Area controller, distributed fault diagnosis system, vehicle and anomaly detection method

Abstract

The invention relates to an area controller, a distributed fault diagnosis system, a vehicle and an anomaly detection method. A central processing unit obtains a data packet of a corresponding area electronic subsystem, and then receives the data packet transmitted by the central processing unit through an artificial intelligence chip which is connected with the central processing unit and carries a pre-trained machine learning model; a machine learning model is operated to identify a communication abnormal mode, generate a health score representing the severity of an abnormal condition and then send the health score to an external central controller, so that a regional controller can perform localization processing and anomaly detection on a data packet of an electronic subsystem in a corresponding region of a vehicle; the data volume needing to be transmitted to an external central controller is reduced so as to reduce the data transmission delay and the processing pressure of the external central controller, and at the same time, with the help of the recognition capability of a machine learning model to a communication abnormal mode, the early abnormality which does not reach a traditional fault threshold is captured. And support is provided for predictive maintenance so as to reduce safety risks caused by potential fault deterioration.

Description

Technical Field

[0001] This application relates to the fields of automotive electronics and network communication, specifically to area controllers, distributed fault diagnosis systems, vehicles, and anomaly detection methods. Background Technology

[0002] As the automotive industry undergoes a profound transformation towards intelligent and electric vehicles, the number of electronic subsystems such as sensors and electronic control units (ECUs) in vehicles has increased dramatically, leading to an explosive growth in the frequency and volume of data interactions between these subsystems. Against this backdrop, the core requirements for vehicle fault diagnosis have become increasingly clear: the need to quickly detect communication anomalies in electronic subsystems, and more importantly, to proactively identify early "soft faults" that have not yet triggered traditional fault thresholds, such as sensor signal drift and slight increases in ECU communication delays. This is crucial to prevent potential fault escalation and the resulting safety risks, and to provide data support for predictive maintenance.

[0003] In related technologies, vehicle fault diagnosis mainly adopts a centralized fault diagnosis system. This system is centered on a central controller such as a central gateway (CGW). The operating data of sensors and ECUs in all areas must be collected by the central processor through vehicle networks such as Controller Area Network (CAN) and Ethernet. The central processor uniformly completes data parsing, fault logic judgment, and diagnostic result output. At the same time, the system relies on a preset fault code (DTC) library to match obvious faults that meet the fault code triggering conditions, so as to realize fault location and alarm prompts.

[0004] However, traditional centralized fault diagnosis systems require diagnostic data to be transmitted to the central processor for analysis. The combined delays in transmission and processing lead to delayed detection of communication anomalies, failing to meet the requirements for rapid fault response. The influx of massive amounts of real-time data into the central processor can exceed its processing capacity, easily causing overload and system crashes, thus compromising the stability of the vehicle's electronic systems. Furthermore, relying on preset fault codes and failing to reach thresholds for soft fault parameters makes it impossible to identify potential faults and hinders predictive maintenance. Summary of the Invention

[0005] This application provides a regional controller, a distributed fault diagnosis system, a vehicle, and an anomaly detection method, which can solve the technical problems that traditional centralized fault diagnosis systems cannot meet the requirements for rapid fault response, are prone to overload and crash, and cannot identify potential faults, making it difficult to achieve predictive maintenance.

[0006] In a first aspect, embodiments of this application provide a region controller, which includes: Ethernet transceiver clusters and controller area network transceivers are used to establish connections with sensors and electronic control units of the electronic subsystems in the corresponding area of ​​the vehicle, respectively. The central processing unit, which is connected to the Ethernet transceiver cluster via an Ethernet switch and to the controller LAN transceiver, is used to acquire data packets from the electronic subsystem. An artificial intelligence chip is connected to the central processing unit. The artificial intelligence chip is equipped with a pre-trained machine learning model. It is used to receive data packets transmitted by the central processing unit, run the machine learning model to identify communication anomaly patterns, generate a health score that characterizes the severity of the anomaly, and send the health score to an external central controller.

[0007] In conjunction with the first aspect, in one embodiment, the area controller further includes: An Ethernet transceiver is provided, which is connected to the Ethernet switch and used to connect to an external central controller.

[0008] In conjunction with the first aspect, in one embodiment, the central processing unit integrates a data packet preprocessing module, which is used to remove invalid and redundant data from the electronic subsystem data packets and calculate the transmission-related parameters of the data packets to generate valid data packet features for anomaly detection by the artificial intelligence chip.

[0009] In conjunction with the first aspect, in one embodiment, the artificial intelligence chip integrates an anomaly information generation module. When the health score generated by the artificial intelligence chip reaches a preset anomaly response condition, the anomaly information generation module generates information including the device identifier of the corresponding electronic subsystem, the anomaly type, and the health score, and transmits the information along with the health score to an external central controller.

[0010] In conjunction with the first aspect, in one implementation, the Ethernet transceiver cluster and the controller LAN transceiver are used to collect statistical attributes of data packets from the electronic subsystem; the statistical attributes include at least three of the following: data transmission rate, data latency, latency fluctuation, data loss, data error, data buffer usage, vehicle mileage, driving scenario information, vehicle speed, and ambient temperature.

[0011] Secondly, embodiments of this application provide a distributed fault diagnosis system, which includes at least one area controller as described above, and a central controller; The area controller is connected to the central controller, and each area controller corresponds to a different functional area of ​​the vehicle. Each of the aforementioned area controllers establishes connections with the electronic subsystems within its corresponding functional area through its own Ethernet transceiver cluster and controller area network transceiver; The central controller can receive health scores sent by each of the regional controllers and generate corresponding alarm information or function restriction instructions based on the health scores.

[0012] Thirdly, embodiments of this application provide a vehicle that includes the distributed fault diagnosis system described above.

[0013] Fourthly, embodiments of this application provide an anomaly detection method for a regional controller, which includes the following steps: Data packets exchanged between the area controller and the electronic subsystems within the corresponding functional area are acquired through Ethernet transceiver clusters and controller LAN transceivers; The AI ​​chip runs a pre-trained machine learning model to analyze the statistical properties of data packets, identify abnormal communication patterns, and generate a health score that characterizes the severity of the anomaly. The data packets are processed by the central processing unit and then transmitted to the AI ​​chip. The central processor sends communication anomaly patterns and health scores, which characterize the severity of the anomalies, to an external central controller.

[0014] In conjunction with the fourth aspect, in one embodiment, prior to acquiring data packets exchanged between the area controller and the electronic subsystems within the corresponding functional area via the Ethernet transceiver cluster and the controller LAN transceiver, the method further includes: Data packets of the electronic subsystems under different usage conditions and different device connection conditions were collected on the test vehicle. Training samples are constructed using time series processing techniques. Each sample includes multiple sets of data packet feature data at different time dimensions, as well as corresponding anomaly severity annotation information. A model architecture including a multi-layer feature learning network and an anti-overfitting layer is adopted to train the model on training samples until the prediction accuracy of the model reaches the preset requirements.

[0015] In conjunction with the fourth aspect, in one embodiment, after sending the communication anomaly pattern and a health score characterizing the severity of the anomaly to an external central controller via the central processor, the method further includes: The central controller executes corresponding response operations based on the received health score; among these response operations, at least one of the following is performed: recording system operation logs, adding targeted inspection prompts to the maintenance plan, triggering alarm information of different levels, restricting vehicle functions that rely on the corresponding electronic subsystems, and recommending service stations.

[0016] The beneficial effects of the technical solutions provided in this application include: The central processing unit (CPU) is connected to an Ethernet transceiver cluster via an Ethernet switch and to a controller area network (CLAN) transceiver to obtain data packets from the electronic subsystems in the area. Then, an AI chip connected to the CPU and equipped with a pre-trained machine learning model receives the data packets transmitted by the CPU. The machine learning model identifies communication anomaly patterns and generates a health score representing the severity of the anomaly before sending it to the external central controller. This allows the area controller to perform localized processing and anomaly detection of data packets from the electronic subsystems within the corresponding vehicle area, reducing the amount of data that needs to be transmitted to the external central controller, thus reducing data transmission latency and the processing load on the external central controller. Simultaneously, by leveraging the machine learning model's ability to identify communication anomaly patterns, early anomalies that have not yet reached traditional fault thresholds can be detected, supporting predictive maintenance and reducing safety risks caused by potential fault escalation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a structural block diagram of the area controller in an embodiment of this application; Figure 2 This is a structural block diagram of the distributed fault diagnosis system in the embodiments of this application; Figure 3 This is a flowchart illustrating the anomaly detection method for the area controller in this application embodiment. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0020] This application provides a regional controller, a distributed fault diagnosis system, a vehicle and anomaly detection method, which can solve the technical problems of traditional centralized fault diagnosis systems failing to meet the requirements for rapid fault response, easily causing overload crashes, failing to identify potential faults, and making it difficult to achieve predictive maintenance.

[0021] Firstly, such as Figure 1As shown in the illustration, this application provides a regional controller, comprising: an Ethernet transceiver cluster (including multiple Ethernet transceivers that collect data from multiple sensors of an electronic subsystem) and a controller area network transceiver, which are respectively used to establish connections with the sensors and electronic control units of the electronic subsystem within the corresponding area of ​​the vehicle; a central processing unit (CPU), which is connected to the Ethernet transceiver cluster via an Ethernet switch and to the CPU, and is used to acquire data packets from the electronic subsystem; and an artificial intelligence chip, which is connected to the CPU and carries a pre-trained machine learning model, which is used to receive data packets transmitted by the CPU, run the machine learning model to identify communication anomaly patterns, generate a health score characterizing the severity of the anomaly, and send the health score to an external central controller.

[0022] In this embodiment, a region controller is set up in the corresponding area of ​​the vehicle. Its Ethernet transceiver group and controller area network transceiver are connected to the sensors and electronic control units of the electronic subsystems within that area. The central processing unit (CPU) connects to the Ethernet transceiver group via an Ethernet switch and to the controller area network transceiver to obtain data packets from the electronic subsystems in that area. An AI chip connected to the CPU and equipped with a pre-trained machine learning model receives the data packets transmitted by the CPU, runs the machine learning model to identify communication anomaly patterns, generates a health score representing the severity of the anomaly, and sends it to the external central controller. This allows the region controller to perform localized processing and anomaly detection of data packets from the electronic subsystems within the corresponding area of ​​the vehicle, reducing the amount of data that needs to be transmitted to the external central controller, thus reducing data transmission latency and the processing pressure on the external central controller. Simultaneously, by leveraging the machine learning model's ability to identify communication anomaly patterns, early anomalies that have not reached traditional fault thresholds can be detected, thereby accelerating the detection and response speed of communication anomalies, preventing the external CPU from being overloaded by massive data inflows, and providing support for predictive maintenance to reduce the safety risks caused by potential fault deterioration.

[0023] In conjunction with the first aspect, in one embodiment, the area controller further includes: a separate Ethernet transceiver connected to an Ethernet switch and used for connecting to an external central controller.

[0024] In conjunction with the first aspect, in one embodiment, the central processing unit integrates a data packet preprocessing module. The data packet preprocessing module is used to remove invalid and redundant data from the data packets of the electronic subsystem and calculate the transmission-related parameters of the data packets to generate valid data packet features for anomaly detection by the artificial intelligence chip.

[0025] In this embodiment, the central processing unit of the area controller integrates a data packet preprocessing module. This data packet preprocessing module processes the acquired electronic subsystem data packets, removes invalid and redundant data, calculates the transmission-related parameters of the data packets, and generates valid data packet features for the artificial intelligence chip to perform anomaly detection. The artificial intelligence chip then runs a machine learning model based on these valid data packet features, reducing the interference of invalid data on the anomaly detection process and lowering the processing load of the artificial intelligence chip.

[0026] In conjunction with the first aspect, in one implementation, the artificial intelligence chip integrates an anomaly information generation module. When the health score generated by the artificial intelligence chip reaches a preset anomaly response condition, the anomaly information generation module generates information including the device identifier of the corresponding electronic subsystem, the anomaly type, and the health score, and transmits this information along with the health score to an external central controller.

[0027] In this embodiment, the AI ​​chip of the area controller integrates an anomaly information generation module. When the health score generated by the AI ​​chip reaches the preset anomaly response conditions, the anomaly information generation module generates information including the device identifier of the corresponding electronic subsystem, the anomaly type and the health score, and transmits the information together with the health score to the external central controller, so that the external central controller can obtain the specific content related to the anomaly based on the information.

[0028] In conjunction with the first aspect, in one implementation, the Ethernet transceiver cluster and the controller area network transceiver are used to collect statistical attributes of data packets from the electronic subsystem; the statistical attributes include at least three of the following: data transmission rate, data latency, latency fluctuation, data loss, data error, data buffer usage, vehicle mileage, driving scenario information, vehicle speed, and ambient temperature.

[0029] In this embodiment, the Ethernet transceiver cluster and controller LAN transceiver of the area controller are used to collect statistical attributes of data packets of the electronic subsystem. These statistical attributes include at least three of the following: data transmission rate, data latency, latency fluctuation, data loss, data error, data cache usage, vehicle mileage, driving scenario information, vehicle speed, and ambient temperature. This provides multi-dimensional data for the artificial intelligence chip to run machine learning models to identify abnormal communication patterns.

[0030] Secondly, such as Figure 2As shown in the figure, this application provides a distributed fault diagnosis system, which includes at least one or more regional controllers and a central controller; the regional controllers are connected to the central controller, and each regional controller corresponds to a different functional area of ​​the vehicle; each regional controller establishes a connection with the electronic subsystem in the corresponding functional area through its own Ethernet transceiver group and controller area network transceiver; the central controller can receive health scores sent by each regional controller and generate corresponding alarm information or function restriction instructions based on the health scores.

[0031] In this embodiment, the distributed fault diagnosis system includes at least one of the aforementioned regional controllers and a central controller. The regional controllers are connected to the central controller, and each regional controller corresponds to a different functional area of ​​the vehicle. Each regional controller establishes a connection with the electronic subsystems in the corresponding functional area through its own Ethernet transceiver group and controller area network transceiver. The central controller can receive the health scores sent by each regional controller and generate corresponding alarm information or function restriction instructions based on the health scores.

[0032] Taking the left front area controller as an example, its hardware configuration and connection relationship are as follows:

[0033] I. Hardware Configuration and Connection Relationship of the Intelligent Left Front Zone Controller The left front area controller uses a multi-core ARM processor as an embedded central processing unit (CPU) and integrates a neural network processor (NPU, as a machine learning accelerator, i.e., an artificial intelligence chip). The controller has an Ethernet port (corresponding to the Ethernet transceiver group in the aforementioned embodiment) and a controller area network (CAN, controller area network) interface (corresponding to the controller area network transceiver in the aforementioned embodiment), which establishes connections with the camera, radar, door electronic control unit (ECU, electronic control unit), and seat electronic control unit (ECU, electronic control unit) in the left front area, respectively. The controller also integrates an Ethernet switch for data interaction and distribution.

[0034] II. Data Processing and Anomaly Detection Procedure for the Left Front Area Controller Taking Ethernet packet processing as an example, the Ethernet switch on the controller mirrors the high-speed data streams from the camera and other peripherals to the embedded central processing unit (CPU) through pre-configured access control list (ACL) rules. The embedded CPU preprocesses the data packets (such as calculating packet rate and packet loss), extracts the data packet feature data, and then transmits it to the artificial intelligence chip (NPU). The artificial intelligence chip (NPU) runs a pre-trained Long Short-Term Memory (LSTM) network model, which is a variant of the recurrent neural network (RNN). It has learned the baseline of normal vehicle communication traffic patterns and can identify abnormal communication patterns such as abnormal decreases in camera frame rate and abnormal increases in radar packet intervals, and outputs a health score of 0-100. When the health score is below 70, the embedded CPU generates a message containing the device identifier (ID), the abnormality type, and the health score, and sends it to the external central controller.

[0035] III. Core Features and Advantages of the LSTM Model The core design of the LSTM model addresses the "long-term dependency" problem that traditional recurrent neural networks (RNNs) tend to forget early information when processing long-sequence data. Compared to simple threshold rules, its advantages are: first, in the context-aware process, it can understand cross-time causal patterns such as "latency increases slowly first, and then the packet loss rate starts to rise," rather than just judging whether a single indicator exceeds the limit; second, it can distinguish between brief network jitter and real fault trends, reducing false alarms and having strong noise resistance; and third, through online learning or periodic fine-tuning, the model can adapt to the normal behavioral baseline drift caused by vehicle aging, maintaining long-term detection accuracy.

[0036] IV. Model Training and Data Processing for Predicting Camera Connector Loosening Faults

[0037] 1. Data Collection

[0038] Data was systematically collected on multiple test vehicles, with a mileage ranging from 50,000 km to 250,000 km. The camera connectors were artificially loosened in four stages (simulating different degrees of connector pull-out). The data collection frequency was 100 milliseconds (ms) to ensure the capture of instantaneous anomalies. The data vector of each time step (every 100ms) contained 10 features: packet rate (fps, frames per second), average latency (ms, milliseconds), latency jitter (ms, milliseconds), number of packet losses, number of cyclic redundancy check (CRC) errors, buffer utilization rate (%), mileage (10,000 km), scene code (0-3, corresponding to highway, urban, off-road, etc., respectively, which is determined and provided by the central controller), vehicle speed (km / h, kilometers per hour), and ambient temperature (°C, degrees Celsius).

[0039] 2. Training Sample Construction

[0040] A sliding window technique is used to construct time-series training samples, with a window size of 300 time steps (i.e., 30 seconds). A single training sample i is constructed as follows: the input X_i is a 300×10 matrix (containing 300 time steps, each time step corresponding to 10-dimensional feature data), and the output Y_i is a scalar (i.e., the true health score corresponding to the last time step t=300 of the window, which is assigned a true health score of 0-100 by engineers based on camera image quality, diagnostic tool reports, and physical examination results). An example is shown below:

[0041] Input X (partial time step):

[0042] Time step 1: [30.0, 10.1, 0.5, 0, 0, 5%, 15.2, 1, 45, 25]

[0043] Time step 2: [29.8, 10.3, 0.6, 0, 0, 6%, 15.2, 1, 44, 25]

[0044] ...

[0045] Time step 300: [25.1, 35.6, 12.4, 15, 8, 65%, 15.2, 1, 42, 26] Output Y: 40 (the actual health score corresponding to this time window)

[0046] 3. Model Architecture

[0047] Input layer: Receives time step tensors of batch size, with dimensions [BatchSize, 300, 10];

[0048] LSTM layer: A two-layer stacked LSTM structure is used, with 128 hidden units in each layer, which is responsible for learning long-term dependencies in time series.

[0049] Dropout layer: Set the dropout rate to 0.3 to prevent the model from overfitting;

[0050] Fully connected layer: The output of the LSTM layer is mapped to a single neuron, and the output is scaled to between 0 and 100 by the Sigmoid activation function as a predicted health score.

[0051] 4. Training Cycle

[0052] The LSTM model gradually builds its fault diagnosis capabilities by analyzing thousands of time series samples mentioned above: In the initial training phase, the model identifies a simple correlation between "increased CRC error count" and "decreased health score"; in the intermediate deepening phase, it can identify a compound fault mode of "increased latency + increased jitter + decreased packet rate"; in the advanced cognitive phase, it can understand the higher risk weight of "continuous increase in CRC error for high-mileage vehicles in urban road conditions"; after 100 training rounds, the model's prediction accuracy on the test set reaches over 90%.

[0053] V. Real Vehicle Data Analysis and System Response Case Studies

[0054] Taking a real-world vehicle scenario as an example, with a time window of 14:30:00-14:30:30, the key indicators collected by the left front area controller changed as follows: packet rate decreased from 29fps to 26fps (continuously decreasing), communication latency increased from 12ms to 28ms (significantly increasing), CRC errors increased from 0 to 5 (continuously occurring), and jitter increased from 1.2ms to 7.5ms (drastic fluctuation). The vehicle was on a highway with a total mileage of 203,000 kilometers. During the LSTM model inference process, the fault characteristics of "continuous occurrence of CRC errors + significant increase in latency" were identified, and the fault risk level was assessed in conjunction with the "high mileage + highway" scenario. Finally, a health score of 58 points was output (belonging to the "warning" level). The system then executed the following response operations: the dashboard displayed "Forward camera performance degraded, please use intelligent driving functions with caution," and a warning message was sent to the fleet management system: "Left front camera connector is suspected to be loose, it is recommended to check it first during the next maintenance." At the same time, the fault mode was recorded for subsequent model optimization and fault analysis.

[0055] Thirdly, embodiments of this application provide a vehicle that includes the above-described distributed fault diagnosis system.

[0056] In this embodiment, the provided vehicle includes the distributed fault diagnosis system of the second aspect mentioned above. Each area controller in the distributed fault diagnosis system corresponds to a different functional area of ​​the vehicle. Fault detection is achieved by connecting with the electronic subsystems in the corresponding functional area. The central controller generates alarm information or function restriction instructions based on the health score sent by the area controller to adapt to the vehicle's operating requirements.

[0057] Fourthly, such as Figure 3 As shown in the figure, this application embodiment provides an anomaly detection method for a regional controller, which includes the following steps: S100: Acquires data packets exchanged between the area controller and the electronic subsystems within the corresponding functional area via Ethernet transceiver clusters and controller area network transceivers; S200: Runs a pre-trained machine learning model through an artificial intelligence chip to analyze the statistical properties of data packets, identify abnormal communication patterns, and generate a health score that characterizes the severity of the anomaly; the data packets are processed by the central processing unit and then transmitted to the artificial intelligence chip. S300: The central processor sends the communication anomaly pattern and the health score representing the severity of the anomaly to the external central controller.

[0058] In this embodiment, data packets between the area controller and the electronic subsystems within the corresponding functional area are acquired through an Ethernet transceiver cluster and a controller LAN transceiver. The data packets acquired by the controller LAN transceiver are sent to the central processing unit via an Ethernet switch. Then, a pre-trained machine learning model is run by an artificial intelligence chip to analyze the statistical attributes of the data packets, identify communication anomaly patterns, and generate a health score that characterizes the severity of the anomaly. Finally, the central processing unit sends the communication anomaly pattern and the health score to the external central controller.

[0059] In conjunction with the fourth aspect, in one embodiment, before S100, there is also S000, which includes the following steps: S000-1: Data packets of electronic subsystems under different usage conditions and different device connection conditions are collected on the test vehicle; S000-2: Training samples are constructed using time series processing techniques. Each sample includes multiple sets of data packet feature data in different time dimensions, as well as corresponding anomaly severity labeling information. S000-3: A model architecture including a multi-layer feature learning network and an anti-overfitting layer is adopted to train the training samples until the prediction accuracy of the model reaches the preset requirements.

[0060] In this embodiment, before acquiring data packets of interaction between the area controller and the electronic subsystems within the corresponding functional area through the Ethernet transceiver cluster and the controller LAN transceiver, a model training step is also included: collecting data packets of the electronic subsystems under different usage states and different device connection states on the test vehicle; constructing training samples using time series processing technology, with each sample including multiple sets of data packet feature data in different time dimensions, as well as corresponding anomaly severity labeling information; and training the training samples using a model architecture including a multi-layer feature learning network and an anti-overfitting processing layer until the model's prediction accuracy reaches the preset requirements.

[0061] In conjunction with the fourth aspect, in one implementation, after S300, the following step is further included: S400: The central controller executes corresponding response operations based on the received health score; wherein, the response operations include at least one of the following: recording system operation logs, adding targeted inspection prompts to the maintenance plan, triggering alarm information of different levels, restricting vehicle functions that depend on the corresponding electronic subsystems, and recommending service stations.

[0062] In this embodiment, after the central processor sends the communication anomaly mode and the health score representing the severity of the anomaly to the external central controller, the method further includes: the central controller performing a corresponding response operation based on the received health score. The response operation includes at least one of the following: recording system operation logs, adding targeted inspection prompts to the maintenance plan, triggering alarm information of different levels, restricting vehicle functions that rely on the corresponding electronic subsystem, and recommending service stations.

[0063] In the description of this application, it should be noted that the terms "upper," "lower," etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Unless otherwise expressly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two elements. For those skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0064] It should be noted that in this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0065] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A regional controller, characterized in that, It includes: Ethernet transceiver clusters and controller area network transceivers are used to establish connections with sensors and electronic control units of the electronic subsystems in the corresponding area of ​​the vehicle, respectively. The central processing unit, which is connected to the Ethernet transceiver cluster via an Ethernet switch and to the controller LAN transceiver, is used to acquire data packets from the electronic subsystem. An artificial intelligence chip is connected to the central processing unit. The artificial intelligence chip is equipped with a pre-trained machine learning model. It is used to receive data packets transmitted by the central processing unit, run the machine learning model to identify communication anomaly patterns, generate a health score that characterizes the severity of the anomaly, and send the health score to an external central controller.

2. The area controller as described in claim 1, characterized in that, The area controller also includes: An Ethernet transceiver is provided, which is connected to the Ethernet switch and used to connect to an external central controller.

3. The area controller as described in claim 1, characterized in that, The central processing unit integrates a data packet preprocessing module, which is used to remove invalid and redundant data from the data packets of the electronic subsystem and calculate the transmission-related parameters of the data packets to generate valid data packet features for the artificial intelligence chip to perform anomaly detection.

4. The area controller as described in claim 1, characterized in that, The artificial intelligence chip integrates an anomaly information generation module. When the health score generated by the artificial intelligence chip reaches the preset anomaly response conditions, the anomaly information generation module generates information including the device identifier of the corresponding electronic subsystem, the anomaly type, and the health score, and transmits this information along with the health score to an external central controller.

5. The area controller as described in claim 1, characterized in that, The Ethernet transceiver cluster and controller area network transceiver are used to collect statistical attributes of data packets from the electronic subsystem; the statistical attributes include at least three of the following: data transmission rate, data latency, latency fluctuation, data loss, data error, data buffer usage, vehicle mileage, driving scenario information, vehicle speed, and ambient temperature.

6. A distributed fault diagnosis system, characterized in that, It includes at least one area controller as claimed in any one of claims 1-5, and a central controller; The area controller is connected to the central controller, and each area controller corresponds to a different functional area of ​​the vehicle. Each of the aforementioned area controllers establishes connections with the electronic subsystems within its corresponding functional area through its own Ethernet transceiver cluster and controller area network transceiver; The central controller can receive health scores sent by each of the regional controllers and generate corresponding alarm information or function restriction instructions based on the health scores.

7. A vehicle, characterized in that, It includes the distributed fault diagnosis system as described in claim 6.

8. An anomaly detection method for a regional controller, characterized in that, It includes the following steps: Data packets exchanged between the area controller and the electronic subsystems within the corresponding functional area are acquired through Ethernet transceiver clusters and controller LAN transceivers; The AI ​​chip runs a pre-trained machine learning model to analyze the statistical properties of data packets, identify abnormal communication patterns, and generate a health score that characterizes the severity of the anomaly. The data packets are processed by the central processing unit and then transmitted to the AI ​​chip. The central processor sends communication anomaly patterns and health scores, which characterize the severity of the anomalies, to an external central controller.

9. The anomaly detection method for a regional controller as described in claim 8, characterized in that, Before acquiring data packets exchanged between the area controller and the electronic subsystems within the corresponding functional area via the Ethernet transceiver cluster and controller area network transceivers, the method further includes: Data packets of the electronic subsystems under different usage conditions and different device connection conditions were collected on the test vehicle. Training samples are constructed using time series processing techniques. Each sample includes multiple sets of data packet feature data at different time dimensions, as well as corresponding anomaly severity annotation information. A model architecture including a multi-layer feature learning network and an anti-overfitting layer is adopted to train the model on training samples until the prediction accuracy of the model reaches the preset requirements.

10. The anomaly detection method for a regional controller as described in claim 8, characterized in that, After the central processor sends the communication anomaly pattern and health score characterizing the severity of the anomaly to the external central controller, the process further includes: The central controller executes corresponding response operations based on the received health score; among these response operations, at least one of the following is performed: recording system operation logs, adding targeted inspection prompts to the maintenance plan, triggering alarm information of different levels, restricting vehicle functions that rely on the corresponding electronic subsystems, and recommending service stations.

Images