosing the cases of the confidence calibration data set by using the current problem processing model, and comparing the manual labeling result with the model diagnosis result: if the structured fields are completely consistent, the structured fields are marked as completely consistent, if the main fault types are consistent but the processing suggestions are different, the structured fields are marked as partially consistent, and the rest of the structured fields are marked as inconsistent; Taking the ratio of the completely consistent cases to the partially consistent cases as the reference consistency p_base of the current problem processing model under the corresponding problem classification; And obtaining semantic relevance scores, compliance and legal risk scores, data rationality and matching degree scores of the current model diagnosis results, respectively carrying out weighted summation according to preset weights w1, w2 and w3, and multiplying the weighted summation by the standard consistency p_base of the category to obtain the overall confidence coefficient.

Description

Automobile intelligent diagnosis method based on artificial intelligence Technical Field The invention belongs to the field of intelligent diagnosis of automobiles, and particularly relates to an intelligent diagnosis method of an artificial intelligence automobile. Background With the rapid development of the new energy automobile industry, the electric, intelligent and networking technologies have become the core trend of industry development. Consumers bring higher expectations to the environmental protection, the intelligent level and the convenience of traveling, and new energy automobiles are taken as important carriers for future traveling, thereby leading the transformation of traffic modes. However, challenges remain in how to efficiently, intelligently, and accurately respond to user problems, and assist vehicle service personnel in quick diagnosis and maintenance. The traditional diagnosis equipment mainly depends on an OBD diagnosis instrument and a remote diagnosis platform, and the tools are mainly used for solving the problems of complex faults and non-faults, which are difficult to effectively solve, and the defects of low response speed, low diagnosis efficiency, high misjudgment rate and the like are overcome, so that the requirements of diversification and instantaneity of users are difficult to meet. Along with the continuous improvement of the requirements of users on the reliability and stability of the vehicle system, the industry still breaks through the limitations of the traditional diagnosis method, and constructs an all-round intelligent solution which covers the vehicle use problems, the function description, the service policies, the maintenance consultation, the fault diagnosis and the like, thereby meeting the multi-dimensional service requirements of the users. Aiming at vehicle maintenance personnel, integrated and systematic vehicle knowledge support is also required to be provided, one-stop service is realized, and maintenance efficiency and accuracy are improved. Based on the above, the intelligent diagnosis system is developed, and the advanced artificial intelligence technology and perfect database construction are relied on to provide an efficient, accurate and convenient problem solution, so that multiple requirements of the new energy automobile on intelligence, instantaneity and high-quality user experience in the new era background are met, and the continuous healthy development of the new energy automobile industry is promoted. However, the current intelligent diagnosis schemes are focused more on solving specific fault types, acquiring related data corresponding to the fault types and designating solutions, and are difficult to provide effectively in the face of complex and numerous faults of an automobile, and are difficult to provide effectively in the face of non-fault types. Therefore, the invention provides an artificial intelligence automobile intelligent diagnosis method. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides an artificial intelligent automobile intelligent diagnosis method which aims to achieve the purposes of realizing automobile intelligent diagnosis and improving the coverage and accuracy of diagnosis service through data standard management, accurate classification of problems, efficient training and application of models and a multi-level auditing mechanism. In order to achieve the purpose, the technical scheme adopted by the invention is that an artificial intelligence automobile intelligent diagnosis method comprises the following steps: s1, constructing a system database containing a case library and vehicle operation data, and preprocessing the data before warehousing; s2, constructing a multi-layer classification system based on the database, and classifying the vehicle problems by adopting a classification model; S3, selecting an adaptive problem processing model according to the problem category, performing scene secondary training and fine tuning in a manner of prompting word template configuration and MCP calling a knowledge base, and finally generating a diagnosis result; S4, checking and confirming the diagnosis result, wherein the checking and confirming comprises semantic association degree evaluation, compliance and legal risk evaluation, data rationality and matching degree verification and integral confidence degree calculation based on a manual labeling case; And S5, periodically refluxing the diagnosis data confirmed by the verification to the system database. Preferably, the multi-level classification system adopts a multi-level tree structure and comprises a plurality of first-level classes, and a plurality of second-level and third-level subclasses are arranged under each first-level class. Preferably, the classification model adopts a text classification architecture based on a pre-training language model, and comprises one or a combination of the two models: (1) A dis