CN-122020469-A - Vehicle fault diagnosis method and device based on deep learning and knowledge graph

CN122020469ACN 122020469 ACN122020469 ACN 122020469ACN-122020469-A

Abstract

The invention discloses a vehicle fault diagnosis method and device based on deep learning and a knowledge graph, and the method comprises the steps of extracting fault text characteristics according to fault description text data, extracting vehicle numerical characteristics according to CAN bus time sequence data, extracting component visual characteristics according to fault component image data, obtaining multi-mode fusion characteristics according to the fault text characteristics, the vehicle numerical characteristics and the component visual characteristics, inputting the multi-mode fusion characteristics into a vehicle fault diagnosis model to obtain first fault root probability distribution, inputting the multi-mode fusion characteristics into the vehicle fault knowledge graph to perform fault reasoning to obtain second fault root probability distribution, performing attention fusion on the first fault root probability distribution and the second fault root probability distribution to obtain target fault root probability distribution, and determining vehicle fault roots according to the target fault root probability distribution. The invention improves the efficiency and accuracy of vehicle fault diagnosis and can be applied to the technical field of artificial intelligence.

Inventors

LIN JIAN
DONG LINGZHI
ZHANG JINMING
LAI JINGYU
LIANG MINHUA
MO RONGSHENG

Assignees

广汽本田汽车有限公司
广汽本田汽车研究开发有限公司

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (10)

1. The vehicle fault diagnosis method based on deep learning and knowledge graph is characterized by comprising the following steps: acquiring fault description text data, CAN bus time sequence data and fault component image data of a target vehicle; Extracting fault text features according to the fault description text data, extracting vehicle numerical features according to the CAN bus time sequence data, extracting component visual features according to the fault component image data, and obtaining multi-mode fusion features according to the fault text features, the vehicle numerical features and the component visual features; inputting the multi-mode fusion characteristics into a pre-trained vehicle fault diagnosis model to obtain a first fault root probability distribution; Inputting the multi-mode fusion characteristics into a pre-constructed vehicle fault knowledge graph for fault reasoning to obtain second fault root probability distribution; And performing attention fusion on the first fault root probability distribution and the second fault root probability distribution to obtain target fault root probability distribution, and determining the vehicle fault root according to the target fault root probability distribution.
2. The vehicle fault diagnosis method based on deep learning and knowledge graph according to claim 1, wherein the extracting fault text features according to the fault description text data, extracting vehicle numerical features according to the CAN bus time sequence data, extracting component visual features according to the fault component image data, and obtaining multi-modal fusion features according to the fault text features, the vehicle numerical features and the component visual features specifically comprises: semantic feature coding is carried out on the fault description text data through a pre-trained text coding model, so that the fault text features are obtained; dividing the CAN bus time sequence data into a plurality of vehicle numerical time sequence data according to the numerical type, extracting dynamic fluctuation characteristics of each vehicle numerical time sequence data, and carrying out weighted fusion on the dynamic fluctuation characteristics based on a hierarchical attention mechanism to obtain the vehicle numerical characteristics; identifying texture abnormal features and/or color abnormal features of the fault component image data through a pre-trained defect identification model, and taking the texture abnormal features and/or the color abnormal features as the component visual features; and performing feature stitching on the fault text feature, the vehicle numerical feature and the component visual feature to obtain the multi-mode fusion feature.
3. The vehicle fault diagnosis method based on deep learning and knowledge graph according to claim 1, wherein the vehicle fault diagnosis model is obtained by training the following steps: acquiring a fault description text sample, a CAN bus time sequence sample and a fault component image sample of a sample vehicle; Extracting multi-mode fusion sample characteristics according to the fault description text sample, the CAN bus time sequence sample and the fault component image sample, and determining corresponding fault root cause labels through manual labeling; Inputting the multi-modal fusion sample characteristics into a pre-constructed multi-branch convolutional neural network to obtain fault root cause prediction probability distribution; determining a loss value according to the fault root cause prediction probability distribution and the fault root cause label; updating parameters of the multi-branch convolutional neural network through a back propagation algorithm according to the loss value to obtain a trained vehicle fault diagnosis model; the multi-branch convolutional neural network comprises a text characteristic branch network, a numerical characteristic branch network, a visual characteristic branch network and a full-connection layer.
4. The vehicle fault diagnosis method based on deep learning and knowledge graph according to claim 1, wherein the vehicle fault knowledge graph is constructed by: Acquiring a preset vehicle fault case database, and extracting fault phenomena, fault codes, fault root causes and fault components corresponding to a plurality of vehicle fault cases according to the vehicle fault case database; Determining a plurality of entity nodes according to the fault phenomenon, the fault code, the fault root cause and the fault component, constructing corresponding relation edges according to relation types among the entity nodes, and determining the confidence degree of the relation edges to obtain the vehicle fault knowledge graph; wherein the relationship types include association, belonging, causing, and exclusion.
5. The vehicle fault diagnosis method based on deep learning and knowledge graph according to claim 1, wherein the inputting the multi-modal fusion feature into a pre-constructed vehicle fault knowledge graph to perform fault reasoning, to obtain a second fault root probability distribution, comprises: Determining node vector characteristics of each entity node of the vehicle fault knowledge graph; Screening a plurality of reasoning starting nodes from the entity nodes according to cosine similarity of the multi-mode fusion characteristics and the node vector characteristics; traversing the connected relation edges from the reasoning initial node based on a breadth-first search algorithm to obtain a plurality of fault reasoning paths; determining a plurality of corresponding target relation edges according to the fault reasoning path, and determining the confidence degree of each target relation edge; determining a corresponding attenuation coefficient according to the path length of the fault reasoning path, and determining the path confidence coefficient of the corresponding fault reasoning path according to the product of the confidence coefficient of each target relation edge and the attenuation coefficient; determining a corresponding target fault root cause according to the fault reasoning path, and determining a fault root cause confidence coefficient of the target fault root cause according to the path confidence coefficient; And determining the probability distribution of the second fault root according to the fault root confidence degrees of the target fault root.
6. The vehicle fault diagnosis method based on deep learning and knowledge graph according to claim 1, wherein the performing attention fusion on the first fault root probability distribution and the second fault root probability distribution to obtain a target fault root probability distribution, and determining a vehicle fault root according to the target fault root probability distribution specifically comprises: determining a first information entropy of the first fault root probability distribution and a second information entropy of the second fault root probability distribution; Determining a first probability distribution confidence coefficient of the first fault root probability distribution according to the first information entropy, and determining a second probability distribution confidence coefficient of the second fault root probability distribution according to the second information entropy; determining a first sum of the first probability distribution confidence and the second probability distribution confidence, determining a first attention weight of the first fault root probability distribution according to the ratio of the first probability distribution confidence to the first sum, and determining a second attention weight of the second fault root probability distribution according to the ratio of the second probability distribution confidence to the first sum; Weighting and summing the first fault root probability distribution and the second fault root probability distribution according to the first attention weight and the second attention weight to obtain the target fault root probability distribution; And determining a plurality of fault roots with the top probability sequence according to the probability distribution of the target fault root, wherein the fault root causes of the vehicle are determined.
7. The vehicle fault diagnosis method based on deep learning and knowledge-graph according to any one of claims 1 to 6, characterized in that it further comprises the steps of: Matching corresponding maintenance operation information from preset maintenance rules according to the vehicle fault root cause, and pushing the maintenance operation information to corresponding maintenance personnel; And acquiring maintenance operation feedback of the maintenance personnel, and incrementally updating the knowledge graph according to the maintenance operation feedback.
8. A vehicle fault diagnosis apparatus based on deep learning and knowledge map, characterized by comprising: The data acquisition module is used for acquiring fault description text data, CAN bus time sequence data and fault component image data of the target vehicle; The feature extraction module is used for extracting fault text features according to the fault description text data, extracting vehicle numerical features according to the CAN bus time sequence data, extracting component visual features according to the fault component image data, and obtaining multi-mode fusion features according to the fault text features, the vehicle numerical features and the component visual features; The fault diagnosis module is used for inputting the multi-mode fusion characteristics into a pre-trained vehicle fault diagnosis model to obtain first fault root probability distribution; The fault reasoning module is used for inputting the multi-mode fusion characteristics into a pre-constructed vehicle fault knowledge graph to perform fault reasoning so as to obtain second fault root probability distribution; The fault root cause determining module is used for carrying out attention fusion on the first fault root cause probability distribution and the second fault root cause probability distribution to obtain target fault root cause probability distribution, and determining the vehicle fault root cause according to the target fault root cause probability distribution.
9. An electronic device, comprising: At least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor is caused to implement a vehicle fault diagnosis method based on deep learning and knowledge-graph as claimed in any one of claims 1 to 7.
10. A computer program product comprising a computer program which, when executed by a processor, implements a vehicle fault diagnosis method based on deep learning and knowledge-graph according to any one of claims 1 to 7.

Description

Vehicle fault diagnosis method and device based on deep learning and knowledge graph Technical Field The invention relates to the technical field of artificial intelligence, in particular to a vehicle fault diagnosis method and device based on deep learning and knowledge maps. Background The current automobile maintenance fault diagnosis technology is mainly divided into three categories, namely a traditional experience-dependent type, a type which relies on a maintenance technician to judge through a fault phenomenon (such as abnormal sound and alarm lamp) and is low in efficiency and limited by experience level, a type which is a single data diagnosis type such as a Bosch FSA740 diagnosis instrument, a type which CAN not process intermittent faults without fault codes only by reading CAN bus fault code positioning problems, and a type which is a preliminary intelligent diagnosis type such as Tesla remote diagnosis system, realizes partial electric fault recognition based on vehicle-mounted data, does not fuse multi-mode data such as maintenance text, part images and the like, and lacks systematic migration of expert experience. In summary, the existing automobile maintenance fault diagnosis scheme has the following problems: 1) The data fracture results in diagnosis on one side, namely, only depending on fault codes or single sensor data, key information such as maintenance text description, visual characteristics of fault parts and the like is ignored, for example, engine shake can be caused by the problems of multiple systems of circuits, fuel oil and machinery, and single data cannot be accurately positioned; 2) The expert experience is difficult to multiplex, the diagnosis experience of the advanced technicians is mostly implicit knowledge, the existing system can not realize structured migration, and the diagnosis accuracy is insufficient when a new technician faces a complex scene; 3) The complex fault processing capability is weak, and the intermittent faults such as accidental flameout, difficult low-temperature start and the like lack of the time sequence data association analysis and reproduction condition prediction capability, so that the misdiagnosis rate of the traditional system is higher. The problems cause long maintenance period and high cost of vehicle faults, and influence user experience. Disclosure of Invention The present invention aims to solve at least one of the technical problems existing in the prior art to a certain extent. Therefore, an object of the embodiment of the present invention is to provide a vehicle fault diagnosis method based on deep learning and knowledge graph, which improves the efficiency and accuracy of vehicle fault diagnosis. Another object of an embodiment of the present invention is to provide a vehicle fault diagnosis device based on deep learning and knowledge maps. In order to achieve the technical purpose, the technical scheme adopted by the embodiment of the invention comprises the following steps: In one aspect, the embodiment of the invention provides a vehicle fault diagnosis method based on deep learning and a knowledge graph, which comprises the following steps: acquiring fault description text data, CAN bus time sequence data and fault component image data of a target vehicle; Extracting fault text features according to the fault description text data, extracting vehicle numerical features according to the CAN bus time sequence data, extracting component visual features according to the fault component image data, and obtaining multi-mode fusion features according to the fault text features, the vehicle numerical features and the component visual features; inputting the multi-mode fusion characteristics into a pre-trained vehicle fault diagnosis model to obtain a first fault root probability distribution; Inputting the multi-mode fusion characteristics into a pre-constructed vehicle fault knowledge graph for fault reasoning to obtain second fault root probability distribution; And performing attention fusion on the first fault root probability distribution and the second fault root probability distribution to obtain target fault root probability distribution, and determining the vehicle fault root according to the target fault root probability distribution. Further, in one embodiment of the present invention, the extracting fault text feature according to the fault description text data, extracting vehicle numerical feature according to the CAN bus time sequence data, extracting component visual feature according to the fault component image data, and obtaining multi-mode fusion feature according to the fault text feature, the vehicle numerical feature and the component visual feature specifically includes: semantic feature coding is carried out on the fault description text data through a pre-trained text coding model, so that the fault text features are obtained; dividing the CAN bus time sequence data into a plurality of vehicle numerical ti