Artificial intelligence agent-based vehicle fault analysis method, system, medium, and device

Abstract

The application provides a vehicle fault analysis method, system, medium and equipment based on an artificial intelligence agent, the method comprising: acquiring multi-source heterogeneous data of a vehicle; constructing a vehicle fault knowledge base and a vehicle maintenance knowledge base; based on the vehicle fault knowledge base, acquiring fault object-fault mode association information corresponding to the multi-source heterogeneous data; performing deduplication and aggregation on the fault object-fault mode association information to acquire effective fault object-fault mode association information; based on the vehicle fault knowledge base, acquiring fault causes corresponding to the effective fault object-fault mode association information; based on the vehicle maintenance knowledge base, acquiring maintenance schemes corresponding to the effective fault object-fault mode association information; and based on the fault causes and the maintenance schemes, generating a vehicle fault analysis report by using a large language model. The vehicle fault analysis method, system, medium and equipment based on the artificial intelligence agent realize automatic analysis of faults based on multi-source heterogeneous data of the vehicle.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of vehicle diagnosis, and particularly relates to a vehicle fault analysis method, system, medium and equipment based on an artificial intelligence agent. BACKGROUND

[0002] With the continuous growth of the number of motor vehicles and the increasing requirements for vehicle emission control, the efficiency of vehicle fault diagnosis and maintenance is of great significance to ensure road traffic safety and improve emission compliance. Currently, motor vehicle remote diagnosis mainly relies on fault code information (such as J2012 and J1939 protocols conforming to SAE standards) provided by the on-board diagnostic system (OBD) for preliminary judgment.

[0003] However, the existing diagnosis system has the following significant problems, which need to be improved by technical innovation.

[0004] (1) The traditional diagnosis method does not make full use of dynamic monitoring data, and the diagnosis ability is limited. The existing system generally focuses on reading fault codes (DTC) as the basis for judgment, but fails to effectively integrate and analyze a large amount of dynamic monitoring data generated during vehicle operation (such as engine speed, load, fuel consumption, aftertreatment efficiency, and emission concentration). This leads to the inability to accurately identify potential faults based on actual operating conditions, especially when the emission aftertreatment system (such as selective catalytic reduction system SCR, diesel particulate filter DPF) has complex problems such as efficiency decline, abnormal regeneration, and urea injection failure. It is difficult to make a comprehensive judgment based on static fault codes alone.

[0005] (2) The enterprise self-defined fault code system is complex and diverse, with low standardization and difficulty in unified identification. In addition to the general J2012 and J1939 fault code standards, different vehicle manufacturers and engine manufacturers (such as Weichai, Yuchai, Cummins, etc.) generally establish their own proprietary fault code systems, which may use different coding rules and meanings for the same fault. This multi-source and heterogeneous fault code system increases the difficulty of diagnosis, and the traditional static fault code table cannot meet the needs of unified identification and interpretation.

[0006] (3) The causes of faults and maintenance recommendations are dynamically changing, and static knowledge bases are difficult to meet actual needs. With the increasing complexity of vehicle operating conditions and the continuous application of new technologies, the causes of some faults may change due to factors such as environment, fuel, and vehicle aging. In addition, maintenance experience and methods are constantly accumulating and updating. Traditional fault handling methods based on fixed rule matching or manual judgment are difficult to reflect new fault patterns and maintenance paths in actual operation in a timely manner.

[0007] (4) Maintenance recommendations still rely on personal experience, lack of systematic knowledge support. In current maintenance practice, maintenance personnel mostly rely on personal experience or subjective judgment to handle problems, lack of systematic integration of knowledge sources such as maintenance manuals, official operation guidelines, network platforms (such as maintenance communities, forums) of the main engine factory. The uneven level of experience leads to unstable maintenance quality, and even may misjudge or cause repeated maintenance, increasing maintenance cost.

[0008] In summary, the existing vehicle fault diagnosis and maintenance recommendation system still has problems such as insufficient information fusion, delayed knowledge update, and insufficient analysis capability. SUMMARY

[0009] In view of the above-mentioned shortcomings of the prior art, the purpose of the present application is to provide a vehicle fault analysis method, system, medium and equipment based on artificial intelligence agent, which realizes the automatic analysis of vehicle faults based on the multi-source heterogeneous data of the vehicle, thereby providing accurate and comprehensive diagnostic reports and maintenance recommendations.

[0010] In a first aspect, the present application provides a vehicle fault analysis method based on an artificial intelligence agent, the method comprising the following steps: acquiring multi-source heterogeneous data of a vehicle; constructing a vehicle fault knowledge base and a vehicle maintenance knowledge base; the vehicle fault knowledge base is used to record the association information of fault object-fault mode-fault cause; the vehicle maintenance knowledge base is used to record the association information of fault object-fault mode-maintenance scheme; based on the vehicle fault knowledge base, acquiring the fault object-fault mode association information corresponding to the multi-source heterogeneous data; de-duplicating and aggregating the fault object-fault mode association information to obtain valid fault object-fault mode association information; based on the vehicle fault knowledge base, acquiring the fault cause corresponding to the valid fault object-fault mode association information; based on the vehicle maintenance knowledge base, acquiring the maintenance scheme corresponding to the valid fault object-fault mode association information; based on the fault cause and the maintenance scheme, generating a vehicle fault analysis report using a large language model.

[0011] In an implementation form of the first aspect, the multi-source heterogeneous data includes remote online monitoring data, fault code list, engine brand, fault protocol type, and textual fault description.

[0012] In an implementation form of the first aspect, based on the vehicle fault knowledge base, acquiring the fault object-fault mode association information corresponding to the multi-source heterogeneous data comprises the following steps:

[0013] Based on the calibration code ID and the calibration verification number in the electronic control unit of the vehicle, it is judged whether the electronic control unit is tampered with, so as to obtain the fault object-fault mode association information of the electronic control unit for tampering;

[0014] extracting corresponding fault object-fault mode association information based on the online monitoring data;

[0015] extracting fault object-fault symptom association information based on the text fault description;

[0016] finding corresponding fault object-fault mode association information in the vehicle fault knowledge base based on the fault code in the fault code list, the engine brand, and the fault protocol type.

[0017] In an implementation form of the first aspect, it is judged whether the electronic control unit is tampered with according to whether an edit distance offset degree of the calibration code ID and the calibration verification number exceeds a preset value.

[0018] In an implementation form of the first aspect, constructing the vehicle fault knowledge base comprises the following steps:

[0019] obtaining a fault code database of the vehicle; the fault code database records fault codes, fault descriptions, and fault causes of each fault;

[0020] based on the fault code database, using a large language model to generate fault object-fault mode-fault cause association information for each fault.

[0021] In an implementation form of the first aspect, constructing the vehicle maintenance knowledge base comprises the following steps:

[0022] obtaining a vehicle maintenance manual, a vehicle technical bulletin, and a maintenance case;

[0023] converting the vehicle maintenance manual, the vehicle technical bulletin, and the maintenance case into pure text information;

[0024] constructing a prompt word template, the prompt word being used to extract fault and maintenance related information in the pure text information and output in a preset format;

[0025] based on the pure text information and the prompt word template, using a large language model to generate fault object-fault mode-maintenance scheme association information.

[0026] In an implementation form of the first aspect, the fault object-fault mode association information is de-duplicated and aggregated to obtain valid fault object-fault mode association information, comprising the following steps:

[0027] vectorizing the fault object-fault mode association information based on an SBERT model or a Transformer model;

[0028] clustering the vectorized fault object-fault mode association information to obtain a plurality of fault clusters;

[0029] For each fault cluster, select the fault object-fault mode association information corresponding to one fault as the effective fault object-fault mode association information.

[0030] In a second aspect, the present application provides a vehicle fault analysis system based on an artificial intelligence agent, the system comprising a first acquisition module, a construction module, a second acquisition module, a third acquisition module, a fourth acquisition module, a fifth acquisition module, and an analysis module.

[0031] The first acquisition module is configured to acquire multi-source heterogeneous data of a vehicle.

[0032] The construction module is configured to construct a vehicle fault knowledge base and a vehicle maintenance knowledge base; the vehicle fault knowledge base is configured to record fault object-fault mode-fault cause association information; and the vehicle maintenance knowledge base is configured to record fault object-fault mode-maintenance scheme association information.

[0033] The second acquisition module is configured to acquire, based on the vehicle fault knowledge base, fault object-fault mode association information corresponding to the multi-source heterogeneous data.

[0034] The third acquisition module is configured to de-duplicate and aggregate the fault object-fault mode association information to acquire effective fault object-fault mode association information.

[0035] The fourth acquisition module is configured to acquire, based on the vehicle fault knowledge base, fault causes corresponding to the effective fault object-fault mode association information.

[0036] The fifth acquisition module is configured to acquire, based on the vehicle maintenance knowledge base, maintenance schemes corresponding to the effective fault object-fault mode association information.

[0037] The analysis module is configured to generate, based on the fault causes and the maintenance schemes, a vehicle fault analysis report using a large language model.

[0038] In a third aspect, the present application provides an electronic device, comprising a processor and a memory.

[0039] The memory is configured to store a computer program.

[0040] The processor is configured to execute the computer program stored in the memory, so that the electronic device executes the vehicle fault analysis method based on an artificial intelligence agent as described above.

[0041] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, the program being executed by an electronic device to implement the vehicle fault analysis method based on an artificial intelligence agent as described above.

[0042] The vehicle fault analysis method, system, medium and equipment based on an artificial intelligence agent according to the present application have the following beneficial effects.

[0043] (1) Based on a large language model and an artificial intelligence agent, the vehicle fault analysis method, system, medium and equipment according to the present application can realize automatic and deep vehicle fault analysis according to multi-source heterogeneous data from a vehicle (including static fault codes, dynamic monitoring data and user descriptions), and can obtain accurate and comprehensive diagnostic reports and maintenance suggestions.

[0044] (2) The vehicle fault analysis method, system, medium and equipment according to the present application can fuse dynamic monitoring data, standard and custom fault codes, real-time semantic analysis and knowledge base reasoning capabilities, realize multi-level and dynamic fault diagnosis, and thus promote the realization of vehicle intelligent diagnosis and remote fine operation and maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 A flowchart showing the vehicle fault analysis method based on an artificial intelligence agent according to the present application in an embodiment is shown.

[0046] Figure 2 A structural schematic diagram showing the vehicle fault analysis system based on an artificial intelligence agent according to the present application in an embodiment is shown.

[0047] Figure 3 A structural schematic diagram showing the electronic device according to the present application in an embodiment is shown. DETAILED DESCRIPTION

[0048] The embodiments of the present application will be described in detail below with reference to specific examples. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. The present application can also be implemented or applied in other different specific embodiments, and the details in the specification can be modified or changed in various ways based on different views and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0049] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present application in a schematic manner, and the diagrams only show the components related to the present application, not the number, shape and size of the components when actually implemented. The actual implementation of each component may be a random change in number, shape and size, and the layout pattern of the components may be more complex.

[0050] Remote online monitoring technology has the potential to intelligently diagnose and push maintenance recommendations for a large number of vehicles in a region. In recent years, with the construction and popularization of remote emission monitoring platforms for heavy diesel vehicles in many places, the running status, emission level and fault information of vehicles have been continuously, real-time and online collected. If these data can be used in combination with the semantic understanding and reasoning ability of artificial intelligence technology and large models, intelligent diagnosis and maintenance recommendation pushing for a large number of vehicles in a region can be realized, breaking through the limitations of traditional manual vehicle-by-vehicle analysis and greatly improving the diagnosis efficiency and accuracy.

[0051] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application.

[0052] As shown in the drawings, Figure 1 In one embodiment, the vehicle fault analysis method based on artificial intelligence agent of the present application includes steps S1-S7.

[0053] Step S1, acquiring multi-source heterogeneous data of the vehicle.

[0054] Specifically, the present application collects vehicle fault data from various sources to form the multi-source heterogeneous data to ensure the comprehensiveness of the diagnosis information. In one embodiment, the multi-source heterogeneous data includes remote online monitoring data, fault code list, engine brand, fault protocol type and textual fault description.

[0055] The fault protocol type and the engine brand are the initial context information of the vehicle fault diagnosis process, such as the communication protocol standard (such as CAN bus protocol) of the vehicle and the specific model of the engine. These information is crucial for limiting the search range of subsequent vehicle fault knowledge base or as the context of large language model to generate prompt words. Table 1 shows the static basic information of the vehicle, based on which the vehicle type, engine model, bus protocol and other information can be obtained.

[0056] Table 1, static basic information of the vehicle

[0057] Serial number Item 1 License plate number 2 VIN code 3 Vehicle model 4 Total mass 5 Vehicle type 6 Engine model 7 Engine power 8 Engine displacement 9 Vehicle CAN bus protocol .

[0058] The standardized (J1939 / J2012) and enterprise-defined fault code list generated by the on-board diagnostic system. The fault codes in it follow industry standards such as SAE J1939 and SAE J2012, as well as enterprise-specific fault codes. The present invention can quickly obtain the fault code-fault object-fault symptom association information by calling an artificial intelligence agent to search for these fault codes one by one in the fault knowledge base. The information provided by the on-board diagnostic system is shown in Table 2. In the present invention, the fault-related information of all vehicles in the region is read according to the requirements of “Heavy-duty Diesel Vehicle Emission Limit Values and Measurement Methods (China Phase VI)” (GB 17691-2018).

[0059] Table 2, information provided by the on-board diagnostic system

[0060] Item Data type 1 OBD diagnostic protocol BYTE 2 MIL status BYTE 3 Diagnostic readiness status WORD 4 Vehicle identification number (VIN) STRING 5 Software calibration identification number (CAL ID) STRING 6 Calibration verification number (CVN) STRING 7 IUPR value DSTRING 8 Total number of fault codes BYTE 9 Fault code information list N*BYTE(4) .

[0061] The remote online monitoring data refers to various performance parameters and sensor readings recorded in real time or historically during the operation of the vehicle. These data reflect the actual operating state of each component of the vehicle. The present invention can convert the problems reflected in these online data into vectorized fault object-fault mode association information by calling an artificial intelligence agent, thereby converting dynamic, numerical data into structured information that can be processed by the diagnostic system. For example, the remote online monitoring data of all vehicles in the region is read according to the requirements of “Heavy-duty Diesel Vehicle Emission Limit Values and Measurement Methods (China Phase VI)” (GB 17691-2018). The remote online monitoring data is shown in Table 3.

[0062] Table 3, remote online monitoring data

[0063] Serial number Data item Unit Data type 1 Vehicle speed km / h WORD 2 Atmospheric pressure kPa BYTE 3 Engine net output torque % BYTE 4 Friction torque % BYTE 5 Engine speed r / min WORD 6 Engine fuel flow L / h WORD 7 SCR upstream NOx sensor output value 10 -6 ]] WORD 8 SCR downstream NOx sensor output value 10 -6 ]] WORD 9 Reagent remaining amount % BYTE 10 Intake air mass kg / h WORD 11 SCR inlet temperature ℃ WORD 12 SCR outlet temperature ℃ WORD 13 DPF differential pressure kPa WORD 14 Engine coolant temperature ℃ BYTE 15 Fuel tank level % BYTE 16 Positioning status BYTE 17 Longitude ° DWORD 18 Latitude ° DWORD 19 Cumulative mileage km DWORD .

[0064] The textual fault description comes from the user's unstructured text description, such as sensory information such as “abnormal noise” and “smoke” input through a guided interface.

[0065] Step S2, constructing a vehicle fault knowledge base and a vehicle maintenance knowledge base; the vehicle fault knowledge base is used to record fault object-fault mode-fault cause association information; the vehicle maintenance knowledge base is used to record fault object-fault mode-maintenance scheme association information.

[0066] Specifically, the semantic understanding ability of a large language model (LLM) is used to process traditional, unstructured vehicle fault and maintenance data into machine-readable, highly structured vehicle fault knowledge base and vehicle maintenance knowledge base.

[0067] In an embodiment, the step of building a vehicle fault knowledge base comprises the following steps.

[0068] 21) Obtain a database of fault codes for the vehicle; the database of fault codes records the fault code, fault description and fault cause for each fault.

[0069] Wherein, the fault code database (including fault code, fault description, fault cause) from SAE standard and enterprise-owned fault code database is taken as the original data.

[0070] For example, for the SAE J1939 database (SAE J1939 fault database, SAE J1939 system enterprise special fault library), it specifically stores fault codes conforming to the SAE J1939 standard and enterprise special fault codes conforming to the system. The fault database contains the following fields: 1, serial number; 2, fault code; 3, fault description; 4, fault cause.

[0071] For the database SAE J2012 (SAE J2012 fault database, SAE J2012 system enterprise special fault library), it is similar to the J1939 database. The fault database stores fault codes conforming to the SAE J2012 standard and enterprise special fault codes conforming to the system. The fault database contains the following fields: 1, serial number; 2, fault code; 3, fault description; 4, fault cause.

[0072] It should be noted that the fault database needs to be formatted before being used as input for the large language model. For example, for the source data: 45, P0135, O2 sensor heater circuit fault (row 1, sensor 1), heater element damage or open circuit, it can be spliced into the following input text: Fault code: P0135. Fault description: O2 sensor heater circuit fault (row 1, sensor 1). Fault cause: heater element damage or open circuit.

[0073] 22) Based on the fault code database, a large language model is used to generate fault object-fault mode-fault cause association information for each fault.

[0074] Wherein, the fault object refers to the physical entity (such as oxygen sensor, EGR valve) or logical entity (such as CAN communication bus, ignition control module software) that occurs a fault. The fault mode refers to the state or behavior (such as signal voltage is too high, stuck in the open position, communication loss, slow response) that the fault object specifically exhibits. The fault cause refers to the reason (such as CAN line breakage, sensor aging) for the fault object to exhibit the fault mode.

[0075] Design a specific prompt template to guide the large language model to automatically decompose and extract standard "fault object-fault mode-fault reason" triplet association information from each fault record. The prompt template needs to teach the large language model how to accurately identify "who is broken (fault object) and "how is broken" (fault mode), "specific reason" (fault reason). For example, the prompt is: You are a professional automobile fault diagnosis engineer, proficient in decomposing complex fault descriptions into standardized "fault object" and "fault mode". Your task is to analyze the fault data I provide and strictly follow the specified JSON format to extract "fault code", "fault object" and "fault mode". Among them, the fault object should be a specific physical or logical component. The fault mode should be the specific failure state of the component. Both should be extracted directly from the provided information, without making assumptions.

[0076] For example, the generation process of fault object-fault mode-fault reason association information is as follows:

[0077] **Case 1:**

[0078] **Input text:**

[0079] "Fault code: P0108. Fault description: manifold absolute pressure / atmospheric pressure sensor circuit high input. Fault reason: internal short circuit of sensor to power supply."

[0080] **Output JSON:**

[0081] {

[0082] "fault_code": "P0108",

[0083] "fault_object": "manifold absolute pressure / atmospheric pressure sensor circuit",

[0084] "fault_mode": "high input (short circuit to power supply)"

[0085] "fault_reason": "power supply short circuit"

[0086] }。

[0087] **Case 2:**

[0088] **Input text:**

[0089] "Fault code: P0404. Fault description: exhaust gas recirculation (EGR) control circuit range / performance. Fault reason: EGR valve stuck due to carbon deposition, unable to reach target position."

[0090] **Output JSON:**

[0091] {

[0092] "fault_code": "P0404",

[0093] "fault_object": "Exhaust Gas Recirculation (EGR) valve",

[0094] "fault_mode": "Control range / performance issue"

[0095] "fault_reason": "Carbon deposit stuck"

[0096] }.

[0097] **Case 3:**

[0098] **Input Text:**

[0099] "Fault Code: U0100. Fault Description: Loss of communication with ECM / PCM 'A'. Fault Cause: CAN bus line open circuit or ECM module failure."

[0100] **Output JSON:**

[0101] {

[0102] "fault_code": "U0100",

[0103] "fault_object": "Communication link with ECM / PCM 'A'",

[0104] "fault_mode": "Loss of communication"

[0105] "fault_reason": "CAN bus line open circuit"}.

[0106] In one embodiment, building a vehicle repair knowledge base includes the following steps.

[0107] 21a) Obtain vehicle repair manuals, vehicle technical service bulletins, and repair cases.

[0108] Among them, the official repair manuals of the target engine and the whole vehicle production enterprise are obtained. These manuals may be in PDF, XML, HTML or proprietary format. Collect relevant technical service bulletins (TSB), repair cases, etc. as supplementary data.

[0109] 22b) Convert the vehicle repair manuals, vehicle technical service bulletins, and repair cases into pure text information.

[0110] wherein the above data is pre-processed: (1) Format conversion, converting all documents into plain text format. For PDF files, high-precision OCR (Optical Character Recognition) tools are needed, and special attention should be paid to preserving the structure of tables and lists. (2) Text cleaning, removing noise information such as headers, footers, watermarks, and irrelevant picture descriptions. (3) Text chunking, dividing lengthy manuals into smaller, more easily processed by LLM text blocks according to chapters, fault codes (DTC) or logical units (such as "cooling system fault diagnosis"). A text block should contain a relatively complete fault description and repair process.

[0111] 23c) Constructing a prompt template, which is a prompt for extracting fault and repair related information from plain text information and outputting in a preset format.

[0112] wherein through a highly structured prompt template containing role definition, format constraints, a small number of sample demonstrations and dynamic input areas, the large language model is guided to accurately extract structured data containing "fault object", "fault mode" and repair scheme (including required tools, spare parts, steps and safety precautions). The structured data extracted by the large language model, such as fault object, fault mode, and required tools, required spare parts, repair steps and safety precautions in the repair scheme, are respectively stored in the corresponding fields of the vehicle repair knowledge base, thereby forming a structured vehicle repair knowledge base.

[0113] The prompt template has the following contents.

[0114] I. Role and Task Definition Zone

[0115] Purpose: To set the precise identity, task goal and behavior criteria for the large language model. This is the highest level of instruction, setting the tone for all subsequent behavior.

[0116] Composition: (a) Persona: Clearly state its expert identity. For example: "You are an experienced automobile repair technical expert and data analyst proficient in various vehicle models."

[0117] (b) Core Task: Clearly describe the work it needs to do. For example: "Your only task is to extract key information about a specific fault from the repair manual text fragments I provide."

[0118] (c) Behavioral Constraints: Set strict rules to exclude unnecessary creativity, inference, or conversational behavior. For example: "You must strictly follow the JSON format as I defined. Absolutely no interpretation, comment, or text unrelated to the JSON format can be added. If the information for a certain field does not exist in the original text, use the specified value (e.g., 'not mentioned'). Do not infer or create information on your own."

[0119] II. Output Format Constraint Zone.

[0120] Purpose: Define the "skeleton" of data extraction, ensuring that the output result is machine-readable, structured data that can be directly parsed by programs. This is the basis for automation.

[0121] Composition:

[0122] Format Declaration: Explicitly require the use of JSON format.

[0123] Schema Definition: Provide a complete JSON structure template with annotations. Annotations are used to explain the meaning of each field to the LLM, ensuring accurate understanding.

[0124] JSON

[0125] {

[0126] / / (a) Macroscopic component or system where the fault occurs, for example: SCR aftertreatment system, engine cooling system

[0127] "fault_object": "string",

[0128] / / (b) Specific description of the fault, which may include fault codes (DTC) and phenomena, for example: DTC P20EE - SCR catalyst efficiency low

[0129] "fault_mode": "string",

[0130] / / (c) Object containing all repair solution information

[0131] "repair_solution": {

[0132] / / (c1) List of all tools needed during repair

[0133] "required_tools": [

[0134] "string"

[0135] ],

[0136] / / (c2) List of spare parts needed for replacement, should include part number if possible

[0137] "required_spare_parts": [

[0138] "string (part number)"

[0139] ],

[0140] / / (c3) List of detailed, step-by-step repair procedures, each step is a separate string

[0141] "repair_steps": [

[0142] "string"

[0143] ],

[0144] / / List of safety precautions that must be followed during the repair process

[0145] "safety_precautions": [

[0146] "string" ]}}.

[0147] 24d) Based on the pure text information and the prompt word template, a large language model is used to generate fault object-fault mode-repair scheme association information.

[0148] In addition, the present application also provides a few-shot demonstration zone (Few-shot Demonstration Zone) for "pre-job training" of the large language model. By providing 1 to 3 high-quality "input -> output" examples, the large language model learns how to apply the rules defined in the output role and task definition area and format constraint area in practice.

[0149] The examples include the following.

[0150] Input text: A real, pre-processed repair manual text.

[0151] Output JSON: "Perfect answer" manually written by human experts according to the input text, strictly following the JSON structure defined in the format constraint area.

[0152] Examples are the key to improving extraction accuracy. The language in the repair manual can have multiple interpretations, and examples can teach large language models how to choose the most appropriate understanding in a specific context. Large language models can learn the names of specific parts, tools, and typical representations of repair steps through examples. Compared to only looking at format definitions, actually seeing a filled-in example can help large language models better internalize and follow the output format.

[0153] In one embodiment, when the "repair solution" based on the enterprise repair manual cannot provide a matching result for a given "fault object-fault mode", the invention will start the secondary reasoning generation mechanism of the vehicle repair knowledge base. This mechanism relies on a diesel vehicle fault knowledge base structured as "fault object-fault mode-fault cause", and uses large language models to intelligently and creatively generate a complete and structured repair recommendation based on known "fault causes".

[0154] In this embodiment, a prompt template is first generated. The core purpose of the prompt template is no longer extraction, but to guide large language models to perform logical reasoning and knowledge creation. The prompt template block includes the following content.

[0155] I. Role and Task Definition Zone.

[0156] Purpose: Upgrade the role of large language models from a "data analyst" to a "diagnostic master and technical trainer", and its task is no longer "extraction", but "reasoning and creation" based on limited information, and set a professional, rigorous and logical tone for subsequent behavior.

[0157] Composition:

[0158] (a) Persona: Give it a higher level of expert identity. For example: "You are a top diesel vehicle diagnostic master and technical trainer with more than 20 years of experience."

[0159] (b) Core Task: Clearly describe the creative work it needs to complete. For example: "Your task is to reason and generate a complete, structured, logical and safe preliminary repair recommendation based on the known 'fault object', 'fault mode' and 'fault cause' I provide."

[0160] (c) Behavioral Constraints & Reasoning Guide: In addition to the existing format constraints, add explicit logical reasoning frameworks to guide the thinking process. For example: "You must strictly follow the JSON format as I defined. In addition, your generation process should follow the following logical guidelines: 1. Confirmation and Diagnosis: First think about how to confirm the cause of the failure. 2. Maintenance Preparation: According to the cause of the failure, infer the required tools and spare parts. 3. Maintenance Steps: Generate steps in the order of 'preparation-disassembly-core maintenance-installation-testing'. 4. Safety First: Consider all potential risks and generate safety precautions."

[0161] II. Output Format Constraint Zone

[0162] Purpose: Ensure that the maintenance suggestions generated by large language models are fully compatible in data structure with standard suggestions extracted from manuals, so that the front-end application can uniformly process and display them. At the same time, add metadata fields to distinguish data sources and credibility.

[0163] Composition:

[0164] Format Declaration: Explicitly require the use of JSON format.

[0165] Schema Definition: Provide a complete JSON structure template with new metadata fields.

[0166] JSON

[0167] {

[0168] / / (a) Clearly indicate the source of this suggestion to distinguish it from official data in the maintenance manual

[0169] "generation_source": "AI-Generated Suggestion",

[0170] / / (b) LLM's reliability assessment based on the explicitness of the input information and its own knowledge base

[0171] "confidence_score": "High / Medium / Low",

[0172] / / (c) An object containing all maintenance plan information, consistent with the extracted template

[0173] "repair_suggestion": {

[0174] / / (c1) List of tools inferred from the cause of the failure and repair steps

[0175] "required_tools": [

[0176] "string"

[0177] ],

[0178] / / (c2) List of spare parts that might be required inferred from the cause of the failure

[0179] "required_spare_parts": [

[0180] "string (part number)"

[0181] ],

[0182] / / (c3) Step-by-step repair procedure generated from the inference logic

[0183] "repair_steps": [

[0184] "string"

[0185] ],

[0186] / / (c4) Safety precautions inferred based on the repair steps and the object

[0187] "safety_precautions": [

[0188] "string" ]

[0190] }

[0191] }。

[0192] Similarly, a small number of examples are also required in this embodiment to provide "advanced training" for the large language model, not only to teach the format, but also to teach a reasoning and creation process from "cause" to "solution". The examples show the large language model how to derive a set of complex "solutions" (such as: the need for cleaning agent, disassembly steps, cleaning methods, installation torque, etc.) from a simple "cause" (such as: too much carbon deposition), helping the large language model to establish the implicit association between "fault cause" and "required tools / spare parts". Through high-quality examples, ensure that the maintenance steps generated by the large language model conform to the actual operation logic and safety specifications. Accordingly, the structured data in the fault object, fault mode, and required tools, required spare parts, maintenance steps, and safety precautions in the derived maintenance solution of the large language model are stored in the corresponding fields of the vehicle maintenance knowledge base, thereby forming a structured vehicle maintenance knowledge base.

[0193] Step S3, based on the vehicle fault knowledge base, obtaining the fault object-fault mode association information corresponding to the multi-source heterogeneous data.

[0194] Specifically, based on the vehicle fault knowledge base, obtaining the fault object-fault mode association information corresponding to the multi-source heterogeneous data includes the following steps.

[0195] 31) Determine whether the electronic control unit (ECU) is tampered based on the calibration code ID and the calibration verification number in the electronic control unit of the vehicle, to obtain the fault object-fault mode association information of the electronic control unit for tampering.

[0196] Wherein, the string corresponding to the calibration code (CAL) ID and the calibration verification number (CVN) of the vehicle control unit is compared with the first recorded string value. If there is a change and the edit distance deviation of the first recorded string exceeds the set value (adjusted according to actual needs), it is considered that the vehicle ECU has been tampered with, forming the fault object-fault mode association information of "ECU tampered". Preferably, the edit distance deviation K(%)=max(Loriginal, Lnew) / D(Soriginal, Snew) x 100%, wherein Soriginal represents the original string recorded for the first time. Snew represents the new string obtained at present. D(Soriginal, Snew) represents the edit distance (Levenshtein Distance), which refers to the number of minimum single-character editing operations (insertion, deletion or replacement) required to convert Soriginal to Snew. max(Loriginal, Lnew) represents the length of the longer of the two strings. Using the length of the longer as the numerator, the result can be normalized to between 0% (completely the same) and 100% (completely different).

[0197] 32) extracting corresponding fault object-fault mode association information based on the online monitoring data.

[0198] In the present application, the online monitoring data is data-extracted in relation to the problems of key components, and then these online data problems are converted into vectorized fault object-fault mode association information by the Agent, so as to ensure that the real-time or historical monitoring data can be effectively integrated into the diagnosis process.

[0199] For the online monitoring data, it can be extracted as corresponding fault object-fault mode association information through single-index statistical rules, multi-parameter ratio rules, one-dimensional feature similarity rules, and two-dimensional feature similarity rules. The single-index statistical rule determines whether a single sensor data is abnormal based on a statistical parameter (such as quantile), for example, the quantile value of the NOx sensor upstream of the SCR is used to determine the failure degree of the SCR carrier. The multi-parameter ratio rule calculates the ratio between cumulative parameters (such as cumulative mileage, cumulative fuel consumption) to form key diagnostic indicators (such as unit power NOx emission factor, NOx conversion rate, urea fuel consumption ratio) to evaluate system performance. The feature similarity rule uses Lin's consistency correlation coefficient (CCC) algorithm to compare the data consistency of vehicle key parameters and standard vehicles under different working conditions to evaluate the performance of thermal management, etc.

[0200] Specifically, the single-index statistical rule counts a piece of data as an index, and determines the severity of the problem according to the value range of the index. Possible statistical values include mean, median, and x quantile value. The statistical indicators and statistical values are shown in Table 4.

[0201] Table 4, statistical indicators and statistical values of single-index statistical rules

[0202]

[0203] Correspondingly, as shown in Tables 5 and 6, in the fault object-fault mode "output rule" of the single-index statistical rule, when the x quantile is less than 1200 ppm, the corresponding symptom judgment is given, such as "SCR-carrier moderate failure"; when the x quantile is less than 2 kPa, the corresponding symptom judgment is given, such as "DPF-carrier moderate blockage". It should be noted that only some parameter rules are exemplarily shown in Tables 5 and 6, and other parameters can be set according to actual application scenarios.

[0204] Table 5, output rule of fault object-fault mode of single-index statistical rule in SCR diagnosis

[0205] Statistical value parameters 99th percentile < 1200 ppm 85% percentile < 1200 ppm 70% percentile < 1200 ppm 70% percentile ≥ 1200 ppm SCR upstream NOx sensor output value SCR normal SCR light failure SCR moderate failure SCR severe failure .

[0206] Table 6, output rule of fault object-fault mode of single-index statistical rule in DPF diagnosis

[0207] 60% fractile < 2 kPa 50% fractile < 2 kPa 40% percentile < 2 kPa 40% percentile ≥ 2 kPa DPF differential pressure DPF normal DPF light clogging DPF moderate clogging DPF severe clogging .

[0208] The parameters targeted by the multi-parameter ratio rule are shown in Table 7.

[0209] Table 7, Parameters corresponding to the multi-parameter ratio rule

[0210]

[0211] The first row and the first column in Table 7 list 7 core basic data accumulated by the system in real time, as follows.

[0212] (1) Cumulative mileage (Mil_AC): The total distance traveled by the vehicle.

[0213] Mil_AC = SUM(V / 3600).

[0214] (2) Cumulative fuel consumption (Mfuel_AC): The total amount of fuel consumed by the engine.

[0215] Mil_AC = SUM(Lfuel*0.85 / 3600).

[0216] (3) Cumulative power (Power_AC): The total power output by the engine during operation.

[0217] Mil_AC = SUM(NE*Tq_max*TQ / 9550 / 3600).

[0218] (4) Cumulative SCR downstream NOx emissions (MNOxD_AC): The cumulative amount of nitrogen oxides (NOx) emitted into the atmosphere after being processed by the SCR (Selective Catalytic Reduction) system. The lower this value is, the better.

[0219] MNOxD_AC = SUM((0.38185 • Lfuel+ 0.44445 • Mair) • PN0x• 10−6).

[0220] (5) Cumulative SCR upstream NOx consumption (MNOxU_AC): The cumulative emission of nitrogen oxides (NOx) before the engine enters the SCR system.

[0221] (6) Cumulative fuel tank level drop percentage (SOT_AC): The cumulative percentage of fuel consumption in the fuel tank. Only when it decreases will it be added.

[0222] SOT_AC = -SUM(ASOT) when ASOT < 0.

[0223] (7) Accumulated Reactant Level Drop Percentage (SOR_AC): Generally refers to the accumulated consumption percentage of the reactant (urea solution) for the SCR system in the urea (AdBlue®) tank. Only when it drops is it accumulated.

[0224] SOR_AC = -SUM(ASOR) when ASOR < 0.

[0225] The middle part of Table 7 is the core "ratio parameter", i.e. the diagnostic criterion. Specifically as follows.

[0226] (1) Unit Mileage Fuel Consumption (Mfuel_AC / Mil_AC): That is, the fuel consumption per 100 kilometers. Its abnormal increase may mean that the engine efficiency is reduced, the tire pressure is insufficient, or the vehicle load is abnormal.

[0227] (2) Unit Mileage Power (Power_AC / Mil_AC): Reflects the work done by the vehicle per unit mileage, which can indirectly reflect the driving road conditions (such as climbing, high-speed driving will be higher).

[0228] (3) Unit Mileage Downstream / Upstream NOx Emission Factor (MNOxD_AC / Mil_AC or MNOxU_AC / Mil_AC): Indicates the amount of NOx emitted per kilometer of driving. This is a key environmental protection indicator for evaluating whether the vehicle emission meets the standard.

[0229] (4) Unit Mileage Urea Tank Level Drop Percentage (RT_AC / Mil_AC): Reflects the percentage of urea consumed per kilometer of driving.

[0230] (5) Specific Fuel Consumption / Thermal Efficiency (Power_AC / Mfuel_AC): This is a very important engine performance indicator. It represents the work that can be generated by consuming a unit mass of fuel. The higher this value, the better the fuel economy (thermal efficiency) of the engine. If this value decreases, it means that the engine's efficiency in converting fuel into power has decreased.

[0231] (6) Unit Fuel Consumption Downstream / Upstream NOx Emission Factor (MNOxD_AC / Mfuel_AC or MNOxU_AC / Mfuel_AC): Indicates the amount of NOx generated per unit of fuel consumed. This indicator eliminates the influence of driving mileage and more directly reflects the NOx generation characteristics of the engine under certain working conditions and the processing efficiency of the SCR system.

[0232] (7) Unit Fuel Consumption Urea Tank Level Drop Percentage (RT_AC / Mfuel_AC): Measures the proportional relationship between fuel and urea consumption.

[0233] (8) Downstream / Upstream NOx emission factor per unit power (MNOxD_AC / Power_AC or MNOxU_AC / Power_AC): represents the amount of NOx emitted per unit of power output by the engine (g / kWh). This is the core standard unit for measuring NOx emissions in heavy-duty diesel vehicle regulations. Monitoring of this indicator is critical for regulatory compliance.

[0234] (9) NOx conversion rate (1-MNOxD_AC / MNOxU_AC): This is the most core indicator for measuring the performance of the SCR system. It directly assesses the working efficiency of the SCR system by calculating the proportion of NOx entering the SCR that is successfully converted. If this value is significantly lower than the design value (usually above 90%), it clearly indicates that the SCR system has a serious fault, such as catalyst aging, urea injection system failure, or sensor misalignment.

[0235] (10) NOx reduction amount per unit percentage of urea level drop ((MNOxU_AC-MNOxD_AC) / SOT_AC): This ratio measures how much NOx reduction can be achieved by consuming a unit percentage of urea. It can be used to diagnose the accuracy of urea injection or the quality of urea solution. If urea is consumed but the NOx reduction amount is low, it may mean that the injection amount is insufficient, the nozzle is blocked, or poor-quality / diluted urea is used.

[0236] (11) Urea fuel consumption ratio (SOR_AC / SOF_AC): This ratio monitors the relative proportion of urea consumption to fuel consumption. Under normal engine operating conditions and SCR system working state, this ratio should be within a relatively fixed range (e.g. 2%-8%, depending on the specific technology). If the ratio is abnormal, such as too high, it may indicate that the urea system is leaking or over-injection. Too low or zero may indicate that the urea injection system is not working at all, the pipeline is blocked, a cheating device (cheater) is used, or the urea tank is empty. This is a parameter that regulatory agencies focus on to prevent emission cheating.

[0237] The value range of "Downstream NOx emission factor per unit power" in SCR diagnosis gives the corresponding symptom judgment. For example, if the downstream NOx emission factor per unit power = 5 g / kWh, then "SCR - mild high emission". The fault object-fault mode output rule of the multi-parameter ratio rule in SCR diagnosis is shown in Table 8. It should be noted that only part of the parameter rules are exemplarily shown in Table 8, and other parameters can be set according to actual application scenarios.

[0238] Table 8, Fault object-fault mode output rule of multi-parameter ratio rule in SCR diagnosis

[0239] Statistical value parameters <2g / kWh <3.5g / kWh <7g / kWh ≥ 7 g / kWh Downstream NOx emission factor per unit power SCR normal SCR light high emission SCR moderate high emission SCR severe high emission .

[0240] One-dimensional and two-dimensional feature similarity rules use Lin's Concordance Correlation Coefficient (CCC) to assess the consistency of two sets of data, which can measure both the correlation (slope) between them and the degree of deviation from the perfect consistency line (y=x) (absolute amount). The value of CCC is also between -1 and 1, the closer to 1, the better the consistency of the two sets of data, that is, very similar in trend and absolute value. The formula of Lin's Concordance Correlation Coefficient is as follows: , wherein x is the set of parameters of the standard vehicle, y is the set of parameters of the vehicle to be evaluated, σ x 2 and σ y 2 is the variance of the two sets of data, μ x and μ y is the mean of the two sets of data.

[0241] The correlation between variables can use rule type 1 or rule type 2, as shown in Tables 9 and 10, respectively. In rule type 1, the correlation of vehicle parameters under different working conditions with the standard vehicle is compared. The better the correlation, the better the health of the vehicle. It can be "input 1-output" mode, or "input 1-input 2-output" mode. In rule type 2, the correlation between vehicle parameters is compared. The better the correlation between some parameters, the worse the performance, and the correlation between some parameters. For example, the better the correlation between instantaneous fuel consumption and NOx, the better the SCR is not started, and when the correlation reaches 0.8 or more, the SCR is completely disabled.

[0242] Table 9, rule type 1

[0243]

[0244] Table 10, rule type 2

[0245]

[0246] Therefore, the fault object-fault mode output rule is obtained based on the relationship between variables. In the exhaust purification device thermal management evaluation, the “speed-engine net output torque-SCR outlet temperature” of the vehicle to be evaluated and the value range of “speed-engine net output torque-SCR outlet temperature” of the standard vehicle in the whole working condition give the corresponding symptom judgment, as shown in Table 11. For example, the “speed-engine net output torque-SCR outlet temperature” of the vehicle to be evaluated and the standard vehicle CCC<0 in the whole working condition, and the output “SCR-thermal management failure”. It should be noted that only part of the parameter rules are exemplarily shown in Table 11, and other parameters can be set according to actual application scenarios.

[0247] Table 11, fault object-fault mode output rule based on variable relationship

[0248] Statistical value parameters CCC < 0 CCC < 0.5 CCC < 0.9 CCC ≥ 0.9 To be evaluated vehicle and standard vehicle "speed-engine net output torque-SCR outlet temperature" full working condition CCC evaluation SCR thermal management failure SCR thermal management poor SCR thermal management normal SCR thermal management excellent .

[0249] 33) Extracting fault object-fault symptom association information based on the textual fault description.

[0250] Among them, the user or the maintenance personnel inputs the “sensory-object-symptom” existing in the vehicle through guidance. For example, the user is guided to the first key decision point. Four options (which can be increased as needed) are provided for the user to choose: “abnormal sound”, “smoke”, “light”, “abnormal smell”. According to the selection, different and targeted guidance paths are entered. The core task of each path is to determine the “object” of the fault and refine the description of the “symptom”. The specific guidance path can be expanded as needed. The guidance process of the specific fault object-fault symptom association information is shown in Table 12.

[0251] Table 12, guidance information of fault object-fault symptom association information

[0252]

[0253] 34) Based on the fault codes in the fault code list, the engine brand and the fault protocol type, find the corresponding fault object-fault mode association information in the vehicle fault knowledge base.

[0254] Among them, according to the fault codes in the fault code list, the engine brand and the fault protocol type, find the corresponding fault object-fault mode association information in the vehicle fault knowledge base, so as to ensure the instant response and matching of standardized fault information.

[0255] For example, when the fault code starts with P, C, B, or U, search the vehicle fault knowledge base that complies with the SAE J2012 specification. When there is none in SAE J2012, search the vehicle fault knowledge base specific to the vehicle or engine manufacturer according to the input information. For another example, when the fault code starts with SPN, search the strategy fault knowledge base that complies with the SAE J1939 specification. When there is none in SAE J2012, search the vehicle fault knowledge base specific to the vehicle or engine manufacturer according to the input information.

[0256] Step S4, deduplicate and aggregate the fault object-fault mode association information to obtain valid fault object-fault mode association information.

[0257] Specifically, deduplicating and aggregating the fault object-fault mode association information aims to solve the problem of repeated fault descriptions or inconsistent expressions caused by multi-source data, and mine the internal association through natural language processing technology and aggregate into a core fault.

[0258] Deduplicating and aggregating the fault object-fault mode association information to obtain valid fault object-fault mode association information includes the following steps.

[0259] 41) Vectorize the fault object-fault mode association information based on an SBERT model or a Transformer model.

[0260] wherein a pre-trained language model f embed For each fault description phrase ti (“fault object-fault mode”), process it to map it into a high-dimensional numerical vector vi. This process can be represented as: vi=f embed (ti). Wherein vi∈Rd, that is, the vector vi is a point in a d-dimensional real number space, and d is a preset vector dimension (for example, d=768). The high-dimensional vector vi is the feature vector or text embedding of the phrase ti. The high-dimensional vector can represent the position of the corresponding phrase in the semantic space defined by the model f embed The geometric distance of the vectors of semantically similar phrases in this space is also close.

[0261] In one embodiment, the pre-trained language model f embed may be a sentence embedding model (Sentence-BERT, SBERT) optimized for sentence similarity tasks. In another embodiment, a general language model based on the Transformer architecture (such as BERT) can also be used, and the output of a specific layer (such as the [CLS] flag of the last layer) is extracted as the feature vector vi representing the text.

[0262] 42) Clustering the vectorized failure object-failure mode association information to obtain a plurality of failure clusters.

[0263] Specifically, the set of feature vectors V = {vi, v2,..., vn} is subjected to clustering analysis. The clustering algorithm groups the vectors according to the spatial distance between them, and each group constitutes a failure cluster. The failure descriptions within the same failure cluster are identified as highly relevant in semantics. First, the semantic similarity between any two vectors vi and vj is defined. In this embodiment, the cosine similarity (Cosine Similarity) is used for calculation, and the formula is:

[0264] .

[0265] where vik is the kth component of the vector vi. The value range of Sim(vi, vj) is [-1, 1], and the closer the value is to 1, the more similar the semantics. Subsequently, the distance D(vi, vj) between vectors is defined based on the similarity, and the following formula can be used: D(vi, vj) = 1 - Sim(vi, vj). The value range of the distance D is [0, 2], and the smaller the value, the closer the distance, and the more relevant the semantics.

[0266] In a preferred embodiment, the clustering algorithm can use the density-based clustering algorithm (DBSCAN). DBSCAN algorithm operates through two core parameters: neighborhood radius and the minimum sample size MinPts constituting the core object. The algorithm considers the vector pair that satisfies D(vi, vj)≤ as density reachable and forms a cluster based on this. The advantage of using such an algorithm is that it does not need to specify the number of clusters in advance, can automatically discover failure clusters Gk according to the distribution characteristics of the data itself, and can effectively identify outliers (marked as noise) as independent failures. For example, by calculating, it is found that the pairwise distances D between the vectors va, vb, vc corresponding to the phrases "1st cylinder - misfire", "engine - shaking" and "engine - acceleration weakness" are all less than the preset threshold , so they are grouped into the same failure cluster.

[0267] 43) For each failure cluster, select the failure object-failure mode association information corresponding to the failure as the effective failure object-failure mode association information.

[0268] where each failure cluster Gk generated is subjected to the following steps: k, apply a preset representative selection strategy to determine a fault from the "fault object-fault mode" {t1, t2,...,tm} contained in the fault cluster as the representative fault of the fault cluster, thereby generating a final fault list aggregated and deduplicated.

[0269] The selection process can be achieved by scoring each fault description within the cluster with a scoring function Score(ti) and selecting the item with the highest score. Wherein The scoring function can be composed of one or more sub-strategies weighted, and its general form can be defined as follows: .

[0270] Wherein, represents a fault description within the fault cluster, represents the weight coefficient of each strategy, f prio ( source(t i ) ) is the scoring function of the data source priority strategy, which returns a preset value according to the source of the fault (such as on-board diagnostic system OBD, on-board monitoring module OBM, owner's oral description). For example, it can be set that fprio (OBD)=10, f prio (OBM)=5, f prio (owner's oral description)=1. is the scoring function of the semantic specificity strategy, which is used to evaluate the detail and clarity of the fault description and returns a quantitative score. For example, a description containing specific component numbers (such as "No. 1 cylinder") or precise values will get a higher score than a macroscopic description (such as "engine").

[0271] Therefore, in the above manner, the fault information from multiple sources that are superficially different but essentially related can be effectively aggregated. For example, multiple symptoms (misfire, shaking, lack of acceleration) are aggregated to "No. 1 cylinder misfire", which greatly improves the accuracy and automation level of fault diagnosis, provides clear and non-redundant input for subsequent maintenance decisions, and overcomes the limitations of relying on manual experience for correlation analysis in the prior art.

[0272] Step S5, based on the vehicle fault knowledge base, obtaining the fault cause corresponding to the effective fault object-fault mode association information.

[0273] Specifically, inputting the effective fault object-fault mode association information into the vehicle fault knowledge base, the corresponding fault cause can be queried.

[0274] Step S6, based on the vehicle repair knowledge base, obtaining the maintenance scheme corresponding to the fault object-fault mode association information.

[0275] Specifically, input the fault object-fault mode association information into the vehicle repair knowledge base, and the corresponding maintenance scheme can be queried.

[0276] Step S7, based on the fault cause and the maintenance scheme, using a large language model to generate a vehicle fault analysis report.

[0277] Specifically, the large language model, based on the fault cause and the maintenance scheme, automatically integrates the basic situation of the vehicle, the main problems (aggregated core faults), the cause analysis of each problem, and the detailed solutions (including maintenance steps, required tools and spare parts), according to the pre-set professional template, to generate a comprehensive diagnostic and maintenance recommendation report for the end user (such as a maintenance technician or car owner), which is professional and easy to understand.

[0278] First, a prompt word template needs to be generated for the large language model to guide the large language model to play the role of a professional technical communicator, and to organically organize scattered data points into a logical and complete technical document. The prompt word template includes the following content.

[0279] I. Role and Task Definition Zone.

[0280] Purpose: Set the final output role of the large language model as a professional automotive technical writer and senior service consultant, whose task is to convert technical data into a smooth and professional written report.

[0281] Composition:

[0282] Persona: Clearly define its expert identity, for example: "You are a professional automotive technical writer and senior service consultant, specializing in converting complex technical diagnostic data into clear, standardized and user-friendly reports."

[0283] Core Task: Clearly describe the work it needs to do, for example: "Your task is to strictly follow the JSON format diagnostic data I provide to generate a complete Vehicle Diagnostic and Maintenance Recommendation Report. The report must be clear in structure, professional in language, and include all the information in the input data."

[0284] Behavioral Constraints: Set strict output rules, such as: "You must strictly follow the Markdown format I defined. Prohibit adding any comments, explanations, or text unrelated to the report format. All information must come from the input data I provided, and no self-censorship or creation is allowed."

[0285] II. Output Format Constraint Zone.

[0286] Purpose: Define the precise "skeleton" of the final report to ensure that the structure, title, and format of all reports remain highly consistent, facilitating reading and archiving.

[0287] Composition: Provide a complete report structure template defined using the Markdown language.

[0288] Similarly, the invention provides a small number of examples for the large language model to help the large language model generate qualified vehicle fault analysis reports. For example, provide a complete "input JSON -> output Markdown report" example to the large language model, so that it can intuitively learn how to fill structured data into the formatted report template. The example includes the following.

[0289] • Composition:

[0290] Input JSON (Input): A complete JSON object containing the output of all diagnostic steps.

[0291] Output Markdown report (Output): A "perfect report example" manually written by a human expert based on the input JSON, strictly following the Markdown template defined by the format constraint zone.

[0292] It should be noted that each step of the vehicle fault analysis method based on the artificial intelligence agent of the present invention can be implemented by a corresponding agent, thereby realizing vehicle fault analysis based on artificial intelligence agents. The protection scope of the vehicle fault analysis method based on the artificial intelligence agent of the present embodiment is not limited to the execution order of the steps listed in the present embodiment. Any scheme realized by increasing, reducing, or replacing the steps of the prior art according to the principle of the present invention is included in the protection scope of the present invention.

[0293] This invention also provides a vehicle fault analysis system based on artificial intelligence agents. The vehicle fault analysis system based on artificial intelligence agents can implement the vehicle fault analysis method based on artificial intelligence agents described in this invention. However, the implementation device of the vehicle fault analysis system based on artificial intelligence agents described in this invention includes, but is not limited to, the structure of the vehicle fault analysis system based on artificial intelligence agents listed in this embodiment. All structural modifications and substitutions of the prior art made in accordance with the principles of this invention are included within the protection scope of this invention.

[0294] like Figure 2 As shown, in one embodiment, the vehicle fault analysis system based on artificial intelligence agent of the present invention includes a first acquisition module 21, a construction module 22, a second acquisition module 23, a third acquisition module 24, a fourth acquisition module 25, a fifth acquisition module 26 and an analysis module 27.

[0295] The first acquisition module 21 is used to acquire multi-source heterogeneous data of the vehicle.

[0296] The construction module 22 is used to construct a vehicle fault knowledge base and a vehicle repair knowledge base; the vehicle fault knowledge base is used to record the association information of fault object-fault mode-fault cause; the vehicle repair knowledge base is used to record the association information of fault object-fault mode-repair plan.

[0297] The second acquisition module 23 is connected to the first acquisition module 21 and the construction module 22, and is used to acquire fault object-fault mode association information corresponding to the multi-source heterogeneous data based on the vehicle fault knowledge base.

[0298] The third acquisition module 24 is connected to the second acquisition module 23 and is used to deduplicate and aggregate the fault object-fault mode association information to obtain valid fault object-fault mode association information.

[0299] The fourth acquisition module 25 is connected to the construction module 22 and the third acquisition module 24, and is used to acquire the fault cause corresponding to the effective fault object-fault mode association information based on the vehicle fault knowledge base.

[0300] The fifth acquisition module 26 is connected to the construction module 22 and the third acquisition module 24, and is used to acquire the repair plan corresponding to the effective fault object-fault mode association information based on the vehicle maintenance knowledge base.

[0301] The analysis module 27 is connected to the fourth acquisition module 25 and the fifth acquisition module 26, and is used to generate a vehicle fault analysis report based on the fault cause and the repair plan using a large language model.

[0302] The first obtaining module 21, the constructing module 22, the second obtaining module 23, the third obtaining module 24, the fourth obtaining module 25, the fifth obtaining module 26 and the analyzing module 27 correspond to the steps of the vehicle fault analysis method based on the artificial intelligence agent one by one in structure and principle, and thus will not be described here.

[0303] In several embodiments provided by the present application, it should be understood that the disclosed system, device or method can be implemented in other ways. For example, the device embodiments described above are only schematic. The division of the modules / units is only a logical function division. There can be another division manner in actual implementation. For example, a plurality of modules or units can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the modules / units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or modules / units, and can be electrical, mechanical or in other forms.

[0304] The modules / units described as separate components can or can not be physically separate, and the components shown as modules / units can or can not be physical modules, i.e., can be located in one place or distributed on a plurality of network units. Some or all of the modules / units can be selected according to actual needs to achieve the purpose of the embodiments of the present application. For example, the functional modules / units in each embodiment of the present application can be integrated in one processing module, or each module / unit can be physically separate, or two or more modules / units can be integrated in one module / unit.

[0305] Those of ordinary skill in the art should further appreciate that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, each example has been described in the above description in terms of its general functionality. Whether the implementation is performed in hardware or software depends on the specific application and design constraints of the technical solution. A skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0306] The embodiments of the present application further provide a computer readable storage medium. A person skilled in the art can understand that all or part of the steps of the method described above can be instructed by a program to complete the processor, and the program can be stored in a computer readable storage medium. The storage medium is a non-transitory medium, for example, a random access memory, a read-only memory, a flash memory, a hard disk, a solid state disk, a magnetic tape, a floppy disk, an optical disc and any combination thereof. The storage medium can be any available medium accessible by a computer or a data storage device such as a server, a data center and the like, which includes one or more available medium sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a digital video disc (DVD)) or a semiconductor medium (for example, a solid state disk (SSD)) and the like.

[0307] The embodiments of the present application further provide an electronic device. The electronic device comprises a processor and a memory.

[0308] The memory is used to store a computer program.

[0309] The memory comprises a ROM, a RAM, a disk, a U disk, a memory card or an optical disc and the like various medium capable of storing program codes.

[0310] The processor is connected with the memory, and is used to execute the computer program stored in the memory, so that the electronic device executes the vehicle fault analysis method based on the artificial intelligence agent described above.

[0311] Preferably, the processor can be a general processor, including a central processing unit (CPU), a network processor (NP) and the like; and can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0312] As Figure 3As shown, the electronic device of the present application is in the form of a general- purpose computing device. The components of the electronic device can include, but are not limited to, one or more processors or processing units 31, a memory 32, and a bus 33 that couples various system components including the memory 32 to the processing unit(s) 31.

[0313] The bus 33 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics bus (e.g., an Accelerated Graphics Port, or AGP bus) and a processor or local bus using any of a variety of bus architectures. By way of example, these architectures include Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MCA) bus, Enhanced ISA (EISA) bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnect (PCI) bus.

[0314] The electronic device typically includes a variety of computer system readable media. Such media can be any available media that is located either internally or externally to the electronic device. It includes storage of removable and non-removable, volatile and non-volatile computer system storage media.

[0315] The memory 32 can include computer system readable media in the form of volatile memory, such as random access memory (RAM) 321 and / or cache memory 322. The electronic device can further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 323 can be provided for reading from and writing to a non-removable, non-volatile magnetic media (e.g., a "hard drive"). Figure 3 not shown, a magnetic hard disk drive for reading from and writing to non-removable, non-volatile magnetic media (e.g., a "hard drive"). Although not specifically shown, such Figure 3 In alternative embodiments, a magnetic hard disk drive, a solid state drive (SSD) which is a non- volatile computer storage media, a floppy disk drive for reading from and writing to a removable, non-volatile magnetic disk (e.g., a "floppy disk"), and an optical disk drive for reading from or writing to a removable, non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, etc.) can be provided. In these instances, each drive can be connected to the bus 33 by one or more data media interfaces. The memory 32 can include a system memory component designed to provide fast storage of instructions and data required to operate the electronic device and other system components.

[0316] A program / utility 324 having a set (at least one) of program modules 3241 can be stored in, for example, memory 32 by way of example, such program modules 3241 include an operating system, one or more application programs, other program modules, and program data, each of which or a combination can include implementation of a network environment. Program modules 3241 generally carry out the functions and / or methodologies of embodiments of the present application as described herein.

[0317] The electronic device can also communicate with one or more external devices such as a keyboard or a pointing device, displays, etc.; other devices that enable a user to interact with the electronic device; and / or any devices (e.g., network card, modem, etc.) that enable the electronic device to communicate with one or more other computing devices. Such communication can occur via Input / Output (I / O) interface 34. Still yet, the electronic device can communicate with one or more networks, such as a local area network (LAN), a wide area network (WAN), and / or the Internet, through network adapter 35. As Figure 3 illustrated, network adapter 35 communicates with other modules of the electronic device through bus 33. It should be appreciated that other hardware and / or software modules can be used in conjunction with the electronic device. Examples, include, but are not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.

[0318] The above-described embodiments are merely illustrative of the principles of the present application and its efficacy, and are not intended to limit the present application. Any modification or variation of the above-described embodiments, which falls within the spirit and scope of the present application, is intended to be included. Accordingly, all such modifications and variations are intended to be included herein as aspects of the present application as defined by the appended claims.

Claims

1. A method for vehicle failure analysis based on artificial intelligence agent, characterized in that, The method comprises the following steps: Obtaining multi-source heterogeneous data of a vehicle; Building a vehicle fault knowledge base and a vehicle maintenance knowledge base; the vehicle fault knowledge base is used to record fault object-fault mode-fault cause correlation information; the vehicle maintenance knowledge base is used to record fault object-fault mode-maintenance scheme correlation information; Based on the vehicle fault knowledge base, obtaining fault object-fault mode correlation information corresponding to the multi-source heterogeneous data; De-duplicating and aggregating the fault object-fault mode correlation information to obtain valid fault object-fault mode correlation information; Based on the vehicle fault knowledge base, obtaining fault causes corresponding to the valid fault object-fault mode correlation information; Based on the vehicle maintenance knowledge base, obtaining maintenance schemes corresponding to the valid fault object-fault mode correlation information; Based on the fault causes and the maintenance schemes, generating a vehicle fault analysis report by using a large language model; The multi-source heterogeneous data comprises remote online monitoring data, a fault code list, an engine brand, a fault protocol type, and a textual fault description; Based on the vehicle fault knowledge base, obtaining fault object-fault mode correlation information corresponding to the multi-source heterogeneous data comprises the following steps: Based on the calibration code ID and the calibration verification number in the electronic control unit of the vehicle, judging whether the electronic control unit is tampered with to obtain fault object-fault mode correlation information of the electronic control unit for tampering; Based on the online monitoring data, extracting corresponding fault object-fault mode correlation information; Based on the textual fault description, extracting fault object-fault symptom correlation information; Based on the fault code in the fault code list, the engine brand, and the fault protocol type, searching for corresponding fault object-fault mode correlation information in the vehicle fault knowledge base. 2.The vehicle fault analysis method based on artificial intelligence agent according to claim 1, characterized in that: Whether the electronic control unit is tampered with is judged according to whether the edit distance offset degree of the calibration code ID and the calibration verification number exceeds a preset value. 3.The vehicle fault analysis method based on artificial intelligence agent of claim 1, wherein: Building a vehicle fault knowledge base comprises the following steps: Obtaining a fault code database of a vehicle; the fault code database records the fault code, fault description, and fault cause of each fault; Based on the fault code database, generating fault object-fault mode-fault cause correlation information for each fault by using a large language model.

4. The method of claim 1, wherein: Building a vehicle maintenance knowledge base comprises the following steps: Obtaining a vehicle maintenance manual, a vehicle technical bulletin, and a maintenance case; Converting the vehicle maintenance manual, the vehicle technical bulletin, and the maintenance case into pure text information; Building a prompt word template, the prompt word being used to extract fault and maintenance related information in the pure text information and output in a preset format; Based on the pure text information and the prompt word template, generating fault object-fault mode-maintenance scheme correlation information by using a large language model.

5. The method of claim 1, wherein: De-duplicating and aggregating the fault object-fault mode correlation information to obtain valid fault object-fault mode correlation information comprises the following steps: Vectorizing the fault object-fault mode correlation information based on an SBERT model or a Transformer model; The vectorized fault object-fault mode association information is clustered to obtain a plurality of fault clusters; For each fault cluster, the fault object-fault mode association information corresponding to one fault is selected as the effective fault object-fault mode association information.

6. An artificial intelligence agent based vehicle failure analysis system, characterized in that, The system comprises a first acquisition module, a construction module, a second acquisition module, a third acquisition module, a fourth acquisition module, a fifth acquisition module, and an analysis module; The first acquisition module is configured to acquire multi-source heterogeneous data of a vehicle; The construction module is configured to construct a vehicle fault knowledge base and a vehicle maintenance knowledge base; the vehicle fault knowledge base is configured to record fault object-fault mode-fault cause association information; and the vehicle maintenance knowledge base is configured to record fault object-fault mode-maintenance scheme association information; The second acquisition module is configured to acquire, based on the vehicle fault knowledge base, fault object-fault mode association information corresponding to the multi-source heterogeneous data; The third acquisition module is configured to de-duplicate and aggregate the fault object-fault mode association information to obtain effective fault object-fault mode association information; The fourth acquisition module is configured to acquire, based on the vehicle fault knowledge base, fault causes corresponding to the effective fault object-fault mode association information; The fifth acquisition module is configured to acquire, based on the vehicle maintenance knowledge base, maintenance schemes corresponding to the effective fault object-fault mode association information; The analysis module is configured to generate, based on the fault causes and the maintenance schemes, a vehicle fault analysis report using a large language model; The multi-source heterogeneous data comprises remote online monitoring data, a fault code list, an engine brand, a fault protocol type, and a textual fault description; Based on the vehicle fault knowledge base, the fault object-fault mode association information corresponding to the multi-source heterogeneous data is acquired by the following steps: Based on a calibration code ID and a calibration verification number in an electronic control unit of the vehicle, it is determined whether the electronic control unit is tampered with to obtain fault object-fault mode association information of the electronic control unit for tampering; Based on the online monitoring data, corresponding fault object-fault mode association information is extracted; Based on the textual fault description, fault object-fault symptom association information is extracted; Based on fault codes in the fault code list, the engine brand, and the fault protocol type, corresponding fault object-fault mode association information is searched in the vehicle fault knowledge base.

7. An electronic device, comprising: The electronic device comprises a processor and a memory; The memory is configured to store a computer program; The processor is configured to execute the computer program stored in the memory to enable the electronic device to perform the vehicle fault analysis method based on an artificial intelligence agent according to any one of claims 1 to 5.

8. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the electronic device to implement the vehicle fault analysis method based on an artificial intelligence agent according to any one of claims 1 to 5.

Images