Vehicle fault determination method, electronic equipment and vehicle

Abstract

The invention relates to the technical field of vehicle safety, and provides a vehicle fault determination method, electronic equipment and a vehicle, and the method comprises the steps: determining that the vehicle has a fault, obtaining vehicle log data and fault text data, and carrying out the coding processing of the vehicle log data, and obtaining a fault signal vector; analyzing the fault text data to obtain a fault semantic vector; determining a target fusion weight, and performing fusion processing on the fault signal vector and the fault semantic vector based on the target fusion weight to obtain a multi-modal fusion vector; and inputting the multi-modal fusion vector into a pre-trained vehicle fault analysis model, processing the multi-modal fusion vector by the vehicle fault analysis model, and outputting a target fault type and a target fault processing mode. According to the method and the device, the target fault type and the target fault processing mode are determined, so that a user can clearly know the fault type of the vehicle fault, and the user can conveniently and correspondingly process the fault of the current vehicle according to the determined target fault processing mode.

Description

Technical Field

[0001] The present disclosure relates to the field of vehicle safety technology, and in particular to a vehicle fault determination method, electronic equipment, and a vehicle. Background Art

[0002] With the rapid development of vehicle technology, vehicles have become an important means of transportation in people's daily lives. During actual use, various vehicle failures may occur.

[0003] When using a vehicle, users may not have enough knowledge about various types of vehicle faults, which may lead to failures that cannot be discovered and resolved in a timely manner, thus affecting the user's normal use of the vehicle. Summary of the Invention

[0004] In view of this, the purpose of the present disclosure is to propose a vehicle fault determination method, electronic equipment and vehicle, so as to solve the current problem that due to users' insufficient understanding of various types of vehicle faults, they are unable to discover and solve vehicle faults in time, affecting users' normal use of the vehicle.

[0005] Based on the above objectives, a first aspect of the present disclosure provides a vehicle fault determination method, the method comprising:

[0006] Determining that a vehicle has a fault, obtaining vehicle log data and fault text data, encoding the vehicle log data to obtain a fault signal vector;

[0007] Parsing the fault text data to obtain a fault semantic vector;

[0008] determining a target fusion weight, and fusing the fault signal vector and the fault semantic vector based on the target fusion weight to obtain a multimodal fusion vector;

[0009] The multimodal fusion vector is input into a pre-trained vehicle fault analysis model, and is processed by the vehicle fault analysis model to output a target fault type and a target fault handling method.

[0010] Specifically, encoding the vehicle log data to obtain a fault signal vector includes:

[0011] Performing time stamp analysis on the vehicle log data to obtain a time series signal matrix;

[0012] The time series signal matrix is ​​input into a pre-trained first language model, and the first language model is used to perform feature encoding processing on the time series signal matrix to obtain a fault signal vector.

[0013] Specifically, parsing the fault text data to obtain a fault semantic vector includes:

[0014] Filtering the fault text data to obtain fault text data to be processed;

[0015] Performing format conversion on the fault text data to be processed to obtain a fault text sequence;

[0016] The fault text sequence is input into a pre-trained second language model, and semantic parsing is performed using the second language model to obtain a fault semantic vector.

[0017] Specifically, the fusing the fault signal vector and the fault semantic vector based on the target fusion weight to obtain a multimodal fusion vector includes:

[0018] Aligning the fault signal vector and the fault semantic vector to obtain a signal embedding vector and a semantic embedding vector;

[0019] The target fusion weight, the signal embedding vector, and the semantic embedding vector are input into a pre-trained vector fusion model, and processed by the vector fusion model to output a multimodal fusion vector.

[0020] Specifically, aligning the fault signal vector and the fault semantic vector to obtain a signal embedding vector and a semantic embedding vector includes:

[0021] Performing a linear transformation on the fault signal vector, mapping the fault signal vector to a preset semantic space, and obtaining a signal embedding vector;

[0022] Performing a linear transformation on the fault semantic vector, mapping the fault semantic vector to a preset semantic space, and obtaining a semantic embedding vector.

[0023] Specifically, the target fusion weight includes the fault signal channel weight corresponding to the fault signal vector and the fault semantic channel weight corresponding to the fault semantic vector;

[0024] Determining the target fusion weight includes:

[0025] Determining a first confidence level corresponding to the fault signal vector and a second confidence level corresponding to the fault semantic vector;

[0026] In response to the first confidence being greater than a first preset confidence threshold and the second confidence being less than the second preset confidence threshold, determining the fault signal channel weight to be a first weight value and determining the fault semantic channel weight to be a second weight value; or,

[0027] In response to the first confidence being less than a first preset confidence threshold and the second confidence being greater than the second preset confidence threshold, determining the fault signal channel weight to be the second weight value and determining the fault semantic channel weight to be the first weight value; or,

[0028] In response to the first confidence being greater than a first preset confidence threshold and the second confidence being greater than the second preset confidence threshold, determining that both the fault signal channel weight and the fault semantic channel weight are a third weight value;

[0029] The first weight value is greater than the third weight value, and the third weight value is greater than the second weight value.

[0030] Specifically, determining the first confidence level corresponding to the fault signal vector includes:

[0031] determining a fluctuation value of the fault signal vector within a preset time window, and determining a first initial confidence level according to the fluctuation value;

[0032] Determining the number of fault features contained in the fault signal vector per unit time, and determining a second initial confidence level according to the number of fault features;

[0033] Calculating a vector distance between the fault signal vector and a preset fault sample, and determining a third initial confidence level according to the vector distance;

[0034] The first initial confidence level, the second initial confidence level, and the third initial confidence level are weighted to obtain a first confidence level corresponding to the fault signal vector.

[0035] Specifically, determining the second confidence level corresponding to the fault semantic vector includes:

[0036] Determining whether the fault semantic vector contains preset semantic structure information, obtaining a first determination result, and determining a syntactic completeness corresponding to the first determination result;

[0037] Determining whether the fault semantic vector contains a preset keyword, obtaining a second judgment result, and determining a keyword coverage rate corresponding to the second judgment result;

[0038] The syntactic completeness and the keyword coverage are weighted to obtain the semantic vector completeness, and the confidence corresponding to the semantic vector completeness is used as the second confidence corresponding to the fault semantic vector.

[0039] Based on the same inventive concept, a second aspect of the present disclosure provides a vehicle fault determination device, comprising:

[0040] a fault signal vector determination module configured to determine whether a vehicle has a fault, obtain vehicle log data and fault text data, and encode the vehicle log data to obtain a fault signal vector;

[0041] a fault semantic vector determination module, configured to parse the fault text data to obtain a fault semantic vector;

[0042] a fusion vector determination module configured to determine a target fusion weight, and fuse the fault signal vector and the fault semantic vector based on the target fusion weight to obtain a multimodal fusion vector;

[0043] The fault analysis module is configured to input the multimodal fusion vector into a pre-trained vehicle fault analysis model, process it through the vehicle fault analysis model, and output a target fault type and a target fault handling method.

[0044] Based on the same inventive concept, the third aspect of the present disclosure proposes an electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable by the processor, wherein the processor implements the vehicle fault determination method as described above when executing the computer program.

[0045] Based on the same inventive concept, a fourth aspect of the present disclosure proposes a non-transitory computer-readable storage medium, which stores computer instructions for causing a computer to execute the vehicle fault determination method as described above.

[0046] Based on the same inventive concept, the fifth aspect of the present disclosure provides a vehicle, comprising the vehicle fault determination device described in the second aspect or the electronic device described in the third aspect or the storage medium described in the fourth aspect.

[0047] As can be seen from the above, the present disclosure proposes a vehicle fault determination method, electronic device and vehicle. When it is determined that a vehicle has a fault, vehicle log data and fault text data are obtained. The fault text data is a warranty text message sent by the user when the vehicle breaks down or voice interaction information between the user and the vehicle customer service. The vehicle log data is encoded to obtain a fault signal vector. The fault text data is parsed to obtain a fault semantic vector. The target fusion weight is determined, and the fault signal vector and the fault semantic vector are fused based on the target fusion weight to obtain a multimodal fusion vector. By fusing the vehicle log data and the fault text data, the accuracy of fault identification is improved when fault analysis is subsequently performed based on the multimodal fusion vector. The multimodal fusion vector is input into a pre-trained vehicle fault analysis model, and is processed by the vehicle fault analysis model to output the target fault type and target fault handling method. The multimodal fusion vector is analyzed and processed through the vehicle fault analysis model to obtain the target fault type and target fault handling method, so that the user can clearly know the fault type of the vehicle fault. At the same time, according to the determined target fault handling method, the user can handle the current vehicle fault accordingly, avoiding panic caused by the user's unfamiliarity with the fault problem, thereby improving driving safety and the user's car experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the present disclosure or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 This is a flow chart of a vehicle fault determination method according to an embodiment of the present disclosure;

[0050] Figure 2 A flowchart of a vehicle fault determination method according to another embodiment of the present disclosure;

[0051] Figure 3 This is a structural block diagram of a vehicle fault determination device according to an embodiment of the present disclosure;

[0052] Figure 4 Schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0053] In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0054] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present disclosure should have the usual meanings understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the embodiments of the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative position relationships. When the absolute position of the described object changes, the relative position relationship may also change accordingly.

[0055] The definitions of the terms and are as follows:

[0056] ECU: An ECU (Electronic Control Unit), also known as a car's "on-board computer," controls the vehicle's driving state and various functions. It primarily uses data collection and exchange between various sensors and buses to determine vehicle status and the driver's intentions, and controls the vehicle through actuators.

[0057] CAN bus: Controller Area Network (CAN) is a serial communication protocol bus used for real-time applications. It can use twisted pair cables to transmit signals and is one of the most widely used field buses in the world.

[0058] DeepSeek-R1: DeepSeek-R1 is an inference model developed by DeepSeek, the AI ​​company under Magic Square Quantitative. DeepSeek-R1 uses reinforcement learning for post-training to improve its reasoning capabilities, particularly for complex tasks such as mathematics, code, and natural language reasoning.

[0059] Qwen3: Qwen3 is the latest member of the Qwen series of language models. It supports multiple thinking modes, multiple languages, multiple modalities, and multiple capabilities, and performs competitively in benchmark tests such as code, mathematics, and general abilities.

[0060] OTA (Over-the-Air Technology, OTA[1]) is a technology that remotely manages mobile terminal devices and SIM card data through the air interface of mobile communications. OTA (Over-the-Air) refers to the technology that remotely manages the firmware, data, and applications on the terminal components of a vehicle through a mobile communication network (2G / 3G / 4G or Wi-Fi).

[0061] TSP: TSP Internet of Vehicles, also known as TSP vehicle information service system, is a system based on the global positioning system and mobile Internet, combining multiple communication technologies to provide real-time information and services to vehicles.

[0062] LoRA: LoRA is a low-resource method for fine-tuning large models that uses low-rank matrices to encode parameter updates, thereby reducing graphics memory and inference time.

[0063] DMS: The DMS system (Driver Monitor System) monitors the driver's status while driving. It includes features such as FaceID, fatigue detection, distraction detection, expression recognition, gesture recognition, dangerous action recognition, and gaze tracking.

[0064] DeepSeek-LLM: A general-purpose large language model with a similar architecture to Llama and multiple optimizations.

[0065] DeepSeek-MoE: China's first large-scale MoE model with excellent generalization effect.

[0066] Gemini model: Gemini is an artificial intelligence model released by Google DeepMind (an artificial intelligence laboratory established under Google's parent company Alphabet) on December 6, 2023. It can simultaneously recognize five types of information: text, images, audio, video, and code. It can also understand and generate high-quality code in mainstream programming languages ​​​​(such as Python, Java, C++) and has comprehensive security assessments.

[0067] DeepSeek model: DeepSeek is a large language model based on the Transformer architecture.

[0068] With the rapid development of vehicle technology, vehicles have become an important means of transportation in people's daily lives. During actual use, various vehicle failures may occur.

[0069] Existing vehicle fault detection systems have problems such as multimodal data fragmentation, weak language comprehension capabilities, delayed fault trend identification, and unexplainable reasoning, making it difficult to meet the needs of intelligent diagnosis.

[0070] That is to say, when using a vehicle, due to the user's lack of knowledge of various types of vehicle faults, it is easy to fail to discover and solve the problem in time when the vehicle breaks down, affecting the user's normal use of the vehicle. Therefore, this embodiment proposes a vehicle fault determination method, such as Figure 1 As shown, the method includes:

[0071] Step 101 : Determine whether a vehicle has a fault, obtain vehicle log data and fault text data, encode the vehicle log data, and obtain a fault signal vector.

[0072] In specific implementation, it is determined whether the vehicle has a fault. When it is determined that the vehicle has a fault, vehicle log data and fault text data are obtained, wherein the vehicle log data is structured log data obtained from the vehicle ECU and CAN bus.

[0073] In this embodiment, the determination of whether the vehicle has a fault may be made by detecting signals inside the vehicle, or by using information output by a user inside the vehicle.

[0074] Specifically, the vehicle is judged whether there is a fault by detecting the signal inside the vehicle, which means that the signal inside the vehicle is compared with the signal when there is no fault in real time. In response to the difference between the signal inside the vehicle and the signal when there is no fault, it indicates that there is currently a fault in the vehicle.

[0075] Specifically, determining whether a vehicle fault exists is based on information provided by the user inside the vehicle. When a user discovers a fault in the vehicle, they often seek help because they don't fully understand the cause of the fault. For example, they may enter a description of the current vehicle fault into the vehicle computer and search for relevant solutions. Alternatively, they may contact customer service to describe the problem and seek assistance, such as through warranty requests. Therefore, when text or voice information related to a fault description is detected, it indicates that the vehicle is currently faulty.

[0076] In this embodiment, the fault text data represents a descriptive text related to the vehicle fault. For example, the fault text data includes the user's warranty text or voice text data when the user communicates with customer service when the vehicle breaks down.

[0077] In this embodiment, when the fault text data is voice text data of a user communicating with customer service when a vehicle malfunctions, the voice text data can be converted into plain text using a speech recognition model. Furthermore, to ensure semantic accuracy, intonation and abnormal tones can be retained as additional semantic references.

[0078] For example, the speech recognition model is the Whisper speech recognition model. The Whisper speech recognition model is a multilingual speech recognition model developed and open-sourced by OpenAI. It boasts high accuracy and robustness, supporting tasks such as speech-to-text, language recognition, and speech translation.

[0079] The acquired vehicle log data is encoded to obtain a fault signal vector. It is understandable that, since the vehicle log data is structured log data, the fault signal vector obtained by encoding the vehicle log data is also a structured signal vector.

[0080] Step 102: parse the fault text data to obtain a fault semantic vector.

[0081] During specific implementation, the acquired fault text data is parsed and processed to obtain a fault semantic vector corresponding to the fault text data.

[0082] In this embodiment, since the fault text data includes text description information of the vehicle fault, when parsing the fault text data, a fault causal structure related to the vehicle fault can be extracted and analyzed.

[0083] It is understandable that, since the fault text data is unstructured text data, the fault semantic vector obtained after parsing the fault text data is also an unstructured semantic vector.

[0084] Step 103 : determining a target fusion weight, and fusing the fault signal vector and the fault semantic vector based on the target fusion weight to obtain a multimodal fusion vector.

[0085] During specific implementation, a target fusion weight is determined, wherein the target fusion weight is the weight value corresponding to the fault signal vector and the fault semantic vector respectively when the fault signal vector and the fault semantic vector are fused.

[0086] Based on the determined target fusion weight, the fault signal vector and the fault semantic vector are fused, that is, the fault signal vector and the fault semantic vector are fused according to their respective corresponding weights to obtain a multimodal fusion vector.

[0087] Step 104 : Input the multimodal fusion vector into a pre-trained vehicle fault analysis model, process the vehicle fault analysis model, and output a target fault type and a target fault handling method.

[0088] In specific implementations, the fused multimodal fusion vector is input into a pre-trained vehicle fault analysis model. The model then processes the vector and outputs a target fault type and a target fault handling method. The target fault type represents the type of vehicle fault, and the target fault handling method represents the possible solutions to the fault.

[0089] In this embodiment, the multimodal fusion vector is integrated with the fault semantic vector, which is derived from parsing the fault text data. Furthermore, the parsing of the fault text data can extract and analyze the fault causal structure associated with the vehicle fault. Therefore, when the multimodal fusion vector is processed using the vehicle fault analysis model, a target fault cause can also be output, where the target fault cause represents the cause of the vehicle fault.

[0090] In this embodiment, the training process of the vehicle fault analysis model specifically includes:

[0091] Step a: obtaining a first training data set and an initial vehicle fault analysis model, wherein the first training data set includes historical multimodal fusion vectors and historical fault types, historical fault handling methods, and historical fault causes.

[0092] Step b: inputting the training data in the first training data set into the initial vehicle fault analysis model for training, determining that a first preset training end condition is met, and obtaining a vehicle fault analysis model.

[0093] In a specific implementation, a first training dataset and an initial vehicle fault analysis model are obtained, wherein the first training dataset includes historical multimodal fusion vectors and historical fault types, historical fault handling methods, and historical fault causes. Training data in the first training dataset is input into the initial vehicle fault analysis model for training, and a first preset training termination condition is determined to be satisfied, thereby obtaining a vehicle fault analysis model.

[0094] The first preset training end condition includes at least one of the following: determining that all data in the first training data set are input into the initial vehicle fault analysis model for training, determining that the loss function of the initial vehicle fault analysis model converges to a first convergence threshold, or determining that the initial vehicle fault analysis model is iteratively trained to a first preset number of iterations.

[0095] Exemplarily, the first preset training end condition is to determine that all data in the first training data set are input into the initial vehicle fault analysis model for training:

[0096] The first training dataset comprises fifty sets of data, each set of which includes a historical multimodal fusion vector and a historical fault type, a historical fault handling method, and a historical fault cause. The first preset training termination condition is to ensure that all data in the first training dataset has been input into the initial vehicle fault analysis model for training. Specifically, when all fifty sets of data have been input into the initial vehicle fault analysis model and no training data in the first training dataset has not yet been input into the initial vehicle fault analysis model, the initial vehicle fault analysis model training is determined to have terminated, thereby obtaining a vehicle fault analysis model.

[0097] In another example, the first preset training end condition is to determine that the loss function of the initial vehicle fault analysis model converges to a first convergence threshold:

[0098] The training data in the first training data set is input into the initial vehicle fault analysis model for training, and a training result is output. A loss function is determined based on the training result and the historical fault type, historical fault handling method, and historical fault cause. The loss function may include at least one of the following: a mean square error loss function, a cross entropy loss function, a logarithmic loss function, an exponential loss function, a square loss function, or an absolute value loss function. When the loss function converges to a first convergence threshold, it is determined that a first preset training termination condition has been met, and a vehicle fault analysis model is obtained.

[0099] In another example, the first preset training end condition is to determine that the initial vehicle fault analysis model is iteratively trained to a first preset number of iterations.

[0100] The training data in the first training data set is input into the initial vehicle fault analysis model for iterative training, the number of iterations is recorded, and when it is determined that the number of iterations is equal to the first preset number of iterations, the first preset training end condition is met, and the vehicle fault analysis model is obtained.

[0101] Through the above scheme, the training data in the first training data set includes historical multimodal fusion vectors and historical fault types, historical fault handling methods and historical fault causes. The training data in the first training data set is used to train the initial vehicle fault analysis model until the first preset training end condition is met, and a vehicle fault analysis model is obtained. The vehicle fault analysis model can be used to subsequently output the target fault type, target fault handling method and target fault cause corresponding to the current multimodal fusion vector, and the determined target fault type, target fault handling method and target fault cause are more accurate.

[0102] Through the above scheme, when it is determined that there is a fault in the vehicle, the vehicle log data and fault text data are obtained. The fault text data is the warranty text information sent by the user when the vehicle breaks down or the voice interaction information between the user and the vehicle customer service. The vehicle log data is encoded and processed to obtain a fault signal vector. The fault text data is parsed and processed to obtain a fault semantic vector. The target fusion weight is determined, and the fault signal vector and the fault semantic vector are fused based on the target fusion weight to obtain a multimodal fusion vector. By fusing the vehicle log data and the fault text data, the accuracy of fault identification is improved when the fault analysis is subsequently performed according to the multimodal fusion vector. The multimodal fusion vector is input into a pre-trained vehicle fault analysis model, and is processed by the vehicle fault analysis model to output the target fault type and target fault handling method. The multimodal fusion vector is analyzed and processed through the vehicle fault analysis model to obtain the target fault type and target fault handling method, so that the user can clearly know the fault type of the vehicle fault. At the same time, according to the determined target fault handling method, the user can handle the current vehicle fault accordingly, avoiding panic caused by the user's unfamiliarity with the fault problem, thereby improving driving safety and the user's car experience.

[0103] In some embodiments, before inputting the multimodal fusion vector into the vehicle fault analysis model to determine the target fault type and target fault handling method, a vehicle dashboard image may be obtained and analyzed to obtain image label data. The image label data and the multimodal fusion vector are input into the vehicle fault analysis model together, specifically including:

[0104] Acquire a vehicle dashboard image and use an image detection model to identify and process it, identifying fault areas and extracting the category information of the fault light. This category information, along with the multimodal fusion vector, is used as image label data and fed into the vehicle fault analysis model to assist the multimodal fusion vector in further determining the vehicle fault type and fault handling method.

[0105] Through the above scheme, since the fault light on the vehicle dashboard can accurately distinguish different vehicle faults, before using the vehicle fault analysis model to determine the vehicle fault type and fault handling method, the fault light type on the vehicle dashboard is identified and then input into the vehicle fault analysis model together with the multimodal fusion vector. The determined fault type and fault solution are more accurate.

[0106] In some embodiments, the vehicle log data may be encoded using a pre-trained language model. That is, the vehicle log data may be encoded in step 101 to obtain a fault signal vector, specifically including:

[0107] Step 1011, performing time stamp analysis on the vehicle log data to obtain a time series signal matrix;

[0108] Step 1012: Input the time series signal matrix into a pre-trained first language model, and use the first language model to perform feature encoding processing on the time series signal matrix to obtain a fault signal vector.

[0109] In a specific implementation, after acquiring vehicle log data, the vehicle log data is timestamp-parsed and organized by channel, outputting a time series signal matrix. The time series signal matrix is ​​a multi-channel time series. Exemplarily, the window length corresponding to the time series signal matrix is ​​128 to 256 frames.

[0110] The time series signal matrix is ​​input into a pre-trained first language model, and the first language model is used to perform feature encoding processing on the time series signal matrix to obtain a fault signal vector.

[0111] In this embodiment, the first language model may be a DeepSeek-R1 model, or a language model of the same type, such as DeepSeek-LLM, DeepSeek-MoE, etc. Taking the DeepSeek-R1 model as an example, the process of encoding and modeling the time series signal matrix using the DeepSeek-R1 model is as follows:

[0112] DeepSeek-R1 uses the autoencoder and Transformer submodules to extract fault semantic features, outputting a high-dimensional semantic embedding vector for each signal window with a dimension of 768. The SimCSE architecture is used for self-supervised training, optimizing the embedding space through positive-negative contrast learning. This further improves the sentence vector representation through supervised data. Positive samples are historical fault samples of the same type, while negative samples are non-fault or abnormal signal samples of different types.

[0113] Through the above scheme, the structured time series signal matrix of any time period can be semantically embedded through the trained first language model as the basis for fault discrimination. The obtained fault signal vector is more accurate, and the fault semantics corresponding to the multimodal fusion vector obtained by subsequent fusion using the fault signal vector are more accurate. The fault type and fault handling method output by the vehicle fault analysis model are more urgent and accurate.

[0114] In some embodiments, the fault text data may be parsed using a pre-trained language model, i.e., the fault text data in step 102 may be parsed to obtain a fault semantic vector, specifically including:

[0115] Step 1021: Filter the fault text data to obtain the fault text data to be processed;

[0116] Step 1022: convert the format of the fault text data to be processed to obtain a fault text sequence;

[0117] Step 1023: Input the fault text sequence into a pre-trained second language model, perform semantic parsing using the second language model, and obtain a fault semantic vector.

[0118] In a specific implementation, the fault text data is filtered and cleaned to obtain unprocessed fault text data. The filtering process includes at least one of the following: sentence segmentation, word segmentation, and semantic disambiguation. The obtained unprocessed fault text data is formatted and converted into a standardized token sequence, thereby obtaining a fault text sequence for use as input to the second language model.

[0119] The fault text sequence is input into a pre-trained second language model, and semantic parsing is performed using the second language model to obtain a fault semantic vector.

[0120] In this embodiment, the second language model may be a Qwen3 model, or a language model of the same type, such as a Gemini model, a DeepSeek model, etc. Taking the Qwen3 model as an example, the process of semantically parsing the fault text sequence using the Qwen3 model is as follows:

[0121] The Qwen3 model fine-tunes the fault text sequence to extract structured semantic causal relationships. The Qwen3 model returns a fault semantic vector, which is a multi-tuple semantic relationship. For example, the fault semantic vectors are [accelerator → muffled] and [engine light → ignition system abnormality].

[0122] In this embodiment, all fault semantic vectors obtained from semantic parsing can be aggregated to form a language semantic tree. Furthermore, the Qwen3 model can be further fine-tuned using a corpus of historical maintenance records from car companies to make its language understanding capabilities more tailored to car company scenarios.

[0123] Through the above solution, the fault text data is filtered to reduce the interference of irrelevant data, so that when the second language model is subsequently processed, the processing efficiency is improved and the obtained fault semantic vector is more accurate.

[0124] In some embodiments, in step 103, the fault signal vector and the fault semantic vector are fused based on the target fusion weight to obtain a multimodal fusion vector, which specifically includes:

[0125] Step 1031: align the fault signal vector and the fault semantic vector to obtain a signal embedding vector and a semantic embedding vector;

[0126] Step 1032: Input the target fusion weight, the signal embedding vector, and the semantic embedding vector into a pre-trained vector fusion model, process them through the vector fusion model, and output a multimodal fusion vector.

[0127] In a specific implementation, the fault signal vector and the fault semantic vector are first projected into the same semantic space for alignment processing to obtain a signal embedding vector and a semantic embedding vector.

[0128] Specifically, a linear transformation is performed on the fault signal vector, and the fault signal vector is mapped to a preset semantic space to obtain a signal embedding vector. A linear transformation is performed on the fault semantic vector, and the fault semantic vector is mapped to a preset semantic space to obtain a semantic embedding vector.

[0129] The aligned fault signal vector and fault language vector are further fused through a multi-head attention mechanism, enabling the model to identify deep relationships and causal associations between them. Specifically, the target fusion weights, signal embedding vector, and semantic embedding vector are input into a pre-trained vector fusion model, which processes them and outputs a multimodal fusion vector.

[0130] In this embodiment, the vector fusion model uses a cross-modal contrastive loss function and is trained on historical signal-language paired data to bring related samples closer together in semantic space. Using historically matched signal-language pairs as positive samples and mismatched pairs as negative samples, a contrastive learning model, or vector fusion model, is constructed to determine the correlation between the two modalities.

[0131] Through the above scheme, the fault signal vector and the fault semantic vector are fused to obtain a multimodal fusion vector, so that the subsequent vehicle fault analysis model can automatically identify the correlation between language and signal when analyzing and processing the multimodal fusion vector, thereby improving the accuracy of fault identification.

[0132] In some embodiments, the target fusion weight includes the fault signal channel weight corresponding to the fault signal vector and the fault semantic channel weight corresponding to the fault semantic vector. Determining the target fusion weight in step 103 specifically includes:

[0133] Step 1031, determining a first confidence level corresponding to the fault signal vector and a second confidence level corresponding to the fault semantic vector;

[0134] Step 1032: In response to the first confidence being greater than a first preset confidence threshold and the second confidence being less than the second preset confidence threshold, determining the fault signal channel weight to be a first weight value and the fault semantic channel weight to be a second weight value; or,

[0135] Step 1033: In response to the first confidence being less than a first preset confidence threshold and the second confidence being greater than the second preset confidence threshold, determining the fault signal channel weight to be a second weight value and determining the fault semantic channel weight to be a first weight value; or,

[0136] Step 1034: In response to the first confidence being greater than a first preset confidence threshold and the second confidence being greater than a second preset confidence threshold, determining that both the fault signal channel weight and the fault semantic channel weight are a third weight value;

[0137] The first weight value is greater than the third weight value, and the third weight value is greater than the second weight value.

[0138] In specific implementation, the target fusion weight includes the fault signal channel weight corresponding to the fault signal vector and the fault semantic channel weight corresponding to the fault semantic vector. That is, when determining the target fusion weight, the fault signal channel weight corresponding to the fault signal vector and the fault semantic channel weight corresponding to the fault semantic vector are determined.

[0139] A first confidence level corresponding to the fault signal vector and a second confidence level corresponding to the fault semantic vector are determined respectively. The first confidence level represents the consistency between the fault signal vector and the original vehicle log data, and the second confidence level represents the consistency between the fault semantic vector and the original fault text data.

[0140] The first confidence level is compared with a first preset confidence level threshold, and the second confidence level is compared with a second preset confidence level threshold. The comparison results can be specifically divided into:

[0141] If the first confidence is greater than a first preset confidence threshold and the second confidence is less than the second preset confidence threshold, the fault signal channel weight is determined to be a first weight value, and the fault semantic channel weight is determined to be a second weight value.

[0142] For example, the first preset confidence threshold is 0.5, and the second preset confidence threshold is 0.5. The first weight value is 0.7, and the second weight value is 0.3. The first confidence corresponding to the fault signal vector is determined to be 0.6, and the second confidence corresponding to the fault semantic vector is 0.3. At this time, the first confidence is greater than the first preset confidence threshold, and the second confidence is less than the second preset confidence threshold. This means that the abnormal signal characteristics are obvious at this time, but the language description is vague, and the fault signal channel weight should be increased. That is, the fault signal channel weight is determined to be the first weight value 0.7, and the fault semantic channel weight is determined to be the second weight value 0.3.

[0143] If the first confidence is less than a first preset confidence threshold and the second confidence is greater than the second preset confidence threshold, the fault signal channel weight is determined to be a second weight value, and the fault semantic channel weight is determined to be a first weight value.

[0144] For example, the first preset confidence threshold is 0.5, and the second preset confidence threshold is 0.5. The first weight value is 0.7, and the second weight value is 0.3. The first confidence corresponding to the fault signal vector is determined to be 0.4, and the second confidence corresponding to the fault semantic vector is 0.8. At this time, the first confidence is less than the first preset confidence threshold, and the second confidence is greater than the second preset confidence threshold. This means that the language description is clear and unambiguous at this time, but the signal change is not obvious, and the fault semantic channel weight should be increased. That is, the fault signal channel weight is determined to be the second weight value 0.3, and the fault semantic channel weight is determined to be the first weight value 0.7.

[0145] If the first confidence is greater than a first preset confidence threshold, and the second confidence is greater than the second preset confidence threshold, it is determined that the fault signal channel weight and the fault semantic channel weight are both third weight values.

[0146] For example, the first preset confidence threshold is 0.5, the second preset confidence threshold is 0.5, and the third weight value is 0.5. The first confidence corresponding to the fault signal vector is determined to be 0.7, and the second confidence corresponding to the fault semantic vector is determined to be 0.8. This satisfies the conditions that the first confidence is greater than the first preset confidence threshold, and the second confidence is greater than the second preset confidence threshold. At this point, the consistency between the fault signal vector and the original vehicle log data, as well as the consistency between the fault semantic vector and the original fault text data, are both high. Therefore, the weight balance can be maintained, that is, the fault signal channel weight and the fault semantic channel weight are both 0.5.

[0147] In this embodiment, if the first confidence level is less than the first preset confidence threshold and the second confidence level is less than the second preset confidence threshold, it indicates that the consistency between the fault signal vector and the original vehicle log data, as well as the consistency between the fault semantic vector and the original fault text data are both low. The default minimum fusion confidence mechanism is adopted, conservative output is given, and manual review is prompted.

[0148] In this embodiment, the lowest fusion confidence mechanism indicates selecting the higher corresponding weight value between the first confidence level and the second confidence level as the fourth weight value and the lower corresponding weight value as the fifth weight value.

[0149] For example, the first preset confidence threshold is 0.5, and the second preset confidence threshold is 0.5. The third weight value is 0.55, and the fourth weight value is 0.45. It is determined that the first confidence corresponding to the fault signal vector is 0.3, and the second confidence corresponding to the fault semantic vector is 0.4. At this time, the first confidence is less than the first preset confidence threshold, and the second confidence is less than the second preset confidence threshold. That is, at this time, the consistency between the fault signal vector and the original vehicle log data, as well as the consistency between the fault semantic vector and the original fault text data are both low, and the second confidence is greater than the first confidence. At this time, it can be determined that the fault signal channel weight is the fourth weight value 0.45, and the fault semantic channel weight is the third weight value 0.55.

[0150] In some embodiments, determining the first confidence level corresponding to the fault signal vector in step 1031 specifically includes:

[0151] Step A, determining a fluctuation value of the fault signal vector within a preset time window, and determining a first initial confidence level according to the fluctuation value;

[0152] Step B: determining the number of fault features contained in the fault signal vector per unit time, and determining a second initial confidence level according to the number of fault features;

[0153] Step C, calculating a vector distance between the fault signal vector and a preset fault sample, and determining a third initial confidence level according to the vector distance;

[0154] Step D: performing weighted processing on the first initial confidence level, the second initial confidence level, and the third initial confidence level to obtain a first confidence level corresponding to the fault signal vector.

[0155] In a specific implementation, a fluctuation value of the fault signal vector within a preset time window is determined, where the fluctuation value represents the degree of fluctuation of the CAN / ECU signal corresponding to the fault signal vector within the time window. A database is searched based on the fluctuation value to determine a first initial confidence level corresponding to the fluctuation value. The more severe the fluctuation, the higher the fluctuation value, the more obvious the abnormality, and the higher the first initial confidence level.

[0156] Determine the number of fault features contained in the fault signal vector per unit time, search a database based on the number of fault features, and determine a second initial confidence level corresponding to the number of fault features. A greater number of fault features indicates a more pronounced fault feature, and a higher corresponding second initial confidence level.

[0157] The vector distance between the fault signal vector and a preset fault sample is calculated, and a database is searched based on the vector distance to determine a third initial confidence level corresponding to the vector distance. The vector distance represents the similarity between the vehicle fault and historical typical faults; a closer vector distance indicates a higher third initial confidence level.

[0158] The first initial confidence level, the second initial confidence level, and the third initial confidence level are weighted to obtain a first confidence level corresponding to the fault signal vector.

[0159] Through the above solution, the first confidence level corresponding to the fault signal vector is determined by combining signal variance volatility, abnormal frequency statistics, and similarity with historical typical faults, and the determination of the first confidence level is more accurate.

[0160] In some embodiments, determining the second confidence level corresponding to the fault semantic vector in step 1031 specifically includes:

[0161] Step a: determining whether the fault semantic vector contains preset semantic structure information, obtaining a first determination result, and determining the syntactic completeness corresponding to the first determination result;

[0162] Step b: determining whether the fault semantic vector contains a preset keyword, obtaining a second judgment result, and determining a keyword coverage rate corresponding to the second judgment result;

[0163] Step c: weighting the syntactic completeness and the keyword coverage to obtain semantic vector completeness, and using the confidence corresponding to the semantic vector completeness as the second confidence corresponding to the fault semantic vector.

[0164] During specific implementation, it is determined whether the fault semantic vector contains preset semantic structure information to obtain a first determination result, wherein the preset semantic structure information includes a subject-predicate structure, a temporal causal relationship, and the like.

[0165] The first judgment result includes including all preset semantic structure information, including partial semantic structure information, and not including preset semantic structure information. The corresponding syntactic completeness is determined based on the first judgment result. It is understandable that because the first judgment result includes including partial semantic structure information, the database can be searched based on the amount of semantic structure information included to further determine the syntactic completeness corresponding to the amount of semantic structure information included.

[0166] It is determined whether the fault semantic vector contains preset keywords to obtain a second determination result, wherein the preset keywords are pre-set keywords or terms related to known faults.

[0167] The second judgment result includes whether the preset keyword is included or not included. The database is searched according to the second judgment result to determine the corresponding keyword coverage.

[0168] The syntactic completeness and the keyword coverage are weighted to obtain a semantic vector completeness. A database is searched based on the semantic vector completeness to determine a confidence level corresponding to the semantic vector completeness, and the determined confidence level is used as a second confidence level corresponding to the fault semantic vector.

[0169] Through the above solution, when determining the first confidence level corresponding to the fault signal vector, the syntactic completeness and keyword coverage are comprehensively considered, thereby making the determination of the first confidence level more accurate.

[0170] Based on the same inventive concept, another embodiment of the present disclosure provides a vehicle fault determination method, such as Figure 2 As shown, the method includes:

[0171] Step 201: multi-source fault data collection and formatting.

[0172] Structured log data obtained from the vehicle's ECU and CAN bus is parsed for timestamps, organized by channel, and output as a time-series signal matrix file. Customer repair report text is preprocessed, including sentence segmentation, word segmentation, and semantic disambiguation, and converted into standardized token sequences for language model input.

[0173] Customer service voice data is converted to plain text using the Whisper speech recognition model, retaining intonation and unusual tones as additional semantic references. Dashboard fault light images are analyzed using the YOLOv8 object detection model to identify the fault light area and extract the fault light category information as auxiliary diagnostic labels.

[0174] Step 202: DeepSeek-R1 model fault signal vector modeling.

[0175] The first language model, such as the DeepSeek-R1 model, is used to encode and model the structured vehicle signal to obtain a fault signal vector.

[0176] The CAN signals / ECU logs collected from each vehicle are processed into multi-channel time series inputs with a window length of 128 to 256 frames. The autoencoder and Transformer submodules in DeepSeek-R1 are used to extract fault semantic features, outputting a high-dimensional semantic embedding vector of 768 dimensions for each signal window.

[0177] The SimCSE structure is used for self-supervised training, and the embedding space is optimized through positive and negative contrast learning. Positive samples are historical fault samples of the same type, and negative samples are non-fault or abnormal signal samples of different types.

[0178] Step 203: Extract Qwen3 model fault language knowledge.

[0179] Use a second language model, such as the Qwen3 large language model, to perform semantic analysis on unstructured language input (text, speech), extract the causal structure related to the fault, and obtain the fault semantic vector.

[0180] Input text includes user repair reports, customer service ticket descriptions, and speech recognition transcriptions. Qwen3 performs supervised fine-tuning on commands, enabling it to extract structured semantic causal relationships. Qwen3 returns results as multi-tuple semantic relationships, such as: "[Step on the gas → Stuffy]" and "[Engine light → Ignition system abnormality]." All extracted triples are constructed into a language semantic tree for subsequent fusion with the structured signal output.

[0181] In step 204, the instrument panel fault light image is used to identify the fault light area through the YOLOv8 target detection model, and the category information of the fault light is extracted as an auxiliary diagnosis label, that is, image label data.

[0182] Step 205: multi-modal vector fusion.

[0183] Using a deep shared attention mechanism, the fault semantic vectors output by Qwen3 and the fault signal vectors output by DeepSeek-R1 are projected into the same semantic space. A cross-modal contrastive loss function is constructed and trained using historical signal-language paired data to bring related samples closer together in the embedding space. The signal embedding vector (e.g., 768-dimensional) and the language embedding vector (e.g., 512-dimensional) are unified into a fusion space through linear projection and then fused using Multi-Head Attention to generate a multimodal fusion vector.

[0184] In step 206 , the multimodal fusion vector and the image label data are input into the vehicle fault analysis model, and the target fault type, target fault cause, and target fault handling method are output.

[0185] The fault semantic vector output by Qwen3 serves as the knowledge graph input, while the fault signal vector output by DeepSeek-R1 serves as the state representation input. The embedding vectors of the two channels are fused through attention, and reasoning is performed in a high-dimensional space. The model's final output includes multi-classification fault type results, possible fault cause chains (structural diagram form), and target fault handling methods.

[0186] In this embodiment, the signal modality confidence index corresponding to DeepSeek-R1 is determined, specifically:

[0187] A fluctuation value of the fault signal vector within a preset time window is determined, wherein the fluctuation value represents the degree of fluctuation of the CAN / ECU signal corresponding to the fault signal vector within the time window. A database is searched based on the fluctuation value to determine a first initial confidence level corresponding to the fluctuation value. Greater fluctuations are associated with higher fluctuation values, indicating a more pronounced anomaly and a higher first initial confidence level.

[0188] Determine the number of fault features contained in the fault signal vector per unit time, search a database based on the number of fault features, and determine a second initial confidence level corresponding to the number of fault features. A greater number of fault features indicates a more pronounced fault feature, and a higher corresponding second initial confidence level.

[0189] The vector distance between the fault signal vector and a preset fault sample is calculated, and a database is searched based on the vector distance to determine a third initial confidence level corresponding to the vector distance. The vector distance represents the similarity between the vehicle fault and historical typical faults; a closer vector distance indicates a higher third initial confidence level.

[0190] The first initial confidence level, the second initial confidence level, and the third initial confidence level are weighted to obtain a first confidence level corresponding to the fault signal vector.

[0191] In this embodiment, the language modality expression integrity index corresponding to Qwen3 is determined, specifically:

[0192] It is determined whether the fault semantic vector contains preset semantic structure information to obtain a first determination result, wherein the preset semantic structure information includes a subject-predicate structure, a temporal causal relationship, and the like.

[0193] The first judgment result includes including all preset semantic structure information, including partial semantic structure information, and not including preset semantic structure information. The corresponding syntactic completeness is determined based on the first judgment result. It is understandable that because the first judgment result includes including partial semantic structure information, the database can be searched based on the amount of semantic structure information included to further determine the syntactic completeness corresponding to the amount of semantic structure information included.

[0194] It is determined whether the fault semantic vector contains preset keywords to obtain a second determination result, wherein the preset keywords are pre-set keywords or terms related to known faults.

[0195] The second judgment result includes whether the preset keyword is included or not included. The database is searched according to the second judgment result to determine the corresponding keyword coverage.

[0196] The syntactic completeness and the keyword coverage are weighted to obtain a semantic vector completeness. A database is searched based on the semantic vector completeness to determine a confidence level corresponding to the semantic vector completeness, and the determined confidence level is used as a second confidence level corresponding to the fault semantic vector.

[0197] The system sets the initial fusion weights of the two modalities and dynamically adjusts the weights of the two modalities when the multimodal fusion vector is determined based on the confidence index calculation results. Specifically:

[0198] If the signal has obvious abnormal characteristics but the language description is vague, increase the signal channel weight, for example, adjusting it to 0.7:0.3. If the language description is clear and unambiguous but the signal change is not obvious, increase the language channel weight, for example, adjusting it to 0.3:0.7. If both parties have high confidence, maintain the weight balance or fine-tune it to 0.6:0.4. If both parties have low confidence, use the default minimum fusion confidence mechanism, output conservatively, and prompt manual review.

[0199] This embodiment supports multi-platform deployment, including embedded and cloud platforms. The interface encapsulation specifically includes a fault identification interface, a fault causal linkage interface, and a maintenance suggestion interface. The fault identification interface inputs vehicle status data and a language description, and outputs the fault type and confidence level. The fault causal linkage interface outputs a structured semantic path tree for engineers to interpret. The maintenance suggestion interface outputs recommended maintenance strategies and spare parts replacement suggestions. This can be integrated into the vehicle manufacturer's TSP platform, IVI pre-installed system, and after-sales DMS system to achieve real-time diagnosis and auxiliary analysis.

[0200] In some embodiments, to ensure the long-term accuracy and adaptability of the model in the automotive industry environment, this step designs the following automotive industry-specific data pipeline. Specifically, it includes:

[0201] The work order system extracts repair types, parts replacement records, and repair conclusions as labeled data for model supervision training. The after-sales voice system converts user repair calls into text, supplemented by manual review to produce high-quality corpus. OTA remote diagnostic data is used to regularly upload ECU logs and combine them with fault labels to build a structured training set. Video maintenance data is used to extract and convert audio tracks and subtitles into text, enhancing Qwen3's semantic understanding capabilities.

[0202] In this embodiment, the above data is updated regularly every month, and the training process adopts incremental learning. Qwen3 updates use LoRA fine-tuning, and DeepSeek-R1 uses sliding window cumulative training to avoid catastrophic forgetting.

[0203] It should be noted that the method of the embodiments of the present disclosure can be performed by a single device, such as a computer or server. The method of the embodiments of the present disclosure can also be applied in a distributed scenario, where multiple devices cooperate to perform the method. In such a distributed scenario, one of the multiple devices may only perform one or more steps of the method of the embodiments of the present disclosure, and the multiple devices will interact with each other to complete the method.

[0204] It should be noted that the above description is limited to some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in an order different from that described in the above embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0205] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present disclosure also provides a vehicle fault determination device.

[0206] refer to Figure 3 , Figure 3 The vehicle fault determination device of the embodiment includes:

[0207] The fault signal vector determination module 301 is configured to determine whether a vehicle has a fault, obtain vehicle log data and fault text data, encode the vehicle log data, and obtain a fault signal vector;

[0208] The fault semantic vector determination module 302 is configured to parse the fault text data to obtain a fault semantic vector;

[0209] A fusion vector determination module 303 is configured to determine a target fusion weight, and fuse the fault signal vector and the fault semantic vector based on the target fusion weight to obtain a multimodal fusion vector;

[0210] The fault analysis module 304 is configured to input the multimodal fusion vector into a pre-trained vehicle fault analysis model, process the vehicle fault analysis model, and output a target fault type and a target fault handling method.

[0211] In some embodiments, the fault signal vector determination module 301 is specifically configured to:

[0212] Performing time stamp analysis on the vehicle log data to obtain a time series signal matrix;

[0213] The time series signal matrix is ​​input into a pre-trained first language model, and the first language model is used to perform feature encoding processing on the time series signal matrix to obtain a fault signal vector.

[0214] In some embodiments, the fault semantic vector determination module 302 is specifically configured to:

[0215] Filtering the fault text data to obtain fault text data to be processed;

[0216] Performing format conversion on the fault text data to be processed to obtain a fault text sequence;

[0217] The fault text sequence is input into a pre-trained second language model, and semantic parsing is performed using the second language model to obtain a fault semantic vector.

[0218] In some embodiments, the fusion vector determination module 303 is specifically configured to:

[0219] Aligning the fault signal vector and the fault semantic vector to obtain a signal embedding vector and a semantic embedding vector;

[0220] The target fusion weight, the signal embedding vector, and the semantic embedding vector are input into a pre-trained vector fusion model, and processed by the vector fusion model to output a multimodal fusion vector.

[0221] In some embodiments, the fusion vector determination module 303 is specifically configured to:

[0222] Performing a linear transformation on the fault signal vector, mapping the fault signal vector to a preset semantic space, and obtaining a signal embedding vector;

[0223] Performing a linear transformation on the fault semantic vector, mapping the fault semantic vector to a preset semantic space, and obtaining a semantic embedding vector.

[0224] In some embodiments, the target fusion weight includes a fault signal channel weight corresponding to the fault signal vector and a fault semantic channel weight corresponding to the fault semantic vector. The fusion vector determination module 303 includes a weight determination unit, and the weight determination unit is specifically configured to:

[0225] Determining a first confidence level corresponding to the fault signal vector and a second confidence level corresponding to the fault semantic vector;

[0226] In response to the first confidence being greater than a first preset confidence threshold and the second confidence being less than the second preset confidence threshold, determining the fault signal channel weight to be a first weight value and determining the fault semantic channel weight to be a second weight value; or,

[0227] In response to the first confidence being less than a first preset confidence threshold and the second confidence being greater than the second preset confidence threshold, determining the fault signal channel weight to be the second weight value and determining the fault semantic channel weight to be the first weight value; or,

[0228] In response to the first confidence being greater than a first preset confidence threshold and the second confidence being greater than the second preset confidence threshold, determining that both the fault signal channel weight and the fault semantic channel weight are a third weight value;

[0229] The first weight value is greater than the third weight value, and the third weight value is greater than the second weight value.

[0230] In some embodiments, the weight determination unit is specifically configured to:

[0231] determining a fluctuation value of the fault signal vector within a preset time window, and determining a first initial confidence level according to the fluctuation value;

[0232] Determining the number of fault features contained in the fault signal vector per unit time, and determining a second initial confidence level according to the number of fault features;

[0233] Calculating a vector distance between the fault signal vector and a preset fault sample, and determining a third initial confidence level according to the vector distance;

[0234] The first initial confidence level, the second initial confidence level, and the third initial confidence level are weighted to obtain a first confidence level corresponding to the fault signal vector.

[0235] In some embodiments, the weight determination unit is specifically configured to:

[0236] Determining whether the fault semantic vector contains preset semantic structure information, obtaining a first determination result, and determining a syntactic completeness corresponding to the first determination result;

[0237] Determining whether the fault semantic vector contains a preset keyword, obtaining a second judgment result, and determining a keyword coverage rate corresponding to the second judgment result;

[0238] The syntactic completeness and the keyword coverage are weighted to obtain the semantic vector completeness, and the confidence corresponding to the semantic vector completeness is used as the second confidence corresponding to the fault semantic vector.

[0239] For the convenience of description, the above devices are described as being functionally divided into various modules. Of course, when implementing the present disclosure, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0240] The device of the above embodiment is used to implement the corresponding vehicle fault determination method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0241] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, the vehicle fault determination method described in any of the above embodiments is implemented.

[0242] Figure 4 10 is a schematic diagram showing a more specific hardware structure of an electronic device provided in this embodiment. The device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are communicatively connected to each other within the device via the bus 1050.

[0243] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0244] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage devices, dynamic storage devices, etc. The memory 1020 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0245] The input / output interface 1030 is used to connect input / output modules to implement information input and output. The input / output modules can be configured as components within the device (not shown in the figure) or can be externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, various sensors, etc., and output devices may include a display, speaker, vibrator, indicator light, etc.

[0246] The communication interface 1040 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via a wired method (such as USB, network cable, etc.) or a wireless method (such as mobile network, WiFi, Bluetooth, etc.).

[0247] The bus 1050 comprises a path for transmitting information between the various components of the device (eg, the processor 1010 , the memory 1020 , the input / output interface 1030 , and the communication interface 1040 ).

[0248] It should be noted that although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040, and the bus 1050, in a specific implementation, the device may also include other components necessary for normal operation. In addition, it will be understood by those skilled in the art that the above device may only include the components necessary to implement the embodiments of this specification, and does not necessarily include all the components shown in the figure.

[0249] The electronic device of the above embodiment is used to implement the corresponding vehicle fault determination method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0250] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present disclosure also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable the computer to execute the vehicle fault determination method described in any of the above embodiments.

[0251] The computer-readable media of this embodiment include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0252] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the vehicle fault determination method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0253] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present application also provides a vehicle, including the vehicle fault determination device in the above-mentioned embodiment, the electronic device in the above-mentioned embodiment, and the computer-readable storage medium in the above-mentioned embodiment, and the vehicle equipment implements the vehicle fault determination method described in any of the above embodiments.

[0254] The vehicle of the above embodiment is used to implement the vehicle fault determination method described in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0255] It is understandable that before using the technical solutions of each embodiment of the present disclosure, the type, scope of use, usage scenarios, etc. of the personal information involved will be informed to the user in an appropriate manner, and the user's authorization will be obtained.

[0256] For example, in response to a user's active request, a prompt message is sent to the user to clearly inform the user that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the electronic device, application, server, storage medium, or other software or hardware that performs the operation of the disclosed technical solution based on the prompt message.

[0257] As an optional but non-limiting implementation, in response to a user's active request, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. Furthermore, the pop-up window may also contain a selection control for the user to select "agree" or "disagree" to provide personal information to the electronic device.

[0258] It is understandable that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0259] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present disclosure (including the claims) is limited to these examples. Within the scope of the present disclosure, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present disclosure as described above, which are not provided in detail for the sake of simplicity.

[0260] In addition, to simplify the description and discussion, and so as not to obscure the embodiments of the present disclosure, known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided figures. In addition, devices may be shown in the form of block diagrams to avoid obscuring the embodiments of the present disclosure, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform on which the embodiments of the present disclosure are to be implemented (i.e., these details should be fully within the purview of those skilled in the art). Where specific details (e.g., circuits) are set forth to describe exemplary embodiments of the present disclosure, it will be apparent to those skilled in the art that the embodiments of the present disclosure may be implemented without these specific details or with variations in these specific details. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0261] Although the present disclosure has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those skilled in the art based on the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

[0262] The embodiments of the present disclosure are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. A vehicle fault determination method, characterized in that: include: Determining that a vehicle has a fault, obtaining vehicle log data and fault text data, encoding the vehicle log data to obtain a fault signal vector; Parsing the fault text data to obtain a fault semantic vector; determining a target fusion weight, and fusing the fault signal vector and the fault semantic vector based on the target fusion weight to obtain a multimodal fusion vector; The multimodal fusion vector is input into a pre-trained vehicle fault analysis model, and is processed by the vehicle fault analysis model to output a target fault type and a target fault handling method.

2. The method according to claim 1, characterized in that The encoding process of the vehicle log data to obtain a fault signal vector includes: Performing time stamp analysis on the vehicle log data to obtain a time series signal matrix; The time series signal matrix is ​​input into a pre-trained first language model, and the first language model is used to perform feature encoding processing on the time series signal matrix to obtain a fault signal vector.

3. The method according to claim 1, characterized in that The parsing process of the fault text data to obtain a fault semantic vector includes: Filtering the fault text data to obtain fault text data to be processed; Performing format conversion on the fault text data to be processed to obtain a fault text sequence; The fault text sequence is input into a pre-trained second language model, and semantic parsing is performed using the second language model to obtain a fault semantic vector.

4. The method according to claim 1, wherein The fusing the fault signal vector and the fault semantic vector based on the target fusion weight to obtain a multimodal fusion vector includes: Aligning the fault signal vector and the fault semantic vector to obtain a signal embedding vector and a semantic embedding vector; The target fusion weight, the signal embedding vector, and the semantic embedding vector are input into a pre-trained vector fusion model, and processed by the vector fusion model to output a multimodal fusion vector.

5. The method according to claim 4, characterized in that The aligning the fault signal vector and the fault semantic vector to obtain a signal embedding vector and a semantic embedding vector includes: Performing a linear transformation on the fault signal vector, mapping the fault signal vector to a preset semantic space, and obtaining a signal embedding vector; Performing a linear transformation on the fault semantic vector, mapping the fault semantic vector to a preset semantic space, and obtaining a semantic embedding vector.

6. The method according to claim 1, wherein The target fusion weight includes the fault signal channel weight corresponding to the fault signal vector and the fault semantic channel weight corresponding to the fault semantic vector; Determining the target fusion weight includes: Determining a first confidence level corresponding to the fault signal vector and a second confidence level corresponding to the fault semantic vector; In response to the first confidence being greater than a first preset confidence threshold and the second confidence being less than the second preset confidence threshold, determining the fault signal channel weight to be a first weight value and determining the fault semantic channel weight to be a second weight value; or, In response to the first confidence being less than a first preset confidence threshold and the second confidence being greater than the second preset confidence threshold, determining the fault signal channel weight to be the second weight value and determining the fault semantic channel weight to be the first weight value; or, In response to the first confidence being greater than a first preset confidence threshold and the second confidence being greater than the second preset confidence threshold, determining that both the fault signal channel weight and the fault semantic channel weight are a third weight value; The first weight value is greater than the third weight value, and the third weight value is greater than the second weight value.

7. The method according to claim 6, characterized in that Determining a first confidence level corresponding to the fault signal vector includes: determining a fluctuation value of the fault signal vector within a preset time window, and determining a first initial confidence level according to the fluctuation value; Determining the number of fault features contained in the fault signal vector per unit time, and determining a second initial confidence level according to the number of fault features; Calculating a vector distance between the fault signal vector and a preset fault sample, and determining a third initial confidence level according to the vector distance; The first initial confidence level, the second initial confidence level, and the third initial confidence level are weighted to obtain a first confidence level corresponding to the fault signal vector.

8. The method according to claim 6, characterized in that Determining the second confidence level corresponding to the fault semantic vector includes: Determining whether the fault semantic vector contains preset semantic structure information, obtaining a first determination result, and determining a syntactic completeness corresponding to the first determination result; Determining whether the fault semantic vector contains a preset keyword, obtaining a second judgment result, and determining a keyword coverage rate corresponding to the second judgment result; The syntactic completeness and the keyword coverage are weighted to obtain the semantic vector completeness, and the confidence corresponding to the semantic vector completeness is used as the second confidence corresponding to the fault semantic vector.

9. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A vehicle, characterized in that: The vehicle includes the electronic device according to claim 9.

Images