Vehicle fault early warning method and device, electronic equipment and storage medium

Abstract

This invention discloses a method, device, electronic device, and storage medium for early warning of vehicle malfunctions. The method includes: acquiring multimodal data of the vehicle; inputting the multimodal data into a deep learning model, using the deep learning model to identify vehicle malfunctions, and obtaining at least one malfunction identification result. The deep learning model is trained based on historical multimodal data and a fault tree, where different nodes in the fault tree represent different causes of malfunctions, and the malfunction identification result reflects the cause of the vehicle malfunction; and generating early warning information based on at least one malfunction identification result. This invention solves the technical problem of low accuracy in early warning of vehicle malfunctions in related technologies.

Description

Technical Field

[0001] This invention relates to the field of vehicles and the field of fault warning, and more specifically, to a method, device, electronic device and storage medium for warning of vehicle faults. Background Technology

[0002] With the rapid development of intelligent connected vehicles, fault warning systems aim to detect anomalies in a timely manner before or in the early stages of vehicle malfunctions, preventing the malfunctions from worsening and causing safety accidents, sudden breakdowns, or significant losses. Accurate fault warnings can improve driving safety, extend component lifespan, reduce maintenance costs, and enhance user trust and service experience.

[0003] Currently, fault assessment often relies on manual methods based on empirical rules or static thresholds, which makes it difficult to accurately identify complex, sporadic, and compound faults. In other words, the relevant technologies still suffer from low accuracy in fault warning.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a method, device, electronic device, and storage medium for early warning of vehicle malfunctions, in order to at least solve the technical problem of low accuracy in early warning of vehicle malfunctions in related technologies.

[0006] According to one aspect of the present invention, a method for early warning of vehicle faults is provided, comprising: acquiring multimodal data of a vehicle, wherein the multimodal data includes at least sensor data, message data, and log data; inputting the multimodal data into a deep learning model, using the deep learning model to identify faults in the vehicle, and obtaining at least one fault identification result, wherein the deep learning model is trained based on historical multimodal data and a fault tree, different nodes in the fault tree represent fault causes corresponding to different fault data, and the fault identification result is used to reflect the fault causes of the vehicle fault; and generating early warning information based on at least one fault identification result.

[0007] Optionally, the multimodal data is input into a deep learning model, and the deep learning model is used to identify vehicle faults to obtain at least one fault identification result, including: preprocessing the multimodal data to obtain processed multimodal data; inputting the processed multimodal data into a deep learning model, and using the deep learning model to output at least one fault identification result.

[0008] Optionally, the multimodal data is preprocessed to obtain processed multimodal data, including: classifying the multimodal data based on the vehicle's electrical circuits to obtain multiple datasets, where different datasets represent data from different electrical circuits; and deleting operational data that deviates from the mode in each dataset to obtain processed multimodal data, where different operational data correspond to different modes.

[0009] Optionally, the deep learning model includes an anomaly detection module and an inference module; the processed multimodal data is input into the deep learning model, and the deep learning model outputs at least one fault identification result, including: using the anomaly detection module to perform anomaly detection on the processed multimodal data to obtain fault data; using the fault tree-guided inference module to infer the fault cause corresponding to the fault data to obtain at least one fault identification result.

[0010] Optionally, at least one fault identification result may also include the confidence level of the fault cause; based on at least one fault identification result, generating early warning information includes: determining the security level corresponding to at least one fault identification result based on the confidence level contained in at least one fault identification result; and generating early warning information based on the security level corresponding to at least one fault identification result.

[0011] Optionally, the above method also includes: pushing the warning information to the vehicle's alert module and / or the target object's maintenance equipment.

[0012] According to another aspect of the present invention, a vehicle fault early warning device is also provided, comprising: a vehicle fault early warning device, optionally comprising: an acquisition module, configured to acquire multimodal data of the vehicle, wherein the multimodal data includes at least sensor data, message data, and log data; a detection module, configured to input the multimodal data into a deep learning model, and use the deep learning model to identify faults in the vehicle to obtain at least one fault identification result, wherein the deep learning model is trained based on historical multimodal data and a fault tree, different nodes in the fault tree represent fault causes corresponding to different fault data, and the fault identification result is used to reflect the fault cause of the vehicle fault; and an early warning module, configured to generate early warning information based on at least one fault identification result.

[0013] According to another aspect of the present invention, an electronic device is also provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods of various embodiments of the present invention during runtime.

[0014] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is executed, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of the present invention.

[0015] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of the present invention.

[0016] According to another aspect of the present invention, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods of various embodiments of the present invention.

[0017] According to another aspect of the present invention, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of the present invention.

[0018] In this embodiment of the invention, multimodal data of a vehicle is acquired, input into a deep learning model, and the deep learning model is used to identify vehicle faults, obtaining at least one fault identification result. Based on the at least one fault identification result, a warning message is generated. This application first acquires multimodal data, and then inputs the multimodal data into a deep learning model to facilitate comprehensive analysis of vehicle faults. Because the deep learning model is jointly trained with historical multimodal data and fault trees, it can accurately pinpoint the cause of the fault while identifying abnormal data patterns. Therefore, it can generate warning messages based on the fault identification results, achieving the goal of accurate alarms. This improves the accuracy of vehicle fault warnings and solves the technical problem of low accuracy in vehicle fault warnings in related technologies. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0020] Figure 1 This is a flowchart of a vehicle fault early warning method according to an embodiment of the present invention;

[0021] Figure 2 This is a schematic diagram of data processing for an optional vehicle data analysis platform according to an embodiment of the present invention;

[0022] Figure 3This is a schematic diagram of an optional vehicle fault early warning method according to an embodiment of the present invention;

[0023] Figure 4 This is a schematic diagram of an optional data preprocessing method according to an embodiment of the present invention;

[0024] Figure 5 This is a schematic diagram of a vehicle malfunction warning device according to an embodiment of the present invention;

[0025] Figure 6 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0027] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] According to an embodiment of the present invention, a method embodiment for a vehicle fault warning method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0029] Figure 1 This is a flowchart of a vehicle fault early warning method according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0030] Step S102: Obtain multimodal data of the vehicle.

[0031] Multimodal data includes at least sensor data, message data, and log data.

[0032] The aforementioned vehicles are integrated information terminals and mechanical systems in the era of intelligent connected vehicles. They are equipped with multiple subsystems, including powertrain, transmission, electronic control, communication, perception, and human-machine interaction. The operational status of each subsystem may be reflected through electrical signals or data. The complex system coupling of vehicles makes it difficult for a single sensor or localized diagnostics to comprehensively capture faults, necessitating the construction of a cross-dimensional collaborative analysis framework. Because the operating environment of vehicles is dynamically changing, covering various scenarios such as urban congestion, high-speed cruising, and low-temperature starts, this enables early warning methods to possess strong robustness and adaptability. The open architecture of vehicles allows data to be uploaded in real time via the in-vehicle network, providing continuously updated training samples for cloud-based deep learning models, thereby achieving a shift from passive response to proactive prediction.

[0033] The aforementioned multimodal data is a heterogeneous collection of information synchronously generated by different sensing channels and information levels during vehicle operation. Multimodal data is characterized by the diversity of its data sources and the complementarity of its representational forms. Multimodal data may originate from analog continuous signals from different sensors, messages from in-vehicle communication networks, or event logs from software systems, etc.

[0034] Single-modal data often suffers from perceptual blind spots or semantic gaps. For example, sensor data may drift due to electromagnetic interference, message data may lose key frames due to bus congestion, and log data may only record results without describing the process. In contrast, the fusion of multimodal data can create a comprehensive understanding of faults. The richness of the data dimensions of multimodal data enables the system to identify hidden faults that are difficult to detect using traditional fault alarm methods, such as signal timing misalignment, control logic conflicts, and hardware-software coordination failures. The quality of multimodal data determines the effectiveness of the input to deep learning models.

[0035] The aforementioned sensor data consists of continuous engineering parameters collected in real time by various physical components of the vehicle through built-in sensing elements. These parameters may include various physical quantities such as temperature, pressure, speed, current, vibration, and oil pressure. Sensor data provides a direct quantitative expression of the operating status of the vehicle's mechanical and electrical systems, characterized by high-frequency sampling, strong temporal sequence, and high resolution. Sensor data may reflect the original record of the fault evolution process; for example, a slow rise in engine temperature may indicate a cooling system leak, and abnormal fluctuations in motor current may suggest a fault in the electronic control module. The accuracy and stability of sensor data reflect the underlying reliability of the diagnostic system.

[0036] The aforementioned message data consists of standardized protocol frames transmitted in the vehicle's in-vehicle communication network, such as controller commands and status information carried on buses like Ethernet. The message data may contain identifiers, data fields, timestamps, and checksums, reflecting the interaction logic and control intentions of various electronic control units within the vehicle. The message data may involve control commands with semantic structures, revealing system-level behavioral anomalies. For example, a brake system not responding to a steering request may indicate controller communication congestion, and frequent start-stop cycles of the air conditioning compressor may stem from control strategy conflicts.

[0037] The structured nature of message data enables the system to trace the root cause of faults, providing crucial evidence for discovering hidden system failures such as signal tampering, bus overload, and control logic disorder. Furthermore, the protocol specifications and transmission timing of message data provide a time reference for data synchronization and event correlation.

[0038] The aforementioned log data consists of event-driven text records generated by the vehicle's software system or controller during operation, such as error codes, warning messages, system startup logs, configuration change records, and abnormal termination records. Log data carries the operational trajectory and decision-making basis of the software layer, serving as the core basis for analyzing intermittent and software-related faults. For example, if a control module is forcibly restarted due to a memory overflow, or a diagnostic protocol is refused execution due to version incompatibility, these events can be precisely pinpointed to the source of the problem through log data. Log data can connect hardware anomalies with software defects. The contextual information in the log data allows the system to understand the specific conditions under which the fault occurred, such as ambient temperature, operating mode, and driving habits.

[0039] In one alternative embodiment, a vehicle-mounted edge computing gateway can be deployed and directly connected to the private bus of each controller to collect custom signals, such as phase current waveforms of the motor controller, cell voltage balancing status in the battery management system, raw radar point cloud data, camera image frames, and LiDAR point cloud streams. The collected signals can be accessed to the domain controller via a data acquisition port or through high-speed protocols such as the vehicle's local area network or Ethernet, achieving millisecond-level synchronous acquisition.

[0040] In another optional embodiment, aggregated data from connected vehicle service providers can be actively retrieved via a cloud platform. This includes multimodal data such as remote diagnostic reports, remote locking records, software upgrade failure logs, emergency call trigger events, abnormal fluctuations in location trajectory, and records of environmental temperature changes. Multimodal data indirectly reflects the vehicle's operating status. For example, multiple failed remote restarts may indicate firmware conflicts in the controller, frequent entry into safe mode may point to power system instability, and significant jumps in location trajectory during non-driving periods may suggest false triggering of the anti-theft system. Leveraging the operator's network and cloud service ecosystem, multimodal data aggregation across vehicle models, brands, and regions can be achieved.

[0041] In another optional embodiment, the in-vehicle wireless communication module can proactively push log data from the in-vehicle mobile terminal or smart cockpit system, such as the central control screen operation trajectory, voice command response records, navigation route changes, air conditioning setting preferences, seat position adjustments, and driving mode switching. It can also acquire message data transmitted in the in-vehicle network. Furthermore, the in-vehicle wireless communication module can connect to various sensors to acquire sensor data.

[0042] Step S104: Input multimodal data into a deep learning model, use the deep learning model to identify vehicle faults, and obtain at least one fault identification result.

[0043] Among them, the deep learning model is trained based on historical multimodal data and fault tree. Different nodes in the fault tree represent the fault causes corresponding to different fault data. The fault identification results are used to reflect the fault causes of vehicle malfunctions.

[0044] The aforementioned deep learning model is an intelligent analysis engine built upon a neural network architecture. Deep learning models can be trained end-to-end, automatically extracting high-dimensional nonlinear features and establishing complex mapping relationships between input data and fault identification results. Deep learning models may integrate temporal convolution and attention mechanisms from graph neural networks, enabling them to discover implicit correlations in massive amounts of historical data. The training process of deep learning models not only relies on multimodal data but also introduces engineering knowledge represented by fault trees as structural constraints. This ensures that the model prioritizes fault paths consistent with expert experience during learning, avoiding spurious correlations. Consequently, the output of the deep learning model provides interpretable fault causes and confidence levels, demonstrating high generalization ability and adaptability through this knowledge-guided data-driven approach.

[0045] The fault identification results described above are diagnostic conclusions output by a deep learning model after comprehensive analysis of multimodal data. These results may include, but are not limited to, specific fault causes, confidence levels, and maintenance recommendations. The fault identification results can clearly reflect specific issues such as "loss of water pump drive signal" or "battery management system communication timeout." The accuracy of the fault identification results determines the reliability of the warning, and the interpretability of the results determines the efficiency of the maintenance.

[0046] The aforementioned historical multimodal data is a complete sample set including sensor data, messages, and logs accumulated by the vehicle during long-term operation. This historical multimodal data may cover normal operating conditions and known fault states. It is used not only to train model parameters but also to construct and validate the structure and node weights of the fault tree. The continuous backhaul and incremental learning mechanism of historical multimodal data enables the system to self-evolve, adapting to the emergence of new vehicle models, new sensors, and new fault modes.

[0047] The aforementioned fault tree is an engineering knowledge graph that uses a tree-like topological structure to express the logical relationship between fault causes and fault phenomena. The root node of the fault tree represents the terminal fault, intermediate nodes represent intermediate causes, leaf nodes represent underlying fault triggers, and edges represent logical combinations. The structure of the fault tree originates from an understanding of real-world fault data, providing semantic constraints and a reasoning framework for deep learning models, making the model output traceable and interpretable. The existence of the fault tree ensures that the model's reasoning path conforms to engineering principles. For example, "engine overheating" may be caused by direct factors such as "insufficient coolant," "water pump failure," or "radiator fan failure," rather than random associations. The fault tree not only guides training but also serves as the structural input during reasoning, forming the foundation for achieving human-machine collaborative intelligent diagnosis.

[0048] In one optional embodiment, a multi-branch temporal convolutional neural network can be constructed. Sensor data, message data, and log data are input into independent convolutional branches. Sensor data undergoes one-dimensional convolution to extract local temporal fluctuation features. Message data is encoded into a structured sequence through a sliding window and then input into a channel attention module. Log data is embedded into semantic vectors and then fed into a bidirectional long short-term memory network to model the event causal chain. Then, the three features are dynamically weighted and fused through a cross-modal attention mechanism, and finally, a fault category probability distribution is output by a fully connected layer. The multi-branch temporal convolutional neural network can effectively identify complex faults caused by the coordinated failure of multiple asynchronous signals, such as increased coolant temperature, abnormal water pump messages, and frequent restarts of air conditioning control logs, all pointing to water pump motor drive failure.

[0049] In another alternative embodiment, a joint modeling method combining graph neural networks and fault tree structures can be employed. The fault tree serves as a prior knowledge graph, mapping each fault node to a vertex in the graph. Multimodal data serves as the initial attribute vectors for the nodes. Message passing and feature aggregation are performed on the topology through a graph convolutional network, enabling the model to automatically learn during training which abnormal combinations should lead to which fault nodes. For example, when "sudden battery voltage drop" and "charging management message timeout" are activated simultaneously, the graph model propagates through neighboring nodes to "charging module communication interruption" rather than "cell aging." This approach encodes expert experience into the network structure, avoiding spurious correlations in purely data-driven models, achieving interpretable and traceable fault attribution, and realizing knowledge-guided intelligent diagnosis.

[0050] In another alternative embodiment, a joint framework of autoencoder and anomaly scoring can be employed. A multimodal joint autoencoder is constructed to compress and reconstruct multimodal data. The training objective is to minimize the reconstruction error. When a hidden fault occurs, if a certain modality causes overall reconstruction distortion due to data anomalies, the system calculates the weighted sum of the reconstruction residuals of each modality as the anomaly score. This score is then combined with cluster analysis to identify typical combinations of anomaly patterns. For example, if "high motor current reconstruction error" and "missing torque command log" occur simultaneously, the system determines it as "abnormal drive control logic." This approach is particularly suitable for zero-shot detection of new vehicle models or new fault modes.

[0051] In another alternative embodiment, fault identification can be achieved through federated learning and an edge-cloud collaborative inference architecture. A lightweight model is deployed on the vehicle to perform preliminary anomaly detection and feature extraction on local multimodal data. The encrypted feature vectors are then uploaded to a cloud server, where a deep learning model performs joint inference, outputs fault identification results, and sends them back to the local machine. The cloud model continuously absorbs new data for incremental updates, while the vehicle-side model only receives updated parameters and does not upload the original data, thus ensuring data security and enabling cross-vehicle and cross-regional swarm intelligent diagnostics.

[0052] Step S106: Generate early warning information based on at least one fault identification result.

[0053] The aforementioned warning information is generated based on the fault identification results. The warning information may include a fault description, severity, handling suggestions, and response priority. Presented in visual, auditory, or digital text format, the warning information aims to guide the driver to take safety measures or notify maintenance personnel, emphasizing operability and urgency. Examples include warnings such as "It is recommended to pull over and check the cooling system as soon as possible" or "For high-risk faults, please contact the service center immediately." These warnings can promptly and accurately convey the fault situation, improving driving safety and user experience.

[0054] In one optional embodiment, warning information can be generated using structured natural language. The fault identification results are input into a pre-trained language generation model, which automatically converts them into natural language descriptions consistent with engineering contexts, such as, "An abnormal rise in drive motor temperature has been detected, with a probability of 93%. Possible cause: malfunction of the coolant circulation pump. It is recommended to immediately stop high-load operation and check the cooling system." The generated content can be defined according to account type, using colloquial language for drivers and professional terminology and technical parameters for maintenance technicians, supporting multi-language switching and customizable terminology databases.

[0055] In another optional embodiment, when multiple fault causes exist, a comprehensive assessment is performed on these causes. This assessment considers the severity, probability of occurrence, repair cost, safety risk, and time sensitivity of the fault. A comprehensive hazard index is calculated using fuzzy decision-making or entropy weighting methods, and the faults are dynamically ranked. The fault cause ranked first is then converted into fault information. For example, if both "abnormal noise from the air conditioning compressor" and "low brake fluid level" are identified simultaneously, the system prioritizes the latter because it is directly related to driving safety, while the former is downgraded to "recommendation for next maintenance check." This engine can combine historical maintenance data to automatically filter recurring faults or known false alarms, avoiding information redundancy and providing a basis for repair shops to predict spare parts needs. Its output not only determines the order of warnings but also affects subsequent diagnostic resource scheduling, thereby achieving intelligent decision-making and improved operational efficiency.

[0056] In another optional embodiment, a cross-terminal collaborative push and closed-loop feedback linkage mechanism is used to generate early warning information. For example, based on the fault identification result, an audible and visual alarm is triggered on the vehicle, the cloud simultaneously archives the information and triggers the early warning pool, and the maintenance platform automatically generates service work orders and other early warning information, which are then pushed to the nearest authorized service center. Early warning information such as "Your vehicle has a potential powertrain system risk. We have booked the nearest repair shop for you. Click to confirm and book with one click" is pushed to the vehicle's mobile terminal. Simultaneously, the system reserves a feedback channel; if the user chooses to ignore it or confirms that it has been processed, the feedback will be sent back to the model to correct parameters.

[0057] In this embodiment of the invention, multimodal data of a vehicle is acquired, input into a deep learning model, and the deep learning model is used to identify vehicle faults, obtaining at least one fault identification result. Based on the at least one fault identification result, a warning message is generated. This application first acquires multimodal data, and then inputs the multimodal data into a deep learning model to facilitate comprehensive analysis of vehicle faults. Because the deep learning model is jointly trained with historical multimodal data and fault trees, it can accurately pinpoint the cause of the fault while identifying abnormal data patterns. Therefore, it can generate warning messages based on the fault identification results, achieving the goal of accurate alarms. This improves the accuracy of vehicle fault warnings and solves the technical problem of low accuracy in vehicle fault warnings in related technologies.

[0058] Optionally, the multimodal data is input into a deep learning model, and the deep learning model is used to identify vehicle faults to obtain at least one fault identification result, including: preprocessing the multimodal data to obtain processed multimodal data; inputting the processed multimodal data into a deep learning model, and using the deep learning model to output at least one fault identification result.

[0059] In one optional embodiment, the multimodal data is preprocessed, such as through systematic cleaning, alignment, denoising, and structural transformation, to remove invalid, redundant, and interfering information, thus constructing high-quality, consistent, and feature-focused data. By addressing issues such as missing values, time asynchrony, inconsistent sampling rates, semantic confusion, and noise interference in the original data, the data input to the model is ensured to possess learnability and stability, thereby reducing the complexity of model training and the risk of misjudgment, improving the accuracy of feature extraction, and preventing subsequent analysis from being misled by low-quality data, thus ensuring the underlying reliability of the intelligent diagnostic system.

[0060] Specifically, preprocessing of multimodal data can be achieved through semantic partitioning alignment based on electrical circuits. For example, according to the automotive electrical schematic, sensor data, message data, and log data can be logically grouped according to their respective functional circuits (such as powertrain circuits, air conditioning control circuits, lighting circuits, etc.) to ensure that data from the same physical system are synchronously aligned in time and semantic space. For instance, engine temperature sensor data, coolant pump current signals, and air conditioning control logs can be uniformly grouped into a "thermal management system" partition, and time-stamp interpolation and sampling rate standardization can reduce timing offsets caused by asynchronous multi-source acquisition.

[0061] For example, this can be achieved through a fusion of multimodal intelligent missing value imputation and interpolation. A context-based deep interpolation model is employed, utilizing the temporal patterns of other related signals within the same loop to predict missing values. For instance, when coolant temperature data is lost, the system infers possible temperature trajectories based on the evolution trends of associated signals such as pump speed, fan duty cycle, and engine load, thereby restoring data integrity. It can also uncover implicit physical coupling relationships between variables, giving the imputation of missing data an engineering logic rather than simple linear interpolation. This improves data continuity and model robustness.

[0062] For example, data preprocessing can be performed through cross-modal normalization and semantic encoding unification. Different types of data are mapped to the same semantic space. For example, through word embedding models, encoding classification, etc., multimodal data is compressed into a high-dimensional vector representation with a unified dimension. This allows information from different sources to be processed equally in the same neural network, which is conducive to the model learning joint patterns and provides structurally consistent and semantically aligned inputs for subsequent deep learning models.

[0063] Then, the processed multimodal data is input into a deep learning model for joint modeling. Through nonlinear mapping and pattern recognition, the model automatically uncovers fault correlation patterns hidden between time-series signals, communication logic, and event logs, ultimately outputting at least one fault identification result. This enables the perception of anomalies and understanding of causes, overcoming the limitations of traditional threshold judgment and rule engines to meet the identification needs of complex, compound, and sporadic faults. This can transform massive amounts of heterogeneous data into interpretable and executable diagnostic conclusions, enabling the system to possess autonomous reasoning and generalization capabilities. It can identify new fault modes even without explicit rules, and promote the transformation of vehicle operation and maintenance systems from passive response to proactive prediction, shifting fault diagnosis from relying on human experience to data-driven intelligent decision-making, thereby improving identification accuracy and response speed.

[0064] Optionally, the multimodal data is preprocessed to obtain processed multimodal data, including: classifying the multimodal data based on the vehicle's electrical circuits to obtain multiple datasets, where different datasets represent data from different electrical circuits; and deleting operational data that deviates from the mode in each dataset to obtain processed multimodal data, where different operational data correspond to different modes.

[0065] The aforementioned electrical circuits are functional units in a vehicle, consisting of wires connecting sensors, actuators, controllers, and power supplies. Each circuit can perform different specific electrical tasks, such as powering the lighting system, power control instruments, etc. Classifying multimodal data through electrical circuits gives data preprocessing engineering and physical meaning, avoids cross-system interference, and improves the targeting of feature extraction. For example, grouping the current signal of the air conditioning compressor and temperature sensor data into the same circuit can accurately identify whether cooling anomalies originate from electrical faults. The division of electrical circuits can be based on automotive electrical schematics and repair manuals, thereby establishing a mapping between abstract data and physical components. This allows diagnostic results to accurately pinpoint specific circuit modules or components, improving repair efficiency and reducing the risk of incorrect disassembly and repair.

[0066] The mode value mentioned above is the statistical central tendency of operating data within a specific electrical circuit under normal conditions. The mode value can be calculated by weighted summation of multiple parameters, representing the typical operating range of the electrical circuit under typical operating conditions. The mode value comprehensively considers environmental variables such as temperature, load, and speed, enabling the baseline value to have dynamic adaptive capabilities. As a dynamic boundary of normal behavior, the mode value is used to identify abnormal samples deviating from normal behavior, adapting to operating changes under different seasons and driving habits, and is a core parameter for achieving unsupervised anomaly detection. The dynamic update mechanism of the mode value allows the system to maintain diagnostic effectiveness even after vehicle aging or component wear, thereby improving system robustness and long-term stability.

[0067] In one optional embodiment, multimodal data from sensors, controllers, and communication messages can be logically categorized and grouped according to their respective functional electrical circuits (such as power drive circuits, thermal management system circuits, lighting control circuits, and vehicle comfort system circuits) based on the design architecture of the automotive electrical system, ensuring that the data organization is consistent with the physical structure. By constructing data units with clear physical meaning, subsequent analysis is ensured to be performed at the functional module level, avoiding cross-system interference and miscorrelation.

[0068] Then, within each electrical circuit, the mode of key signals is calculated based on historical normal operation data. The mode is the range of values ​​that frequently appear under specific operating conditions, rather than a fixed threshold or mean. The system compares the current data point in real time to see if it deviates from this mode and removes abnormal sampling points that exceed the tolerance, thereby reducing the interference of sensor noise, transient interference, and occasional anomalies on the diagnostic model, retaining the core data reflecting the true operating state, and avoiding the obliteration of the dynamic characteristics of the data by simple averaging. This achieves "adaptive denoising," giving data cleaning the ability to perceive operating conditions. For example, when the mode of engine coolant temperature rises under high load, the system automatically adjusts the tolerance range to avoid misjudgment. This allows for the construction of high-quality data input with environmental adaptability, reducing the model's false alarm rate.

[0069] Furthermore, each signal exhibits different normal behavior patterns under various operating conditions (such as vehicle speed, ambient temperature, and load status), therefore the mode needs to be dynamically calculated rather than globally uniform. This establishes a normal baseline for context awareness, ensuring that the data cleaning logic is highly aligned with the actual operating patterns of the vehicle.

[0070] Optionally, the deep learning model includes an anomaly detection module and an inference module; the processed multimodal data is input into the deep learning model, and the deep learning model outputs at least one fault identification result, including: using the anomaly detection module to perform anomaly detection on the processed multimodal data to obtain fault data; using the fault tree-guided inference module to infer the fault cause corresponding to the fault data to obtain at least one fault identification result.

[0071] The aforementioned anomaly identification module is a front-end component of the deep learning model. It automatically extracts abnormal feature signals from preprocessed multimodal data, ignores normal fluctuations, and focuses on potential fault areas. This module can use time-series analysis clustering algorithms and autoencoder structures to establish dynamic normal boundaries by learning the distribution of historical normal data. It does not rely on manually set thresholds and can identify small but continuous abnormal trends, such as slow shifts in sensor signals or slight extensions in message intervals. Its output is a high-dimensional feature vector rather than a single alarm signal, providing accurate input for subsequent inference modules and realizing the transition from "detecting anomalies" to "analyzing anomalies".

[0072] The aforementioned inference module is a backend component of the deep learning model. It matches the feature vectors output by the anomaly detection module with the topology of the fault tree, uses the logical relationships and weight distribution between nodes to perform probabilistic deduction, and outputs the cause of the fault and the confidence level. The inference module integrates data-driven discovery capabilities and knowledge-driven guidance capabilities, making the diagnostic process interpretable and engineering-reasonable. For example, when the feature vector points to "increased coolant temperature" and "decreased water pump current," the inference module infers "water pump failure" as the primary cause based on the fault tree, rather than "radiator blockage," and its inference path is traceable and verifiable.

[0073] In one optional embodiment, an anomaly detection module is used to perform unsupervised or semi-supervised analysis on preprocessed multimodal data, automatically detect abnormal patterns that deviate from the normal operating mode, and output data segments or feature vector sets marked as "potential faults" to proactively discover hidden abnormal behaviors, filter out a large amount of fault-free information, improve diagnostic efficiency, and thus achieve proactive perception of unknown anomalies.

[0074] Then, for the fault data, joint reasoning is performed with a pre-constructed structured fault tree based on engineering knowledge and historical maintenance cases. Through mechanisms such as graph neural networks or Bayesian networks, abnormal features are backpropagated within the fault tree topology to infer possible root cause nodes and confidence levels. This embodiment integrates data-driven anomaly detection results with expert experience and knowledge, making the diagnostic conclusions interpretable, traceable, and engineering-reasonable, thereby identifying specific fault causes. For example, based on fault data such as abnormal coolant temperature and water pump current fluctuations, water pump motor drive failure can be analyzed, and multiple candidate causes ranked by priority can be output.

[0075] Optionally, at least one fault identification result may also include the confidence level of the fault cause; based on at least one fault identification result, generating early warning information includes: determining the security level corresponding to at least one fault identification result based on the confidence level contained in at least one fault identification result; and generating early warning information based on the security level corresponding to at least one fault identification result.

[0076] The confidence level mentioned above is a probability assessment value for each fault cause, reflecting the degree of match between the fault cause and the observed data. The confidence level can range from zero to 1, used to quantify the uncertainty of the diagnosis. As a quantitative basis for implementing graded early warning and priority ranking, the introduction of confidence level enables the system to distinguish between high-probability primary causes and low-probability secondary causes, avoiding false alarms. For example, when the confidence level for "water pump failure" is 0.92, while that for "temperature control valve failure" is 0.31, the system can prioritize pushing the former. Confidence level is not only a label of the credibility of the diagnostic results, but also a feedback signal for model self-calibration and continuous learning. By comparing with maintenance feedback data, the system can correct the model's judgment bias on certain fault modes.

[0077] The aforementioned safety levels are warning classifications derived from a comprehensive assessment of the confidence level and impact range in the fault identification results. Safety levels can be divided into three categories: minor, moderate, and severe, or represented numerically to determine the intensity of the warning and the response method. The determination of safety levels can be combined with vehicle operating scenarios and safety standards to ensure timely intervention for high-risk faults and appropriate reminders for low-risk faults. For example, "brake system failure" is a severe level triggering an emergency stop warning, while "weak air conditioning cooling" is a minor level only displaying a warning icon. Safety levels serve as a decision-making basis connecting diagnostic results and user behavior to balance safety and user experience, avoiding information overload leading to neglect.

[0078] In one optional embodiment, the deep learning model, while outputting the cause of the fault, can simultaneously generate a probability value indicating that the fault is actually present, reflecting the system's certainty regarding the diagnostic results. This quantifies the reliability of the diagnostic results, avoiding the misinterpretation of low-confidence false alarms or vague speculations as definitive faults, enhancing the prudence of system decision-making, and providing quantifiable evidence for subsequent early warning grading. This effectively reduces false alarm rates and user fatigue, achieving a perception of the uncertainty in diagnostic results.

[0079] Then, based on the confidence level and the inherent safety impact weight of the fault type (e.g., braking faults have a higher weight than air conditioning faults), a single fault identification result can be mapped to a safety level through preset rules or a weighted scoring model. This transforms the technical probability value into a risk level that users can understand and act upon. By establishing a two-dimensional assessment mechanism of confidence and severity, it ensures that even faults with low confidence but high hazards (such as battery thermal runaway) are handled with high priority, while anomalies with high confidence but low hazards (such as abnormal window noises) are only alerted, thus achieving reasonable resource allocation.

[0080] Furthermore, based on the determined safety level, preset tiered early warning strategies can be invoked to dynamically generate matching prompts, presentation formats, and interaction methods, or the safety level can be manually analyzed. For example, an emergency level might trigger a voice alarm, flashing red lights on the instrument panel, or automatic emergency calls; a high-risk level might display graphic or textual prompts or suggest nearby repairs; while low- and medium-risk levels would remain silently displayed on the information screen. This ensures accurate expression and scenario adaptation of warning information, avoiding information overload or ignoring the real threat. This improves human-computer interaction efficiency, ensuring drivers receive safety guidance quickly, reducing cognitive load, and increasing response speed.

[0081] Optionally, the above method also includes: pushing the warning information to the vehicle's alert module and / or the target object's maintenance equipment.

[0082] The aforementioned alert modules are information interaction terminals installed in the vehicle's cabin, such as dashboard displays, central control screens, head-up displays, or voice broadcast systems. These modules can present warning information to the driver in the form of graphics, text, and sound, ensuring the information is clear, easy to read, and does not distract the driver. For example, a flashing red icon and voice warning may be used when driving at high speeds, while a silent pop-up window may be used when parking. The level of intelligence of the alert module determines the user's trust in the system and is the core carrier for achieving human-machine collaborative driving and safety assistance.

[0083] The aforementioned repair equipment refers to external systems such as remote repair stations, mobile diagnostic terminals, or service cloud platforms that receive early warning information and support repair operations. Repair equipment receives early warning information, retrieves historical data, and generates repair work orders to guide disassembly and assembly processes and parameter calibration. Repair equipment can extend cloud-based diagnostic capabilities to the offline service chain. The data interoperability between repair equipment and vehicle systems determines repair efficiency. For example, repair technicians can use augmented reality glasses to overlay virtual faults to guide them to the actual component locations, or the system can automatically push spare parts inventory and scheduled services. Repair equipment translates diagnostic results into executable operational instructions to achieve predictive maintenance and service upgrades.

[0084] In one optional embodiment, the warning information can be pushed to the vehicle's alert module. For example, through the in-vehicle smart cockpit's dashboard, central control screen, or voice system, fault alarms and operational instructions can be delivered to the driver in real time, ensuring that the driver is aware of the vehicle's fault status immediately. This improves driving safety and prevents the driver from continuing to drive without knowing the fault, thus avoiding increased risk. For instance, when an abnormality in the braking system is detected, the system can immediately display "Please slow down and stop for repairs as soon as possible" accompanied by audio and visual reminders, guiding the driver to take safe operating procedures.

[0085] In another alternative embodiment, the warning information can be pushed to the target repair equipment. For example, fault details, confidence levels, suggested repair items, and historical data can be automatically synchronized to the terminal system of the repair station or remote diagnostic platform, such as to the repair technician's tablet, work order system, or AR repair glasses. This achieves a closed loop of diagnosis and repair, allowing repair personnel to have a complete fault profile before the vehicle arrives at the shop, enabling them to prepare tools, spare parts, and work plans in advance, shortening diagnosis time and waiting periods. This breaks down information silos between the vehicle and repair ends, promoting proactive prediction and accurate response in after-sales service, improving user satisfaction and repair efficiency, and reducing operating costs.

[0086] In another optional embodiment, the warning information can be pushed to the vehicle's alert module and the target object's maintenance equipment to build a three-in-one collaborative warning mechanism of "vehicle-cloud-maintenance end". This enables drivers and maintenance personnel to be informed and respond in a coordinated manner, thereby forming a complete service loop. For example, drivers receive immediate safety alerts, maintenance personnel prepare resources in advance, and the system automatically records diagnostic paths and feedback data, providing closed-loop samples for subsequent model improvements. This can create an intelligent, full-link, and traceable vehicle operation and maintenance ecosystem, improve the efficiency of single maintenance, and promote the systemic upgrade of service capabilities.

[0087] The technical solution proposed in this application will be described below with reference to an optional embodiment. This application proposes a method for processing vehicle data.

[0088] The transmission, processing, and analysis of vehicle data is a core component in building a next-generation intelligent vehicle fault diagnosis system. This process begins with collecting raw data from various onboard sensors (such as temperature, pressure, and speed sensors) and driver alerts (such as the instrument panel and central control interface). This data is then initially integrated and converted via onboard networks (such as controller area networks, local interconnect networks, and Ethernet), and subsequently synchronized to cloud or edge data centers in near real-time or batch processing. This data is used to comprehensively monitor vehicle status, identify potential faults, and analyze root causes. Once the system detects an anomaly, it translates the warning information into easily understandable instructions or prompts and feeds them back to the driver alert, guiding the driver to perform appropriate actions, such as limiting speed, seeking nearby repairs, or making an emergency stop.

[0089] However, current data transmission and analysis systems suffer from bottlenecks. Data accuracy is limited by multiple factors, including sensor performance, signal interference, and network transmission quality. Raw data often contains noise, incomplete data, and even format conflicts. Reliance on manual operation and offline processing tools results in low visualization and poor interactivity, and is unable to handle the automatic parsing and deep fusion of multi-source heterogeneous data (such as time-series data, event logs, images, and video streams). With the widespread adoption of intelligent connected functions, the dimensions and volume of data have increased dramatically. Traditional spreadsheet processing methods are no longer sufficient to effectively support real-time or near-real-time analysis needs, and are unable to quickly and accurately locate fault loops or identify the root causes of complex faults.

[0090] To address the aforementioned challenges, this embodiment provides a next-generation vehicle data analysis platform with multi-format compatibility, automated processing, and intelligent visualization capabilities.

[0091] This platform enables adaptive access and parsing of multimodal data. For example, its built-in intelligent parser automatically identifies data structures, extracts valid fields, and performs unified spatiotemporal alignment and standardization on data of different formats, reducing the complexity of multi-source data fusion. It also achieves high-performance real-time processing and fault feature extraction. Leveraging a stream processing engine and time-series database, it enables online computation and feature generation on massive amounts of vehicle data. By integrating signal processing algorithms (such as wavelet transform and filtering / denoising) and machine learning models, it automatically identifies abnormal patterns in the data, extracts fault-sensitive features, and generates preliminary fault hypotheses. This allows for interactive visualization and insight generation, providing rich visualization components (such as parallel coordinate graphs, heatmaps, 3D signal simulation, and fault tree walkthroughs). Users can explore data through drag-and-drop, drill-down, and filtering methods to quickly locate abnormal time periods, related parameters, and potential fault sources.

[0092] The platform can also automatically generate multi-dimensional diagnostic dashboards, integrating key indicators, fault confidence levels, and repair suggestions to improve the intuitiveness of data analysis and operational efficiency. This leads to the generation of automated reports and closed-loop feedback. Based on the analysis results, it can automatically generate structured diagnostic reports, including fault descriptions, occurrence times, impact ranges, processing priorities, and repair guidelines, which are then simultaneously delivered to vehicle terminals or repair stations via communication or the cloud. Simultaneously, the system supports the transmission of on-site repair results and new data to improve deep learning models and knowledge bases, forming a continuous improvement closed-loop learning mechanism.

[0093] This platform will not only be applicable to passenger and commercial vehicle fault diagnosis scenarios, but can also be extended to multiple fields such as fleet management, intelligent transportation, remote monitoring and preventive maintenance, providing the industry with standardized, intelligent and visualized data decision support, and ultimately promoting the intelligent transformation of vehicle operation and maintenance models from "post-event handling" to "pre-event early warning" and "in-event guidance".

[0094] like Figure 2 As shown, the data processing flow of the platform is illustrated. The flow includes: real-time data upload followed by data preprocessing, obtaining a fault tree from historical fault data, and then performing deep learning to achieve fault analysis and fault display.

[0095] The platform can process vehicle data through the following methods: processing, automatic identification, conversion, and display.

[0096] First, acquire vehicle background data, and then perform data cleaning, data integration, data transformation, feature selection and feature dimensionality reduction on the background data in sequence. Identify and remove invalid data, and automatically sort the valid data to form a data matrix.

[0097] Then, thresholds or data ranges under normal vehicle conditions are assigned to keywords in the big data list as the basis for binary classification. If the data deviates from the normal threshold, a deviation warning signal is generated as an alert.

[0098] Furthermore, relevant historical faults or other fault analysis data are uploaded. The uploaded content is then identified to obtain the possible causes corresponding to different faults, forming a fault tree list. Based on deep learning algorithms, the data matrix of different fault phenomena is analyzed. Data irrelevant to the current fault is excluded based on data link or fault analysis, and useful data information is identified. Specifically, different rows and columns of the database are identified, and the position of the first row keyword in the fault tree list is detected. It is then checked whether the data in that column conforms to the data range under normal vehicle conditions. If it is within the normal range, it is classified as 0; if the detection result is outside the normal data range, it returns 1. For data with a return value of 1, possible fault causes are searched in the fault tree.

[0099] After identifying potential causes of a fault, the cause is displayed on the vehicle's driver information display, and the severity level of the fault is assessed and displayed on the vehicle's monitor. This method improves the efficiency of data-driven fault detection, effectively filters fault causes, and provides significant assistance in troubleshooting and restoring the vehicle to a healthy state.

[0100] In summary, this embodiment provides an optional vehicle fault early warning method, such as... Figure 3 As shown, the method includes: acquiring vehicle data; cleaning and preprocessing the vehicle data; uploading relevant historical faults or other fault analysis data; drawing fault trees for historical faults; analyzing the data matrix of different fault phenomena based on deep learning algorithms, excluding data irrelevant to the current fault based on data links or fault analysis, and identifying useful data information; fault analysis; and visualization.

[0101] In the data preprocessing step, the collected data is sequentially cleaned, integrated, transformed, and feature selected and reduced in dimensionality to obtain complete, reliable, and high-quality data.

[0102] Furthermore, since the data extracted from the vehicle end often has many dimensions, but the proportion of effective data is actually very small, the burden of this invalid data can be significant during data analysis. Therefore, data cleaning is necessary before data analysis. This includes first identifying and deleting empty sets in the original dataset. Then, binary classification is performed on rows and columns with data, grouping similar labels or identical loops into one class. The data for each class of labels is then aggregated using a percentage system and compared against parameters to obtain the mode of the dataset for that class of labels. Offset values ​​are then discarded based on the mode, effectively removing deviations from the dataset. In particular, after removing deviations, the dataset needs to be further restored to preserve the original characteristics of the data.

[0103] To preserve the true values ​​of the original data, the original dataset will not be altered for null or off-target values. For datasets where not all features result in null or off-target values, data imputation is required after transformation to protect the validity and consistency of the data; for example, mode imputation can be selected.

[0104] Finally, feature selection and dimensionality reduction are performed. Since the original vehicle dataset contains a large number of features, only a small subset is effective for a specific fault. Therefore, some feature values ​​can be removed. For example, following historical fault troubleshooting methods, features with higher frequencies are prioritized for output, those with moderate or low frequencies are given secondary priority, and features with zero frequency are hidden. This effectively reduces the dimensionality of the data features, lowers the difficulty of data investigation, and prevents the loss of some feature values.

[0105] like Figure 4 As shown, after data preprocessing, a dataset with fewer feature dimensions and more valid values ​​will be obtained. Data preprocessing methods can include data cleaning, data integration, data transformation, feature selection, and feature dimensionality reduction.

[0106] In summary, during the data preprocessing stage, the system sequentially performs data cleaning, data integration, data transformation, feature selection, and feature dimensionality reduction operations on the raw data collected from the vehicle, aiming to build a complete, reliable, and high-quality dataset to provide a solid foundation for subsequent deep learning and analysis.

[0107] During the data cleaning phase, missing values ​​and outlier data are identified and processed.

[0108] Specifically, the system automatically detects and removes null records in the original data to avoid interference from invalid data. Then, non-null data undergoes binary classification based on label category or signal circuit, grouping data with similar semantic labels or belonging to the same electrical circuit into the same category. Within each category, the system uses a weighted cumulative percentage and multi-parameter comparison strategy to calculate the mode of the data for that category, and uses the mode to identify and remove outliers that deviate from the mode. In the data transformation stage, the system processes missing values ​​and identified outliers without altering the values ​​of the valid original data. Finally, the system performs feature selection and feature dimensionality reduction. This approach reduces data dimensionality without significant information loss, improving the efficiency of subsequent fault identification and reasoning. Through this systematic preprocessing, the final dataset exhibits low feature dimensionality and high effective information concentration.

[0109] Then, deep learning models can be used, based on deep neural network architectures (such as transformers or graph neural networks), to jointly model historical datasets and fault trees that have undergone cleaning, standardization, and feature enhancement. The system learns fault evolution patterns, solution decision paths, and complex nonlinear relationships between parameters through multi-layer attention mechanisms and temporal modeling capabilities.

[0110] When new fault data is input, the model can not only identify explicit fault characteristics, but also generate a variety of implicit causal hypotheses through latent space reasoning and Bayesian inference. It outputs a multi-level fault attribution analysis including primary causes, concurrent factors and indirect causes, and dynamically generates multiple troubleshooting suggestion sequences with probability assessment based on reinforcement learning strategies.

[0111] Finally, the visualization engine utilizes structured labels and multi-dimensional feature vectors to construct high-dimensional fault profiles through clustering analysis and dimensionality reduction techniques. Each fault profile can integrate multi-source heterogeneous data, including spatiotemporal attributes, signal characteristics, failure modes, maintenance history, and environmental factors, to form a unified and queryable dynamic map. Using real-time rendering technology and an interactive visualization interface, the system presents fault profiles in the form of 3D radar maps, heat map topology maps, and cross-dimensional parallel coordinates. It also features real-time drift analysis, correlation penetration, and hypothetical simulation functions. Users can dynamically filter, drill down, and backtrack through the profiles via touch, voice, or gestures.

[0112] Ultimately, the generated fault analysis view and decision suggestions can be pushed to the vehicle's smart cockpit display through a low-latency, high-reliability in-vehicle embedded system in the form of hierarchical alarms and auxiliary diagnostic interfaces. The information granularity and presentation can be adaptively adjusted according to the driving scenario, and it can work in conjunction with the vehicle control system to achieve closed-loop support for diagnosis, strategy, and execution, thereby comprehensively improving the accuracy of fault handling and system robustness.

[0113] In addition, this embodiment can also provide a data processing system. When an intelligent connected vehicle experiences an unknown or complex fault, the system can rely on the "end-edge-cloud" collaborative intelligent fault diagnosis architecture to build a full-link diagnostic system covering data perception, multimodal fusion analysis, dynamic reasoning and decision-making.

[0114] By introducing multi-level neural networks, knowledge graphs, and uncertainty reasoning techniques, the system can automatically identify fault modes in high-dimensional heterogeneous data, generate multivariate fault hypotheses with probabilistic interpretations, and realize the dynamic sorting and continuous evolution of diagnostic paths based on multi-objective optimization criteria, ultimately outputting an interpretable, executable, and verifiable accurate diagnostic solution.

[0115] Specifically, the system can achieve multimodal perception and high-dimensional data fusion. For example, it can collect sensor data, controller status, network communication messages, and environmental context information in real time through onboard edge computing nodes, and construct a spatiotemporally consistent vehicle state representation by leveraging cross-domain data alignment and semantic fusion technologies. For data streams of different modalities (such as continuous signals, discrete events, images, and text logs), the system adopts a partitioned attention mechanism and adaptive feature weighting to achieve deep fusion of multi-source information and noise suppression, providing high-quality input for subsequent diagnostic reasoning.

[0116] Secondly, the system can perform deep reasoning and fault hypothesis generation. For example, based on a pre-trained deep spatiotemporal model and vehicle knowledge graph, the system performs joint encoding and pattern recognition on real-time data streams to capture explicit features and implicit correlations of faults. The inference engine integrates data-driven and knowledge-guided methods, combining hidden Markov models, Bayesian networks, and domain expert rules to generate a multi-level set of fault hypotheses, including primary faults, concurrent causes, derivative anomalies, and inter-system coupling effects. Each hypothesis is accompanied by probability confidence, evidence support, uncertainty interval, and traceable diagnostic basis.

[0117] Furthermore, the system can implement multi-objective optimization and dynamic ranking strategies. For example, it incorporates multi-criteria decision theory to comprehensively analyze the statistical likelihood of hypotheses, historical case matching degree, feature significance, maintenance costs, safety risk weights, and real-time environmental context (such as vehicle operating conditions, geographical climate, and load status). Through fuzzy comprehensive evaluation, entropy weighting, and reinforcement learning strategies, it dynamically calculates the "comprehensive suspicion index" of each fault cause and generates a priority investigation sequence. This ranking supports real-time updates and can automatically adjust hypothesis priorities as new evidence is input, forming a progressively accurate diagnostic guidance.

[0118] Furthermore, maintenance personnel can receive diagnostic results through the in-vehicle human-machine interface, augmented reality-assisted diagnostic equipment, or mobile terminals. The system provides an interactive visual interface, presenting a ranked list of fault hypotheses, confidence distribution, correlation characteristic curves, and a 3D component influence domain simulation. Utilizing augmented reality technology, the system can overlay virtual guidance information onto the actual vehicle structure, identifying inspection points, disassembly / reassembly paths, and expected parameter ranges for highly suspected components, thereby improving first-time repair rates and operational efficiency.

[0119] In addition, the system constructs a continuous closed loop of diagnosis, verification, and learning. During maintenance, test results, human feedback, and expert corrections are transmitted back to the cloud-based diagnostic platform in real time, triggering incremental learning and model evolution. Based on federated learning and swarm intelligence frameworks, the system can integrate multi-vehicle collaborative information and cross-model fault knowledge to achieve joint reasoning and early warning of intermittent faults and system-level risks, gradually improving its ability to recognize and respond to complex faults.

[0120] Therefore, the system's data processing and transformation methods, as well as fault tree construction, provide a foundation for the effective detection and attribution of data faults. For example, the system constructs a multi-level, multi-granular data fault tree based on historical fault cases and domain knowledge. To achieve intuitive understanding and interactive analysis of complex data faults, the system introduces multimodal visualization technology, transforming high-dimensional, dynamic, and multi-source data into understandable graphical and tabular forms, realizing visualized data faults and multi-dimensional data analysis. Furthermore, since raw vehicle data typically contains a large amount of noise, missing values, and irrelevant information, the system performs data cleaning and intelligent filtering processes to reduce the interference of invalid data on the fault analysis process. This improves data quality and reduces the computational burden and risk of misleading data in fault attribution and pattern recognition.

[0121] This system can be applied to after-sales service systems for intelligent electric vehicles, large commercial fleet operation management platforms, and remote technical support centers, reducing average fault handling time, labor costs, and misjudgment rates. With the continuous development of digital twin and edge computing capabilities, the system can evolve towards predictive maintenance, fault self-healing, and full lifecycle health management, ultimately forming a vehicle-cloud-human collaborative intelligent fault management ecosystem, providing support for high availability and autonomous operation and maintenance of vehicles in the era of autonomous driving.

[0122] According to an embodiment of the present invention, a device embodiment for a vehicle fault warning device is provided. It should be noted that the device can be used to execute the above-described vehicle fault warning method. The specific implementation scheme and application scenario in this embodiment are the same as those in the above embodiment, and will not be repeated here.

[0123] Figure 5 This is a schematic diagram of a vehicle malfunction warning device according to an embodiment of this application, such as... Figure 5 As shown, the device includes the following:

[0124] The acquisition module 52 is used to acquire multimodal data of the vehicle, wherein the multimodal data includes at least sensor data, message data, and log data.

[0125] The detection module 54 is used to input multimodal data into a deep learning model, use the deep learning model to identify vehicle faults, and obtain at least one fault identification result. The deep learning model is trained based on historical multimodal data and a fault tree. Different nodes in the fault tree represent different fault data corresponding to fault causes. The fault identification result is used to reflect the fault causes of vehicle faults.

[0126] The early warning module 56 is used to generate early warning information based on at least one fault identification result.

[0127] Optionally, the detection module is also used to: preprocess the multimodal data to obtain processed multimodal data; input the processed multimodal data into a deep learning model, and use the deep learning model to output at least one fault identification result.

[0128] Optionally, the detection module is also used to: classify multimodal data based on the vehicle's electrical circuits to obtain multiple datasets, wherein different datasets represent data from different electrical circuits; and for the operational data in each dataset, delete operational data that deviates from the mode to obtain processed multimodal data, wherein different operational data correspond to different modes.

[0129] Optionally, the deep learning model includes an anomaly detection module and an inference module; the detection module is also used to: use the anomaly detection module to perform anomaly detection on the processed multimodal data to obtain fault data; and use the fault tree-guided inference module to infer the fault causes corresponding to the fault data to obtain at least one fault detection result.

[0130] Optionally, at least one fault identification result may also include the confidence level of the fault cause; the early warning module is further configured to: determine the security level corresponding to at least one fault identification result based on the confidence level contained in at least one fault identification result; and generate early warning information based on the security level corresponding to at least one fault identification result.

[0131] Optionally, the above device also includes a push module for: pushing the warning information to the vehicle's alert module and / or the maintenance equipment of the target object.

[0132] This application also provides an electronic device 90, please refer to... Figure 6 It includes a memory 910 and a processor 920, wherein the memory 910 is used to store computer programs; and the processor 920 is used to execute the programs stored in the memory 910 to implement the methods in the various embodiments of this application.

[0133] Embodiments of this application also provide a computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of the present invention.

[0134] Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the methods of various embodiments of the present invention.

[0135] Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium for storing a computer program that, when executed by a processor, implements the methods in various embodiments of the present invention.

[0136] Embodiments of this application also provide a computer program that, when executed by a processor, implements the methods described in the various embodiments of the present invention.

[0137] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0139] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0140] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0141] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0142] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for early warning of vehicle malfunctions, characterized in that, include: Acquire multimodal data of the vehicle, wherein the multimodal data includes at least sensor data, message data, and log data; The multimodal data is input into a deep learning model, and the deep learning model is used to identify faults in the vehicle to obtain at least one fault identification result. The deep learning model is trained based on historical multimodal data and a fault tree. Different nodes in the fault tree represent different fault data corresponding to fault causes. The fault identification result is used to reflect the fault causes of the vehicle's fault. Based on the at least one fault identification result, an early warning message is generated.

2. The method according to claim 1, characterized in that, The multimodal data is input into a deep learning model, and the deep learning model is used to identify faults in the vehicle, obtaining at least one fault identification result, including: The multimodal data is preprocessed to obtain processed multimodal data; The processed multimodal data is input into the deep learning model, and the deep learning model is used to output the at least one fault identification result.

3. The method according to claim 2, characterized in that, The multimodal data is preprocessed to obtain processed multimodal data, including: Based on the vehicle's electrical circuits, the multimodal data is classified to obtain multiple datasets, where different datasets represent data from different electrical circuits; For each dataset, the running data that deviates from the mode is removed, resulting in processed multimodal data, where different running data correspond to different modes.

4. The method according to claim 2, characterized in that, The deep learning model includes an anomaly detection module and an inference module; the processed multimodal data is input into the deep learning model, and the deep learning model outputs the at least one fault detection result, including: The anomaly identification module is used to identify anomalies in the processed multimodal data to obtain fault data. The fault tree is used to guide the reasoning module to reason about the fault causes corresponding to the fault data, thereby obtaining at least one fault identification result.

5. The method according to claim 1, characterized in that, The at least one fault identification result also includes the confidence level of the fault cause; Based on the at least one fault identification result, an early warning message is generated, including: Based on the confidence level contained in the at least one fault identification result, the security level corresponding to the at least one fault identification result is determined; The warning information is generated based on the security level corresponding to the at least one fault identification result.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: The warning information is pushed to the vehicle's alert module and / or the target object's maintenance equipment.

7. A vehicle malfunction early warning device, characterized in that, include: The acquisition module is used to acquire multimodal data of the vehicle, wherein the multimodal data includes at least sensor data, message data, and log data; The detection module is used to input the multimodal data into a deep learning model, use the deep learning model to identify faults in the vehicle, and obtain at least one fault identification result. The deep learning model is trained based on historical multimodal data and a fault tree. Different nodes in the fault tree represent different fault data corresponding to fault causes. The fault identification result is used to reflect the fault causes of the vehicle's faults. The early warning module is used to generate early warning information based on the at least one fault identification result.

8. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the storage medium is located to perform the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images