Function navigation method and vehicle diagnostic equipment

Abstract

The present application relates to the technical field of diagnostic equipment and discloses a function navigation method and vehicle diagnostic equipment. The function navigation method includes: obtaining first vehicle data and a real-time diagnostic data stream of a target vehicle, as well as user behavior data; obtaining second vehicle data supported by the diagnostic equipment, as well as diagnostic steps associated with each second vehicle data and fault information; mapping the first vehicle data classification to the second vehicle data through a semantic mapping engine to determine the target vehicle data of the target vehicle in the diagnostic equipment; generating an optimized diagnostic path for the target vehicle through a dynamic logic generator; and generating recommended diagnostic steps by combining the recommendation results of vehicle model collaborative filtering, user behavior prediction, and expert mode. The present application effectively solves the problem of the inability to align actual vehicle data with diagnostic equipment data, provides users with an optimized diagnostic path and personalized recommended diagnostic steps, simplifies the operating steps of the vehicle diagnostic equipment, and improves the user experience.

Description

Technical Field

[0001] The present application relates to the field of vehicle diagnosis technology, and in particular to a function navigation method and vehicle diagnosis equipment. Background Art

[0002] A common pain point in the current industry is that after users connect the target vehicle to the vehicle diagnostic equipment, they need to click through multiple pages before jumping to the page they need. This is extremely unfriendly for novices or newbies. Based on this pain point, an AI large model application is proposed to improve the functional navigation experience on automotive diagnostic equipment. However, many problems still exist. After all, the information scanned by the vehicle diagnostic equipment, such as brand, model and year, may not necessarily align with the data on the diagnostic equipment. Summary of the Invention

[0003] In view of this, an embodiment of the present application provides a function navigation method and a vehicle diagnostic device, which can effectively solve the problem of the inability to align actual vehicle data with diagnostic device data.

[0004] In a first aspect, an embodiment of the present application provides a function navigation method, comprising:

[0005] Acquire first vehicle data and real-time diagnostic data stream of the target vehicle, as well as user behavior data;

[0006] Acquire second vehicle data supported by the diagnostic device, and diagnostic steps associated with each of the second vehicle data and fault information;

[0007] Classify and map the first vehicle data into the second vehicle data through a semantic mapping engine to determine target vehicle data of the target vehicle in the diagnostic device;

[0008] generating an optimized diagnostic path for the target vehicle by a dynamic logic generator based on the target vehicle data, the real-time diagnostic data stream, and the diagnostic steps;

[0009] Combine the recommendation results of vehicle model collaborative filtering, user behavior prediction and expert model to generate recommended diagnostic steps.

[0010] In a second possible embodiment of the first aspect, obtaining the first vehicle data and the real-time diagnostic data stream of the target vehicle, as well as the user behavior data, includes:

[0011] Collecting and parsing the real-time data stream of the target vehicle through an OBD data collector to obtain the first vehicle data and the real-time diagnostic data stream, wherein the real-time diagnostic data stream includes fault codes and status information of the target component;

[0012] The user behavior data of the user when using the diagnostic device is recorded through the user behavior log.

[0013] In a second possible embodiment of the first aspect, the fault information includes a fault code and a fault symptom, and the obtaining of second vehicle data supported by the diagnostic device and the diagnostic steps associated with each of the second vehicle data and the fault information include:

[0014] Performing document recognition on the input manufacturer's technical bulletin document using a multi-document summarization model to obtain the second vehicle data and a knowledge triple of the second vehicle data, the fault information, and the diagnostic steps;

[0015] Performing target detection on the maintenance manual illustration to identify and mark the target component in the maintenance manual illustration;

[0016] A diagnostic reference threshold of the target component in the maintenance manual text is extracted through a regular expression template, and a knowledge triple of the second vehicle data, the fault information and the diagnostic steps is generated, wherein the fault phenomenon of the target vehicle is determined based on the status information of the target component and the diagnostic reference threshold.

[0017] In a third possible embodiment of the first aspect, the target vehicle, in the target vehicle data, includes target brand data, target vehicle model code, and target model year data, and the classifying and mapping the first vehicle data to the second vehicle data by using a semantic mapping engine includes:

[0018] calculating a brand association between first brand data in the first vehicle data and a plurality of second brand data in the second vehicle data;

[0019] Mapping the first brand data to one of the second brand data according to the brand association to obtain the target brand data;

[0020] Calculating semantic similarity or visual similarity between first vehicle model data in the first vehicle data and a plurality of second vehicle model data in the second vehicle data;

[0021] Mapping the first vehicle model data to one of the second vehicle model data according to the semantic similarity or visual similarity, and outputting a vehicle model code corresponding to the second vehicle model data to obtain the target vehicle model code;

[0022] Calculating a matching weight between the first model year data in the first vehicle data and the second model year data in the second vehicle data according to an exponential decay function of a time decay model, wherein a decay coefficient of the exponential decay function is determined by fitting a historical model year mapping result;

[0023] The first model year data is mapped to one of the second model year data according to the matching weight to obtain the target model year data.

[0024] In a fourth possible embodiment of the first aspect, calculating the brand association between the first brand data in the first vehicle data and the plurality of second brand data in the second vehicle data includes:

[0025] Building a brand knowledge graph based on the multidimensional relationship types and brand embedding vectors of the first brand data and each of the second brand data;

[0026] Based on the brand embedding vector of each brand node in the brand knowledge graph, the brand association between the first brand data and each of the second brand data is calculated through a multi-head self-attention mechanism.

[0027] In a fifth possible embodiment of the first aspect, calculating the semantic similarity or visual similarity between the first vehicle model data in the first vehicle data and the plurality of second vehicle model data in the second vehicle data includes:

[0028] Converting the vehicle type text in the input first vehicle type data into a first semantic vector using a pre-trained language model;

[0029] Calculating the semantic similarity based on the first semantic vector and the second semantic vector of the second vehicle type data;

[0030] Multi-scale feature extraction is performed on the first vehicle model image in the first vehicle model data, so as to calculate the visual similarity based on the first vehicle model image after feature extraction and the second vehicle model image in the second vehicle model data.

[0031] In a sixth possible embodiment of the first aspect, generating an optimized diagnostic path for the target vehicle by a dynamic logic generator based on the target vehicle data, the real-time diagnostic data stream, and the diagnostic step includes:

[0032] Constructing a diagnostic decision tree to select the corresponding diagnostic steps according to the target vehicle data and the fault information;

[0033] The diagnostic steps are path optimized by a heuristic search algorithm to obtain the optimized diagnostic path.

[0034] In a seventh possible embodiment of the first aspect, the step of generating a recommended diagnosis by combining vehicle model collaborative filtering, user behavior prediction, and expert model recommendation results includes:

[0035] Extracting user behavior features based on the user behavior data to build a user behavior model;

[0036] Recommending a first initial diagnosis step corresponding to the second vehicle data having the highest vehicle model matching degree between the target vehicle data and other second vehicle data based on the vehicle model matching degree between the target vehicle data and the second vehicle data;

[0037] Analyzing the user behavior characteristics through a long short-term memory network to generate a second initial diagnosis step consistent with the user behavior data;

[0038] Determining a third initial diagnostic step based on the correspondence between the target vehicle data, the fault code, and the diagnostic steps in the expert knowledge graph;

[0039] At least one of the initial diagnostic steps is determined to be the recommended diagnostic step based on the vehicle model collaborative filtering, the user behavior prediction, and the recommendation weight of the expert mode.

[0040] In an eighth possible embodiment of the first aspect, the function navigation method further includes:

[0041] Based on the vehicle model feature vector, fault code feature vector and user behavior data of the target vehicle data, each diagnostic step is predicted through a pre-trained language model, and the diagnostic step with corresponding confidence and temperature map is output.

[0042] In a second aspect, an embodiment of the present application provides a vehicle diagnostic device, which includes a processor and a memory, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement the above-mentioned function navigation method.

[0043] The embodiments of the present application have the following beneficial effects:

[0044] A function navigation method of an embodiment of the present application includes: obtaining first vehicle data and real-time diagnostic data stream of a target vehicle, as well as user behavior data; obtaining second vehicle data supported by a diagnostic device, and diagnostic steps associated with each of the second vehicle data and fault information; mapping the first vehicle data classification to the second vehicle data through a semantic mapping engine to determine the target vehicle data of the target vehicle in the diagnostic device; generating an optimized diagnostic path for the target vehicle through a dynamic logic generator based on the target vehicle data, real-time diagnostic data stream and diagnostic steps; and generating recommended diagnostic steps by combining the recommendation results of vehicle model collaborative filtering, user behavior prediction and expert mode. Based on the above scheme, the present application can map the actual vehicle data of the target vehicle to the vehicle data supported by the diagnostic device, and can provide users with target vehicles and optimized diagnostic paths, as well as personalized recommended diagnostic steps, which greatly simplifies the operating steps of the vehicle diagnostic device and improves the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0046] Figure 1 A first flow chart of the function navigation method according to an embodiment of the present application is shown;

[0047] Figure 2 A second flow chart of the function navigation method according to an embodiment of the present application is shown;

[0048] Figure 3 A third flow chart of the function navigation method according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0050] The components of the embodiments of the present application generally described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but rather merely represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0051] Hereinafter, the terms "including", "having" and their cognates used in various embodiments of the present application are intended only to indicate specific features, numbers, steps, operations, elements, components or combinations of the aforementioned items, and should not be understood as excluding the existence of one or more other features, numbers, steps, operations, elements, components or combinations of the aforementioned items or adding the possibility of one or more features, numbers, steps, operations, elements, components or combinations of the aforementioned items. In addition, the terms "first", "second", "third" and the like are only used to distinguish descriptions and should not be understood as indicating or implying relative importance.

[0052] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art to which the various embodiments of the present application belong. Terms (such as those defined in generally used dictionaries) will be interpreted as having the same meaning as in the context of the relevant technical field and will not be interpreted as having an idealized meaning or an overly formal meaning unless clearly defined in the various embodiments of the present application.

[0053] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments may be combined with each other.

[0054] Figure 1 A flow chart of a function navigation method according to an embodiment of the present application is shown. Exemplarily, the function navigation method includes the following steps:

[0055] S110, obtaining first vehicle data, real-time diagnostic data stream, and user behavior data of a target vehicle.

[0056] In an embodiment of this application, an OBD data collector is used to collect and parse the target vehicle's real-time data stream to obtain first vehicle data and a real-time diagnostic data stream. OBD (On-Board Diagnostics) is a standardized system used in modern vehicles to monitor and diagnose vehicle performance. Through the OBD interface, various vehicle operating parameters and fault information can be obtained in real time. In this application, the OBD data collector is the core component for collecting data from the target vehicle. Connecting the OBD data collector to the vehicle's OBD-II interface extracts key first vehicle data and a real-time diagnostic data stream from the target vehicle.

[0057] For example, in vehicle diagnostic equipment, the OBD data collector needs to process data from different bus protocols, such as CAN (Controller Area Network) and LIN (Local Interconnect Network). To ensure that the data can be correctly parsed and understood, the OBD data collector is equipped with a multi-protocol parsing module. This multi-protocol parsing module includes an adapter, a listener, and an adapter, which are responsible for extracting, monitoring, and adapting different types of real-time diagnostic data streams, respectively.

[0058] In one embodiment, the first vehicle data is the target vehicle's model information, including first brand data, first model data, and first model year data. The first vehicle data is typically derived from the vehicle's VIN (Vehicle Identification Number). The VIN is a unique identifier for each vehicle and includes information such as the manufacturer code, place of production, and model year. An OBD data collector can read the VIN data stored by the ECU (Electronic Control Unit) via the OBD protocol. A decoder is used to parse the brand, model, and model year fields in the VIN.

[0059] Exemplarily, the real-time diagnostic data stream includes the target vehicle's status data and fault codes. For example, status data can include engine speed, vehicle speed, door status, and light status. Diagnostic Trouble Codes (DTCs) are standardized codes used to describe potential problems or faults in vehicle systems. These fault codes are generated by the on-board diagnostics (OBD) system and stored in the vehicle's electronic control unit. In this application, by reading these fault codes, corresponding diagnostic steps can be executed to locate the problem and perform precise repairs. By reading status data, the current vehicle fault symptom can be determined based on diagnostic reference thresholds.

[0060] In one embodiment, a decoder is used to parse the data stream of the CAN / LIN bus protocol and extract specific data fields and attribute values ​​of the real-time diagnostic data stream, such as engine speed and vehicle speed. For example, the decoder can be a CAN / LIN bus decoder that parses 11 / 29-bit identifiers according to the ISO 15765-2 standard. A listener is used to monitor and collect real-time diagnostic data streams on the LIN bus, including periodic status information such as door status and light status. For example, the listener can be a LIN bus listener that uses master-slave polling to collect periodic data. An adapter (such as a custom protocol adapter) is used to load a manufacturer-specific DLL dynamic link library to parse the manufacturer-specific diagnostic protocol and data format and extract the corresponding data fields and attribute values ​​of the real-time diagnostic data stream, such as fault codes and inverter temperature thresholds for Toyota hybrid systems.

[0061] In one embodiment, the present application records user behavior data when using diagnostic equipment through user behavior logging. User behavior data refers to the user's current or historical operations and interactions with automotive diagnostic equipment. User behavior logging is a technical means of recording user behavior, system status, or business processes within the vehicle diagnostic equipment application.

[0062] In one embodiment, the user behavior log burial point can be triggered when using the diagnosis, including the user's function call sequence and parameter setting values. The user behavior log burial point can be triggered during interface operation, and the user interface operation can be recorded in real time at a set frequency, recording the user's click position on the screen and the time the user stays on a certain interface or function module. The user behavior log burial point can be triggered when the user performs a voice command operation, and the voice command record can be recorded. For example, the user enters the voice command "Check the high-pressure oil pump circuit of a certain brand, a certain model and a certain year", and the current user's command text and the historical user's command text and the corresponding operations performed are recorded.

[0063] S120: Acquire second vehicle data supported by the diagnostic device, and diagnostic steps associated with each of the second vehicle data and fault information.

[0064] In one embodiment, the present application performs document recognition on the input manufacturer's technical announcement document through a multi-document summary model to obtain the second vehicle data, as well as the knowledge triples of vehicle data, fault information, and diagnostic steps. The present application understands the semantic relationship of the document through a multi-document summary model based on BERT (Bidirectional Encoder Representations from Transformers). The BERT-based multi-document summary model is a text summary generation tool that combines natural language processing (NLP) and deep learning technology. It understands and extracts key information from multiple documents through pre-trained language models (such as BERT) and generates concise and accurate summaries. In the present application, the BERT-based multi-document summary model is used to extract key information from the manufacturer's technical announcement document, determine the second vehicle data supported by the diagnostic equipment, and generate structured knowledge triples.

[0065] Manufacturer Technical Bulletins (MTBs) are technical documents published by automakers to provide detailed guidance to repair technicians, 4S dealerships, and third-party diagnostic equipment regarding vehicle system failures, maintenance, and repairs. These documents, typically published in PDF or HTML format, cover various vehicle systems (such as the engine, transmission, and chassis) and provide solutions to specific issues.

[0066] In one embodiment, a manufacturer's technical bulletin document in PDF / HTML format is input into a BERT-based multi-document summarization model, and a tool (such as LayoutParser) is used to identify the document structure, extract fault information, diagnostic steps, and applicable vehicle models, and output a structured knowledge triple: <applicable vehicle model, fault information, solution>.

[0067] In one embodiment, the present application performs target detection on maintenance manual illustrations to identify and label target components in the maintenance manual illustrations. For example, by training a YOLOv5 model to identify target components (such as engines), the positions of these target components can be quickly located and labeled.

[0068] In another embodiment, the present application uses a regular expression template to extract diagnostic reference thresholds for target components from the repair manual text and generate a knowledge triple consisting of second vehicle data, fault information, and diagnostic steps. Regular expression (Regex) templates are tools for matching strings, efficiently extracting information with specific patterns from unstructured text. In this application, regular expression templates are used to extract diagnostic steps and diagnostic reference thresholds corresponding to key vehicle models and fault information from the repair manual text.

[0069] In this embodiment, the present application determines the fault phenomenon of the target vehicle based on the state information of the target component and the diagnostic reference threshold. For example, the engine power of the target vehicle is matched according to the power diagnostic reference threshold to determine whether the actual engine power is lower than the standard value and the extent of the lower standard value; the engine torque of the target vehicle is matched according to the torque diagnostic reference threshold to determine whether the actual engine torque is lower than the standard value and the extent of the lower standard value; the engine pressure of the target vehicle is matched according to the pressure threshold diagnostic reference threshold to determine whether the actual engine pressure is lower than the standard value and the extent of the lower standard value. After the above matching, the engine fault phenomenon can be determined.

[0070] In one embodiment, when defining the diagnostic steps and diagnostic reference thresholds extracted from the standard expression template, the corresponding indicators may include various technical parameters, fault codes, and key measurement values ​​in the diagnostic steps mentioned in the repair manual. The specific content of these indicators depends on the specific content and format of the repair manual, as well as the vehicle model and fault type. In the standard expression template, key measurement values ​​are defined based on the textual characteristics of the repair manual and generally correspond to the following indicator types: power parameters: such as engine power (unit: kW / PS), torque (unit: N·m), displacement (unit: L); dimensional parameters: wheelbase (mm), vehicle length / width / height (mm); electrical parameters: rated voltage (V), current (A), sensor range (e.g., temperature range °C); diagnostic thresholds: fault code trigger conditions (e.g., speed threshold rpm, pressure threshold bar); time parameters: maintenance interval (km / month), test duration (s).

[0071] In another embodiment, the present application establishes a dynamic data lake architecture, providing an efficient, flexible, and scalable data storage and processing platform to support the optimization of the functional navigation experience of large AI models on automotive diagnostic equipment. The dynamic data lake can integrate heterogeneous data from multiple sources, including first-vehicle data, real-time diagnostic data streams, second-vehicle data supported by diagnostic equipment, and diagnostic steps. Through steps such as data cleaning, feature extraction, and knowledge graph construction, it provides high-quality data support for subsequent deep semantic mapping, adaptive navigation engines, and continuous learning systems.

[0072] Specifically, the dynamic data lake architecture comprises a three-tier storage structure: a raw data layer using Parquet columnar storage to preserve raw byte streams; a feature data layer using the Delta Lake format to store vectorized features; and a semantic data layer using the Neo4j graph database to store knowledge graph relationships. The dynamic data lake architecture can be used to implement a data cleansing pipeline, which includes raw data, timestamp alignment, field masking, and outlier removal. Raw data refers to unprocessed data, including diagnostic data streams, technical bulletin documents, repair manual illustrations and text, and user behavior logs. Before entering the dynamic data lake, this data requires preprocessing, including timestamp alignment to ensure temporal consistency, field masking to protect user privacy and data security, and outlier removal (e.g., using the Laida principle) to improve data quality and accuracy. Compared to collected real-time diagnostic data streams, raw data is more extensive and diverse, including non-real-time and historical data, while real-time diagnostic data streams focus on data streams generated during the real-time diagnostic process.

[0073] S130 , classify and map the first vehicle data into the second vehicle data through a semantic mapping engine to determine target vehicle data of the target vehicle in the diagnostic device.

[0074] In the embodiments of this application, the first vehicle data includes the first brand data, first model data, and first model year data of the target vehicle. Because the description or expression of the second vehicle data in the diagnostic device differs from the actual first vehicle data of the target vehicle, it is easy for the first vehicle data to not align with the second vehicle data on the diagnostic device. To address this issue, this application classifies and maps the first vehicle data to the second vehicle data to resolve the problem of vehicle data misalignment.

[0075] In one embodiment, if Figure 2 As shown, the present application maps the first vehicle data classification to the second vehicle data, including the following steps:

[0076] S131, calculating the brand association between the first brand data in the first vehicle data and the plurality of second brand data in the second vehicle data.

[0077] In one embodiment, the present application constructs a brand knowledge graph based on the multidimensional relationship types and brand embedding vectors of the first brand data and each second brand data. The multidimensional relationship types include holding, technology alliance, platform sharing, supplier supply, patent cross-licensing, and market area overlap. The definition of these relationship types enables the brand knowledge graph to more comprehensively capture the complex associations between brands. Each brand can be regarded as a brand knowledge graph node, and different relationship types can be used as edges to construct an initial brand knowledge graph. The present application can learn the low-dimensional embedding representation of nodes from the brand knowledge graph through the GraphSAGE (Graph Sample and Aggregate) algorithm, and generate the brand embedding vector of the brand node by sampling and aggregating the information of neighbor nodes. The brand embedding vector retains the topological information and semantic features of the brand in the knowledge graph.

[0078] In another embodiment, the present application also calculates the brand association of the first brand data and each second brand data based on the brand embedding vector of each brand node in the brand knowledge graph through a multi-head self-attention mechanism.

[0079] The formula for calculating the attention weight in the multi-head self-attention mechanism is as follows:

[0080] ;

[0081] Represents query (Query), key (Key) and value (Value), respectively. A vector representing the target information that currently needs to be focused on or retrieved, Is a vector representing the characteristics of each position (or element) to respond to the "question" of Query, Is the data vector that actually needs to be extracted or used. In the multi-head self-attention mechanism, The inputs to multiple heads are generated through multiple independent linear transformations. Each head learns different subspace features, and the output is ultimately concatenated and linearly transformed. This approach enhances the model's expressiveness and better captures the complex dependencies between first- and second-brand data.

[0082] Expressed in the form of matrices, these three matrices are obtained by combining the brand embedding vector e of the first brand data with the linear transformation matrix 、 、 Multiply them together to get . ; ; . 、 、 is a linear transformation matrix used to map the brand embedding vector of the first brand data to a different space to better capture the association between the first brand data and the second brand data. Indicates the dimension of the key. When calculating the attention weight, in order to prevent the inner product from being too large and causing the Softmax function to saturate, the inner product result is divided by Softmax is a normalization function used to convert each element in the association score matrix into a probability value so that the sum of the association strengths between each first brand data and all second brand data is 1. Represents the final attention output, which is obtained by adding the normalized correlation score matrix to the value The output matrix obtained by matrix multiplication contains the association information between the first brand data and the second brand data, which is used for subsequent brand association calculations.

[0083] The final correlation calculation formula is:

[0084]

[0085] Indicates the final brand association between the first brand data and the second brand data, Represents the number of attention heads. The multi-head attention mechanism captures various correlation patterns between the first brand data and the second brand data by using multiple attention heads. Indicates the The correlation strength between the first brand data i and the second brand data j under the attention head. This value is obtained by the above attention weight calculation formula. Represents the controlling relationship weight. If there is a controlling relationship between the first brand data and the second brand data, then Equal to 1, otherwise Equal to 0. Used to quantify the impact of this controlling relationship on brand relevance. Represents the technical alliance weight. If there is a technical alliance relationship between the first brand data and the second brand data, then Equal to 1, otherwise Equal to 0. Used to quantify the impact of this technology alliance relationship on brand relevance. and It is a binary identifier of the graph correspondence relationship. If there is a controlling relationship or a technical alliance relationship between the first brand data and the second brand data, the corresponding relationship identifier is 1, otherwise it is 0.

[0086] It can be understood that by substituting the above indicators into the final correlation calculation formula, the final brand correlation between the first brand data and the second brand data can be obtained.

[0087] S132 : Map the first brand data to one of the second brand data according to the brand association to obtain target brand data.

[0088] Exemplarily, if the first brand data has the highest correlation with one of the second brand data, the first brand data may be mapped to one of the second brand data, and the target brand data is the second brand data having the highest correlation with the first brand data.

[0089] S133 , calculating semantic similarity or visual similarity between the first vehicle model data in the first vehicle data and the plurality of second vehicle model data in the second vehicle data.

[0090] In one embodiment, the present application converts the vehicle type text in the input first vehicle type data into a first semantic vector through a pre-trained language model; and calculates the semantic similarity based on the first semantic vector and the second semantic vector of the second vehicle type data.

[0091] In one embodiment, the present application extracts vehicle model text information through regularized point detection. Regularized point detection is a rule-based method that uses a predefined set of vehicle model parsing rules to extract specific information (such as model year, drive type, powertrain configuration, etc.) from the vehicle model description text. The pre-trained language model can be the ELECTRA model, a pre-trained language model based on the Transformer architecture that excels at capturing semantic information in text. In this application, the ELECTRA model is used to generate a first semantic vector for the vehicle model description text, thereby supporting fuzzy matching tasks.

[0092] In another embodiment, the present application performs multi-scale feature extraction on the first vehicle model image in the first vehicle model data to calculate visual similarity based on the first vehicle model image after feature extraction and the second vehicle model image in the second vehicle model data.

[0093] In this embodiment, the present application uses ResNet-50 to extract the global features of the first vehicle model image. ResNet-50 is an architecture based on a deep convolutional neural network that excels at extracting high-level semantic features from images. The first vehicle model image is fed into the ResNet-50 model as input. The ResNet-50 model extracts the features of the first vehicle model image layer by layer through a series of convolutional layers, pooling layers, and residual connections, and finally outputs a first vehicle model vector, which represents the global features of the image. The present application determines the visual similarity by calculating the similarity between the first vehicle model vector of the first vehicle model image and the vehicle model vector of the second vehicle model image.

[0094] S134 , mapping the first vehicle model data to one of the second vehicle model data according to semantic similarity or visual similarity, and outputting a vehicle model code corresponding to the second vehicle model data to obtain a target vehicle model code.

[0095] Exemplarily, if the first vehicle model data has the greatest correlation with one of the second vehicle model data, the first vehicle model data can be mapped to one of the second vehicle model data, where different second vehicle model data correspond to different vehicle model codes. This application uses the vehicle model code corresponding to the second vehicle model data as the target vehicle model code.

[0096] S135 , calculating matching weights of the first model year data in the first vehicle data and the second model year data in the second vehicle data according to an exponential decay function of the time decay model, wherein the decay coefficient of the exponential decay function is determined by fitting the historical model year mapping results.

[0097] In one embodiment, the formula for the exponential decay function is:

[0098]

[0099] in, represents the matching weight of the first vehicle data and the second vehicle data, , Indicates the current first year model data Compared with the second year data The time difference, in years, is the attenuation coefficient, which is determined by fitting the historical year mapping results. The typical value of the attenuation coefficient is 0.2~0.5.

[0100] For example, if the diagnostic device does not support the 2002 model, the 2001 model is selected when mapping the first model year data. =1, if =0.3, then the matching weight is As can be understood, this time decay model ensures that the closest year matches have the highest priority while avoiding completely eliminating old data.

[0101] In one embodiment, the present application collects the historical model year mapping results of each first model year data and combines the maximum likelihood function fitting to determine the attenuation coefficient of the exponential decay function. Our core idea is to find a value, so that under given data conditions, the probability of observing the current data (i.e. likelihood) reaches the maximum. For the exponential decay function , set N observation samples, where each sample corresponds to a time difference and binary labels , Indicates the The time difference of the observation samples, Indicates the Binary labels for observation samples.

[0102] This application describes the likelihood of observation data through a probability model based on exponential decay weights. Under this probability model, the probability of successful matching can be associated with the exponential decay weights. For example, assuming the probability of successful matching is and is proportional to (or related to after some transformation), where The time difference between the first year model data and the second year model data is The corresponding matching weight when ; and the probability of matching failure is .

[0103] In this embodiment, for N independent and identically distributed samples, the likelihood function is the product of the joint probabilities of all samples:

[0104]

[0105] When the probability of successful matching is assumed , the probability of matching failure , then the likelihood function can be expressed as:

[0106]

[0107] In order to facilitate calculation, this application takes the logarithm of the likelihood function and obtains the log-likelihood function :

[0108]

[0109] The goal of this application is to find a function that makes the log-likelihood function The largest The value can be achieved by taking the derivative and setting it to zero. about Derivative:

[0110]

[0111] The above equations usually have no analytical solutions and need to be solved using numerical optimization methods (such as Newton-Raphson method, gradient ascent method, etc.) It can be understood that the maximum likelihood function provides a way to estimate the model parameters (here is the attenuation coefficient) under given data. ) framework, by constructing a likelihood function based on an exponential decay function and maximizing it, the decay coefficient that best fits the observed data can be found. , which enables the model to better fit the time decay pattern in the data.

[0112] S136 , mapping the first model year data to one of the second model year data according to the matching weight to obtain the target model year data.

[0113] Exemplarily, if the matching weight between the first-year model data and one of the second-year model data is the largest, the first-year model data can be mapped to one of the second-year model data, and this application uses the second-year model data as the target vehicle model code.

[0114] S140 , generating an optimized diagnostic path for the target vehicle through a dynamic logic generator based on the vehicle mapping information, the real-time diagnostic data stream, and the fault diagnosis data.

[0115] In one embodiment, the present application constructs a diagnostic decision tree to select corresponding diagnostic steps based on the target vehicle data and fault information. The diagnostic decision tree is a hierarchical diagnostic model that, through gradual screening and refinement, ultimately locates specific diagnostic steps and forms an initial diagnostic path. Each node contains preconditions and execution actions, and branches point to the next node or final conclusion based on the diagnostic results. Among them, the preconditions of the diagnostic decision tree include the target vehicle data and fault information, and the execution actions are the set of all diagnostic steps corresponding to the target vehicle data and fault information.

[0116] In another embodiment, the present application uses a heuristic search algorithm (such as the A* search algorithm) to optimize the diagnostic steps and obtain an optimized diagnostic path. This heuristic search algorithm uses a specific cost function, F(n) = G(n) + 0.7H(n), to guide the path optimization process. G(n) represents the time taken to execute the step; H(n) represents the estimated complexity of the remaining steps from the current node to the target node (heuristic function). F(n) represents the total cost, which measures the total cost from the starting node to the target node through the current node n. The weight of 0.7 reduces the influence of the heuristic estimate H(n), making the algorithm more focused on the actual time taken to execute the step, G(n).

[0117] It can be understood that this application compares the F(n) values ​​of different nodes, and the A* algorithm selects the node with the smallest F(n) for expansion, gradually approaching the target node. In multiple diagnostic steps, adjusting the weights can optimize the execution order of the diagnostic steps and reduce overall time consumption. For example, based on the collected fault information and mapped target vehicle data, an optimized diagnostic path for the fuel system of a certain brand, model, and year of vehicle is dynamically generated. This optimized diagnostic path may include steps such as reading fault codes, performing a fuel pressure test, checking the high-pressure fuel pump circuit, and refreshing the ECU calibration.

[0118] S150 combines the recommendation results of vehicle model collaborative filtering, user behavior prediction, and expert mode to generate recommended diagnostic steps.

[0119] In one embodiment, this application extracts user behavior features from user behavior data to construct a user behavior model. These features include average step duration, operation correction rate, and brand preferences. The extracted behavioral features comprehensively reflect user operating habits and preferences. These features provide important input for subsequent recommendations, making them more personalized.

[0120] In another embodiment, Figure 3 As shown, the application generates a recommended diagnosis step by comprehensively recommending the results, including the following steps:

[0121] S151 , based on the vehicle type matching degree between the target vehicle data and other second vehicle data, recommending a first initial diagnosis step corresponding to the second vehicle data having the highest vehicle type matching degree between the target vehicle data and the second vehicle data.

[0122] For example, this application can calculate vehicle model similarity based on the feature vectors of the target vehicle data and other second vehicle data using similarity calculation methods (such as cosine similarity and Euclidean distance). For a new vehicle model or new user's target vehicle data, the application can find the most similar second vehicle data. The application can also analyze user behavior (such as common diagnostic steps) associated with the second vehicle data to generate a first initial diagnostic step.

[0123] It can be understood that vehicle model-based collaborative filtering is an effective way to solve the cold start problem. Its core idea is to use the user behavior of similar vehicle models to infer the diagnostic steps of the target vehicle.

[0124] S152, analyzing the user behavior characteristics through the long short-term memory network to generate a second initial diagnosis step consistent with the user behavior data.

[0125] In the embodiments of this application, a Long Short-Term Memory (LSTM) network is a special type of recurrent neural network (RNN) that is used to process and predict long-term dependencies in time series data. In this application, the LSTM is used to predict user behavior characteristics and, based on the user's behavioral data, predict the next diagnostic steps the user is likely to take.

[0126] In one embodiment, an LSTM model is trained using historical user behavior data to learn to extract patterns from past user behavior. When a user is using a diagnostic device, the system inputs the current user's operation sequence into the LSTM model to obtain a possible second initial diagnostic step.

[0127] S153, determining a third initial diagnostic step based on the correspondence between the target vehicle data, fault codes and diagnostic steps in the expert knowledge graph.

[0128] In one embodiment, the data source of the expert knowledge graph is the second vehicle data and the diagnostic steps corresponding to each vehicle data and fault information. When the diagnostic requirements are input (such as selecting the target vehicle data and fault code), the system will search for relevant nodes and relationships in the knowledge graph and output the third initial diagnostic step.

[0129] S154 , determining at least one initial diagnostic step as a recommended diagnostic step based on the vehicle model collaborative filtering, user behavior prediction, and recommendation weights of the expert model.

[0130] In this embodiment, user behavior and needs may vary depending on the scenario. For example, when a new user first uses the diagnostic device, the system needs to quickly adapt to their needs (cold start problem); during the diagnostic process, the user's behavior pattern may change, and the system needs to adjust the recommended content in real time (continuous diagnosis); for complex problems, the system needs to draw on professional knowledge for reasoning and judgment (expert mode). To address these diverse needs, this application combines three modes through a hybrid recommendation strategy, dynamically adjusting the recommendation logic based on the specific scenario.

[0131] In one embodiment, recommendations from vehicle model collaborative filtering, user behavior prediction, and expert mode can be assigned different weights. The weight indicates the importance of the recommendation, with the higher the weight, the more likely it is to be the primary recommended diagnostic step. For example, cold start-based collaborative filtering, weighted 70%; continuous diagnosis, based on LSTM behavior prediction, weighted 85%; and expert mode, weighted knowledge graph reasoning, weighted 60%. For example, recommendations from vehicle model collaborative filtering, user behavior prediction, and expert mode might prioritize diagnostic steps or solutions such as "fuel rail pressure sensor calibration."

[0132] Exemplarily, the present application predicts each diagnostic step based on the vehicle model feature vector, fault code feature vector and user behavior data of the target vehicle data through a pre-trained language model, and outputs the diagnostic steps with corresponding confidence levels and temperature maps.

[0133] In one embodiment, the pre-trained language model is a fine-tuned model based on BERT (Bidirectional Encoder Representations from Transformers), a pre-trained language model that excels at capturing semantic information in text. The BERT-based fine-tuned model in this application can combine multimodal inputs (vehicle model feature vectors, fault code feature vectors, and user behavior data) to achieve more intelligent recommendations and predictions.

[0134] For example, the input for a BERT-based fine-tuned model is a vehicle feature vector (brand / model / year code), a fault code vector (fault code + snapshot data, such as engine speed and fuel pressure), and a user action sequence (click coordinates / parameter setting values). The output layer outputs the diagnostic steps with corresponding confidence levels and a temperature map. The temperature map is a visualization tool used to directly demonstrate the reliability of the model's predictions.

[0135] The temperature map generation logic maps the confidence scores output by a fine-tuned BERT-based model into the HSV color space. The color or temperature gradient in the temperature map represents the reliability of the model's predictions. For example, high temperatures (or dark colors) indicate high reliability, while low temperatures (or light colors) indicate low reliability. This allows users to intuitively understand the reliability of the model's predictions and make decisions accordingly. The correspondence between the temperature map and prediction reliability can be defined and adjusted using a color mapping table or a temperature gradient mapping table. For example, high confidence (>0.9) is mapped to red (H=0°, S=100%, V=100%); medium confidence (0.7-0.9) to yellow (H=60°, S=80%); and low confidence (<0.7) to blue (H=240°, V=70%).

[0136] In another embodiment, the present application generates a top 3 recommended diagnostic step based on the confidence level of the diagnostic step, recommending the three most likely top 3 recommended diagnostic steps to the user. This recommendation mechanism can automatically generate a ranked list of recommendations based on user input and contextual information, as well as the model's understanding and analysis of the target vehicle data and diagnostic steps. The user can select or take further action based on the options in the recommendation list to improve diagnostic efficiency and accuracy. Top 3 recommendation generation is an important function in the adaptive navigation engine, which can provide users with personalized diagnostic advice and support.

[0137] The present application also provides a vehicle diagnostic device. Exemplarily, the vehicle diagnostic device includes a processor and a memory, wherein the memory stores a computer program, and the processor runs the computer program, thereby enabling the vehicle diagnostic device to execute the above-mentioned function navigation method.

[0138] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a central processing unit (CPU), a graphics processing unit (GPU), a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, etc., and can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0139] The memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), and electrically erasable programmable read-only memory (EEPROM). The memory is used to store computer programs, and the processor can execute the computer programs accordingly after receiving execution instructions.

[0140] This application also provides a computer-readable storage medium for storing a computer program used in the vehicle diagnostic device. For example, the computer-readable storage medium may include, but is not limited to, a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, among other media capable of storing program code.

[0141] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and structure diagrams in the accompanying drawings show the possible architectures, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in an alternative implementation, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the structure diagram and / or flowchart, and the combination of boxes in the structure diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0142] In addition, the functional modules or units in the various embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0143] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a smart phone, personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0144] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A function navigation method, characterized in that: include: Acquire first vehicle data and real-time diagnostic data stream of the target vehicle, as well as user behavior data; Acquire second vehicle data supported by the diagnostic device, and diagnostic steps associated with each of the second vehicle data and fault information; Classify and map the first vehicle data into the second vehicle data through a semantic mapping engine to determine target vehicle data of the target vehicle in the diagnostic device; generating an optimized diagnostic path for the target vehicle by a dynamic logic generator based on the target vehicle data, the real-time diagnostic data stream, and the diagnostic steps; Combine vehicle model collaborative filtering, user behavior prediction, and expert model recommendation results to generate recommended diagnostic steps; The target vehicle includes target brand data, target vehicle model code, and target model year data in the target vehicle data, and the classification and mapping of the first vehicle data to the second vehicle data by the semantic mapping engine includes: calculating a brand association between first brand data in the first vehicle data and a plurality of second brand data in the second vehicle data; Mapping the first brand data to one of the second brand data according to the brand association to obtain the target brand data; Calculating semantic similarity or visual similarity between first vehicle model data in the first vehicle data and a plurality of second vehicle model data in the second vehicle data; Mapping the first vehicle model data to one of the second vehicle model data according to the semantic similarity or visual similarity, and outputting a vehicle model code corresponding to the second vehicle model data to obtain the target vehicle model code; Calculating a matching weight between the first model year data in the first vehicle data and the second model year data in the second vehicle data according to an exponential decay function of a time decay model, wherein a decay coefficient of the exponential decay function is determined by fitting a historical model year mapping result; The first model year data is mapped to one of the second model year data according to the matching weight to obtain the target model year data.

2. The function navigation method according to claim 1, characterized in that: The obtaining of the first vehicle data and the real-time diagnostic data stream of the target vehicle, as well as the user behavior data, includes: Collecting and parsing the real-time data stream of the target vehicle through an OBD data collector to obtain the first vehicle data and the real-time diagnostic data stream, wherein the real-time diagnostic data stream includes fault codes and status information of the target component; The user behavior data of the user when using the diagnostic device is recorded through the user behavior log.

3. The function navigation method according to claim 2, characterized in that: The fault information includes a fault code and a fault phenomenon. The second vehicle data supported by the diagnostic device is obtained, and the diagnostic steps associated with each of the second vehicle data and the fault information include: Performing document recognition on the input manufacturer's technical bulletin document using a multi-document summarization model to obtain the second vehicle data and a knowledge triple of the second vehicle data, the fault information, and the diagnostic steps; Performing target detection on the maintenance manual illustration to identify and mark the target component in the maintenance manual illustration; A diagnostic reference threshold of the target component in the maintenance manual text is extracted through a regular expression template, and a knowledge triple of the second vehicle data, the fault information and the diagnostic steps is generated, wherein the fault phenomenon of the target vehicle is determined based on the status information of the target component and the diagnostic reference threshold.

4. The function navigation method according to claim 1, characterized in that: The calculating the brand association between the first brand data in the first vehicle data and the plurality of second brand data in the second vehicle data includes: Building a brand knowledge graph based on the multidimensional relationship types and brand embedding vectors of the first brand data and each of the second brand data; Based on the brand embedding vector of each brand node in the brand knowledge graph, the brand association between the first brand data and each of the second brand data is calculated through a multi-head self-attention mechanism.

5. The function navigation method according to claim 1, characterized in that: The calculating the semantic similarity or visual similarity between the first vehicle model data in the first vehicle data and the plurality of second vehicle model data in the second vehicle data includes: Converting the vehicle type text in the input first vehicle type data into a first semantic vector using a pre-trained language model; Calculating the semantic similarity based on the first semantic vector and the second semantic vector of the second vehicle type data; Multi-scale feature extraction is performed on the first vehicle model image in the first vehicle model data, so as to calculate the visual similarity based on the first vehicle model image after feature extraction and the second vehicle model image in the second vehicle model data.

6. The function navigation method according to claim 1, characterized in that: The step of generating an optimized diagnostic path for the target vehicle by a dynamic logic generator based on the target vehicle data, the real-time diagnostic data stream, and the diagnostic steps includes: Constructing a diagnostic decision tree to select the corresponding diagnostic steps according to the target vehicle data and the fault information; The diagnostic steps are path optimized by a heuristic search algorithm to obtain the optimized diagnostic path.

7. The function navigation method according to claim 1, characterized in that: The steps of generating a recommended diagnosis by combining the recommendation results of vehicle model collaborative filtering, user behavior prediction, and expert mode include: Extracting user behavior features based on the user behavior data to build a user behavior model; Recommending a first initial diagnosis step corresponding to the second vehicle data having the highest vehicle model matching degree between the target vehicle data and other second vehicle data based on the vehicle model matching degree between the target vehicle data and the second vehicle data; Analyzing the user behavior characteristics through a long short-term memory network to generate a second initial diagnosis step consistent with the user behavior data; Determining a third initial diagnostic step based on the correspondence between the target vehicle data, the fault code, and the diagnostic steps in the expert knowledge graph; At least one of the initial diagnostic steps is determined to be the recommended diagnostic step based on the vehicle model collaborative filtering, the user behavior prediction, and the recommendation weight of the expert mode.

8. The function navigation method according to claim 1, characterized in that: Also includes: Based on the vehicle model feature vector, fault code feature vector and user behavior data of the target vehicle data, each diagnostic step is predicted through a pre-trained language model, and the diagnostic step with corresponding confidence and temperature map is output.

9. A vehicle diagnostic device, characterized in that: The vehicle diagnostic device includes a processor and a memory, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement the function navigation method according to any one of claims 1 to 8.

Images