Automobile fault diagnosis method, system, product, equipment and storage medium

Abstract

The invention discloses an automobile fault diagnosis method and system, a product, equipment and a storage medium, and the method comprises the steps: collecting and continuously learning historical fault data, real-time Internet of Vehicles data and multi-source information, and training a large language model capable of understanding a complex fault mode. And when the vehicle initiates a diagnosis request, acquiring corresponding state data and a historical report according to the vehicle identification code, and inputting the state data and the historical report into the large model for analysis so as to position a fault reason and generate a maintenance suggestion. A vehicle is continuously monitored through a remote platform, a trained model is utilized to scan, predict and analyze real-time data flow, once a risk mode conforming to potential fault features is identified, early warning information is actively generated, and an effective prompt is sent to a user or a server before a fault is actually triggered. According to the intelligent closed loop of the whole process, the dependence on manual troubleshooting is greatly reduced, the labor cost is remarkably reduced, and the risk of serious faults of the vehicle is effectively reduced through prospective intervention.

Description

Technical Field

[0001] This application relates to the field of automotive fault diagnosis, and in particular to an automotive fault diagnosis method, system, product, device and storage medium. Background Technology

[0002] In the field of automotive fault diagnosis, especially in the after-sales repair of new energy vehicles, the coexistence of multiple fault codes and chain reactions are common phenomena. A single underlying fault in a vehicle may trigger multiple related electronic control units to report errors simultaneously, resulting in diagnostic equipment reading a large number of historical and current fault codes, making it difficult to accurately locate the source of the fault. In addition, intermittent faults such as poor wiring harness contact or momentary failure of internal circuitry in the controller are often impossible to reproduce when the vehicle is brought in for inspection, further increasing the complexity and uncertainty of diagnosis.

[0003] Current mainstream diagnostic methods primarily rely on fault code parsing and pre-defined fault tree logic. Their diagnostic rules are fixed, and their data dimensions are limited. They typically only combine real-time sensor data with fault codes for judgment, lacking the fusion analysis of multi-source information such as vehicle history repair records, design parameters, and user behavior habits. This leads to a high misdiagnosis rate when faced with complex situations involving the same fault code but different causes, failing to effectively distinguish between related faults and the root cause, significantly limiting diagnostic accuracy. Furthermore, most existing methods are reactive diagnostics, unable to provide early warnings of potential faults, and hindering the continuous accumulation and iteration of repair knowledge. This results in low diagnostic efficiency, high reliance on manual intervention, and an inability to meet the demands of modern automotive intelligent and precise diagnostics.

[0004] Therefore, how to overcome the limitations of fixed diagnostic logic and achieve effective fusion and analysis of multi-source heterogeneous data has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] In view of the above problems, this application provides a method, system, product, device and storage medium for diagnosing automotive faults.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] The first aspect of this application provides a method for diagnosing automotive faults, including:

[0008] Obtain a fault diagnosis request, which includes a vehicle identification number and fault description information;

[0009] Based on the vehicle identification code and the fault description information, multi-source data related to the target vehicle indicated by the vehicle identification code is obtained from the database;

[0010] The fault description information and the multi-source data are input into a fully trained large language model, so that the fully trained large language model can analyze the fault type and the data related to the vehicle under test, and output at least one fault cause, as well as the corresponding solution suggestions and occurrence probability of the fault cause.

[0011] A diagnostic report is generated based on the cause of the fault, the suggested solutions, and the probability of occurrence.

[0012] In one possible implementation, the fully trained large language model is determined in the following ways:

[0013] Obtain historical diagnostic reports and vehicle networking data for the sample vehicles;

[0014] Data extraction and cleaning are performed on the historical diagnostic reports and vehicle network data to obtain the fault type, fault cause and corresponding solution;

[0015] The fault type, fault cause, and corresponding solution are fused with data from at least one of the following systems: vehicle production management system, dealer management system, customer relationship management system, and technical information system.

[0016] The fused multi-source data is input into the large language model for training. Data with confidence levels below a preset threshold during training is corrected, and the corrected data is then re-input into the large language model for iterative training until a fully trained large language model is obtained.

[0017] In one possible implementation, the method further includes:

[0018] Select the two most probable causes of failure from the output results and calculate the sum of their probabilities.

[0019] If the calculation result is lower than a preset threshold, a diagnostic case is generated based on the fault diagnosis request and the output of the large language model.

[0020] The diagnostic cases are pushed to the human end so that the human can make a judgment based on the production batch and the frequency of similar failures, and obtain a human judgment result including the fault location conclusion.

[0021] Based on the results of the manual judgment, the causes of the faults and the solutions suggested in the diagnostic report are adjusted.

[0022] In one possible implementation, the step of extracting and cleaning data from the historical diagnostic reports and vehicle network data to obtain the fault type, fault cause, and corresponding solution includes:

[0023] The historical diagnostic reports and vehicle network data are processed by structured element extraction to obtain structured data elements related to the fault.

[0024] Extract key information related to the fault from the unstructured text content in the historical diagnostic reports and vehicle network data;

[0025] The structured data elements are associated and integrated with the key information to reconstruct and generate a structured data record containing the fault type, fault cause, and corresponding solution.

[0026] In one possible implementation, the method further includes:

[0027] After the diagnostic report is adopted or corrected, the corrected feedback data of the diagnostic results is used as training samples to train the large language model.

[0028] In one possible implementation, the method further includes:

[0029] Real-time monitoring data of the target vehicle is obtained based on the vehicle-to-everything (V2X) platform;

[0030] The real-time monitoring data is analyzed to obtain the current health status indicators;

[0031] The current health status indicators are compared and analyzed with the historical health status of the target vehicle and / or the benchmark health status of the same or similar vehicle models.

[0032] When the comparative analysis results exceed the preset threshold, an early warning message is generated and output.

[0033] This application provides a second embodiment of an automotive fault diagnosis system, including:

[0034] The first acquisition unit is used to acquire a fault diagnosis request, the fault diagnosis request including a vehicle identification code and fault description information;

[0035] The second acquisition unit is used to acquire multi-source data related to the target vehicle indicated by the vehicle identification code from the database based on the vehicle identification code and the fault description information.

[0036] The output unit is used to input the fault description information and the multi-source data into a fully trained large language model, so that the fully trained large language model can analyze the fault type and the data related to the vehicle under test, and output at least one fault cause, as well as the corresponding solution suggestion and occurrence probability of the fault cause.

[0037] The generation unit is used to generate a diagnostic report based on the cause of the fault, the suggested solutions, and the probability of occurrence.

[0038] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the vehicle fault diagnosis method as described in the first aspect above.

[0039] The fourth aspect of this application provides a computer program product that, when run on a computer, executes the vehicle fault diagnosis method as described in the first aspect above.

[0040] A fifth aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the vehicle fault diagnosis method as described in the first aspect above.

[0041] Compared with the prior art, this application has the following beneficial effects:

[0042] This system retrieves comprehensive fault requests that include a vehicle identification number (VIN) and a natural language description, overcoming the limitation of only being able to parse fault codes as a single input. Based on this unique identifier, it retrieves diverse and heterogeneous data associated with the target vehicle, effectively integrating fault information that is traditionally scattered across different stages. By inputting fault description information and integrated multi-source data into a fully trained large language model, and relying on the large language model to conduct flexible and in-depth comprehensive analysis based on fault type and relevant data of the vehicle under test, it is no longer limited by a fixed diagnostic logic framework, breaking through the limitations of traditional diagnostic logic. Furthermore, leveraging the efficient analytical capabilities of the fully trained large language model, coupled with a standardized multi-source data retrieval and integration process, it can quickly complete comprehensive fault-related analysis and output corresponding fault causes, matching solution suggestions, and probability of occurrence. Based on the relevant analysis results, a diagnostic report is efficiently generated, significantly reducing the processing time of the entire diagnostic process and effectively solving the problem of slow diagnostic response in traditional remote diagnostic modes. At the same time, the probability of occurrence of various fault causes output by the large language model can accurately quantify the likelihood of different faults occurring. Combined with multi-source vehicle operating data and historical fault-related data, it can predict potential vehicle fault risks in advance, moving beyond passive diagnosis based solely on already occurred faults, thus solving the technical problem of difficulty in achieving fault early warning in traditional diagnostic modes. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 A flowchart of a vehicle fault diagnosis method provided in an embodiment of this application;

[0045] Figure 2 A flowchart illustrating the construction of a fault diagnosis system provided in this application embodiment;

[0046] Figure 3 A flowchart of a vehicle fault diagnosis method provided in an embodiment of this application;

[0047] Figure 4 A flowchart of multi-source information processing required for training a large language model is provided for embodiments of this application.

[0048] Figure 5 An iterative optimization flowchart provided for an embodiment of this application;

[0049] Figure 6 This is a structural diagram of an automotive fault diagnosis system provided in an embodiment of this application;

[0050] Figure 7 This is a schematic diagram of the architecture of a computer hardware system provided in an embodiment of this application. Detailed Implementation

[0051] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0052] To facilitate understanding of the technical solutions provided in the embodiments of this application, the background technology involved in the embodiments of this application will be described below.

[0053] As mentioned earlier, during after-sales diagnostics, some vehicle models may have multiple fault codes coexisting in their diagnostic reports. These reports are used to guide repair personnel in determining whether there are faults in the vehicle's controller and circuitry. However, for intermittent issues such as poor wiring harness contact, internal controller circuit faults, or soldering failures, the fault is often not immediately reproducible when the vehicle arrives at the repair shop and is not presented to the repair personnel in its current form. In such cases, using diagnostic equipment to locate the fault may result in reading a large number of historical fault codes, making it difficult to directly and effectively pinpoint the source of the fault.

[0054] Fault codes are complex and can trigger chain reactions; a single underlying fault can cause errors in multiple related modules, making it difficult to determine the starting point for diagnosis. Repair work is highly dependent on the technician's experience; new technicians need a long period of experience to handle complex faults in the three-electric systems. Furthermore, the colloquial and inconsistent language used to describe faults also affects the overall efficiency of diagnosis.

[0055] Beyond this, the current automotive diagnostics and maintenance field faces four core pain points that hinder improvements in after-sales efficiency and user experience. The most prominent issue is the limited data dimensions. Traditional diagnostics rely solely on fault codes and basic sensor data, lacking crucial information such as historical repair experience, vehicle design parameters (e.g., ECU hardware thresholds), owner operating habits, and frequent short-distance driving. This results in a 15% to 20% misdiagnosis rate for cases with the same fault code but different causes. Poor troubleshooting coordination is another pain point. Information is fragmented between the repair and user ends; users only receive the fault result, not step-by-step operational guidance. Complex faults require on-site support from repair personnel, with an average cost exceeding 500 yuan per instance and a response time of 24 to 48 hours. Predictive maintenance is relatively weak. Existing systems mostly diagnose after a fault occurs, lacking early warnings for potential failures. For example, a fault might only be triggered 30 days after the battery voltage difference exceeds the standard, by which time charging efficiency has already declined, making early intervention impossible. The efficiency of knowledge accumulation is also low. Repair experience relies heavily on technicians' personal memories and has not formed a structured knowledge system. The training period for new technicians is as long as 6 months. Moreover, the extraction of fault characteristics still needs to be done manually, which further reduces the overall efficiency.

[0056] To address the aforementioned issues, this application proposes an intelligent diagnostic system based on a large language model and multi-source data fusion, aiming to fundamentally improve the accuracy, efficiency, and foresight of automotive fault diagnosis. The system first integrates multi-dimensional data from the entire vehicle lifecycle, including real-time sensor data, historical maintenance records, vehicle design parameters, production batch information, and owner operating habits, overcoming the problems of single data dimensions and high misdiagnosis rates due to different causes for the same fault code in traditional diagnostics. Leveraging a well-trained large language model, the system can perform deep correlation analysis and chain reasoning on complex scenarios with multiple coexisting fault codes, effectively distinguishing between current and historical faults, accurately locating the root cause, and reducing over-reliance on the personal experience of repair personnel. Simultaneously, the system achieves collaboration between the repair and user ends through a cloud platform, providing users with structured troubleshooting guidance and offering remote assistance to repair personnel, significantly shortening response times and reducing on-site support costs.

[0057] In predictive maintenance, the system utilizes real-time vehicle network data and machine learning models to identify and warn of potential faults such as battery voltage differences, shifting from passive diagnosis to proactive intervention. Furthermore, the system automatically accumulates case studies after each diagnosis and repair, transforming experience into structured knowledge through expert review and model iteration. This continuously optimizes diagnostic capabilities, thereby improving repair efficiency, shortening technician training cycles, and forming a closed-loop intelligent diagnosis and maintenance system.

[0058] It should be noted that the automotive fault diagnosis method, system, product, equipment, and medium provided in this application can be applied to the field of computer technology. The above are merely examples and do not limit the application areas of the automotive fault diagnosis method, system, product, equipment, and medium provided in this application. Furthermore, the embodiments of this application may not limit the executing entity of the automotive fault diagnosis; for example, the automotive fault diagnosis method of this application embodiment can be applied to data processing devices such as terminal devices or servers. The terminal device can be an electronic device such as a computer or a personal digital assistant (PDA). The server can be a standalone server, a cloud server, or a cluster server composed of multiple servers.

[0059] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0060] The following embodiment illustrates a vehicle fault diagnosis method provided in this application.

[0061] See Figure 1 , Figure 1 This diagram illustrates an application scenario provided by an embodiment of this application, depicting the application of the automotive diagnostic and predictive maintenance system constructed by this solution throughout the entire vehicle lifecycle. This scenario demonstrates the system's end-to-end diagnostic management capabilities, from manufacturing to after-sales operation. Its core lies in using a unified cloud platform to link multiple business systems, achieving full controllability and traceability of diagnostic data.

[0062] This application's diagnostic management capabilities cover key stages from production to after-sales service. Through a cloud platform linking multiple systems, it achieves end-to-end controllability and traceability of diagnostic data. Specifically, number 1 corresponds to an automotive diagnostic and predictive maintenance system based on multi-source data fusion and end-to-end intelligence; numbers 2 to 9 correspond to after-sales vehicle data diagnostics, offline device data diagnostics such as diagnostic instruments, WEB cloud diagnostics, engineering cloud diagnostics, program upgrade management, production vehicle data diagnostics, off-line vehicle data diagnostics, and entertainment system data diagnostics, respectively. The entire lifecycle covers EOL, manufacturing, and after-sales operation stages, involving multiple carriers such as TBOX, HUD, OBD, and after-sales vehicles.

[0063] During the production phase, the MES system synchronizes vehicle ECU part information, the software management system synchronizes ECU software packages, and the cloud platform performs OTA tasks on the assembly line according to the vehicle model. After the vehicle enters the factory mode, it automatically connects to the factory WiFi, installs and starts the flashing APP, completes the production line OTA task detection, download and multi-station execution, and finally cleans up the environment to exit the factory mode. This process solves the problems of many physical workstations and high labor costs in traditional production lines.

[0064] During the after-sales phase, vehicles can be accessed via a cloud platform. The diagnostic tool, combined with the VCI device, acquires vehicle data. After inputting basic vehicle diagnostic information, the system retrieves the corresponding diagnostic data and repair scripts based on the VIN, linking with the DMS to achieve a closed-loop repair process. Simultaneously, FOTA upgrade tasks and diagnostic data are managed synchronously to ensure that the after-sales vehicle software version matches the diagnostic data, improving repair accuracy.

[0065] The entire system is centered on a cloud platform, connecting security and management systems such as MES, DMS, and PKI / KMS to achieve end-to-end flow and management of diagnostic data from production to after-sales service. This ensures efficient and accurate diagnostic work at each stage, adapting to the diagnostic needs throughout the vehicle's entire lifecycle. The system is user-centric and service quality-oriented, covering the entire service chain from vehicle monitoring to after-sales maintenance, and linking multiple systems to achieve a closed-loop business process.

[0066] The business system comprises four core segments: virtual service center, TSP system, after-sales products, and user products, with a wide range of specific application scenarios. First, diagnostic tools, such as cloud-based diagnostic instruments, flexible data acquisition, and one-click quick checks, meet the diagnostic needs of different roles. Second, service response tools, such as fault alerts and one-click assistance, enhance user experience. Third, system integration tools, linking with DMS, CRM, and TIS to achieve collaborative management of repair processes, customer data, and technical information.

[0067] This application focuses on two aspects: firstly, creating user-centric data service products that provide personalized services through data collection and analysis; and secondly, building after-sales tools that prioritize service quality and efficiency, providing intelligent diagnostic support for repair personnel. Simultaneously, the system offers proactive services around the work order process, driving a shift in business models from passively responding to faults to proactively predicting and intervening, covering multiple roles including car owners, repair technicians, and service stations, ultimately forming a complete business loop encompassing monitoring, early warning, diagnosis, repair, and evaluation.

[0068] See Figure 2 ,Should Figure 2 A flowchart illustrating the construction of a fault diagnosis system provided in this application embodiment, the method comprising:

[0069] The process begins with the acquisition of two types of data. First, it involves collecting historical diagnostic reports and vehicle data from all vehicles, covering past fault cases, repair records, and other information. Second, it involves acquiring the diagnostic report and vehicle data of the vehicle under test, including current fault symptoms and basic parameters. These two types of data together form the basis of the diagnostic analysis, covering both historical experience cases and clarifying the specific information of the vehicle currently under inspection.

[0070] After acquiring the data, the information is first organized and structured: redundant content in the data is cleaned, the specific type of fault is identified, such as distinguishing between front-end faults and historical faults; at the same time, the definition of fault codes, the causes of the faults, and the corresponding solutions in the past are clarified, so that the scattered raw data can be transformed into usable diagnostic materials.

[0071] Subsequently, the large model system uses the vehicle identification code to link the MES production management system, DMS dealer management system, CRM customer relationship management system, and TIS technical information system to trace the original data stored in these systems, such as vehicle production parameters, dealer maintenance records, customer usage feedback, and official technical documents, further enriching the information dimensions required for diagnosis.

[0072] Next, the system acquires vehicle data from the vehicle networking platform and inputs fault data, fault causes, and corresponding solutions into a large model for deep learning. This information allows the large model to continuously accumulate relevant fault diagnosis experience, gradually improving its analysis and matching capabilities. For complex data encountered in the process, an expert verification stage is initiated. Professionals check and correct this data to reduce the interference of erroneous data on diagnostic results and ensure the accuracy and reliability of the data. The verified data is then fed back into the large model, allowing it to absorb these validated and accurate diagnostic experiences, continuously optimizing its analysis logic and matching mechanisms, and constantly improving diagnostic accuracy.

[0073] Ultimately, through the information accumulation, processing, and model optimization of the above multiple stages, a large-scale fault diagnosis system with multi-data fusion is generated. With the help of integrated information and trained models, more efficient and accurate fault diagnosis support is achieved.

[0074] In one possible implementation, the method for determining the training of a complete large language model involves using real vehicle-related data as the core foundation, relying on standardized data processing and multiple rounds of model training iterations to ultimately complete the training of the large language model.

[0075] The first step is to comprehensively acquire basic data, with the core task being the collection of historical diagnostic reports and vehicle-to-everything (V2X) data from the sample vehicles. This type of data is the core raw material for model training. The historical diagnostic reports can fully present the diagnostic records related to various faults that have occurred in the past of the sample vehicles, while the V2X data can accurately reflect the various parameters and operating performance of the sample vehicles in different operating scenarios and states. Comprehensive and detailed collection of raw data can lay a solid foundation for subsequent full-process data processing and model training.

[0076] After acquiring the raw data, professional data extraction and cleaning operations are performed on the collected historical diagnostic reports and vehicle networking data. The data extraction stage precisely filters out core information with practical application value from the massive amount of raw data. The data cleaning stage removes invalid and redundant content, distorted and abnormal data, and irrelevant interference data from the raw data. Through these two core operations, clear fault types can be accurately extracted from the raw data, the specific causes of each fault can be identified, and corresponding solutions adapted to various faults can be integrated, forming structured and accurate core data content.

[0077] The next step will be the crucial multi-source data fusion, which involves deeply integrating the extracted core data on fault types, causes, and corresponding solutions with data from various vehicle-related business systems. These systems include at least one of the following: vehicle production management systems, dealer management systems, customer relationship management systems, and technical information systems. These different types of systems can supplement data from various dimensions, such as vehicle production process parameters, after-sales maintenance records, customer feedback, and professional technical support materials. Multi-source data fusion effectively enriches the dimensions and content of the data, creating effective correlations and complementarities between different types of data. This makes the fused dataset more closely aligned with the actual business scenarios of vehicle fault diagnosis and provides a more comprehensive reference for subsequent model training.

[0078] After the fusion of multi-source data is completed, the training and iterative optimization phase of the large language model begins. The fused multi-source data is then fully input into the large language model, officially initiating the basic training process. Throughout the training process, the confidence level of the model's output is continuously monitored and quantified. For data with confidence levels below a preset threshold, targeted and precise verification is performed to correct biases and errors, ensuring the accuracy and validity of the data. The verified data is then re-input into the large language model for a new round of iterative training. This iterative training process continues cyclically, with each iteration optimizing the model based on the corrected, high-quality data. This process continues until the model's output consistently reaches the preset confidence level, and its fault diagnosis capabilities demonstrate stable and accurate performance. At this point, the entire training process is complete, and a fully trained large language model is obtained.

[0079] See Figure 3 ,Should Figure 3 A flowchart of an automotive fault diagnosis method provided in this application embodiment, wherein the execution subject of the method is the large-model fault diagnosis system constructed in the aforementioned steps, and the method includes:

[0080] S101, Obtain fault diagnosis request.

[0081] The process of obtaining a fault diagnosis request involves two core elements: a vehicle identification number (VIN) and a fault description. The VIN is used to accurately locate the corresponding vehicle, while the fault description clarifies the actual fault situation. Both elements serve as the core basis for subsequent fault diagnosis throughout the entire process.

[0082] S102. Based on the vehicle identification code and the fault description information, obtain multi-source data related to the target vehicle indicated by the vehicle identification code from the database.

[0083] The multi-source data includes historical diagnostic reports of the target vehicle, vehicle networking data, and raw data associated with the vehicle production management system, dealer management system, customer relationship management system, and technical information system.

[0084] Based on the acquired vehicle identification number (VIN) and fault description information, relevant data is precisely retrieved from the corresponding database. Specifically, this involves obtaining multi-source data matching the target vehicle indicated by the VIN. This multi-source data has comprehensive coverage, including historical diagnostic reports retained by the target vehicle, real-time vehicle network data collected during vehicle operation, and various raw data retrieved from vehicle production management systems, dealer management systems, customer relationship management systems, and technical information systems. This multi-dimensional data source provides detailed and comprehensive reference support for subsequent fault analysis.

[0085] S103. Input the fault description information and the multi-source data into a fully trained large language model, so that the fully trained large language model can analyze the fault type and the data related to the vehicle under test, and output at least one fault cause, as well as the corresponding solution suggestions and occurrence probability of the fault cause.

[0086] The core data range of the input model covers two types of key information. The first is fault description information that directly reflects the current problem of the vehicle. This type of information is a direct clue to locate the fault direction. The second is multi-source data collected and organized in the early stage. This type of data includes the historical fault records of the vehicle under test, real-time operating parameters, and data related to the entire life cycle of the vehicle, such as production, after-sales service, and customer feedback. This provides rich background support for the model to comprehensively analyze the fault.

[0087] Subsequently, the fully trained large language model will conduct a systematic analysis based on the input information, combined with the identified fault types and various data directly related to the vehicle under test. Leveraging its accumulated knowledge of vehicle fault diagnosis and the correlation patterns between various faults and their causes during training, the model can perform in-depth mining and logical matching of the input data to accurately pinpoint factors that may lead to the current fault. Ultimately, the model outputs at least one clear fault cause and generates corresponding solution suggestions for each cause, providing specific guidance for subsequent maintenance operations. Furthermore, it outputs the probability of each fault cause occurring, helping relevant personnel assess the likelihood of different fault causes and providing a basis for prioritizing high-probability faults.

[0088] S104. Based on the cause of the fault, the proposed solution, and the probability of occurrence, generate a diagnostic report.

[0089] Based on the three core information output by S103, namely the clearly defined cause of the fault, the corresponding solution suggestions for each type of fault, and the probability of occurrence of each fault cause, the report generation process first logically organizes and systematically integrates these three types of information. The fault causes are sorted according to their probability of occurrence, with high-probability fault causes and corresponding solution suggestions placed at the core of the report. This allows relevant personnel to prioritize key fault points. Simultaneously, the process ensures that the correspondence between various types of information is clear and accurate, avoiding misalignments between fault causes, solution suggestions, and occurrence probabilities.

[0090] The integration process balances report standardization and readability, presenting information in a standardized manner. This includes clearly indicating the specific description of the fault's cause, suggested operational steps or key points for resolution, and the specific numerical value or range of the probability of occurrence. Furthermore, the diagnostic report may supplement necessary basic information, such as the corresponding vehicle identification number (VIN) and the time the fault diagnosis request was received, making the report more complete.

[0091] The final diagnostic report provides clear decision-making basis and operational guidance for maintenance personnel, vehicle management personnel and other relevant parties, helping them to quickly grasp the core situation of vehicle faults, efficiently formulate and implement repair plans, and also provide standardized document support for subsequent fault tracing and maintenance record retention.

[0092] Therefore, this embodiment of the application obtains a fault diagnosis request containing vehicle identification number (VIN) and fault description information, and then retrieves multi-source data corresponding to the target vehicle from the database based on the VIN and fault description information. This integrates various scattered vehicle-related data to form a complete and comprehensive dataset, solving the problem of scattered fault data and ineffective integration in traditional diagnostic models. Simultaneously, it abandons the traditional diagnostic approach that relies on the fixed logic of fault trees for analysis. Instead, it inputs the fault description information and the integrated multi-source data into a well-trained large language model. The large language model, combined with the fault type and relevant data of the vehicle under test, performs comprehensive analysis, no longer limited by a fixed diagnostic logic framework, breaking through the limitations of traditional diagnostic logic. Furthermore, it relies on a well-trained large language model... The efficient analytical capabilities of the large language model, combined with a standardized multi-source data retrieval and integration process, can quickly complete comprehensive fault analysis and output corresponding fault causes, matching solution suggestions, and occurrence probabilities. Based on the relevant analysis results, a diagnostic report is generated efficiently, reducing the processing time of the entire diagnostic process and effectively solving the problem of slow diagnostic response in the traditional remote diagnostic mode. At the same time, the occurrence probabilities of various fault causes output by the large language model can accurately quantify the likelihood of different faults occurring. Combined with the vehicle's multi-source operating data and historical fault-related data, potential fault risks of the vehicle can be predicted in advance, rather than just passively diagnosing faults that have already occurred. This effectively solves the technical problem of difficulty in achieving fault early warning in the traditional diagnostic mode.

[0093] In one possible implementation, the process of extracting and cleaning historical diagnostic reports and vehicle network data to obtain fault types, fault causes, and corresponding solutions is a key step in building training data for a large language model for vehicle fault diagnosis. The core objective is to filter and integrate accurate, standardized, and practically valuable structured fault-related information from massive amounts of raw data.

[0094] First, the structured elements of historical diagnostic reports and vehicle network data are extracted and processed to quickly capture fault-related structured data elements with fixed formats and clear identifiers from these two types of data. Historical diagnostic reports typically contain many standardized fields, such as fault occurrence time, vehicle mileage, fault codes, and diagnostic equipment models. These fields are all structured elements and can be accurately captured directly using preset data extraction rules. Vehicle network data covers various sensor data and system logs collected in real time during vehicle operation. The structured elements in this data include core parameters related to vehicle operating status and faults, such as engine speed, vehicle speed, oil pressure, battery voltage, and fault alarm trigger time.

[0095] This step utilizes professional data extraction tools or algorithms to traverse and filter the two types of data according to preset field rules, ensuring that the extracted structured data elements are complete and accurate, and avoiding the impact of missing key structured information on the subsequent data integration effect. These extracted structured data elements will serve as the basic framework for subsequently constructing fault-related data records, providing support for accurately locating the core fault information.

[0096] After extracting the structured elements, it is necessary to extract key fault-related information from the unstructured text content in historical diagnostic reports and vehicle networking data. Both types of data contain a large amount of unstructured text without a fixed format and scattered distribution. For example, in historical diagnostic reports, there may be textual content such as descriptions of fault phenomena by repair personnel, records of fault investigation processes, and summaries of repair experience; in vehicle networking data, there may be textual feedback from drivers on abnormal vehicle conditions and non-standardized fault prompt text generated by the system.

[0097] Because unstructured text information is fragmented and flexibly expressed, it is necessary to utilize tools such as natural language processing techniques, keyword matching algorithms, and semantic understanding models to accurately extract key information related to the fault from the text. The core information extracted includes the specific manifestation of the fault, the environmental conditions at the time of the fault, key points in the fault investigation process, and the maintenance personnel's initial judgment of the fault cause. This step effectively supplements the deficiencies of structured elements, uncovers deep fault information hidden in the text, and makes the fault-related data more complete and richer, providing a more comprehensive basis for subsequently identifying the fault type and analyzing the fault cause.

[0098] After extracting structured data elements and key information from unstructured text, the next step is association, integration, and reconstruction of structured data records. The core of this step is establishing logical connections between the two types of extracted information. This is achieved through common identifier fields, such as vehicle identification numbers or fault occurrence times, binding structured data elements corresponding to the same fault event with the extracted key information.

[0099] During the association process, the two types of information need to be cross-validated and logically sorted out to eliminate contradictory or unreasonable data content, ensuring the consistency and accuracy of the information. Subsequently, based on the sorted association information, a structured reconstruction is performed according to unified specifications and formats to generate standardized data records containing fault types, fault causes, and corresponding solutions. Fault types are categorized and defined based on extracted fault codes, fault phenomena, and other information; fault causes are summarized by combining structured parameter anomalies with troubleshooting conclusions from unstructured text; and solutions are organized and refined based on maintenance records, experience summaries, and other information. The resulting structured data records will achieve a systematic and standardized presentation of fault-related information, providing high-quality core data support for subsequent multi-source data fusion and large language model training.

[0100] See Figure 4 , Figure 4 The flowchart for multi-source information processing required for training the large language model provided in this application embodiment is based on three types of core data sources. Through information processing along different paths, comprehensive learning materials are finally delivered to the large language model. The specific process is as follows:

[0101] First, there is the input of the data source. The process covers three core information sources: one is technical information, including vehicle-related technical documents, maintenance manuals, etc.; the second is diagnostic-related data, such as diagnostic terminology, CAD data, signal matrix, and other information directly related to vehicle fault diagnosis; and the third is the after-sales problem historical experience database, which is a collection of experience records of handling vehicle after-sales faults in the past.

[0102] For technical data, the first step is to process the content into elements, breaking it down into the smallest independent content units, such as fault terminology and component description fragments. Then, the content is extracted using an AI model, and effective information is filtered out from the elementized content using artificial intelligence tools. After that, the data is atomized, breaking down the extracted information into indivisible atomic-level data units, such as precise minimum information fragments like faulty components like the braking system.

[0103] For diagnostic data, we first sort out the relationships between elements, analyze the logical connections between information elements in diagnostic statements, CAD data, and signal matrices, such as the correspondence between CAD components and diagnostic signals; then we structure the data, organizing these relationships into a standardized structured data format, such as a table structure that corresponds to components and signals.

[0104] Based on the historical experience database of after-sales issues, the system first processes the data using fault trees and guided troubleshooting methods. The historical after-sales issues are organized into a fault tree structure according to the fault level, clarifying the hierarchical relationship between fault phenomena, related components, and potential causes. Then, the troubleshooting approach is streamlined, transforming the fault tree into a clear logical chain for fault troubleshooting, such as a step-by-step troubleshooting path from fault phenomena to core causes.

[0105] Finally, the atomic data, structured data, and troubleshooting information processed through the above path will be aggregated and fed into the large language model, allowing the model to perform deep learning on this comprehensive information, thereby enriching the model's knowledge reserves and analytical capabilities in scenarios such as vehicle fault diagnosis and maintenance guidance.

[0106] In one possible implementation, see Figure 5 , Figure 5 This application provides an iterative optimization flowchart illustrating the result processing and iterative optimization process of a large language model in a vehicle fault diagnosis scenario. After the large language model completes deep learning by combining previously integrated comprehensive information and outputs corresponding comparative analysis results, these results proceed to the subsequent judgment stage. Combining expert judgment, the production batch of the target vehicle, and the frequency of similar faults, the probability of the corresponding fault is calculated, and appropriate fault diagnosis and repair suggestions are output. In other words, the various fault causes output by the large language model are obtained according to their corresponding probabilities of occurrence.

[0107] Next, the various fault causes output by the large language model will be sorted in descending order of their respective probabilities of occurrence. The two fault causes with the highest probabilities will be selected, and their probabilities will be summed to obtain the corresponding probability sum. If the calculated probability sum is lower than a preset threshold, a corresponding diagnostic case will be generated based on the relevant information from this fault diagnosis request, combined with all fault causes output by the large language model, corresponding solution suggestions, and their probabilities of occurrence.

[0108] The preset threshold can be set to 80%. If the calculated result of the sum of these probabilities is lower than the preset threshold, a corresponding diagnostic case is generated based on the relevant information from this fault diagnosis request, combined with all fault causes, corresponding solution suggestions, and occurrence probabilities output by the large language model. After the diagnostic case is generated, it is pushed to the human end. Professional human analysts conduct a comprehensive analysis based on the production batch information of the target vehicle and the actual frequency of similar faults in similar vehicles. Through professional analysis and verification, accurate fault location is achieved, ultimately yielding a human judgment result containing a clear fault location conclusion.

[0109] In one possible implementation, after the diagnostic report has undergone adoption or correction processing, the correction feedback data corresponding to the diagnostic results is included in the training samples. This correction feedback data is then used to conduct targeted training and optimization of the large language model. Simultaneously, based on manually completed fault-related judgments, technical experts will manually correct the current diagnostic cases, simultaneously and accurately labeling the corresponding data. All this corrected and labeled content is fed back into the large language model's training system for a new round of deep learning. Relying on this type of authentic and verified high-quality data, the large language model can continuously iterate and optimize itself, constantly improving its internal fault diagnosis logic and data matching system, gradually enhancing the accuracy and effectiveness of its analysis in subsequent vehicle fault diagnosis scenarios.

[0110] In summary, in this embodiment, the user can directly input a fault description and fault code. After receiving this input information, the system's diagnostic big data model quickly completes predictive analysis and then returns the corresponding fault cause and solution suggestions to the user. The response time of this diagnostic process is no more than 3 seconds, and the fault diagnosis accuracy is no less than 85%, providing users with reliable fault handling guidance quickly. Simultaneously, the system is designed with a continuous iterative optimization mechanism. Data generated throughout the diagnostic process is collected via a remote platform and fed back to the diagnostic big data model. Relying on deep learning algorithms, the big data model can continuously learn from the feedback data. During this process, expert diagnostic intervention is introduced to correct and verify the diagnostic results, ensuring the accuracy of the model's learning data. As data accumulates, the fault diagnosis capability of the diagnostic big data model will gradually improve, continuously optimizing the system's diagnostic performance and applicability.

[0111] The above are some specific implementations of the vehicle fault diagnosis method provided in the embodiments of this application. Based on this, this application also provides a corresponding vehicle fault diagnosis system. The system provided in the embodiments of this application will be described below from the perspective of functional modularity. Figure 3 This is a structural diagram of an automotive fault diagnosis system provided in an embodiment of this application.

[0112] The system includes:

[0113] The first acquisition unit 110 is used to acquire a fault diagnosis request, the fault diagnosis request including a vehicle identification code and fault description information;

[0114] The second acquisition unit 111 is used to acquire multi-source data related to the target vehicle indicated by the vehicle identification code from the database based on the vehicle identification code and the fault description information.

[0115] Output unit 112 is used to input the fault description information and the multi-source data into a fully trained large language model, so that the fully trained large language model can analyze the fault type and the data related to the vehicle under test, and output at least one fault cause, as well as the corresponding solution suggestion and occurrence probability of the fault cause.

[0116] The generation unit 113 is used to generate a diagnostic report based on the cause of the fault, the suggested solution, and the probability of occurrence.

[0117] This application also provides corresponding devices and computer storage media for implementing the vehicle fault diagnosis solution provided in this application.

[0118] The device includes a memory and a processor. The memory is used to store instructions or code, and the processor is used to execute the instructions or code to enable the device to perform the vehicle fault diagnosis method according to any embodiment of this application.

[0119] The computer storage medium stores code, and when the code is run, the device running the code implements the vehicle fault diagnosis method described in any embodiment of this application.

[0120] See Figure 7 , Figure 7 The schematic diagram of the computer hardware system provided in this application illustrates the basic architecture of the computer hardware system. The functions and positions of each component in the diagram include:

[0121] The memory is the core of a computer's storage system, consisting of two parts: internal memory and storage modules. Storage modules are responsible for carrying the operating system, computer programs, and non-volatile storage media. Non-volatile storage media can retain various types of data for a long time, preventing information loss due to power outages. Internal memory, on the other hand, is used to temporarily cache intermediate data generated during computer operation and currently executing program instructions, helping the processor retrieve information more efficiently and improving processing speed.

[0122] The processor is the central hub for computation and control of the entire computer system, equivalent to the "brain" of the device. It undertakes core tasks such as instruction parsing, data processing, and coordinated scheduling of hardware components, and is the core component that drives the computer to complete various functions.

[0123] The system bus is a communication link connecting core hardware such as the processor, internal memory, and memory. It is equivalent to a data channel between various components, responsible for transmitting data, instructions, and control signals between different hardware components, ensuring that each component can work together in an orderly manner.

[0124] Input / output interfaces are the central hub for interaction between computers and external devices. They play a crucial role in coordinating data transmission between external devices and the core components of the computer, serving as the key connection point for external information input and internal processing results output.

[0125] Among the external devices connected by the input / output interface, the input device is a tool for transmitting information to the computer, such as a keyboard and mouse, which can input user operation commands or external data into the computer; the communication interface is used to realize the connection between the computer and other devices or networks, supporting remote data transmission and interaction between devices; the display unit is the visual information output carrier of the computer, such as a monitor, which can present the results processed by the computer to the user in an intuitive form, making it easy for the user to obtain and view information.

[0126] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0127] It should be understood that in this application, "at least one" refers to one or more items, and "more" refers to two or more items. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one" of a, b, or c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0128] It should be understood that the terms center, longitudinal, transverse, up, down, front, back, left, right, vertical, horizontal, top, bottom, inside, outside, etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0129] It should be noted that, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0130] It should also be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the statement "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0131] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0132] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for diagnosing automotive faults, characterized in that, include: Obtain a fault diagnosis request, which includes a vehicle identification number and fault description information; Based on the vehicle identification code and the fault description information, multi-source data related to the target vehicle indicated by the vehicle identification code is obtained from the database; The fault description information and the multi-source data are input into a fully trained large language model, so that the fully trained large language model can analyze the fault type and the data related to the vehicle under test, and output at least one fault cause, as well as the corresponding solution suggestions and occurrence probability of the fault cause. A diagnostic report is generated based on the cause of the fault, the suggested solutions, and the probability of occurrence.

2. The method according to claim 1, characterized in that, The methods for determining the fully trained large language model include: Obtain historical diagnostic reports and vehicle networking data for the sample vehicles; Data extraction and cleaning are performed on the historical diagnostic reports and vehicle network data to obtain the fault type, fault cause and corresponding solution; The fault type, fault cause, and corresponding solution are fused with data from at least one of the following systems: vehicle production management system, dealer management system, customer relationship management system, and technical information system. The fused multi-source data is input into the large language model for training. Data with confidence levels below a preset threshold during training is corrected, and the corrected data is then re-input into the large language model for iterative training until a fully trained large language model is obtained.

3. The method according to claim 1, characterized in that, The method further includes: Select the two most probable causes of failure from the output results and calculate the sum of their probabilities. If the calculation result is lower than a preset threshold, a diagnostic case is generated based on the fault diagnosis request and the output of the large language model. The diagnostic cases are pushed to the human end so that the human can make a judgment based on the production batch and the frequency of similar failures, and obtain a human judgment result including the fault location conclusion. Based on the results of the manual judgment, the causes of the faults and the solutions suggested in the diagnostic report are adjusted.

4. The method according to claim 2, characterized in that, The process of extracting and cleaning data from the historical diagnostic reports and vehicle network data to obtain fault types, fault causes, and corresponding solutions includes: The historical diagnostic reports and vehicle network data are processed by structured element extraction to obtain structured data elements related to the fault. Extract key information related to the fault from the unstructured text content in the historical diagnostic reports and vehicle network data; The structured data elements are associated and integrated with the key information to reconstruct and generate a structured data record containing the fault type, fault cause, and corresponding solution.

5. The method according to claim 1, characterized in that, The method further includes: After the diagnostic report is adopted or corrected, the corrected feedback data of the diagnostic results is used as training samples to train the large language model.

6. The method according to claim 1, characterized in that, The method further includes: Real-time monitoring data of the target vehicle is obtained based on the vehicle-to-everything (V2X) platform; The real-time monitoring data is analyzed to obtain the current health status indicators; The current health status indicators are compared and analyzed with the historical health status of the target vehicle and / or the benchmark health status of the same or similar vehicle models. When the comparative analysis results exceed the preset threshold, an early warning message is generated and output.

7. A vehicle fault diagnosis system, characterized in that, include: The first acquisition unit is used to acquire a fault diagnosis request, the fault diagnosis request including a vehicle identification code and fault description information; The second acquisition unit is used to acquire multi-source data related to the target vehicle indicated by the vehicle identification code from the database based on the vehicle identification code and the fault description information. The output unit is used to input the fault description information and the multi-source data into a fully trained large language model, so that the fully trained large language model can analyze the fault type and the data related to the vehicle under test, and output at least one fault cause, as well as the corresponding solution suggestion and occurrence probability of the fault cause. The generation unit is used to generate a diagnostic report based on the cause of the fault, the suggested solutions, and the probability of occurrence.

8. A computer program product, characterized in that, The computer program product stores instructions that, when executed on a terminal device, cause the terminal device to perform the vehicle fault diagnosis method as described in any one of claims 1-6.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the vehicle fault diagnosis method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the vehicle fault diagnosis method as described in any one of claims 1-6.

Images