Sensor data fused automobile fault detection method and system

Abstract

The invention relates to the technical field of fault detection control, and discloses an automobile fault detection method and system fusing sensor data. The method comprises the following steps: modeling a multi-system sensor network through an automobile system topology mapping algorithm to obtain a sensor topology incidence matrix; performing synchronous processing on the sensor topological incidence matrix according to a vehicle dynamics constraint equation to obtain a synchronous multi-sensor data set; inputting the synchronous multi-sensor data set into a fault propagation path sensing network for fusion to obtain a vehicle system fault feature vector; detecting the fault feature vector based on an automobile physical model constraint optimizer to obtain a fault state recognition result; and tracing and controlling a fault state identification result according to the fault propagation atlas model to obtain a system adjustment control signal. According to the invention, the source identification accuracy of the complex coupling fault and the pertinence of control and adjustment are improved.

Description

Technical Field

[0001] This application relates to the technical field of fault detection and control, and particularly to a method and system for detecting vehicle faults by fusing sensor data. Background Art

[0002] As a complex mechatronic product, modern vehicles are equipped with a large number of sensors to monitor the operating states of various subsystems such as engines, brakes, steering systems, and transmissions. The massive data generated by these sensors provides a rich information source for vehicle fault detection. Traditional vehicle fault detection methods mainly rely on single sensors or simple combinations of multiple sensors for fault identification, and judge system anomalies by setting fixed thresholds or using basic statistical analysis methods. Existing fault detection technologies for fusing sensor data include model-based methods, signal processing-based methods, and machine learning-based methods. These methods can identify common vehicle fault types under specific conditions. Model-based methods use mathematical models and physical constraints of vehicle systems for fault detection. Signal processing-based methods extract fault features through techniques such as frequency domain analysis and wavelet transform. Machine learning-based methods use algorithms such as neural networks and support vector machines to establish fault classification models. Currently, multi-sensor data fusion mainly uses traditional fusion algorithms such as weighted average, Kalman filtering, and Bayesian inference, and improves the fusion quality by time alignment and data cleaning in the data preprocessing stage.

[0003] However, the existing technologies have significant deficiencies, mainly manifested in the lack of in-depth modeling of the complex coupling relationships between vehicle systems and effective analysis of the multi-system fault propagation mechanism. Traditional methods regard each sensor as an independent information source, ignoring the physical connection relationships and functional dependencies between vehicle subsystems, and are unable to fully utilize the correlation information between systems to improve the accuracy of fault detection. Existing time synchronization methods mostly adopt fixed timestamp alignment strategies, without considering the influence of changes in vehicle dynamics characteristics on sensor response delays under different driving conditions, resulting in a significant decline in the fusion effect under complex conditions. There is a lack of modeling and analysis of the propagation paths and cascading effects of faults between multiple systems during the fault detection process, and it is impossible to accurately identify the root causes of faults and predict the development trends of faults. In addition, existing methods lack the constraint verification of vehicle physical models during feature extraction and fusion processes, and are prone to generating detection results that violate physical laws, affecting the reliability and credibility of fault diagnosis. Summary of the Invention

[0004] This application provides a method and system for detecting vehicle faults by fusing sensor data, which is used to improve the accuracy of identifying the root causes of complex coupling faults and the pertinence of control regulation.

[0005] In a first aspect, this application provides a method for detecting vehicle faults by fusing sensor data. The method for detecting vehicle faults by fusing sensor data includes: Perform correlation modeling on the multi-system sensor network through the automotive system topology mapping algorithm to obtain a sensor topology correlation matrix; Perform adaptive time window synchronization processing on the sensor topology correlation matrix according to the vehicle dynamics constraint equation to obtain a synchronized multi-sensor dataset; Input the synchronized multi-sensor dataset into the fault propagation path perception network for multi-level feature fusion processing to obtain a vehicle system fault feature vector; Perform anomaly detection processing on the vehicle system fault feature vector based on the automotive physical model constraint optimizer to obtain a fault status recognition result; Perform traceability positioning and control instruction generation processing on the fault status recognition result according to the fault propagation map model to obtain a system adjustment control signal.

[0006] Optionally, the performing correlation modeling on the multi-system sensor network through the automotive system topology mapping algorithm to obtain a sensor topology correlation matrix includes: Perform physical connection relationship recognition processing on the sensor nodes of the engine, brake, steering, and transmission systems through CAN bus network topology scanning to obtain an inter-system connection topology diagram; Perform weight assignment processing on the inter-system connection topology diagram according to the functional dependency relationship of each subsystem to obtain a system correlation weight topology diagram; Perform sensor node mapping processing on the system correlation weight topology diagram based on the three-dimensional space coordinates and signal propagation path of the sensors to obtain a sensor spatial distribution topology diagram; Perform numerical quantization processing on the sensor spatial distribution topology diagram with the sensor physical correlation strength and signal propagation delay parameters to obtain a sensor correlation metric matrix; Perform matrix normalization and symmetry adjustment processing on the sensor correlation metric matrix to obtain the sensor topology correlation matrix.

[0007] Optionally, the performing adaptive time window synchronization processing on the sensor topology correlation matrix according to the vehicle dynamics constraint equation to obtain a synchronized multi-sensor dataset includes: Perform real-time acquisition processing on the current vehicle operating state parameters through vehicle speed, acceleration, and angular velocity sensors to obtain a vehicle dynamics state vector; Perform dynamic adjustment calculation processing on the preset basic time window length according to the vehicle dynamics state vector to obtain an adaptive time window parameter; Perform propagation delay compensation calculation processing on each sensor data based on the distance and signal propagation path between sensors in the sensor topology correlation matrix to obtain a sensor delay correction coefficient; Apply the sensor time delay correction coefficient to the raw data of the multi-system sensors for time series alignment calibration processing to obtain time series synchronized sensor data; Perform segmented cutting and data integrity verification processing on the time series synchronized sensor data according to the adaptive time window parameter to obtain the synchronized multi-sensor data set.

[0008] Optionally, inputting the synchronized multi-sensor data set into a fault propagation path perception network for multi-level feature fusion processing to obtain a vehicle system fault feature vector, including: Input the synchronized multi-sensor data set into the sensor topology convolution layer of the fault propagation path perception network for convolution feature extraction processing based on the automotive system topology structure to obtain a topology-aware sensor feature map; Perform time series dependent fault propagation path prediction processing on the topology-aware sensor feature map through the fault propagation modeling layer of the fault propagation path perception network to obtain a fault propagation path probability matrix; Based on the multi-system coupling analysis layer of the fault propagation path perception network, perform system-to-system coupling strength calculation and fault cascade effect modeling processing on the fault propagation path probability matrix to obtain a multi-system coupling fault feature tensor; Input the multi-system coupling fault feature tensor into the adaptive feature fusion layer of the fault propagation path perception network for dynamic feature weighted fusion processing based on fault severity to obtain a fault severity weighted feature vector; Perform vehicle physical constraint verification and feature validity screening processing on the fault severity weighted feature vector to obtain the vehicle system fault feature vector.

[0009] Optionally, the multi-system coupling analysis layer based on the fault propagation path perception network performs system-to-system coupling strength calculation and fault cascade effect modeling processing on the fault propagation path probability matrix to obtain a multi-system coupling fault feature tensor, including: Quantify the coupling strength between the engine - transmission, brake - ABS, and steering - suspension systems of the fault propagation path probability matrix through an inter-system physical coupling degree calculation module to obtain a system coupling strength coefficient matrix; Calculate the propagation delay and attenuation coefficient of faults between different systems for the system coupling strength coefficient matrix according to the fault cascade propagation timing model to obtain a fault cascade propagation parameter matrix; Based on a multi-system fault interaction impact analysis algorithm, perform fault interaction impact and amplification effect modeling processing on the fault cascade propagation parameter matrix to obtain a fault interaction impact feature matrix; Perform three-dimensional tensor reconstruction and multi-dimensional fault feature encoding processing on the fault interaction impact feature matrix to obtain an initial multi-system coupling fault feature tensor; Perform tensor decomposition and eigen-dimension optimization on the initial multi-system coupling fault feature tensor to obtain the multi-system coupling fault feature tensor.

[0010] Optionally, the vehicle system fault feature vector is processed for anomaly detection by the vehicle physics model-based constraint optimizer to obtain a fault state recognition result, including: Perform physical rationality verification on the vehicle system fault feature vector through an engine thermodynamics model, a braking system dynamics model, and an electrical system Ohm's law constraint model to obtain a physical constraint violation metric parameter; Construct a constraint penalty term for the anomaly detection objective function according to the physical constraint violation metric parameter to obtain a physical constraint anomaly detection loss function; Perform a global optimal solution search on the physical constraint anomaly detection loss function based on an improved particle swarm optimization algorithm to obtain an optimal set of anomaly detection threshold parameters; Input the vehicle system fault feature vector into an anomaly detection classifier configured with the optimal set of anomaly detection threshold parameters for fault type discrimination to obtain a preliminary fault state classification result; Perform confidence evaluation and uncertainty quantification on the preliminary fault state classification result to obtain the fault state recognition result.

[0011] Optionally, trace the source and generate control instructions for the fault state recognition result according to the fault propagation graph model to obtain a system adjustment control signal, including: Locate the root node of the fault for the fault state recognition result through a fault propagation path backtracking algorithm to obtain the identifier of the root sensor node of the fault; Calculate the fault influence range and trace the propagation link of the fault propagation graph model according to the identifier of the root sensor node of the fault to obtain a fault influence system mapping table; Perform a fault risk level assessment and priority sorting on the fault influence system mapping table based on a fault severity grading rule to obtain a graded fault handling strategy table; Input the graded fault handling strategy table into a control strategy library for corresponding control instruction matching and parameter adjustment to obtain an initial system adjustment instruction set; Perform execution feasibility verification and instruction optimization on the initial system adjustment instruction set to obtain the system adjustment control signal.

[0012] In a second aspect, the present application provides a vehicle fault detection system integrating sensor data. The vehicle fault detection system integrating sensor data includes: A modeling module for performing correlation modeling processing on a multi-system sensor network through an automotive system topology mapping algorithm to obtain a sensor topology correlation matrix; A synchronization module for performing adaptive time window synchronization processing on the sensor topology correlation matrix according to vehicle dynamics constraint equations to obtain a synchronized multi-sensor data set; A fusion module for inputting the synchronized multi-sensor data set into a fault propagation path perception network for multi-level feature fusion processing to obtain a vehicle system fault feature vector; A detection module for performing anomaly detection processing on the vehicle system fault feature vector based on an automotive physical model constraint optimizer to obtain a fault state recognition result; A tracing module for performing tracing location and control instruction generation processing on the fault state recognition result according to a fault propagation graph model to obtain a system adjustment control signal.

[0013] In a third aspect, there is provided an automotive fault detection device for fusing sensor data, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the automotive fault detection device for fusing sensor data executes the above-mentioned automotive fault detection method for fusing sensor data.

[0014] In a fourth aspect, there is provided a computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when it runs on a computer, it causes the computer to execute the above-mentioned automotive fault detection method for fusing sensor data.

[0015] In the technical solution provided by this application, through the automotive system topology mapping algorithm, the multi-system sensor network is subjected to correlation modeling to obtain the sensor topology correlation matrix, which overcomes the limitation of the prior art that regards sensors as independent information sources, fully explores the physical connection relationships and functional dependencies among various subsystems of the vehicle, establishes a complete sensor spatial correlation model, provides a system-level topology foundation for multi-sensor data fusion, and significantly improves the utilization efficiency of correlation information between sensors and the accuracy of fault detection. According to the vehicle dynamics constraint equation, the sensor topology correlation matrix is subjected to adaptive time window synchronization processing to obtain a synchronized multi-sensor data set, which solves the deficiency of the prior art that uses a fixed timestamp alignment strategy. By dynamically adjusting the time window length and propagation delay compensation, the influence of sensor response delay differences under different driving conditions on the data fusion quality is effectively eliminated, ensuring a high degree of consistency of multi-sensor data in the time dimension and laying a reliable data foundation for subsequent feature extraction and fault detection. The synchronized multi-sensor data set is input into the fault propagation path perception network for multi-level feature fusion processing to obtain the vehicle system fault feature vector, which breaks through the singularity limitation of the traditional method in the feature extraction level. Through the collaborative action of the sensor topology convolution layer, the fault propagation modeling layer, the multi-system coupling analysis layer, and the adaptive feature fusion layer, the propagation law and coupling mechanism of faults among multiple automotive systems are deeply explored, forming a fault feature representation with rich semantic information and greatly improving the feature recognition ability and classification accuracy of complex coupling faults. Based on the automotive physical model constraint optimizer, the vehicle system fault feature vector is subjected to anomaly detection processing to obtain the fault state recognition result, effectively avoiding the physically unreasonable detection results that may be generated by the pure data-driven method in the prior art. Through the joint verification of the engine thermodynamics model, the brake system dynamics model, and the electrical system Ohm's law constraint model, it is ensured that the fault detection result conforms to the automotive physical laws and engineering common sense, significantly improving the reliability and credibility of fault diagnosis. According to the fault propagation map model, the fault state recognition result is subjected to traceability positioning and control instruction generation processing to obtain the system adjustment control signal, making up for the deficiency of the prior art in lacking fault root cause positioning and active control capabilities. By accurately positioning the fault root cause through the fault propagation path backtracking algorithm and generating targeted control instructions based on the fault severity grading and the control strategy library matching, a closed-loop processing from fault detection to active control is realized, providing a complete solution for automotive fault management and greatly improving the safety and reliability of the automotive system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 Schematic diagram of an embodiment of the method for detecting vehicle faults by fusing sensor data in the embodiments of the present application; Figure 2 Schematic diagram of an embodiment of the system for detecting vehicle faults by fusing sensor data in the embodiments of the present application; Figure 3 Block diagram showing the structure of the vehicle fault detection device by fusing sensor data in the embodiments of the present invention. Detailed implementation manners

[0018] The embodiments of the present application provide a method and system for detecting vehicle faults by fusing sensor data. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned drawings of the present application are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order different from that shown or described here. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these process, method, product or device.

[0019] For ease of understanding, the specific processes of the embodiments of the present application are described below. Please refer to Figure 1 An embodiment of the method for detecting vehicle faults by fusing sensor data in the embodiments of the present application includes: Step S101: Perform correlation modeling on the multi-system sensor network through the vehicle system topology mapping algorithm to obtain a sensor topology correlation matrix; Step S102: Perform adaptive time window synchronization processing on the sensor topology correlation matrix according to the vehicle dynamics constraint equation to obtain a synchronized multi-sensor data set; Step S103: Input the synchronized multi-sensor data set into the fault propagation path perception network for multi-level feature fusion processing to obtain a vehicle system fault feature vector; Step S104: Perform anomaly detection on the vehicle system fault feature vector based on the vehicle physical model constraint optimizer to obtain a fault state recognition result; Step S105: Perform traceability positioning and control instruction generation processing on the fault state recognition result according to the fault propagation map model to obtain a system adjustment control signal.

[0020] It can be understood that the execution entity of this application can be an automotive fault detection system that fuses sensor data, or it can also be a terminal or a server. Specifically, no limitation is imposed here. In the embodiments of this application, the server is taken as an example of the execution entity for illustration.

[0021] Specifically, the automotive system topology mapping algorithm first identifies the physical connection relationships of sensor nodes in the engine, braking, steering, and transmission systems through CAN bus network topology scanning, and constructs an inter-system connection topology graph. This topology graph reflects the functional dependency relationships of each subsystem, and forms a system association weight topology graph through weight assignment. Based on the three-dimensional space coordinates of the sensors and the signal propagation paths, the sensor nodes are mapped into the topology graph to form a sensor spatial distribution topology graph. Subsequently, the physical association strength and signal propagation delay parameters of the sensors are numerically quantified, and a sensor topology association matrix is obtained through matrix normalization and symmetry adjustment. The elements of this matrix represent the degree of association between sensors, solving the problem that traditional methods ignore the spatial relationship of sensors.

[0022] The vehicle dynamics constraint equation is used to perform adaptive time window synchronization processing on the sensor topology association matrix. The operating state parameters collected in real time by the vehicle speed, acceleration, and angular velocity sensors constitute the vehicle dynamics state vector, and this vector dynamically adjusts the preset basic time window length to form adaptive time window parameters. Based on the distances between sensors and the signal propagation paths in the sensor topology association matrix, the propagation delay compensation coefficients of the data of each sensor are calculated. This correction coefficient is applied to the original data of multi-system sensors for time series alignment and calibration, and the data integrity is segmented and verified according to the adaptive time window parameters, and finally a synchronized multi-sensor data set is obtained. This dynamic synchronization mechanism effectively solves the problem of sensor response delay differences under different working conditions.

[0023] The fault propagation path perception network uses a multi-level architecture for feature fusion processing. The sensor topology convolution layer extracts convolution features based on the automotive system topology structure to obtain a topology-aware sensor feature map. The fault propagation modeling layer predicts the fault propagation path with time series dependence for this feature map to generate a fault propagation path probability matrix. The multi-system coupling analysis layer quantifies the coupling strengths between the engine - transmission, braking - ABS, and steering - suspension systems through the inter-system physical coupling degree calculation module, combines the fault cascade propagation time series model to calculate the fault propagation delay and attenuation coefficients, and models the mutual influence and amplification effects between faults through the multi-system fault interaction influence analysis algorithm, and forms a multi-system coupling fault feature tensor through three-dimensional tensor reconstruction and feature encoding. The adaptive feature fusion layer performs dynamic feature weighted fusion based on the fault severity, and obtains the vehicle system fault feature vector through vehicle physical constraint verification and feature effectiveness screening.

[0024] The automotive physical model constraint optimizer uses the engine thermodynamics model, the brake system dynamics model, and the electrical system Ohm's law constraint model to verify the physical rationality of the vehicle system fault feature vector, and obtains the physical constraint violation metric parameter. This parameter constructs the constraint penalty term of the anomaly detection objective function to form the physical constraint anomaly detection loss function. The improved particle swarm optimization algorithm searches for the global optimal solution of this loss function to obtain the optimal anomaly detection threshold parameter set. The anomaly detection classifier configured with this parameter set discriminates the fault type of the vehicle system fault feature vector, and obtains the fault status recognition result through confidence evaluation and uncertainty quantification. The physical constraint mechanism ensures that the detection result conforms to the automotive physical laws, avoiding the physically unreasonable results of pure data-driven methods.

[0025] The fault root node is located for the fault status recognition result through the fault propagation path backtracking algorithm to obtain the fault root sensor node identifier. Based on this identifier, the fault influence range is calculated and the propagation link is traced for the fault propagation graph model to form the fault influence system mapping table. The fault severity grading rule evaluates the risk level and prioritizes the mapping table to generate the graded fault handling strategy table. The control strategy library matches the corresponding control instructions according to this strategy table and adjusts the parameters, and obtains the system adjustment control signal through the execution feasibility verification and instruction optimization. This traceability mechanism overcomes the limitation of the existing technology that can only identify a single fault type, accurately locates the fault root through the fault propagation chain analysis, and generates targeted control strategies.

[0026] In a specific embodiment, the process of executing step S101 may specifically include the following steps: The physical connection relationship of the sensor nodes of the engine, brake, steering, and transmission systems is identified through the CAN bus network topology scan to obtain the inter-system connection topology diagram; The weight assignment process is performed on the inter-system connection topology diagram according to the functional dependency of each subsystem to obtain the system association weight topology diagram; The sensor node mapping process is performed on the system association weight topology diagram based on the three-dimensional space coordinates of the sensors and the signal propagation path to obtain the sensor spatial distribution topology diagram; The physical association strength of the sensors and the signal propagation delay parameters are used to numerically quantify the sensor spatial distribution topology diagram to obtain the sensor association metric matrix; The matrix normalization and symmetry adjustment processes are performed on the sensor association metric matrix to obtain the sensor topology association matrix.

[0027] Specifically, the CAN bus network topology scan identifies the distribution of sensor nodes in various systems such as the engine control module, anti-lock braking system (ABS) module, electric power steering (EPS) module, and transmission control module by reading the device identifiers and network addresses of automotive electronic control units (ECUs). The scan process traverses all node addresses on the CAN bus, obtains information such as the device ID, system type to which it belongs, and communication protocol version of each sensor node, and at the same time records the data frame transmission paths and communication dependencies between nodes, constructing an inter-system connection topology graph containing nodes and connection edges, where the nodes represent each subsystem and the connection edges represent the communication links between systems.

[0028] The analysis of the functional dependencies between subsystems is based on automotive dynamics and control theory to determine the coupling degree between different systems. The engine system directly affects the operating state of the transmission through torque output. The braking system and the ABS work together to control the wheel slip ratio. The steering system and the suspension system jointly affect the handling stability of the vehicle. The weight assignment process assigns numerical weights according to the functional coupling strength between systems. A higher functional dependency degree between the engine and the transmission is assigned a weight value of 0.9. The collaborative control relationship between the braking system and the ABS is assigned a weight value of 0.8. The handling coupling between the steering and suspension systems is assigned a weight value of 0.7. The system association weight topology graph after weight assignment contains weighted connection edges, and the weight value reflects the tightness of the functional dependencies between systems.

[0029] The sensor node mapping process maps specific sensor nodes to the system association weight topology graph based on the physical installation positions of the sensors in the vehicle's three-dimensional coordinate system. The water temperature sensor, intake pressure sensor, oxygen sensor, etc. in the engine compartment are mapped to the engine system node. The wheel speed sensor and brake pressure sensor near the wheels are mapped to the braking system node. The steering angle sensor and torque sensor of the steering wheel assembly are mapped to the steering system node. The signal propagation path analysis considers the physical path length and transmission delay of the sensor signal transmitted to the data acquisition point through hardware paths such as wire harnesses, connectors, and ECUs. The sensor spatial distribution topology graph adds the spatial position information of the sensor nodes and the signal propagation path data on the basis of the system association weight topology graph.

[0030] The numerical quantization process converts the physical association strength of sensors and the signal propagation delay parameters into computable numerical matrix elements. The physical association strength of sensors is calculated based on factors such as the physical distance between sensors, the number of shared hardware connections, and the degree of signal coupling. The closer the distance, the more shared connections, and the stronger the signal coupling, the greater the association strength. The signal propagation delay parameter is calculated based on the propagation speed and distance of the signal in different media. The propagation speed of an electrical signal in a copper wire is about two-thirds of the speed of light, and the propagation speed of an optical signal in an optical fiber is close to the speed of light. Taking the physical association strength between each pair of sensors as the element value at the intersection of the matrix row and column, and the signal propagation delay as the time correction parameter, a sensor association metric matrix is formed.

[0031] The matrix normalization process uniformly scales the element values in the sensor association metric matrix to the standard range between 0 and 1, and the normalization operation is performed by dividing by the maximum element value in the matrix. The symmetry adjustment process ensures that the matrix satisfies the mathematical properties of a symmetric matrix, that is, the element value in the i-th row and j-th column of the matrix is equal to the element value in the j-th row and i-th column. The symmetric matrix is obtained by calculating the average of the matrix and its transpose matrix. The sensor topology association matrix obtained after normalization and symmetry adjustment has a standardized numerical range and good mathematical properties. The matrix element value represents the degree of association between the corresponding sensor pairs. The larger the value, the stronger the association, and the smaller the value, the weaker the association.

[0032] In a specific embodiment, the process of executing step S102 may specifically include the following steps: Real-time acquisition and processing of the current vehicle operating state parameters are performed through vehicle speed, acceleration, and angular velocity sensors to obtain a vehicle dynamics state vector; Based on the vehicle dynamics state vector, dynamic adjustment and calculation processing are performed on the preset basic time window length to obtain an adaptive time window parameter; Based on the distance between sensors and the signal propagation path in the sensor topology association matrix, propagation delay compensation calculation processing is performed on each sensor data to obtain a sensor delay correction coefficient; The sensor delay correction coefficient is applied to the original data of the multi-system sensors for time series alignment and calibration processing to obtain time series synchronized sensor data; The time series synchronized sensor data is segmented and cut and data integrity verification processing is performed according to the adaptive time window parameter to obtain a synchronized multi-sensor data set.

[0033] Specifically, the vehicle speed sensor converts the rotational speed signal detected from the output shaft of the transmission or the wheels to obtain the vehicle driving speed data. The acceleration sensor detects the acceleration changes of the vehicle in three axial directions based on the principle of inertial measurement. The angular velocity sensor measures the rotational angular velocity of the vehicle around the vertical axis through a gyroscope. The analog signals of these three types of sensors are converted into digital signals through an analog-to-digital converter in real-time acquisition and processing, and the sensor values are continuously read at a fixed sampling frequency to form a vehicle dynamics state vector containing parameters such as vehicle speed, longitudinal acceleration, lateral acceleration, vertical acceleration, and yaw angular velocity.

[0034] The parameters in the vehicle dynamics state vector directly reflect the characteristics of the current vehicle operating conditions. There are differences in the sensor response characteristics and data synchronization requirements under different operating conditions. The dynamic adjustment calculation process judges the current vehicle operating condition type according to the combined state of the vehicle speed and acceleration. When driving at a low speed and uniformly, the sensor data changes slowly and a longer time window is required to capture complete information. When accelerating suddenly or turning sharply, the sensor data changes violently and a shorter time window is required to maintain timeliness. A preset basic time window length is used as a standard reference value, and the window length is dynamically adjusted through the weighted calculation of the speed change rate and acceleration amplitude in the vehicle dynamics state vector. The more violently the speed changes and the greater the acceleration, the shorter the time window. Finally, an adaptive time window parameter suitable for the current operating condition is obtained.

[0035] The sensor topology correlation matrix contains the physical distance information between each sensor and the signal propagation path length data. The propagation delay compensation calculation process calculates the delay based on the propagation speed characteristics of the signal in different transmission media. The propagation speed of the electrical signal in the copper wire is about two hundred million meters per second, and the propagation speed of the optical signal in the optical fiber is close to three hundred million meters per second. When the digital signal propagates in the CAN bus, the protocol processing delay also needs to be considered. The propagation delay compensation calculation obtains the theoretical propagation delay by dividing the physical distance between each pair of sensors by the signal propagation speed of the corresponding transmission media, and then adds the hardware processing delay and protocol conversion delay to obtain the total propagation delay. The difference in the propagation delay of different sensors to the data acquisition center constitutes the sensor delay correction coefficient, which represents the time offset of each sensor relative to the reference sensor.

[0036] The time sequence alignment and calibration process applies the sensor delay correction coefficient as the time offset to the original data of the multi-system sensors, and adjusts the time reference of the data by adding or subtracting the corresponding correction coefficient value to the time stamps of the data of each sensor. In the calibration process, the sensor with the minimum propagation delay is selected as the time reference, and the data of other sensors are adjusted to move forward or backward according to their delay correction coefficients to ensure that all sensor data correspond to the same physical moment. The sensor data after time sequence alignment is consistent in the time dimension, eliminating the time deviation caused by the difference in the signal propagation path, and forming time sequence synchronized sensor data.

[0037] The segmented cutting process divides the continuous time-series synchronized sensor data into segments of a fixed time length according to the adaptive time window parameter. Each data segment contains the synchronized data of all sensors within that time window. The data integrity verification process checks whether there are missing sensor data, outliers, or communication errors in each data segment, and judges the data quality by verifying the continuity, range rationality, and checksum correctness of the data. The integrity verification includes multiple dimensions such as detecting whether the number of data sampling points meets the expectation, whether the values are within the sensor measurement range, and whether the change between adjacent sampling points exceeds the physical limit. The data segments that pass the verification form a synchronized multi-sensor data set, which ensures time synchronization and data integrity.

[0038] In a specific embodiment, the process of executing step S103 may specifically include the following steps: Input the synchronized multi-sensor data set into the sensor topology convolution layer of the fault propagation path perception network for convolution feature extraction processing based on the vehicle system topology structure to obtain a topology-aware sensor feature map; Perform temporal dependence fault propagation path prediction processing on the topology-aware sensor feature map through the fault propagation modeling layer of the fault propagation path perception network to obtain a fault propagation path probability matrix; Based on the multi-system coupling analysis layer of the fault propagation path perception network, perform system-to-system coupling strength calculation and fault cascade effect modeling processing on the fault propagation path probability matrix to obtain a multi-system coupling fault feature tensor; Input the multi-system coupling fault feature tensor into the adaptive feature fusion layer of the fault propagation path perception network for dynamic feature weighted fusion processing based on the fault severity to obtain a fault severity weighted feature vector; Perform vehicle physical constraint verification and feature validity screening processing on the fault severity weighted feature vector to obtain a vehicle system fault feature vector.

[0039] Specifically, the sensor topology convolution layer uses a convolution kernel based on the automotive system topology structure to perform feature extraction on the synchronous multi-sensor dataset. The design of this convolution kernel follows the physical connection relationship and functional dependence relationship among the various subsystems of the vehicle. Different from the two-dimensional convolution in traditional image processing, the convolution operation performs graph convolution calculations based on the adjacency relationship defined by the sensor topology association matrix. The eigenvalue of each sensor node is obtained by aggregating the weighted eigenvalues of its topological neighbor nodes, and the weight coefficient comes from the corresponding association strength value in the sensor topology association matrix. The convolution feature extraction process performs a weighted sum of the original data values of each sensor and the data values of its topological neighbor sensors. The weight reflects the physical association degree between sensors. The aggregated eigenvalue contains both the local information of the sensor itself and the association information of the neighbor sensors. The topology-aware sensor feature map maintains the same topology structure as the original sensor network, but the eigenvalue of each node has fused the association information of the surrounding sensors, forming a feature representation with topology awareness. The fault propagation modeling layer performs temporal dependence analysis on the topology-aware sensor feature map based on the recurrent neural network structure to mine the propagation law and path characteristics of faults in the time dimension. The temporal dependence fault propagation path prediction process inputs the topology-aware sensor feature map into the recurrent neural network according to the time series. The network memorizes the feature information of historical moments through the hidden state and learns the temporal pattern of fault diffusion from the initial abnormal sensor node to other relevant sensor nodes. The prediction process calculates the probability value of each sensor node having a fault at a future moment and the transition probability of the fault propagating from one sensor node to another based on the feature map states at the current moment and historical moments. The rows of the fault propagation path probability matrix represent the fault source sensor nodes, the columns represent the fault target sensor nodes, and the matrix element values represent the probability of the fault propagating from the source node to the target node. The larger the probability value, the higher the propagation possibility.

[0040] The multi-system coupling analysis layer deeply analyzes the fault propagation path probability matrix through the inter-system physical coupling degree calculation module, and quantifies the coupling strength between different systems such as engine - transmission, braking - ABS, and steering - suspension. The calculation of the inter-system coupling strength is based on the statistical analysis of the fault propagation probability values between sensor nodes belonging to different systems. The average fault propagation probability between sensor nodes within the same system is calculated as the in-system coupling strength, and the average fault propagation probability between sensor nodes of different systems is calculated as the inter-system coupling strength. The fault cascade effect modeling processes and analyzes the chain propagation mechanism of faults between multiple systems, and identifies possible cascade fault paths by constructing an inter-system fault propagation link diagram. The cascade effect modeling considers the time delay, attenuation coefficient, and amplification factor of fault propagation, and establishes a complete propagation link model from the initial fault system to the final affected system. The multi-system coupling fault feature tensor combines the fault propagation path probability matrix, the system coupling strength matrix, and the cascade effect parameters into a three-dimensional tensor structure. The first dimension represents the sensor node index, the second dimension represents the time step, and the third dimension represents the fault feature type.

[0041] The adaptive feature fusion layer dynamically weights the multi-system coupling fault feature tensor according to the fault severity. The severity assessment is graded based on the impact of the fault on vehicle safety and performance. Braking system faults, which are safety-critical faults, obtain the highest weight coefficient. Engine performance degradation, which is a performance-impacting fault, obtains a medium weight coefficient. Air conditioning system anomalies, which are comfort faults, obtain a lower weight coefficient. The dynamic feature weighted fusion process sums up different feature dimensions in the multi-system coupling fault feature tensor according to the corresponding severity weight coefficients, highlighting the importance of safety-critical fault features and suppressing the interference effects of secondary fault features. The attention mechanism is used in the fusion process to dynamically calculate the weight coefficients of each feature dimension, and the weight calculation is comprehensively evaluated based on the magnitude, change trend, and historical fault patterns of the eigenvalues. The fault severity weighted feature vector reduces the three-dimensional tensor to a one-dimensional vector form, and each element in the vector corresponds to the weighted fault feature value of a sensor node.

[0042] The vehicle physical constraint verification process conducts a rationality check on the weighted feature vector of fault severity based on automotive engineering principles to ensure that the fault detection results conform to vehicle physical laws and engineering common sense. Physical constraint verification includes multiple dimensions such as checking the matching relationship between engine speed and vehicle speed, the corresponding relationship between braking pressure and deceleration, and the consistency between steering angle and vehicle trajectory. In the verification process, the fault feature values in the feature vector are compared with the preset physical constraint conditions to identify abnormal feature values that violate physical laws and correct or eliminate them. Feature validity screening processes evaluate the contribution of each feature dimension to fault detection through statistical analysis and information gain calculation, eliminate redundant and noise features, and retain effective features with significant discrimination ability for fault identification. The screening process uses a recursive feature elimination algorithm to gradually eliminate feature dimensions with lower contribution until the remaining feature set reaches the preset number of features or contribution threshold. The vehicle system fault feature vector contains high-quality fault features that have undergone physical constraint verification and validity screening, and each feature dimension has a clear physical meaning and strong fault identification ability.

[0043] In a specific embodiment, the process of performing the steps of calculating the system - to - system coupling strength and modeling the fault cascade effect on the fault propagation path probability matrix based on the multi - system coupling analysis layer of the fault propagation path perception network may specifically include the following steps: The system - to - system physical coupling degree calculation module quantifies the coupling strength between the engine - transmission, braking - ABS, and steering - suspension systems for the fault propagation path probability matrix to obtain a system coupling strength coefficient matrix; According to the fault cascade propagation timing model, calculate the propagation delay and attenuation coefficient of faults between different systems for the system coupling strength coefficient matrix to obtain a fault cascade propagation parameter matrix; Based on the multi - system fault interaction impact analysis algorithm, model the mutual influence and amplification effect between faults on the fault cascade propagation parameter matrix to obtain a fault interaction impact feature matrix; Perform three - dimensional tensor reconstruction and multi - dimensional fault feature encoding on the fault interaction impact feature matrix to obtain an initial multi - system coupling fault feature tensor; Perform tensor decomposition and feature dimension optimization on the initial multi - system coupling fault feature tensor to obtain a multi - system coupling fault feature tensor.

[0044] Specifically, the inter-system physical coupling degree calculation module quantifies the physical coupling strength between systems by analyzing the fault propagation probability values between sensor nodes of different systems in the fault propagation path probability matrix. The calculation of the coupling strength between the engine and transmission systems is based on the statistical analysis of the fault propagation probability from the engine sensor nodes to the transmission sensor nodes. All sensor nodes in the engine system are used as source nodes, and all sensor nodes in the transmission system are used as target nodes. The average value of the corresponding propagation probabilities is calculated to obtain the engine-transmission coupling strength coefficient. The coupling strength between the braking and ABS systems is calculated through the fault propagation probability from the braking system sensors such as the brake pressure sensor and the brake pedal position sensor to the ABS system sensors such as the wheel speed sensor and the pressure regulator. The coupling strength between the steering and suspension systems is obtained based on the fault propagation probability analysis from the steering angle sensor and the steering torque sensor to the suspension position sensor and the shock absorber sensor. The system coupling strength coefficient matrix is organized in the form of rows representing the source system and columns representing the target system. The matrix element values reflect the physical coupling degree between the corresponding systems. The larger the value, the closer the coupling between the systems and the higher the probability of fault propagation.

[0045] The fault cascading propagation timing model establishes a mathematical model of time delay and intensity attenuation for fault propagation between different systems based on system dynamics theory. This model considers the time characteristics of system response characteristics and physical propagation mechanisms. The propagation delay calculation processes and analyzes the time interval required for a fault to be detected as abnormal by the source system sensor and for the target system sensor to show a corresponding change. The delay time is related to factors such as the physical distance between systems, the signal propagation path, and the response speed of the control system. The delay time for the engine system fault to propagate to the transmission system is calculated by the time difference between the engine torque change and the transmission input shaft torque response. The delay time for the braking system fault to propagate to the ABS system is determined based on the response time from the brake pedal operation to the activation of the ABS system. The attenuation coefficient calculation processes and quantifies the degree of weakening of the fault signal intensity during the propagation between systems. The attenuation degree is related to factors such as the buffer mechanism between systems, the damping characteristics, and the compensation ability of the control algorithm. The fault cascading propagation parameter matrix organizes the propagation delay and attenuation coefficient into a matrix form according to the system correspondence. The matrix rows and columns correspond to the propagation relationship between systems, and the element values contain two parameters: the delay time and the attenuation coefficient.

[0046] The multi-system fault interaction impact analysis algorithm establishes a mathematical model of the interaction between multi-system faults based on graph theory and complex network theory. This algorithm takes into account complex interaction relationships such as the synergistic effect, competitive effect, and amplification effect between faults. The analysis of the mutual influence between faults is carried out by constructing a system fault impact graph, with systems as nodes and fault propagation relationships as edges, to analyze how multiple simultaneous faults influence and interact with each other. The synergistic effect modeling analyzes the superimposed impact when multiple systems fail simultaneously. For example, when the engine overheating fault and the cooling system fault occur simultaneously, they will reinforce each other, resulting in the fault impact degree exceeding the simple superposition of individual faults. The competitive effect modeling analyzes the competition relationship of different system faults for shared resources. For example, when the electrical system fault and the engine control fault occur simultaneously, they will compete for limited power resources and controller processing capabilities. The amplification effect modeling analyzes how certain system faults cause other system faults to be amplified. For example, a braking system fault causes an increase in the engine braking load, amplifying the working pressure and fault risk of the engine system. The fault interaction impact characteristic matrix organizes the intensity parameters of various interaction effects according to the system pairing relationship, and the matrix element values quantify the impact degree of the fault interaction between the corresponding systems.

[0047] The three-dimensional tensor reconstruction process expands the fault interaction impact characteristic matrix into a three-dimensional data structure that includes the time dimension, system dimension, and fault type dimension. The first dimension of the tensor corresponds to different automotive subsystems, the second dimension corresponds to the time series, and the third dimension corresponds to the fault feature types. The multi-dimensional fault feature encoding process encodes different types of fault feature information into numerical elements in the tensor, including feature parameters such as the fault occurrence probability, fault severity, fault duration, and fault impact range. The encoding process combines one-hot encoding and numerical normalization to convert discrete fault type labels into numerical vectors and normalize continuous fault parameter values to the standard numerical range. The tensor structure maintains the multi-dimensional correlation relationship between fault features. The position index of each tensor element corresponds to a specific system-time-feature combination, and the element value reflects the fault feature intensity of the corresponding combination. The initial multi-system coupled fault feature tensor contains complete multi-system fault interaction information, but has a high dimension and contains redundant information.

[0048] The tensor decomposition process uses the singular value decomposition or non - negative matrix factorization algorithm to decompose the high - dimensional initial tensor into a combination of multiple low - dimensional tensors. The decomposition process retains the main fault feature information while reducing the data dimension and computational complexity. The decomposition algorithm finds the optimal low - dimensional tensor representation through iterative optimization, minimizing the error between the reconstructed tensor and the original tensor. The feature dimension optimization process determines the number of retained feature dimensions based on feature importance assessment and information entropy calculation, eliminating redundant feature dimensions with low contribution to fault detection. The optimization process uses principal component analysis and feature selection algorithms to identify the most representative fault feature dimensions, ensuring that the optimized tensor not only maintains the integrity of the original fault information but also has high computational efficiency. After decomposition and optimization, the multi - system coupled fault feature tensor has a compact data structure and optimized feature representation, and each element in the tensor corresponds to important fault feature information.

[0049] In a specific embodiment, the process of executing step S104 may specifically include the following steps: Perform physical rationality verification processing on the vehicle system fault feature vector through the engine thermodynamics model, the brake system dynamics model, and the electrical system Ohm's law constraint model to obtain the physical constraint violation metric parameter; Construct a constraint penalty term for the anomaly detection objective function according to the physical constraint violation metric parameter to obtain the physical constraint anomaly detection loss function; Perform a global optimal solution search processing on the physical constraint anomaly detection loss function based on the improved particle swarm optimization algorithm to obtain the optimal anomaly detection threshold parameter set; Input the vehicle system fault feature vector into the anomaly detection classifier configured with the optimal anomaly detection threshold parameter set for fault type discrimination processing to obtain the preliminary fault state classification result; Perform confidence evaluation and uncertainty quantification processing on the preliminary fault state classification result to obtain the fault state recognition result.

[0050] Specifically, the engine thermodynamic model establishes the energy conservation and entropy change relationships of the engine working process based on the first and second laws of thermodynamics. The model includes physical constraint conditions such as the relationship equation between the combustion chamber temperature and pressure, the corresponding relationship between the coolant temperature and the engine load, and the correlation law between the exhaust temperature and the combustion efficiency. The braking system dynamics model establishes the relationship between the braking force and the vehicle deceleration according to Newton's laws of motion. The model covers dynamic constraints such as the linear relationship between the braking pressure and the braking torque, the non-linear relationship between the wheel slip ratio and the road adhesion coefficient, and the square relationship between the braking distance and the square of the initial speed. The Ohm's law constraint model of the electrical system establishes the relationship between voltage, current, and resistance based on the basic laws of circuits. The model includes electrical constraints such as the corresponding relationship between the battery voltage and the load current, the proportional relationship between the line resistance and the voltage drop, and the relationship between the power consumption and the square of the current. The physical rationality verification process compares the eigenvalues in the vehicle system fault feature vector with the constraint conditions of the corresponding physical model, and calculates the degree to which the eigenvalues deviate from the physical constraint range. The verification process checks whether the matching relationship between the engine speed and torque conforms to the engine characteristic curve, whether the ratio of the braking pressure to the deceleration is within a reasonable range, and whether the product of the voltage and current is equal to the actual power consumption. The physical constraint violation metric parameter quantifies the severity of the violation of physical constraints in each feature dimension. The greater the degree of violation, the less the eigenvalue conforms to the physical law, and more attention needs to be paid in fault detection.

[0051] The constraint penalty term construction process incorporates the physical constraint violation metric parameter into the objective function of anomaly detection based on the Lagrange multiplier method and the penalty function theory. The role of the penalty term is to impose an additional cost on the detection results that violate physical constraints. The mathematical form of the penalty term uses a square penalty function, and the penalty intensity is proportional to the square of the physical constraint violation metric parameter. The greater the degree of violation, the more severe the penalty. The anomaly detection objective function originally only considered the data-driven classification error. After adding the constraint penalty term, a balance needs to be sought between classification accuracy and physical rationality. The physical constraint anomaly detection loss function performs a weighted sum of the original detection loss and the constraint penalty term. The weight coefficient controls the importance of physical constraints in the overall optimization objective. The gradient calculation of the loss function considers both the classification error gradient and the constraint violation gradient to ensure that the optimization process improves both detection accuracy and meets physical constraint requirements.

[0052] The improved particle swarm optimization algorithm incorporates an adaptive inertia weight and a diversity preservation mechanism on the basis of the standard particle swarm algorithm, enhancing the global search ability and convergence stability. In the algorithm initialization stage, a particle swarm is randomly generated within the parameter space, and each particle represents a candidate solution for a set of anomaly detection threshold parameters. The particle velocity update formula incorporates physical constraint gradient information, enabling the particles to consider both the historical optimal position and the physical constraint requirements during the search process. The adaptive inertia weight is dynamically adjusted according to the current iteration number and fitness value. A larger inertia weight is adopted in the initial stage of the search to maintain the global exploration ability, and the inertia weight is reduced in the later stage of the search to enhance the local convergence accuracy. The diversity preservation mechanism monitors the dispersion degree of the particle swarm. When the particles are overly aggregated, random perturbations are introduced to re-disperse the particle distribution, preventing the algorithm from falling into local optima. The global optimal solution search process finds the parameter combination that minimizes the physical constraint anomaly detection loss function through iterative optimization. The optimal anomaly detection threshold parameter set includes discrimination thresholds and confidence thresholds for various fault types.

[0053] The anomaly detection classifier uses machine learning algorithms such as support vector machines or random forests. After configuring the optimal anomaly detection threshold parameter set, it has the ability to distinguish between normal states and various fault states. The fault type discrimination process inputs the vehicle system fault feature vector into the classifier, and the classifier calculates the classification probability based on the similarity between the feature vector and the samples of each fault type. The discrimination process is based on the position of the feature vector in the high-dimensional feature space, calculating the distance to the decision boundary of each fault type, and the fault type with the closest distance is used as the preliminary classification result. The classifier output includes the classification probability values for each fault type, and the probability values reflect the likelihood of the input feature vector belonging to the corresponding fault type. The role of the optimal threshold parameter is to determine the critical condition for classification decisions. When the classification probability of a certain fault type exceeds the corresponding threshold, it is determined as that type of fault; otherwise, it is determined as the normal state or other fault types. The preliminary fault state classification result includes the fault type label and the corresponding classification probability value, laying the foundation for subsequent confidence evaluation.

[0054] The confidence evaluation process calculates the credibility of the classification result based on the probability distribution and decision boundary distance output by the classifier. A high confidence indicates that the classification result is relatively reliable, and a low confidence indicates the existence of classification uncertainty. The evaluation methods include calculating the difference between the maximum probability and the second-largest probability, analyzing the entropy value of the probability distribution, measuring the distance from the feature vector to the decision boundary, and other indicators. The uncertainty quantification process uses Bayesian inference or Monte Carlo methods to estimate the uncertainty level of the classification result, quantifying the classification uncertainty caused by factors such as data noise, model limitations, and parameter estimation errors. The quantification result is expressed in the form of a confidence interval or an uncertainty score, providing a reliability reference for fault diagnosis decisions. The fault state recognition result integrates the preliminary classification result, confidence evaluation, and uncertainty quantification information, forming a comprehensive diagnosis conclusion that includes the fault type, occurrence probability, confidence level, and uncertainty range.

[0055] In a specific embodiment, the process of executing step S105 may specifically include the following steps: Perform fault root node location processing on the fault status recognition result through the fault propagation path backtracking algorithm to obtain the fault root sensor node identifier; Calculate the fault influence range and perform propagation link tracing processing on the fault propagation graph model according to the fault root sensor node identifier to obtain the fault influence system mapping table; Based on the fault severity grading rule, perform fault risk level assessment and priority sorting processing on the fault influence system mapping table to obtain the graded fault handling strategy table; Input the graded fault handling strategy table into the control strategy library for corresponding control instruction matching and parameter adjustment processing to obtain the initial system adjustment instruction set; Perform execution feasibility verification and instruction optimization processing on the initial system adjustment instruction set to obtain the system adjustment control signal.

[0056] Specifically, the fault propagation path backtracking algorithm is based on the reverse search principle in graph theory. Starting from the sensor node corresponding to the fault status recognition result, it traverses backward along the directed edges in the fault propagation graph to find the initial occurrence location of the fault. The backtracking process adopts a depth-first search strategy. Starting from the current fault sensor node, it recursively visits all possible fault source nodes, and determines the true origin of the fault by analyzing the timing relationship and causal relationship of fault propagation. The algorithm maintains a node access status table to record the access status and fault occurrence timestamp of each sensor node, and identifies the direction and path of fault propagation by comparing the sequence of timestamps. The fault root node location processing is based on a comprehensive judgment of fault propagation probability and timing logic, and selects the sensor node with the earliest fault timestamp and the highest outward fault propagation probability as the root node. The location process also considers the reliability and historical fault records of the sensor nodes to exclude misjudgments caused by the sensor's own faults. The fault root sensor node identifier contains information such as the unique number of the node, the system type it belongs to, the physical location coordinates, and the fault occurrence time, providing an accurate fault source location for subsequent impact analysis and control decisions.

[0057] The fault propagation graph model describes the fault propagation relationships among automotive systems in the form of a directed graph. The nodes in the graph represent sensors or subsystems, and the directed edges indicate the direction and intensity of fault propagation. The calculation and processing of the fault impact range are based on the breadth-first search algorithm of the graph. Starting from the fault root sensor node, it expands and searches layer by layer for all possible downstream nodes that may be affected. During the search process, the impact degree and propagation delay time of each affected node are recorded. The impact degree is calculated based on the fault propagation probability and the length of the propagation path. The propagation link tracing process constructs a complete propagation path from the fault root to each affected node. Each path includes intermediate transfer nodes and the corresponding propagation time delay. The tracing algorithm uses the dynamic programming method to find the shortest path and the highest probability path of fault propagation, providing multiple propagation scenario analyses for fault impact assessment. The fault impact system mapping table organizes the impact analysis results in tabular form. The rows in the table correspond to the affected systems or sensor nodes, and the columns correspond to attribute information such as the impact degree, propagation delay, and propagation path. The mapping table also includes impact type classification, distinguishing direct impact, indirect impact, and potential impact, providing detailed impact analysis data for formulating targeted treatment strategies.

[0058] The fault severity grading rule establishes a multi-level fault hazard level assessment system based on automotive safety standards and engineering experience. The grading criteria consider the degree of impact of faults on vehicle safety, performance, and comfort. Safety-critical faults include faults that directly threaten driving safety, such as brake failure, steering loss of control, and sudden loss of power. The priority is set to the highest level. Performance-impacting faults include faults that affect the normal performance of the vehicle but do not directly threaten safety, such as engine power decline, abnormal gearbox shifting, and reduced fuel efficiency. The priority is set to the medium level. Comfort-impacting faults include faults that only affect the driving and riding comfort, such as abnormal air conditioning, audio system failure, and seat adjustment failure. The priority is set to the lower level. The fault hazard level assessment process determines the hazard level of each affected system according to the impact range and impact degree in the fault impact system mapping table, combined with the grading rule. The priority sorting process arranges all affected systems according to the hazard level and urgency to ensure that safety-critical faults receive the highest treatment priority. The graded fault treatment strategy table integrates the results of hazard level assessment and priority sorting, formulating corresponding treatment strategies and resource allocation plans for each fault impact system.

[0059] The control strategy library stores standard control response schemes for various fault scenarios, including the mapping relationship between fault types and control instructions, parameter adjustment rules, and execution timing arrangements. The control instruction matching process looks up the corresponding control response scheme in the control strategy library according to the fault types and severity levels in the hierarchical fault handling strategy table. The matching process uses a combination of pattern matching and similarity calculation. When no exactly matching strategy is found, the most similar strategy scheme is selected as a reference. The parameter adjustment process makes personalized adjustments to the parameters in the standard control scheme according to the specific characteristics of the current fault and the vehicle operating state. The adjustment process takes into account factors such as the current speed, load, and environmental conditions of the vehicle to ensure that the control instructions adapt to the actual operating conditions. The adjusted parameters include key control parameters such as the controller gain coefficient, actuator action amplitude, and response time constant. The initial system adjustment instruction set contains specific control instructions for each affected system. Each instruction includes the target system identifier, control action type, parameter setting value, and execution time requirement.

[0060] The execution feasibility verification process checks whether each instruction in the initial system adjustment instruction set can be executed safely and effectively under the current conditions. The verification content includes actuator availability check, system state compatibility analysis, and safety constraint satisfaction evaluation. The availability check confirms that the target actuator is in a normal working state and has the ability to execute the specified action. The state compatibility analysis ensures that there are no conflicts or interferences between multiple control instructions. The safety constraint evaluation verifies that the execution of the control instructions will not trigger new safety risks or exacerbate existing faults. The instruction optimization process further optimizes the instructions that pass the feasibility verification, including execution timing optimization, parameter fine-tuning, and conflict resolution. The timing optimization determines the optimal execution order and time interval of each control instruction to avoid system overload caused by simultaneous execution of multiple instructions. The parameter fine-tuning makes fine adjustments to the instruction parameters based on the real-time system state and expected control effect. The conflict resolution process coordinates potential conflicts between multiple instructions. The system adjustment control signal integrates the optimized control instructions to form a standard control signal that can be directly sent to each subsystem controller. The signal format conforms to the communication protocol and interface standard of the automotive electronic system.

[0061] The above describes the method for detecting vehicle faults by fusing sensor data in the embodiments of the present application. Next, the system for detecting vehicle faults by fusing sensor data in the embodiments of the present application will be described. Please refer to Figure 2 One embodiment of the system for detecting vehicle faults by fusing sensor data in the embodiments of the present application includes: A modeling module, configured to perform correlation modeling processing on the multi-system sensor network through the vehicle system topology mapping algorithm to obtain a sensor topology correlation matrix; A synchronization module, configured to perform adaptive time window synchronization processing on the sensor topology association matrix according to vehicle dynamics constraint equations, to obtain a synchronized multi-sensor data set; A fusion module, configured to input the synchronized multi-sensor data set into a fault propagation path perception network for multi-level feature fusion processing, to obtain a vehicle system fault feature vector; A detection module, configured to perform anomaly detection processing on the vehicle system fault feature vector based on an automotive physical model constraint optimizer, to obtain a fault status recognition result; A traceability module, configured to perform traceability positioning and control instruction generation processing on the fault status recognition result according to a fault propagation graph model, to obtain a system adjustment control signal.

[0062] above Figure 2 The automotive fault detection system for fusing sensor data in the embodiments of the present invention is described in detail from the perspective of modular functional entities. Below, the automotive fault detection device for fusing sensor data in the embodiments of the present invention is described in detail from the perspective of hardware processing.

[0063] Refer to Figure 3 , in the embodiments of the present invention, there is also provided an automotive fault detection device for fusing sensor data. The automotive fault detection device for fusing sensor data may be a server, and its internal structure may be as Figure 3 shown. The automotive fault detection device for fusing sensor data includes a processor, a memory, a display screen, an input device, a network interface, and a database connected through a system bus. Among them, the processor of this computer design is used to provide computing and control capabilities. The memory of the automotive fault detection device for fusing sensor data includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the automotive fault detection device for fusing sensor data is used to store the corresponding data in this embodiment. The network interface of the automotive fault detection device for fusing sensor data is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0064] Those skilled in the art can understand that Figure 3 the structure shown in

[0065] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer is caused to execute the steps of the method for detecting vehicle faults by fusing sensor data.

[0066] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, systems, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0067] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an 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. The computer software product is stored in a storage medium and includes several instructions to cause a vehicle fault detection device that fuses sensor data (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0068] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An automobile fault detection method for fusing sensor data, characterized in that, The method includes: Performing correlation modeling processing on the multi-system sensor network through an automotive system topology mapping algorithm to obtain a sensor topology correlation matrix; Performing adaptive time window synchronization processing on the sensor topology correlation matrix according to the vehicle dynamics constraint equation to obtain a synchronized multi-sensor data set; Inputting the synchronized multi-sensor data set into a fault propagation path perception network for multi-level feature fusion processing to obtain a vehicle system fault feature vector; Performing anomaly detection processing on the vehicle system fault feature vector based on an automotive physical model constraint optimizer to obtain a fault state recognition result; Performing traceability positioning and control instruction generation processing on the fault state recognition result according to the fault propagation map model to obtain a system adjustment control signal.

2. The method for detecting vehicle faults by fusing sensor data according to claim 1, wherein The performing correlation modeling processing on the multi-system sensor network through an automotive system topology mapping algorithm to obtain a sensor topology correlation matrix includes: Performing physical connection relationship recognition processing on the sensor nodes of the engine, braking, steering, and transmission systems through CAN bus network topology scanning to obtain an inter-system connection topology diagram; Performing weight assignment processing on the inter-system connection topology diagram according to the functional dependency relationships of each subsystem to obtain a system correlation weight topology diagram; Performing sensor node mapping processing on the system correlation weight topology diagram based on the three-dimensional space coordinates of the sensors and the signal propagation path to obtain a sensor spatial distribution topology diagram; Performing numerical quantization processing on the sensor spatial distribution topology diagram with the sensor physical correlation strength and signal propagation delay parameters to obtain a sensor correlation metric matrix; Performing matrix normalization and symmetry adjustment processing on the sensor correlation metric matrix to obtain the sensor topology correlation matrix.

3. The method for detecting vehicle faults by fusing sensor data according to claim 1, wherein, The performing adaptive time window synchronization processing on the sensor topology correlation matrix according to the vehicle dynamics constraint equation to obtain a synchronized multi-sensor data set includes: Performing real-time acquisition processing on the current vehicle operation state parameters through vehicle speed, acceleration, and angular velocity sensors to obtain a vehicle dynamics state vector; Performing dynamic adjustment calculation processing on a preset basic time window length according to the vehicle dynamics state vector to obtain an adaptive time window parameter; Performing propagation delay compensation calculation processing on the data of each sensor based on the distance between sensors and the signal propagation path in the sensor topology correlation matrix to obtain a sensor delay correction coefficient; Applying the sensor delay correction coefficient to the original data of the multi-system sensors for time series alignment calibration processing to obtain time series synchronized sensor data; Performing segmented cutting and data integrity verification processing on the time series synchronized sensor data according to the adaptive time window parameter to obtain the synchronized multi-sensor data set.

4. The method for detecting vehicle faults by fusing sensor data according to claim 1, wherein The inputting the synchronized multi-sensor data set into a fault propagation path perception network for multi-level feature fusion processing to obtain a vehicle system fault feature vector includes: Inputting the synchronized multi-sensor data set into the sensor topology convolution layer of the fault propagation path perception network for convolution feature extraction processing based on the automotive system topology structure to obtain a topology-aware sensor feature map; The fault propagation modeling layer of the fault propagation path perception network performs time-series dependent fault propagation path prediction processing on the topology-aware sensor feature map to obtain a fault propagation path probability matrix; Based on the multi-system coupling analysis layer of the fault propagation path perception network, the system coupling strength calculation and fault cascade effect modeling processing are performed on the fault propagation path probability matrix to obtain a multi-system coupling fault feature tensor; The multi-system coupling fault feature tensor is input into the adaptive feature fusion layer of the fault propagation path perception network for dynamic feature weighted fusion processing based on the fault severity to obtain a fault severity weighted feature vector; The vehicle physical constraint verification and feature validity screening processing are performed on the fault severity weighted feature vector to obtain the vehicle system fault feature vector.

5. The method for detecting vehicle faults by fusing sensor data according to claim 4, wherein The multi-system coupling analysis layer based on the fault propagation path perception network performs system coupling strength calculation and fault cascade effect modeling processing on the fault propagation path probability matrix to obtain a multi-system coupling fault feature tensor, including: The system coupling strength coefficient matrix is obtained by quantifying the coupling strength between the engine - transmission, brake - ABS, and steering - suspension systems of the fault propagation path probability matrix through the inter-system physical coupling degree calculation module; According to the fault cascade propagation time series model, the propagation delay and attenuation coefficient of the fault between different systems are calculated for the system coupling strength coefficient matrix to obtain a fault cascade propagation parameter matrix; Based on the multi-system fault interaction influence analysis algorithm, the fault interaction influence and amplification effect modeling processing are performed on the fault cascade propagation parameter matrix to obtain a fault interaction influence feature matrix; The fault interaction influence feature matrix is subjected to three-dimensional tensor reconstruction and multi-dimensional fault feature encoding processing to obtain an initial multi-system coupling fault feature tensor; The initial multi-system coupling fault feature tensor is subjected to tensor decomposition and feature dimension optimization processing to obtain the multi-system coupling fault feature tensor.

6. The method for detecting vehicle faults by fusing sensor data according to claim 1, wherein, The vehicle physical model constraint optimizer performs anomaly detection processing on the vehicle system fault feature vector to obtain a fault state recognition result, including: The physical rationality verification processing is performed on the vehicle system fault feature vector through the engine thermodynamics model, brake system dynamics model, and electrical system Ohm's law constraint model to obtain a physical constraint violation metric parameter; According to the physical constraint violation metric parameter, the constraint penalty term construction processing is performed on the anomaly detection objective function to obtain a physical constraint anomaly detection loss function; Based on the improved particle swarm optimization algorithm, the global optimal solution search processing is performed on the physical constraint anomaly detection loss function to obtain an optimal anomaly detection threshold parameter set; The vehicle system fault feature vector is input into an anomaly detection classifier configured with the optimal anomaly detection threshold parameter set for fault type discrimination processing to obtain a preliminary fault state classification result; The confidence evaluation and uncertainty quantification processing are performed on the preliminary fault state classification result to obtain the fault state recognition result.

7. The method for detecting vehicle faults by fusing sensor data according to claim 1, wherein Performing traceability positioning and control instruction generation processing on the fault status recognition result according to the fault propagation graph model to obtain a system regulation control signal, including: Performing fault root node positioning processing on the fault status recognition result through a fault propagation path backtracking algorithm to obtain a fault root sensor node identifier; Calculating the fault influence range and tracing the propagation link of the fault propagation graph model according to the fault root sensor node identifier to obtain a fault influence system mapping table; Performing fault risk level assessment and priority sorting processing on the fault influence system mapping table based on the fault severity grading rule to obtain a graded fault handling strategy table; Inputting the graded fault handling strategy table into a control strategy library for corresponding control instruction matching and parameter adjustment processing to obtain an initial system regulation instruction set; Performing execution feasibility verification and instruction optimization processing on the initial system regulation instruction set to obtain the system regulation control signal.

8. An automotive fault detection system that fuses sensor data, characterized in that, For implementing the vehicle fault detection method for fusing sensor data according to any one of claims 1-7, the vehicle fault detection system for fusing sensor data includes: A modeling module, configured to perform correlation modeling processing on a multi-system sensor network through a vehicle system topology mapping algorithm to obtain a sensor topology correlation matrix; A synchronization module, configured to perform adaptive time window synchronization processing on the sensor topology correlation matrix according to a vehicle dynamics constraint equation to obtain a synchronized multi-sensor data set; A fusion module, configured to input the synchronized multi-sensor data set into a fault propagation path perception network for multi-level feature fusion processing to obtain a vehicle system fault feature vector; A detection module, configured to perform anomaly detection processing on the vehicle system fault feature vector based on a vehicle physical model constraint optimizer to obtain a fault status recognition result; A traceability module, configured to perform traceability positioning and control instruction generation processing on the fault status recognition result according to the fault propagation graph model to obtain a system regulation control signal.

9. An automotive fault detection device that fuses sensor data, characterized in that, Including a memory and a processor, the memory stores a computer program that can run on the processor, and when the processor executes the computer program, it implements the vehicle fault detection method for fusing sensor data according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, the processor is caused to execute the vehicle fault detection method for fusing sensor data according to any one of claims 1 to 7.

Images