An Artificial Intelligence-Based Method for Vehicle Fault Diagnosis and Maintenance Assessment

Abstract

This invention discloses an artificial intelligence-based method for vehicle fault diagnosis and maintenance assessment. The method includes deploying an onboard sensor network to collect multiple heterogeneous data sets and processing the heterogeneous data; constructing a fault detection model based on the processed heterogeneous data; making fault judgments based on the results of the fault detection model; and constructing a vehicle knowledge graph to generate maintenance assessment schemes based on the results of the fault detection model. This invention achieves comprehensive collection of multi-source heterogeneous data by deploying an advanced onboard sensor network and establishing a comprehensive historical maintenance database, providing a rich and reliable data foundation for subsequent analysis. Furthermore, by employing edge computing technology and a multi-level adaptive fusion strategy, combined with improved Dempster-Shafer evidence theory and Kalman filters, it achieves efficient real-time processing and fusion of heterogeneous data, significantly improving the accuracy and real-time performance of data processing.

Description

Technical Field

[0001] This invention relates to the field of vehicle intelligent diagnosis and maintenance technology, and in particular to a vehicle fault diagnosis and maintenance assessment method based on artificial intelligence. Background Technology

[0002] With the rapid development of the automotive industry and the continuous improvement of its intelligence, vehicle fault diagnosis and maintenance assessment technologies have evolved from traditional manual experience-based judgment to data-driven intelligent diagnosis. Early fault diagnosis relied primarily on the experience of mechanical engineers and simple diagnostic instruments, a method that was often inefficient and had limited accuracy when dealing with complex faults. Subsequently, expert system-based fault diagnosis methods were widely used, simulating the diagnostic process of human experts by encoding expert knowledge into a rule base. However, this method's adaptability and scalability were severely limited when facing novel faults or complex systems. In recent years, with the rapid development of artificial intelligence technology, machine learning-based fault diagnosis methods have emerged. These methods can learn fault patterns from large amounts of historical data, improving the accuracy and efficiency of diagnosis.

[0003] Despite these shortcomings, existing vehicle fault diagnosis and maintenance assessment technologies still have some significant limitations. First, most systems fail to fully utilize the diverse and heterogeneous data generated during vehicle operation, affecting the comprehensiveness and accuracy of diagnostic results. Second, traditional fault detection models are often static, lacking the ability to adapt to the dynamic operating state of vehicles in real time, making it difficult to cope with complex and changing driving environments and usage conditions. Third, existing maintenance assessment schemes are usually based on pre-set rules or simple statistical models, failing to fully consider individual vehicle differences and historical maintenance records, resulting in insufficient targeting and effectiveness of maintenance recommendations. Furthermore, most systems lack structured representation and utilization of knowledge, failing to effectively integrate and apply the experiential knowledge of domain experts, thus limiting the system's reasoning ability and interpretability. Summary of the Invention

[0004] In view of the problems existing in the above-mentioned vehicle fault diagnosis and maintenance assessment technologies, this invention is proposed.

[0005] Therefore, the problem to be solved by this invention is how to provide a vehicle fault diagnosis and maintenance assessment technology based on artificial intelligence.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, embodiments of the present invention provide a vehicle fault diagnosis and maintenance assessment method based on artificial intelligence, comprising,

[0008] Deploy an onboard sensor network to collect multiple heterogeneous data sets and process the heterogeneous data.

[0009] Based on the processed heterogeneous data, a fault detection model is constructed, and fault judgment is made based on the results of the fault detection model.

[0010] A vehicle knowledge graph is constructed, and a maintenance assessment plan is generated based on the results of the fault detection model.

[0011] As a preferred embodiment of the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence described in this invention, the heterogeneous data is processed using multi-source heterogeneous data based on the evidence theory framework and filters.

[0012] The theoretical framework of the evidence is shown in the following equation:

[0013] ;

[0014] in, and , respectively, are the weights of the first and second data sources at time t, m1 and m2 are the confidence functions of the two independent evidence sources for the hypothesis, A is the set of possible hypotheses, and B and C are subsets in the basic probability assignment, representing different hypotheses about the proposition.

[0015] As a preferred embodiment of the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence described in this invention, the filter fuses rapidly changing sensor data as follows:

[0016] ;

[0017] in, Let k be the predicted value of the system state at time k. The estimated value of the system state at time k-1. For dynamic weights, Let be the estimated value of the system state at time k-1. Let be the error covariance matrix. Let be the estimated value of the covariance matrix of the state estimation error at time k−1. For Kalman gain, Let k be the observation value at time k. The mapping of the predicted value at time k in the measurement space, where A, B, H, Q, R, and I are the state transition matrix, control input matrix, observation matrix, process noise covariance matrix, observation noise covariance matrix, and identity matrix, respectively. The state update results are for the improved Kalman filter. This is the updated error covariance matrix.

[0018] As a preferred embodiment of the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence described in this invention, the processing of the heterogeneous data further includes:

[0019] Based on historical data and current environmental conditions, the initial weights of each data source are calculated using a dynamic weighting formula. A filter is then applied to each data source individually to obtain preliminary state predictions and update results. Based on the preliminary filter results, the weights of each data source are adjusted to obtain dynamic weights. Evidence is combined using these dynamic weights. The improved evidence combination results are then used to further optimize the state transition matrix, control input matrix, observation matrix, process noise covariance matrix, and observation noise covariance matrix of the data sources. Finally, the fusion result is obtained by combining the optimized data source states and weights from multiple optimizations.

[0020] As a preferred embodiment of the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence described in this invention, the fault detection model is a two-stream neural network architecture;

[0021] First-class neural networks process spatiotemporally relevant sensor data, while second-class neural networks process highly sequential data.

[0022] The outputs of the first-stream neural network and the second-stream neural network are dynamically fused through an attention mechanism.

[0023] As a preferred embodiment of the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence described in this invention, the fault judgment includes:

[0024] If the predicted probability of the first fault is greater than the high probability threshold and the uncertainty is less than the low uncertainty threshold LU, report the fault directly and recommend immediate repair, providing a detailed fault description and possible impact; if the uncertainty is greater than or equal to the low uncertainty threshold but less than the medium uncertainty threshold, report the fault and recommend repair as soon as possible; if a second fault exists and its probability is greater than the high probability threshold, and the probability of the first fault minus the probability of the second fault is less than the probability difference threshold, then detect the second fault; if the probability of the first fault minus the probability of the second fault is greater than the probability difference threshold, then the second fault cannot be determined.

[0025] If there are no faults with a probability greater than the high probability threshold, but there are multiple faults with a probability greater than the medium probability threshold, or if multiple faults with a probability greater than the medium probability threshold are less than the probability difference threshold or the uncertainty is greater than the medium uncertainty threshold, report multiple fault types. If the uncertainty is greater than the high uncertainty threshold, warn that multiple faults exist. Otherwise, provide a detailed description of the fault type and check them one by one according to their probability.

[0026] As a preferred embodiment of the artificial intelligence-based vehicle fault diagnosis and maintenance assessment method of the present invention, wherein: the vehicle knowledge graph includes,

[0027] When a specific fault symptom is input, the message is propagated in the vehicle knowledge graph. An attention mechanism is used to dynamically adjust the importance of different nodes and edges, and then multi-hop reasoning is used to locate the fault cause with the highest probability.

[0028] After the cause of the fault is located, the case reasoning engine searches for similar cases. If a similar case is found, it provides a reference for the current repair decision. If no similar case is found, a repair plan is generated based on various factors of the vehicle.

[0029] Secondly, to further address the safety issues existing in artificial intelligence-based vehicle fault diagnosis and maintenance assessment methods, the present invention provides an artificial intelligence-based vehicle fault diagnosis and maintenance assessment system, comprising:

[0030] The data processing module is used to deploy an on-board sensor network, collect multiple heterogeneous data, and process the heterogeneous data;

[0031] The fault diagnosis module is used to construct a fault detection model based on the processed heterogeneous data, and to diagnose faults based on the results of the fault detection model.

[0032] The maintenance assessment module is used to construct a vehicle knowledge graph and generate a maintenance assessment plan based on the results of the fault detection model.

[0033] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the artificial intelligence-based vehicle fault diagnosis and maintenance assessment method as described in the first aspect of the present invention.

[0034] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence as described in the first aspect of the present invention.

[0035] The beneficial effects of this invention are as follows: First, by deploying an advanced vehicle-mounted sensor network and establishing a comprehensive historical maintenance database, it achieves comprehensive collection of multi-source heterogeneous data, providing a rich and reliable data foundation for subsequent analysis. Second, by employing edge computing technology and a multi-level adaptive fusion strategy, combined with an improved Dempster-Shafer evidence theory and Kalman filter, it achieves efficient real-time processing and fusion of heterogeneous data, significantly improving the accuracy and real-time performance of data processing. Third, by designing an innovative dual-stream neural network architecture, combined with 3D-CNN and LSTM, it achieves effective processing of spatiotemporally related and sequential data, greatly improving the accuracy of fault diagnosis and prediction. Furthermore, by constructing a complex vehicle maintenance knowledge graph and combining it with graph neural network technology, it achieves accurate fault cause localization and personalized maintenance plan generation. Finally, by introducing meta-learning, reinforcement learning, and case-based reasoning techniques, it further enhances the system's adaptability, decision-making ability, and interpretability. This comprehensive approach not only significantly improves the accuracy, efficiency, and personalization of vehicle fault diagnosis and maintenance assessment, but also provides strong support for preventive maintenance and full lifecycle management of vehicles, possessing significant practical application value. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0037] Figure 1 This is a flowchart of an AI-based vehicle fault diagnosis and maintenance assessment method. Detailed Implementation

[0038] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0039] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0040] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0041] Example 1:

[0042] Reference Figure 1 This is the first embodiment of the present invention, which provides a vehicle fault diagnosis and maintenance assessment method based on artificial intelligence, including the following steps:

[0043] First, an advanced onboard sensor network is designed and deployed, including various types of sensors such as engine temperature sensors, vibration sensors, and tire pressure monitors. These sensors achieve real-time data acquisition through a carefully designed low-power, high-reliability network architecture. Simultaneously, a comprehensive vehicle history maintenance database is established, integrating maintenance records from vehicle manufacturers, repair shops, and users, and a standardized data format is designed to ensure information consistency and availability.

[0044] To capture the details of driving behavior, a specialized data acquisition system was developed, including in-vehicle behavior monitoring devices and mobile applications, to record operations such as acceleration, braking, and steering, as well as information such as driving mileage, frequency, and road conditions. To enable real-time data processing and preliminary analysis, an edge computing platform was deployed in the in-vehicle system to run data preprocessing algorithms for noise filtering and outlier detection.

[0045] Furthermore, in the process of real-time processing and fusion of heterogeneous data using edge computing technology, a multi-layered, adaptive fusion strategy is adopted. First, the system preprocesses inputs from multiple heterogeneous data sources, such as onboard sensor networks, vehicle historical maintenance records, and driving behavior data, including data cleaning, normalization, and time synchronization.

[0046] Subsequently, an improved Dempster-Shafer evidence theory framework is applied to address data uncertainty and conflicts, where each data source is assigned a dynamic weight that is adjusted in real time based on the data's historical reliability and current environmental conditions. Building upon this, a Kalman filter is introduced to optimize the prediction and updating of time-series data, particularly for rapidly changing sensor data.

[0047] The performance of edge computing nodes is taken into consideration, and the allocation of computing resources is evaluated and optimized using a kernel function K(x, y). Furthermore, information entropy H(X) is introduced to quantify data uncertainty, and a hyperbolic tangent function is used to smooth the impact of data processing speed. This comprehensive approach not only effectively fuses heterogeneous data from different sources but also dynamically adjusts the fusion strategy based on real-time conditions, ultimately forming a highly accurate and real-time updated vehicle "digital twin" model. The entire process is iterative; the system continuously learns and optimizes the fusion algorithm to adapt to different vehicle types, usage environments, and data characteristics, thereby providing the most accurate vehicle state representation.

[0048] For example, in a feasible implementation, the improved Dempster-Shafer evidence theory framework is as follows:

[0049] Assuming the weights of each data source are dynamically adjusted based on historical reliability and current environmental conditions, a weight function wi(t) is introduced, representing the weight of the i-th data source at time t. A Kalman filter is used to optimize the time-series data. Specifically, the combination rules are as follows:

[0050] ;

[0051] in, and , respectively, are the weights of the first and second data sources at time t, m1 and m2 are the confidence functions of the two independent evidence sources for the hypothesis, A is the set of possible hypotheses, and B and C are subsets in the basic probability assignment, representing different hypotheses about the proposition.

[0052] Specifically, the weights are dynamically adjusted as shown in the following formula:

[0053] ;

[0054] in, Indicates the historical reliability of the i-th data source. This represents the impact of the current environmental conditions on i data sources. It's about adjusting the parameters.

[0055] Furthermore, combining the prediction and update steps of the Kalman filter, the improved fusion strategy for rapidly changing sensor data is as follows:

[0056] ;

[0057] in, Let k be the predicted value of the system state at time k. The estimated value of the system state at time k-1. For dynamic weights, Let be the estimated value of the system state at time k-1. Let be the error covariance matrix. Let be the estimated value of the covariance matrix of the state estimation error at time k−1. For Kalman gain, Let k be the observation value at time k. The mapping of the predicted value at time k in the measurement space, where A, B, H, Q, R, and I are the state transition matrix, control input matrix, observation matrix, process noise covariance matrix, observation noise covariance matrix, and identity matrix, respectively. The state update results are for the improved Kalman filter. This is the updated error covariance matrix.

[0058] Specifically, based on historical data and current environmental conditions, preliminary weights for each data source are calculated using a dynamically adjusted weighting formula. A Kalman filter is then applied individually to each data source to obtain preliminary state predictions and updates. Based on these preliminary Kalman filter results, the weights for each data source are adjusted to obtain dynamic weights. These dynamic weights are then used for evidence combination. The improved evidence combination results are then used to further optimize the state transition matrix, control input matrix, observation matrix, process noise covariance matrix, and observation noise covariance matrix of the data sources. Combining the optimized data source states and weights from these multiple optimizations yields the final fusion result. Through this multi-layered nested interaction logic, complex fusion and optimization processing of heterogeneous data sources can be achieved, ensuring the accuracy and real-time performance of data processing.

[0059] Furthermore, by fully leveraging the heterogeneous data sources fused and optimized in the previous stage, a highly accurate and adaptive deep learning model was constructed. First, the system receives heterogeneous data that has been processed and fused in real time using edge computing technology. This data includes real-time data from the vehicle sensor network, long-term trend information from vehicle historical maintenance records, and driving behavior data. The data preprocessing stage further standardizes and times-aligns this fused data, ensuring that data from different sources can be seamlessly input into the deep learning model.

[0060] Subsequently, an innovative dual-stream neural network architecture was designed. One stream uses a 3D convolutional neural network (3D-CNN) to process spatiotemporally relevant sensor data, such as engine vibration spectrum and temperature distribution, to extract high-level spatial-temporal features. The other stream uses a long short-term memory network (LSTM) to process highly sequential data, such as historical maintenance records and driving behavior data, to capture long-term dependencies.

[0061] The outputs of the two streams are dynamically fused through an attention mechanism, enabling the model to adaptively focus on key information from different data sources. To handle potential uncertainties in the fused data, Bayesian deep learning techniques are introduced, using variational inference to estimate the posterior distribution of the model parameters, thereby providing reliable uncertainty estimates.

[0062] The model also integrates a multi-task learning framework, simultaneously performing current fault diagnosis and future fault risk prediction. Through residual connections and gating mechanisms, the model can effectively utilize information at different time scales, thereby improving the accuracy and robustness of predictions. Finally, the model's output passes through a post-processing module based on a graph neural network (GNN). This module utilizes the relationship graph between various system components of the vehicle to further optimize the fault diagnosis results, providing more interpretable fault localization and propagation analysis. This deep learning approach, which comprehensively integrates heterogeneous data sources, not only significantly improves the accuracy of fault identification and prediction but also significantly enhances the system's adaptability to complex and dynamic vehicle environments, laying a solid foundation for achieving intelligent and personalized preventative maintenance.

[0063] For example, in a feasible implementation, the two-stream neural network architecture is as follows:

[0064] The 3D-CNN stream processes spatiotemporally relevant sensor data. 3D-CNN can capture spatial and temporal patterns in the data. For example, it might identify a specific vibration pattern that frequently appears before a vehicle is about to fail. The LSTM stream processes highly sequential data. LSTM can capture long-term dependencies. For example, it might detect a trend of gradually shortening intervals between the three most recent maintenance visits, which could indicate a decline in vehicle performance. Specifically, this can be expressed using the following formula:

[0065] ;

[0066] The specific expression of the two-stream neural network architecture is as follows:

[0067] ;

[0068] in, F 3D-CNN The output features of a 3D convolutional neural network F LSTM Here, represents the output features of the Long Short-Term Memory network, Attention is the attention mechanism function, MLP is a multilayer perceptron, α is the attention weight coefficient, G is the non-linear activation function, and H is the Kullback-Leibler divergence. q φ(w) Let be the variational posterior distribution of the model parameters w, p(w) be the prior distribution of the model parameters w, λ be the entropy regularization coefficient, S be the entropy function, ReLU be the modified linear unit activation function, and σ be the sigmoid activation function. W 3D 3D convolution kernel, b 3D For 3D convolution bias. W f Here is the weight matrix of the LSTM.b f This is the bias vector of the LSTM. x spatial-temporal For spatiotemporally relevant sensor data, x sequential This is data with strong sequential characteristics.

[0069] Specifically, the attention mechanism function is shown in the following equation:

[0070] ;

[0071] in, W Q , W K , W V This is the weight matrix in the attention mechanism. d k This is the scaling factor for the attention mechanism.

[0072] ;

[0073] ;

[0074] in, F 3D-CNN The output features of a 3D convolutional neural network F LSTM This represents the output characteristics of the Long Short-Term Memory network.

[0075] Furthermore, define the key parameters:

[0076] High probability threshold (HP), medium probability threshold (MP), low probability threshold (LP), high uncertainty threshold (HU), medium uncertainty threshold (MU), low uncertainty threshold (LU), probability difference threshold (PD).

[0077] If the predicted probability of the first fault is greater than the high probability threshold and the uncertainty is less than the low uncertainty threshold LU, report the fault directly and recommend immediate repair, providing a detailed fault description and possible impact. If the uncertainty is greater than or equal to the low uncertainty threshold but less than the medium uncertainty threshold, report the fault and recommend repair as soon as possible. If a second fault exists and its probability is greater than the high probability threshold, and the difference between the probability of the first fault and the probability of the second fault is less than the probability difference threshold, then detect the second fault. If the difference between the probability of the first fault and the probability of the second fault is greater than the probability difference threshold, then the second fault cannot be determined.

[0078] If there are no faults with a probability greater than the high probability threshold, but there are multiple faults with a probability greater than the medium probability threshold, or if multiple faults with a probability greater than the medium probability threshold are less than the probability difference threshold or the uncertainty is greater than the medium uncertainty threshold, report multiple fault types. If the uncertainty is greater than the high uncertainty threshold, warn that multiple faults exist. Otherwise, provide a detailed description of the fault type and check them one by one according to their probability.

[0079] If the probability of a fault is less than the medium probability threshold and the probability of a fault is greater than the low probability threshold, then the vehicle has a minor fault. If the fault is safety-related (such as brakes), then the problem is reported.

[0080] Furthermore, a vast and complex vehicle maintenance knowledge graph was constructed. This graph not only includes the hierarchical structure of the vehicle system, the interrelationships between components, common fault symptoms and their possible causes, but also integrates a large amount of historical maintenance records and expert experience. This knowledge graph uses ontology modeling and natural language processing techniques to transform unstructured maintenance manuals, technical documents, and expert opinions into structured knowledge representations. When a specific fault symptom is input, message propagation occurs on the graph, using an attention mechanism to dynamically adjust the importance of different nodes and edges, and then multi-hop reasoning is used to locate possible fault causes.

[0081] To improve the accuracy of inference, a meta-learning module is introduced, enabling the model to quickly adapt to different vehicle models and newly emerging fault modes. After locating the cause of the fault, the system activates a reinforcement learning-based decision-making module. This module considers multiple factors such as the vehicle's usage history, current state, repair costs, and future usage plans to generate a personalized repair plan. This plan not only includes specific repair steps but also provides priority suggestions and expected repair results.

[0082] To further enhance the system's interpretability and credibility, a case-based reasoning engine was integrated. This engine retrieves similar cases from historical repair records, providing reference and evidence for current repair decisions. Finally, the system also features an interactive interface that allows human experts to review and adjust AI-generated repair plans, enabling intelligent decision-making through human-machine collaboration. This approach, combining knowledge graphs and deep learning, not only significantly improves the accuracy of fault diagnosis and the personalization of repair plans but also substantially enhances the efficiency and reliability of the repair process, providing robust support for preventative vehicle maintenance and full lifecycle management.

[0083] This embodiment also provides an artificial intelligence-based vehicle fault diagnosis and maintenance assessment system, including:

[0084] The data processing module is used to deploy an on-board sensor network, collect multiple heterogeneous data, and process the heterogeneous data;

[0085] The fault diagnosis module is used to construct a fault detection model based on the processed heterogeneous data, and to diagnose faults based on the results of the fault detection model.

[0086] The maintenance assessment module is used to construct a vehicle knowledge graph and generate a maintenance assessment plan based on the results of the fault detection model.

[0087] This embodiment also provides a computer device applicable to the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence as proposed in the above embodiment.

[0088] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0089] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0090] Example 2:

[0091] As a second embodiment of the present invention, experimental data of an artificial intelligence-based vehicle fault diagnosis and maintenance assessment method are provided to further verify the beneficial effects of the present invention.

[0092] To verify the effectiveness of this invention, 100 vehicles of different models, years of use, and mileage were selected as test subjects. First, an advanced onboard sensor network containing 20 sensors was installed on each vehicle, including engine temperature sensors, vibration sensors, tire pressure monitors, and oil pressure sensors. These sensors were connected to the onboard edge computing unit via Bluetooth Low Energy 5.0 technology to achieve real-time data acquisition and preliminary processing. Simultaneously, a database containing the maintenance records of these 100 vehicles over the past five years was established, including more than 4,500 standardized maintenance records.

[0093] To capture driving behavior, a dedicated mobile application was developed and installed on the driver's smartphone to record parameters such as acceleration, braking force, and steering angle.

[0094] In addition, the vehicle's GPS module records driving routes and road conditions, and all data is sampled at a frequency of 5Hz and transmitted to the cloud server in real time via a 4G network.

[0095] In terms of data processing, the raw data is first preprocessed, including outlier removal (using the 3σ rule), missing value imputation (using forward imputation), and standardization (using Z-score standardization). Then, an improved Dempster-Shafer evidence theory framework is applied to fuse multi-source data. Specifically, the historical reliability of each data source is calculated based on its accuracy over the past 30 days, and the weights are dynamically adjusted according to current environmental conditions (such as weather and road conditions). For example, the weight of the tire pressure sensor is reduced in rainy weather.

[0096] Next, a Kalman filter is used to process the rapidly changing sensor data. The state transition matrix A and the observation matrix H are obtained by analyzing historical data, while the process noise covariance matrix Q and the observation noise covariance matrix R are determined based on sensor specifications and actual test results. Through iterative optimization, a stable state estimate is finally obtained.

[0097] In terms of deep learning model construction, a dual-stream neural network architecture was adopted. The 3D-CNN stream used four 3D convolutional layers, each followed by BatchNormalization and ReLU activation functions, and finally a fully connected layer.

[0098] The LSTM stream consists of two bidirectional LSTM layers, each with 128 units, followed by an attention layer. The outputs of the two streams are fused through a multilayer perceptron (MLP) containing three hidden layers with 256, 128, and 64 neurons, respectively.

[0099] To evaluate the model's performance, 100 vehicles were randomly divided into a training set (80 vehicles) and a test set (20 vehicles) for a six-month experiment.

[0100] During this period, all actual failures were recorded and compared with the model's predictions. Simultaneously, the results of traditional rule-based diagnostic methods and machine learning methods based on a single data source were recorded as a comparison, as follows:

[0101] Table 1 Experimental Data

[0102] Vehicle number Actual fault Model prediction Rule-based prediction Single-source machine learning prediction Prediction accuracy Reduced maintenance costs Reduced downtime V001 3 3 2 2 100% 25% 30% V002 2 2 1 2 100% 20% 25% V003 4 4 3 3 100% 30% 35% V004 1 1 1 0 100% 15% 20% V005 3 3 2 2 100% 22% 28% V006 2 2 1 1 100% 18% 23%

[0103] By analyzing the data in the table above, it is clear that the AI-based vehicle fault diagnosis and maintenance assessment method proposed in this invention has significant advantages over traditional methods:

[0104] First, the model of this invention demonstrates superior performance in terms of fault prediction accuracy. For all six test vehicles, the model accurately predicted the actual number of faults, achieving 100% prediction accuracy. In contrast, traditional rule-based methods and machine learning methods based on a single data source exhibit varying degrees of prediction bias in most cases.

[0105] For example, for vehicle V003, four faults actually occurred, while the traditional method and the single data source method only predicted three, showing a clear deficiency.

[0106] Secondly, this invention excels in reducing maintenance costs. Data shows that using models for fault diagnosis and maintenance assessment can reduce maintenance costs by an average of 15% to 30%. This significant cost saving is mainly due to the model's accurate predictive capabilities, which allows maintenance work to be carried out more effectively, avoiding unnecessary inspections and repairs. For example, for vehicle V003, the maintenance cost reduction reached 30%, which is a considerable figure in industrial applications.

[0107] Furthermore, this invention also demonstrates significant advantages in reducing vehicle downtime: data shows that using the model can reduce downtime by an average of 20% to 35%. This improvement is of great significance for increasing vehicle utilization efficiency and reducing operational losses. In particular, for vehicles V003 and V005, downtime is reduced by 35% and 28% respectively, which can bring significant economic benefits to commercial fleets or public transportation systems.

[0108] Furthermore, by comparing data from different vehicles, it can be found that the model of the present invention exhibits stable high performance when dealing with different fault complexities. Whether it is vehicle V004 with only one fault or vehicle V003 with four faults, the model can make accurate predictions, which demonstrates the robustness and adaptability of the method.

[0109] In summary, the AI-based vehicle fault diagnosis and maintenance assessment method proposed in this invention has demonstrated significant technical advantages and economic benefits in practical applications. It not only greatly improves the accuracy of fault diagnosis but also reduces maintenance costs and vehicle downtime through precise maintenance assessment. This innovative method provides a new solution for vehicle preventive maintenance and intelligent management, and is expected to have a wide and far-reaching impact on the automotive industry and transportation sector.

[0110] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A vehicle fault diagnosis and maintenance assessment method based on artificial intelligence, characterized in that: include: Deploy an onboard sensor network to collect multiple heterogeneous data sets and process the heterogeneous data. Based on the processed heterogeneous data, a fault detection model is constructed on the vehicle edge computing platform, and fault judgment is made based on the results of the fault detection model. Construct a vehicle knowledge graph and generate a maintenance assessment plan based on the results of the fault detection model; The heterogeneous data is processed using a multi-source heterogeneous data processing method based on the evidence theory framework and filters. The theoretical framework of the evidence is shown in the following equation: ； in, and , respectively, are the weights of the first and second data sources at time t, m1 and m2 are the confidence functions of the two independent evidence sources for the hypothesis, A is the set of possible hypotheses, and B and C are subsets in the basic probability assignment, representing different hypotheses about the proposition; The filter fuses rapidly changing sensor data as follows: ； in, Let k be the predicted value of the system state at time k. The estimated value of the system state at time k-1. For dynamic weights, Let be the estimated value of the system state at time k-1. Let be the error covariance matrix. Let be the estimated value of the covariance matrix of the state estimation error at time k−1. For Kalman gain, Let k be the observation value at time k. The mapping of the predicted value at time k in the measurement space, where A, B, H, Q, R, and I are the state transition matrix, control input matrix, observation matrix, process noise covariance matrix, observation noise covariance matrix, and identity matrix, respectively. The state update results are for the improved Kalman filter. The updated error covariance matrix is ​​used for processing the heterogeneous data. This process further includes calculating the initial weights of each data source based on historical data and current environmental conditions using a dynamic weighting formula, applying filters to each data source individually to obtain preliminary state predictions and update results, adjusting the weights of each data source based on the preliminary filter results to obtain dynamic weights, combining evidence using the dynamic weights, and further optimizing the state transition matrix, control input matrix, observation matrix, process noise covariance matrix, and observation noise covariance matrix of the data sources using the improved evidence combination results. Finally, the fusion result is obtained by combining the optimized data source states and weights from multiple optimizations. The fault detection model is a two-stream neural network architecture; the first-stream neural network processes spatiotemporally correlated sensor data, and the second-stream neural network processes data with strong sequentiality; the outputs of the first-stream neural network and the second-stream neural network are dynamically fused through an attention mechanism.

2. The vehicle fault diagnosis and maintenance assessment method based on artificial intelligence as described in claim 1, characterized in that: The fault diagnosis includes, If the predicted probability of the first fault is greater than the high probability threshold and the uncertainty is less than the low uncertainty threshold LU, report the fault directly and recommend immediate repair, providing a detailed fault description and possible impact; if the uncertainty is greater than or equal to the low uncertainty threshold but less than the medium uncertainty threshold, report the fault and recommend repair as soon as possible; if a second fault exists and its probability is greater than the high probability threshold, and the probability of the first fault minus the probability of the second fault is less than the probability difference threshold, then detect the second fault; if the probability of the first fault minus the probability of the second fault is greater than the probability difference threshold, then the second fault cannot be determined. If there are no faults with a probability greater than the high probability threshold, but there are multiple faults with a probability greater than the medium probability threshold, or if multiple faults with a probability greater than the medium probability threshold are less than the probability difference threshold or the uncertainty is greater than the medium uncertainty threshold, report multiple fault types. If the uncertainty is greater than the high uncertainty threshold, warn that multiple faults exist. Otherwise, provide a detailed description of the fault type and check them one by one according to their probability.

3. The vehicle fault diagnosis and maintenance assessment method based on artificial intelligence as described in claim 2, characterized in that: The vehicle knowledge graph includes... When a specific fault symptom is input, the message is propagated in the vehicle knowledge graph. An attention mechanism is used to dynamically adjust the importance of different nodes and edges, and then multi-hop reasoning is used to locate the fault cause with the highest probability. After the cause of the fault is located, the case reasoning engine searches for similar cases. If a similar case is found, it provides a reference for the current repair decision. If no similar case is found, a repair plan is generated based on various factors of the vehicle.

4. A vehicle fault diagnosis and maintenance assessment system based on artificial intelligence, based on the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence as described in any one of claims 1 to 3, characterized in that: include: The data processing module is used to deploy an on-board sensor network, collect multiple heterogeneous data, and process the heterogeneous data; The fault diagnosis module is used to construct a fault detection model based on the processed heterogeneous data, and to diagnose faults based on the results of the fault detection model. The maintenance assessment module is used to construct a vehicle knowledge graph and generate a maintenance assessment plan based on the results of the fault detection model.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence as described in any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the vehicle fault diagnosis and maintenance assessment method based on artificial intelligence as described in any one of claims 1 to 3.

Images