Vehicle fault diagnosis method and device based on general-purpose collaborative decision-making

By adopting MBSE ideas and collaborative decision-making methods in vehicle fault diagnosis, combining generative artificial intelligence models and specialized models for fault diagnosis, the problem of overall architecture coordination and insufficient computing resources in vehicle fault diagnosis is solved, and more efficient faults are achieved Diagnosis and vehicle health assessment.

CN118938868APending Publication Date: 2024-11-123RD GENERAL DESIGN DEPT CHINA AEROSPACE SCI & IND CORP
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411118099.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing vehicle fault diagnosis methods are difficult to comprehensively consider the overall vehicle architecture, resulting in insufficient computing resources and poor diagnostic results, and it is difficult to evaluate the overall health of the vehicle.

Method used

Using the MBSE idea-based collaborative decision-making method, we will build a dedicated domain knowledge data set of structured and unstructured data, combine the generative artificial intelligence model and the specialized fault diagnosis model, and fuse the diagnostic results of multiple models through the Kalman filtering method. , realize vehicle health assessment.

Benefits of technology

It improves the effect and accuracy of vehicle fault diagnosis, optimizes the model fusion results, provides error covariance of calculation results, and enhances the reliability of diagnosis and vehicle health assessment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118938868A_ABST
    Figure CN118938868A_ABST
Patent Text Reader

Abstract

The invention provides a vehicle fault diagnosis method and device based on general-special collaborative decision-making. The method comprises the following steps: constructing a special domain knowledge data set; constructing an instruction control data set; constructing a fault diagnosis data set; constructing a professional domain knowledge question and answer data set; constructing an intention recognition data set; obtaining the model accuracy of the trained generative artificial intelligence model; obtaining respective model accuracy of the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power supply system fault diagnosis models; the probabilities of the first suspension system fault, the first bearing fault and the first power system fault are output; fault sensitive features are obtained; the probabilities of N second suspension system faults, N second bearing faults and N second power system faults are output; and obtaining the probability of the suspension system fault, the bearing fault and the power system fault after fusion. According to the invention, the problem of overall planning of various fault diagnosis models under the condition of considering the overall architecture of the vehicle is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of automobile fault diagnosis, and in particular to a vehicle fault diagnosis method and device based on general-specialist collaborative decision-making. Background Art

[0002] Vehicle fault diagnosis is a key link in the process of automobile repair and maintenance. Timely fault diagnosis of vehicles can avoid potential safety risks, reduce maintenance costs and increase service life. Diagnostic analysis methods are usually based on signal characteristics and state recognition methods to analyze the collected signals and locate vehicle faults.

[0003] At present, the commonly used vehicle fault diagnosis and analysis methods include artificial experience method, fault code reading method, expert system diagnosis and deep learning. With the rapid development of big data, artificial intelligence and information network technology, modern vehicles are equipped with a large number of sensors, which continuously integrate advanced theoretical methods and technologies to realize the real-time collection of vehicle operation data, and identify and predict the potential faults and abnormal conditions of the vehicle through data analysis and deep learning algorithms. Although vehicle fault diagnosis technology has made significant progress, there are still some limitations. First, the amount of data generated by the sensors equipped in modern vehicles is increasing, and the fault diagnosis algorithms are becoming more and more complex, often requiring a lot of computing resources, and the limited computing power of the on-board computing platform may not be able to meet the needs of real-time data processing. Second, it is difficult to coordinate the various fault diagnosis methods equipped in modern vehicles, and it is difficult to consider the overall system architecture of the vehicle in the overall health and reliability assessment of the vehicle. Summary of the invention

[0004] The present invention provides a vehicle fault diagnosis method and device based on general-specialist collaborative decision-making, which can solve the coordination problem of various fault diagnosis models of existing vehicles under consideration of the overall vehicle architecture, use large model technology to expand the diagnosis range, and improve the diagnosis effect.

[0005] According to one aspect of the present invention, a vehicle fault diagnosis method based on general-specialist collaborative decision-making is provided, the method comprising:

[0006] Based on MBSE ideas, structured data and unstructured data, a dedicated domain knowledge dataset is constructed, where the output of the dedicated domain knowledge dataset is structured data;

[0007] Constructing an instruction control data set, wherein the input of the instruction control data set is an instruction control text, and the output is an instruction code corresponding to the instruction control text;

[0008] Constructing a fault diagnosis data set, wherein the fault diagnosis data set includes a first fault diagnosis data set for training a generative artificial intelligence model and a second fault diagnosis data set for training a dedicated fault diagnosis model, the first fault diagnosis data set is in the form of X1-Y, the second fault diagnosis data set is in the form of X2-Y, X1 represents the external manifestation of the fault, X2 represents the sensor acquisition signal, and Y represents the fault mode probability;

[0009] Construct a professional domain knowledge question-answering dataset, where the professional domain knowledge question-answering dataset includes documents and questions and answers in the form of AB, where A represents the fault problem input by the user and B represents the troubleshooting solution related to the problem;

[0010] Construct an intention recognition dataset, where the input of the intention recognition dataset is natural language text, and the output is the intention corresponding to the natural language text. The intention corresponding to the natural language text includes three, corresponding to the command control mode, fault diagnosis mode, and professional knowledge question and answer mode;

[0011] Pre-training the generative artificial intelligence model based on the dedicated domain knowledge data set, re-training the generative artificial intelligence model based on the instruction control data set, the first fault diagnosis data set in the fault diagnosis data set, the professional domain knowledge question and answer data set, and the intention recognition data set to obtain a trained generative artificial intelligence model, and obtaining the model accuracy of the trained generative artificial intelligence model;

[0012] Based on the second fault diagnosis data set in the fault diagnosis data set, N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models in the fault diagnosis special model are trained respectively to obtain the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models in the fault diagnosis special model, and obtain the model accuracy of each of the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models, where N≥1 and is an integer;

[0013] The user's description of the external manifestation of the fault is input into the trained generative artificial intelligence model for reasoning and evaluation. The trained generative artificial intelligence model recognizes the intention of the user's description of the external manifestation of the fault, performs diagnosis when the intention is recognized as a fault diagnosis mode, and outputs the probabilities of the first suspension system fault, the first bearing fault, and the first power supply system fault;

[0014] Collect vehicle-side sensor data, pre-process, extract features and filter the vehicle-side sensor data to obtain fault-sensitive features;

[0015] The fault-sensitive features are respectively input into the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models for inference evaluation, and the probabilities of N second suspension system faults, N second bearing faults and N second power system faults are output;

[0016] Through the Kalman filtering method, the model accuracy of the trained generative artificial intelligence model, the probabilities of the first suspension system failure, the first bearing failure and the first power supply system failure, the model accuracy of the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power supply system fault diagnosis models, and the probabilities of N second suspension system failures, N second bearing failures and N second power supply system failures are fused to obtain the fused probabilities of suspension system failures, bearing failures and power supply system failures.

[0017] Preferably, the following formula is used to fuse the probabilities of suspension system failure, bearing failure and power system failure:

[0018]

[0019] P k|k-1 =P k-1 +Q

[0020]

[0021] P k =(1-K k )·P k|k-1

[0022] in, m 1-2 or m 1-3 , P0=(1-a1) 2 ,

[0023] S1=m 2-1 、m 2-2 or m 2-3 , R = (1-b1) 2 、(1-b2) 2 or (1-b3) 2 ,

[0024] In the formula, is the probability estimate of the k-th prediction, k = 1 ~ N, is the probability estimate after the k-1th and kth updates, Δt is the time interval, P k|k-1 is the error covariance of the k-th prediction, P k-1 , P k is the error covariance after the k-1th and kth updates, Q is the process noise covariance, K kis the Kalman gain, R is the observation noise covariance, S k is the kth measurement probability result, is the initial probability estimate, m 1-1 、m 1-2 、m 1-3 are the probabilities of the first suspension system failure, the first bearing failure, and the first power system failure, respectively. P0 is the initial estimation error covariance. a1 is the model accuracy of the trained generative artificial intelligence model. S1 is the first set of measurement probability results. m 2-1 、m 2-2 、m 2-3 They are respectively the probabilities of the first suspension system fault, the first bearing fault, and the first power system fault output by the first group of trained any suspension system fault diagnosis model, any bearing fault diagnosis model, and any power system fault diagnosis model participating in the fusion; b1, b2, and b3 are the model accuracies of the first group of trained any suspension system fault diagnosis model, any bearing fault diagnosis model, and any power system fault diagnosis model participating in the fusion.

[0025] Preferably, the ratio of the number of samples in the instruction control data set, the first fault diagnosis data set in the fault diagnosis data set, and the professional field knowledge question and answer data set is 1:1:1.

[0026] Preferably, the structured data is pseudo code or markup language.

[0027] According to another aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above methods when executing the computer program.

[0028] By applying the technical solution of the present invention, the health status of the whole vehicle is evaluated in a collaborative decision-making manner. Multiple fault diagnosis-specific algorithm models are used to process the collected signals, and a large model is used to process the text description. The predicted fault probability and model test accuracy of the generative artificial intelligence model and the fault diagnosis professional model are combined. The collaborative decision-making of general and specialist is realized based on the Kalman filtering method, the model fusion result is optimized, the model fault diagnosis effect is improved, and the error covariance of the calculation result is given. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The included drawings are used to provide a further understanding of the embodiments of the present invention, which constitute a part of the specification, are used to illustrate the embodiments of the present invention, and together with the text description, explain the principles of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1 A flow chart of a vehicle fault diagnosis method based on general-specialist collaborative decision-making provided according to an embodiment of the present invention is shown;

[0031] Figure 2 A schematic diagram of a vehicle architecture design based on MBSE according to an embodiment of the present invention is shown;

[0032] Figure 3 A schematic diagram of the architecture of a fault diagnosis system provided according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0033] It should be noted that, in the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0035] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, numerical expressions and numerical values ​​do not limit the scope of the present invention. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be regarded as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0036] like Figure 1As shown, the present invention provides a vehicle fault diagnosis method based on general and professional collaborative decision-making, the method comprising:

[0037] Based on MBSE ideas, structured data and unstructured data, a dedicated domain knowledge dataset is constructed, where the output of the dedicated domain knowledge dataset is structured data;

[0038] Constructing an instruction control data set, wherein the input of the instruction control data set is an instruction control text, and the output is an instruction code corresponding to the instruction control text;

[0039] Constructing a fault diagnosis data set, wherein the fault diagnosis data set includes a first fault diagnosis data set for training a generative artificial intelligence model and a second fault diagnosis data set for training a dedicated fault diagnosis model, the first fault diagnosis data set is in the form of X1-Y, the second fault diagnosis data set is in the form of X2-Y, X1 represents the external manifestation of the fault, X2 represents the sensor acquisition signal, and Y represents the fault mode probability;

[0040] Construct a professional domain knowledge question-answering dataset, where the professional domain knowledge question-answering dataset includes documents and questions and answers in the form of AB, where A represents the fault problem input by the user and B represents the troubleshooting solution related to the problem;

[0041] Construct an intention recognition dataset, where the input of the intention recognition dataset is natural language text, and the output is the intention corresponding to the natural language text. The intention corresponding to the natural language text includes three, corresponding to the command control mode, fault diagnosis mode, and professional knowledge question and answer mode;

[0042] Pre-training the generative artificial intelligence model based on the dedicated domain knowledge data set, re-training the generative artificial intelligence model based on the instruction control data set, the first fault diagnosis data set in the fault diagnosis data set, the professional domain knowledge question and answer data set, and the intention recognition data set to obtain a trained generative artificial intelligence model, and obtaining the model accuracy of the trained generative artificial intelligence model;

[0043] Based on the second fault diagnosis data set in the fault diagnosis data set, N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models in the fault diagnosis special model are trained respectively to obtain the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models in the fault diagnosis special model, and obtain the model accuracy of each of the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models, where N≥1 and is an integer;

[0044] The user's description of the external manifestation of the fault is input into the trained generative artificial intelligence model for reasoning and evaluation. The trained generative artificial intelligence model recognizes the intention of the user's description of the external manifestation of the fault, performs diagnosis when the intention is recognized as a fault diagnosis mode, and outputs the probabilities of the first suspension system fault, the first bearing fault, and the first power supply system fault;

[0045] Collect vehicle-side sensor data, pre-process, extract features and filter the vehicle-side sensor data to obtain fault-sensitive features;

[0046] The fault-sensitive features are respectively input into the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models for inference evaluation, and the probabilities of N second suspension system faults, N second bearing faults and N second power system faults are output;

[0047] Through the Kalman filtering method, the model accuracy of the trained generative artificial intelligence model, the probabilities of the first suspension system failure, the first bearing failure and the first power supply system failure, the model accuracy of the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power supply system fault diagnosis models, and the probabilities of N second suspension system failures, N second bearing failures and N second power supply system failures are fused to obtain the fused probabilities of suspension system failures, bearing failures and power supply system failures.

[0048] The present invention evaluates the health of the whole vehicle in a collaborative decision-making manner, uses multiple fault diagnosis-specific algorithm models to process collected signals, uses a large model to process text descriptions, integrates the predicted fault probability and model test accuracy of a generative artificial intelligence model and a fault diagnosis professional model, and implements collaborative decision-making based on the Kalman filtering method, which optimizes the model fusion results, improves the model fault diagnosis effect and gives the error covariance of the calculation results.

[0049] According to an embodiment of the present invention, in the process of constructing a special domain knowledge data set, in order to combine structured data with unstructured data, in addition to using unstructured document data, a vehicle system architecture is constructed based on the MBSE concept, and its results are output as structured data such as pseudocode or markup language (XML / JSON / YAML). The specific process of the special domain knowledge data set is as follows: 1) System requirements analysis: comprehensively consider elements such as diagnostic speed, fault coverage, data acquisition frequency and quality, and algorithm model maintenance scalability. For example, after determining the fault coverage, the diagnostic speed can be determined according to the real-time requirements according to the fault type, and a suitable fault diagnosis professional model can be selected. In addition, considering the scalability of the subsequent algorithm model, an overall architecture such as microservices can be adopted. Finally, through indicator decomposition and management, the system performance is converted into specific system performance parameter indicators. 2) System architecture and function analysis: The whole vehicle architecture is divided into physical architecture and logical architecture. The physical architecture is further divided into mechanical system, electrical system and hydraulic system. The logical architecture includes algorithm software architecture, control system architecture and data relationship. The key architecture of the whole vehicle can be extracted by referring to the pictures in the description document, etc., to realize the analysis of the vehicle hardware layout and the definition of the interaction status of system components. After modeling, the system architecture decomposition diagram, vehicle subsystem state machine diagram, and vehicle operation activity diagram are output. 3) System model simulation: Based on the constructed system architecture and system requirements, the interaction and response analysis of each subsystem, the performance evaluation of the fault diagnosis system, the analysis of the information flow relationship of the diagnostic algorithm, and the verification of the consistency and correctness of the model are carried out to ensure the rationality and accuracy of the model architecture. 4) System model file output: The description results are converted into pseudocode or markup language (structured data such as XML / JSON / YAML, combined with unstructured document data in professional fields, to realize the construction of a dedicated domain knowledge data set as the pre-learning domain knowledge of the generative artificial intelligence model.

[0050] According to an embodiment of the present invention, in the process of constructing the instruction control data set, the instruction control text input by the user is output as the instruction code corresponding to the text, such as querying the AP01 instruction code corresponding to the current power. In order to ensure the coverage of the instruction semantics, one instruction can be labeled with multiple different input samples. The formed data set can be used for training the generative artificial intelligence model.

[0051] According to an embodiment of the present invention, in the process of constructing a professional knowledge question-and-answer dataset, the dataset consists of two parts: one is a PDF or word document such as a maintenance manual or instruction manual; the other is questions and answers in the form of AB. A is the fault problem input by the user, and B is the troubleshooting solution related to the problem. Some user questions can be annotated for each fault mode, and one fault mode corresponds to multiple questions. The formed dataset can be used for training a generative artificial intelligence model.

[0052] According to an embodiment of the present invention, in the process of constructing an intent recognition data set, the input is a natural language text input by the user, and the output is the intent corresponding to the text. There are three intents in total: command control mode, fault diagnosis mode, and professional field knowledge question and answer mode. After the command control, fault diagnosis, and professional field knowledge question and answer data sets are constructed, the training samples of the intent recognition module are directly synthesized. In order to ensure the recognition accuracy of the three intents by the intent recognition module, the training samples should include samples of these three intents, and the proportions of the three samples in the training set cannot differ too much. Specifically, the ratio of the number of samples in the command control data set, the first fault diagnosis data set in the fault diagnosis data set, and the professional field knowledge question and answer data set is 1:1:1.

[0053] According to one embodiment of the present invention, in the generative artificial intelligence model training process, the generative artificial intelligence model is first pre-trained using a dedicated domain knowledge data set, and then the training samples of four tasks (intention recognition, command control, fault diagnosis, and professional domain knowledge question and answer) are merged to retrain the generative artificial intelligence model to obtain a model that can perform multiple tasks simultaneously. Considering that the model has a large number of parameters and a small amount of data, full training is prone to overfitting, and the P-Tuning v2 fine-tuning method can be used to fix the parameters of the backbone model unchanged, and only a small number of new parameters are added to each layer for learning new tasks. During training, the Adam learning algorithm can be used for model training, and only the newly introduced small number of parameters are updated.

[0054] According to one embodiment of the present invention, in the process of collecting vehicle-side sensor data, data is collected from multiple vehicle-side sensors to obtain engine speed, oil pressure, battery voltage, vehicle network data flow, vehicle internal and external environment, etc., and the collected signals are pre-processed such as data cleaning, synchronization, conversion and normalization, and feature extraction and screening are performed. After the fault-sensitive features are obtained, they can be used for training and reasoning of professional fault diagnosis models.

[0055] According to an embodiment of the present invention, the following formula is used to fuse the probabilities of suspension system failure, bearing failure and power system failure:

[0056]

[0057] P k|k-1 =P k-1 +Q

[0058]

[0059] P k =(1-K k )·P k|k-1

[0060] in, m 1-2 or m 1-3 , P0=(1-a1) 2 ,

[0061] S1=m 2-1 、m 2-2 or m 2-3 , R = (1-b1) 2 、(1-b2) 2 or (1-b3) 2 ,

[0062] In the formula, is the probability estimate of the k-th prediction, is the probability estimate after the k-1th and kth updates, Δt is the time interval, P k|k-1 is the error covariance of the k-th prediction, P k-1 , P k is the error covariance after the k-1th and kth updates, Q is the process noise covariance, K k is the Kalman gain, R is the observation noise covariance, S k is the kth measurement probability result, is the initial probability estimate, m 1-1 、m 1-2 、m 1-3 are the probabilities of the first suspension system failure, the first bearing failure, and the first power system failure, respectively. P0 is the initial estimation error covariance. a1 is the model accuracy of the trained generative artificial intelligence model. S1 is the first set of measurement probability results. m 2-1 、m 2-2 、m 2-3 They are respectively the probabilities of the first suspension system fault, the first bearing fault, and the first power system fault output by the first group of trained any suspension system fault diagnosis model, any bearing fault diagnosis model, and any power system fault diagnosis model participating in the fusion; b1, b2, and b3 are the model accuracies of the first group of trained any suspension system fault diagnosis model, any bearing fault diagnosis model, and any power system fault diagnosis model participating in the fusion.

[0063] Below is m 1-1 and m 2-1 Take a fusion process as an example to illustrate:

[0064] 1) Kalman filter initialization:

[0065] Initial probability estimate

[0066] Initial estimation error covariance P0 = (1-a1) 2

[0067] 2) Prediction step:

[0068] The probability estimate for the current forecast

[0069] The current prediction error covariance P 1|0 =P0+Q

[0070] 3) Update steps

[0071] Observation noise covariance: R = (1-b1) 2

[0072] Calculate the current Kalman gain:

[0073] The current measurement probability result:

[0074] Updated error covariance: P1 = (1-K1)·P 1|0

[0075] The probability of the failure occurring is The estimated error covariance is P1.

[0076] Compared with using a single model, the accuracy of the prediction results is improved. In the fusion process, the test accuracy of the model after training is considered, and the error covariance after model fusion is calculated, which enhances the reliability of the prediction results.

[0077] According to an embodiment of the present invention, the cloud-based big model automatically calls the deployed fault diagnosis knowledge graph for analysis based on the current fault mode. If the knowledge graph fails to find an effective means, a comprehensive analysis is performed on the big model's understanding of the vehicle architecture and network information, and a recommended solution is given based on a professional knowledge question-and-answer model. At the same time, according to the user's wishes and based on the command control mode, the diagnostic information, vehicle health assessment results, and recommended maintenance methods are pushed to the vehicle, mobile phone, or 4S shop, etc., to generate customized maintenance methods and plans.

[0078] In order to have a further understanding of the present invention, the vehicle fault diagnosis method based on general and specialized collaborative decision-making of the present invention is described in detail below.

[0079] In this embodiment, the vehicle system architecture is constructed based on the MBSE concept, and the description results are converted into output model generation documents, pseudocode or markup languages ​​(XML / JSON / YAML, etc.) to facilitate the understanding and learning of the generative artificial intelligence model. The vehicle architecture design process based on MBSE is as follows: Figure 2As shown. The system sorts out the vehicle fault diagnosis requirements, comprehensively considers accuracy, diagnostic accuracy, diagnostic speed, fault coverage, data collection frequency and quality, algorithm model maintenance and scalability and other elements, uses SysML language to model the typical vehicle architecture, analyzes user needs, scenarios, and performance requirements, and refines the basic functional requirements of the system through SysML activity diagrams. Through indicator decomposition and management, the system performance is converted into specific system performance parameter indicators. To facilitate system management, the whole vehicle architecture is divided into physical architecture and logical architecture, among which the physical architecture is further subdivided into mechanical system, electrical system and hydraulic system, etc., which can realize the analysis of the layout of vehicle hardware (such as sensors and actuators). The logical architecture includes software algorithm architecture, control system architecture and data relationship, etc., which can realize the interaction and communication protocol definition between system components. In order to further sort out the whole vehicle architecture and functional operation logic, the system architecture decomposition diagram, vehicle subsystem state machine diagram and vehicle operation process activity diagram are used to model and explain the operation logic, related interfaces and transmission relationships between systems. Verify the integrity of the system function model, automatically generate sequence diagrams through simulation of each state machine diagram and activity diagram, conduct interaction and response analysis of each subsystem, performance evaluation of the fault diagnosis system, analysis of the information flow relationship of the diagnosis algorithm, and verification of model consistency and correctness, and save the generated vehicle internal and external system interface diagrams and the operation sequence diagrams of each system module of the vehicle. Generate documents, pseudocodes and markup languages ​​according to the vehicle overall architecture description file, and upload them to the cloud for fine-tuning of the large model so that the large model can understand the entire vehicle architecture, the composition of internal components and the signal flow relationship.

[0080] Fault diagnosis system architecture Figure 3As shown. Deploying a generative artificial intelligence model in the cloud as a cloud diagnosis center can realize algorithm scheduling management, data coordinated operation and maintenance, and centralized resource control. The sensors deployed on the vehicle side can obtain the parameters of the launch vehicle, vehicle speed, oil pressure, water temperature, vibration signals, etc. in real time. After data preprocessing such as cleaning, denoising and normalization, the fault characteristics are extracted. The screened features are used for intelligent algorithm reasoning and uploaded to the cloud for advanced data information archiving. A variety of intelligent diagnostic algorithms can be deployed on the vehicle side or in the cloud, such as suspension system fault diagnosis and life prediction models, bearing fault diagnosis and life prediction models, and power system diagnosis and life prediction models. These algorithm models can be used in the disconnected state to give prediction results for key components. The large model can be used as the comprehensive control center, model interaction center and diagnostic control center of the intelligent algorithm model in the networked state. When different intelligent algorithm modules need to perform data interaction operations, the large model can realize information understanding, transfer and processing, which is convenient for expanding large models of different forms at any time. The fault diagnosis knowledge graph can be deployed in the cloud or on the vehicle side. The fault diagnosis knowledge graph stores recommended solutions for common faults and faults with higher risk levels. The large model can comprehensively judge the current fault level based on the prediction results of the fault diagnosis model, automatically call the fault diagnosis knowledge graph, and find specific maintenance suggestions and countermeasures. If no effective measures are found in the knowledge graph, the relevant information is automatically stored and the results are queried online. The knowledge graph can provide auxiliary capabilities for the large model to reduce the impact of hallucination problems. Multiple intelligent diagnosis models of the same type with different implementation methods can be deployed in the cloud as redundant backups. When the confidence of the prediction results of the fault diagnosis or life prediction model deployed on the vehicle side is low, the large model can automatically call multiple cloud models to improve the prediction confidence and accuracy in an integrated learning manner.

[0081] The large model can combine its understanding of the overall architecture of the vehicle, integrate the prediction results of multiple intelligent diagnostic models to make judgments, and generate targeted vehicle maintenance recommendations, fault diagnosis lists and prediction results, and overall health assessment results. After the diagnostic information is summarized, it can be automatically output to the vehicle, mobile phone, or 4S store, etc. according to the user's choice. The present invention can effectively integrate the general knowledge and multi-source professional field knowledge in the generative artificial intelligence model, use the cloud-based generative artificial intelligence model to build a comprehensive processing, model interaction and diagnostic control center, coordinate multiple diagnostic and prediction algorithms, realize the health assessment of the whole vehicle and recommend maintenance methods.

[0082] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above methods when executing the computer program.

[0083] In summary, the present invention provides a vehicle fault diagnosis method and device based on general and professional collaborative decision-making, which has the following beneficial effects:

[0084] 1. When performing vehicle fault diagnosis, the present invention constructs the system architecture of the whole vehicle based on the MBSE concept, including the electrical system, hydraulic system and mechanical system, etc., and uses the SysML language to describe them. The system architecture is uploaded to the big model in the cloud in the form of text description, pseudo code or markup language as a training sample, so that it can understand the whole vehicle architecture, the composition of each internal component and the signal flow relationship, etc.;

[0085] 2. The present invention uses a collaborative decision-making method between generalists and professionals to evaluate the health of the entire vehicle, uses multiple fault diagnosis-specific algorithm models to process the collected signals, uses a large model to process text descriptions, integrates the predicted fault probability and model test accuracy of the generative artificial intelligence model and the fault diagnosis professional model, and implements collaborative decision-making between generalists and professionals based on the Kalman filtering method, optimizes the model fusion results, improves the model fault diagnosis effect, and provides the error covariance of the calculation results;

[0086] 3. The present invention uses a generative artificial intelligence model deployed in the cloud as the comprehensive processing center, model interaction center and diagnostic control center of the entire fault diagnosis system. It can realize the intelligent scheduling management of multiple fault diagnosis-specific algorithm models, and conduct data coordination management and automatic archiving of the features after vehicle-side signal collection. It can also provide targeted maintenance suggestions and solutions with the assistance of the fault diagnosis knowledge graph.

[0087] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0088] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. If not otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of the present invention.

[0089] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A vehicle fault diagnosis method based on collaborative decision-making between general and professional personnel. It is characterized in that The method comprises: Based on MBSE ideas, structured data and unstructured data, a dedicated domain knowledge dataset is constructed, where the output of the dedicated domain knowledge dataset is structured data; Constructing an instruction control data set, wherein the input of the instruction control data set is an instruction control text, and the output is an instruction code corresponding to the instruction control text; Constructing a fault diagnosis data set, wherein the fault diagnosis data set includes a first fault diagnosis data set for training a generative artificial intelligence model and a second fault diagnosis data set for training a dedicated fault diagnosis model, the first fault diagnosis data set is in the form of X1-Y, the second fault diagnosis data set is in the form of X2-Y, X1 represents the external manifestation of the fault, X2 represents the sensor acquisition signal, and Y represents the fault mode probability; Construct a professional domain knowledge question-answering dataset, where the professional domain knowledge question-answering dataset includes documents and questions and answers in the form of AB, where A represents the fault problem input by the user and B represents the troubleshooting solution related to the problem; Construct an intention recognition dataset, where the input of the intention recognition dataset is natural language text, and the output is the intention corresponding to the natural language text. The intention corresponding to the natural language text includes three, corresponding to the command control mode, fault diagnosis mode, and professional knowledge question and answer mode; Pre-training the generative artificial intelligence model based on the dedicated domain knowledge data set, re-training the generative artificial intelligence model based on the instruction control data set, the first fault diagnosis data set in the fault diagnosis data set, the professional domain knowledge question and answer data set, and the intention recognition data set to obtain a trained generative artificial intelligence model, and obtaining the model accuracy of the trained generative artificial intelligence model; Based on the second fault diagnosis data set in the fault diagnosis data set, N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models in the fault diagnosis special model are trained respectively to obtain the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models in the fault diagnosis special model, and obtain the model accuracy of each of the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models, where N≥1 and is an integer; The user's description of the external manifestation of the fault is input into the trained generative artificial intelligence model for reasoning and evaluation. The trained generative artificial intelligence model recognizes the intention of the user's description of the external manifestation of the fault, performs diagnosis when the intention is recognized as a fault diagnosis mode, and outputs the probabilities of the first suspension system fault, the first bearing fault, and the first power supply system fault; Collect vehicle-side sensor data, pre-process, extract features and filter the vehicle-side sensor data to obtain fault-sensitive features; The fault-sensitive features are respectively input into the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power system fault diagnosis models for inference evaluation, and the probabilities of N second suspension system faults, N second bearing faults and N second power system faults are output; Through the Kalman filtering method, the model accuracy of the trained generative artificial intelligence model, the probabilities of the first suspension system failure, the first bearing failure and the first power supply system failure, the model accuracy of the trained N suspension system fault diagnosis models, N bearing fault diagnosis models and N power supply system fault diagnosis models, and the probabilities of N second suspension system failures, N second bearing failures and N second power supply system failures are fused to obtain the fused probabilities of suspension system failures, bearing failures and power supply system failures.

2. The method according to claim 1, It is characterized in that The following formula is used to combine the probabilities of suspension system failure, bearing failure and power system failure: P k|k-1 =P k-1 +Q P k =(1-K k )·P k|k-1 in, m 1-2 or m 1-3 , P 0 =(1-a 1 ) 2 , S 1 =m 2-1 、m 2-2 or m 2-3 , R=(1-b 1 ) 2 、(1-b 2 ) 2 or (1-b 3 ) 2 , In the formula, is the probability estimate of the k-th prediction, k = 1 ~ N, is the probability estimate after the k-1th and kth updates, Δt is the time interval, P k|k-1 is the error covariance of the k-th prediction, P k-1 , P k is the error covariance after the k-1th and kth updates, Q is the process noise covariance, K k is the Kalman gain, R is the observation noise covariance, S k is the kth measurement probability result, is the initial probability estimate, m 1-1 、m 1-2 、m 1-3 are the probabilities of the first suspension system failure, the first bearing failure, and the first power supply system failure, respectively, P 0 is the initial estimation error covariance, a 1 is the model accuracy of the trained generative AI model, S 1 is the first set of measurement probability results, m 2-1 、m 2-2 、m 2-3 are the probabilities of the first suspension system fault, the first bearing fault, and the first power system fault output by any of the trained suspension system fault diagnosis models, any of the bearing fault diagnosis models, and any of the power system fault diagnosis models participating in the first group of fusion, respectively. 1 , b 2 , b 3 They are respectively the model accuracy of any suspension system fault diagnosis model, any bearing fault diagnosis model and any power system fault diagnosis model trained in the first group participating in the fusion.

3. The method according to claim 1 or 2, It is characterized in that The ratio of the number of samples in the instruction control data set, the first fault diagnosis data set in the fault diagnosis data set, and the professional field knowledge question and answer data set is 1:1:

1.

4. The method according to any one of claims 1 to 3, It is characterized in that Structured data is in pseudocode or markup language.

5. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the method according to any one of claims 1 to 4 is implemented.

Citation Information

Cited By

  • Method, system and equipment for automatically generating fault diagnosis rule of signal equipment

    CN121188212A