Vehicle remote intelligent diagnosis method, electronic equipment and program product

Abstract

The invention discloses a vehicle remote intelligent diagnosis method, electronic equipment and a program product, and the method comprises the steps: controlling a vehicle end to collect the multi-modal data of a vehicle through a multi-source sensor, and carrying out the edge processing of the multi-modal data; the control cloud receives the encrypted data from the vehicle end, performs multi-modal data fusion and feature extraction, performs anomaly detection and prediction based on an AI model, and generates an early warning; the control cloud aggregates local model parameters of a plurality of vehicles through a federated learning technology, updates a global AI model, and issues the global AI model to the vehicles through OTA; the control cloud generates an optimal maintenance strategy through a dynamic maintenance decision optimization engine on the basis of a fault diagnosis result in combination with real-time information; and pushing the maintenance strategy to a user side and a server side, and executing corresponding maintenance operation. According to the invention, the potential risk of the vehicle can be predicted through the AI model, and predictive maintenance is realized.

Description

Technical Field

[0001] This invention relates to the field of vehicle remote diagnostics technology, and in particular to a vehicle remote intelligent diagnostics method, electronic device, and program product. Background Technology

[0002] Remote vehicle diagnostics is one of the core functions of modern automobiles, especially intelligent connected vehicles. It enables real-time or near real-time monitoring, analysis, and early warning of vehicle health status through integrated sensors, communication modules, and cloud platforms.

[0003] However, in current technologies, most systems only upload the code and a simple data snapshot after a vehicle fault code occurs. Diagnostic action begins after the fault has occurred, by which time the user has already perceived the problem. Furthermore, diagnosis relies on single-dimensional CAN bus data, lacking correlation analysis with multi-dimensional information such as the vehicle environment (video, audio, geolocation) and user driving behavior, resulting in a high false alarm rate. Moreover, remote center technicians still need to manually analyze data, which is inefficient and heavily reliant on personal experience, making it difficult to quickly locate complex or intermittent faults. Additionally, the inability to accurately predict component lifespan leads to inconsistencies in spare parts inventory management and technician dispatch, resulting in high repair costs. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a remote intelligent diagnostic method, electronic device and program product for vehicles.

[0005] The present invention provides a method for remote intelligent diagnosis of vehicles, comprising: The vehicle control unit collects multimodal data of the vehicle through multiple source sensors and performs edge processing on the multimodal data; The cloud controls the reception of encrypted data from the vehicle, performs multimodal data fusion and feature extraction, and uses AI models to detect and predict anomalies, generating early warnings. The cloud-based control system aggregates local model parameters from multiple vehicles using federated learning technology, updates the global AI model, and sends the updated data to the vehicles via OTA. Based on fault diagnosis results and combined with real-time information, the cloud-based control system generates the optimal maintenance strategy through a dynamic maintenance decision optimization engine. The maintenance strategy is pushed to the user terminal and the server terminal, and the corresponding maintenance operations are executed.

[0006] In one of the optional technical solutions, the control vehicle terminal collects multimodal data of the vehicle through multi-source sensors and performs edge processing on the multimodal data, including: The system collects structured and unstructured data, including CAN bus data and GPS location data, and unstructured data including camera footage, microphone audio, and vibration signals. Using the lightweight AI chip built into the vehicle's T-Box, a local AI model is run to perform real-time cleaning, compression, and preliminary analysis of the raw data.

[0007] In one of the optional technical solutions, the multimodal data fusion and feature extraction in the cloud, and the anomaly detection and prediction based on the AI ​​model, include: By aligning data from different sources and frequencies on a timeline, a unified panoramic view of vehicle status is formed. Using deep recurrent neural networks or Transformer models, compare current data with health benchmarks generated by vehicle digital twin models, or compare them with fleet historical data, to identify deviations and issue early warnings; Among them, the anomaly detection methods include at least one of reconstruction-based anomaly detection, time-series prediction-based anomaly detection, or ultra-lightweight dynamic threshold monitoring.

[0008] In one of the optional technical solutions: The reconstruction-based anomaly detection includes: Use an autoencoder or variational autoencoder to learn a compressed representation of normal data; During inference, the reconstruction error is used to determine whether the data is abnormal; The anomaly detection based on time-series prediction includes: Use LSTM or temporal convolutional networks to predict future data based on past sequence data; Anomalies are determined by the deviation between the actual value and the predicted value.

[0009] In one of the alternative technical solutions, the control cloud aggregates local model parameters from multiple vehicles using federated learning technology, updates the global AI model, and distributes it to the vehicles via OTA, including: Each vehicle trains a global model locally using the new data and obtains local model parameter updates. The encrypted local model parameters are updated and uploaded to the cloud aggregation server; The cloud uses a secure aggregation protocol to calculate the average update and update the global model. The updated global model will be delivered to the vehicle via OTA.

[0010] In one of the alternative technical solutions, the federated learning includes the following steps: When the k-th vehicle uses new data locally, the global model is trained to obtain a local update; The vehicle control unit uploads the encrypted model update to the cloud aggregation server; Control the cloud to use a secure aggregation protocol to calculate average updates and update the global model; The updated global model is then distributed to the fleet vehicles via OTA, completing one iteration.

[0011] In one of the optional technical solutions: In the process of generating the optimal maintenance strategy through the dynamic maintenance decision optimization engine, the model is a Markov decision process, and the state includes fault type, fault severity level, vehicle location, remaining warranty period, list of nearby service centers and their real-time status. Actions include recommending immediate repair, recommending repair within one week, recommending a repair center, and triggering an OTA software update; The reward function calculates the total reward based on successful repair, user feedback, repair delay costs, and repair economic costs.

[0012] The present invention provides an electronic device, including a memory, a processor, and an electronic device program on the memory, wherein the processor executes the electronic device program to implement the steps of any of the aforementioned vehicle remote intelligent diagnostic methods.

[0013] The present invention provides an electronic device readable storage medium storing an electronic device program / instructions thereon, which, when executed by a processor, implements the steps of any of the aforementioned vehicle remote intelligent diagnostic methods.

[0014] The present invention provides an electronic device program product, including an electronic device program / instruction, which, when executed by a processor, implements the steps of any of the aforementioned vehicle remote intelligent diagnostic methods.

[0015] The above technical solution has the following beneficial effects: In the vehicle remote intelligent diagnostic method provided by this invention, various types of data are collected at the vehicle end through multiple different sensors, which can comprehensively grasp the vehicle status. In addition, edge processing is performed at the vehicle end first, so that not all raw data is transmitted to the cloud, which greatly reduces the pressure of data transmission. Then, anomaly detection and prediction are performed through AI model to generate early warnings, so as to predict vehicle risks in advance. The repair plan is pushed to both the car owner and the repair shop at the same time. Both parties can obtain information synchronously. The repair shop can prepare parts and work positions in advance, and the car owner can arrange time in time without repeated communication, making the repair operation more efficient. Attached Figure Description

[0016] The disclosure of this invention will become more readily understood by referring to the accompanying drawings. It should be understood that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings: Figure 1 A flowchart illustrating the process of a remote intelligent vehicle diagnostic method according to an embodiment of the present invention; Figure 2 A schematic diagram of the vehicle-mounted terminal, cloud platform, and repair shop terminal provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0018] like Figure 1 The image shows a remote intelligent diagnostic method for vehicles provided by an embodiment of the present invention, comprising: Step S101: Control the vehicle to collect multimodal data of the vehicle through multi-source sensors, and perform edge processing on the multimodal data; Step S102: Control the cloud to receive encrypted data from the vehicle, perform multimodal data fusion and feature extraction, perform anomaly detection and prediction based on the AI ​​model, and generate early warnings; Step S103: The cloud control system aggregates local model parameters of multiple vehicles through federated learning technology, updates the global AI model, and sends it to the vehicles via OTA. Step S104: The cloud control system generates the optimal maintenance strategy based on the fault diagnosis results and real-time information through a dynamic maintenance decision optimization engine. Step S105: Push the maintenance strategy to the user terminal and the server terminal, and execute the corresponding maintenance operation.

[0019] Specifically, the method of the present invention can be applied to a vehicle remote diagnostic system with a "vehicle-edge-cloud" collaborative architecture, covering multiple terminals such as vehicle terminal, cloud platform, user terminal and server terminal. Through the collaborative work of multiple modules, it can realize the proactive prediction, accurate diagnosis and intelligent decision-making of vehicle faults, and complete the closed-loop process from data collection to maintenance execution without human intervention.

[0020] like Figure 2The diagram illustrates the structure of the vehicle-mounted terminal, cloud platform, and repair shop terminal. The vehicle-mounted terminal is responsible for multi-range data acquisition, edge computing nodes, and local diagnostic agents. The cloud platform comprises a digital twin layer, an intelligent diagnostic core layer, and an automation layer. After initial anomaly detection and early warning at the vehicle-mounted terminal, encrypted feature data is transmitted to the digital twin layer. The digital twin layer then transmits residual / anomaly data to the intelligent diagnostic core layer. The intelligent diagnostic core layer interacts and updates with the automation layer. After optimizing the diagnostic decision, the intelligent diagnostic core layer pushes a diagnostic report to the repair shop terminal. The repair shop terminal then provides the final repair result to the automation layer. The automation layer distributes the latest model to the digital twin layer.

[0021] Therefore, it can be seen that in the vehicle remote intelligent diagnostic method provided by the embodiments of the present invention, various types of data are collected at the vehicle end through multiple different sensors, which can comprehensively grasp the vehicle status. In addition, edge processing is performed at the vehicle end first, so that not all raw data is transmitted to the cloud, which greatly reduces the pressure of data transmission. Then, anomaly detection and prediction are performed through AI models to generate early warnings, thereby realizing the early prediction of vehicle risks. The repair plan is pushed to both the car owner and the repair shop at the same time, and both parties can obtain information synchronously. The repair shop can prepare parts and work positions in advance, and the car owner can also arrange time in a timely manner without repeated communication, making the repair operation more efficient.

[0022] Optionally, step S101 includes the following sub-steps: The system collects structured and unstructured data, including CAN bus data and GPS location data, and unstructured data including camera footage, microphone audio, and vibration signals. Using the lightweight AI chip built into the vehicle's T-Box, a local AI model is run to perform real-time cleaning, compression, and preliminary analysis of the raw data.

[0023] Optionally, step S102 includes the following sub-steps: By aligning data from different sources and frequencies on a timeline, a unified panoramic view of vehicle status is formed. Using deep recurrent neural networks or Transformer models, compare current data with health benchmarks generated by vehicle digital twin models, or compare them with fleet historical data, to identify deviations and issue early warnings; Among them, the anomaly detection methods include at least one of reconstruction-based anomaly detection, time-series prediction-based anomaly detection, or ultra-lightweight dynamic threshold monitoring.

[0024] Optional: The aforementioned reconstruction-based anomaly detection includes: Use an autoencoder or variational autoencoder to learn a compressed representation of normal data; During inference, the reconstruction error is used to determine whether the data is abnormal; The anomaly detection based on time-series prediction includes: Use LSTM or temporal convolutional networks to predict future data based on past sequence data; Anomalies are determined by the deviation between the actual value and the predicted value.

[0025] Optionally, step S103 includes the following sub-steps: Each vehicle trains a global model locally using the new data and obtains local model parameter updates. The encrypted local model parameters are updated and uploaded to the cloud aggregation server; The cloud uses a secure aggregation protocol to calculate the average update and update the global model. The updated global model will be delivered to the vehicle via OTA.

[0026] Optionally, the aforementioned federated learning includes the following steps: When the k-th vehicle uses new data locally, the global model is trained to obtain a local update; The vehicle control unit uploads the encrypted model update to the cloud aggregation server; Control the cloud to use a secure aggregation protocol to calculate average updates and update the global model; The updated global model is then distributed to the fleet vehicles via OTA, completing one iteration.

[0027] Preferably, in the process of generating the optimal maintenance strategy through the dynamic maintenance decision optimization engine, the model is a Markov decision process, and the state includes fault type, fault severity level, vehicle location, remaining warranty period, list of nearby service centers and their real-time status. Actions include recommending immediate repair, recommending repair within one week, recommending a repair center, and triggering an OTA software update; The reward function calculates the total reward based on successful repair, user feedback, repair delay costs, and repair economic costs.

[0028] In summary, when the present invention is running, step S101 is executed first: the vehicle end is controlled to collect multimodal data of the vehicle through multi-source sensors, and the multimodal data is processed by edge processing.

[0029] In this step, the vehicle's multi-source sensors will continuously and synchronously collect two types of data: One type is structured data, including core vehicle operating parameters such as engine speed, brake pressure, battery SOC, and transmission gear position transmitted via the CAN bus, as well as location data such as real-time vehicle coordinates, driving speed, and driving route collected by the GPS module.

[0030] Another type is unstructured data, including road environment images captured by vehicle cameras, images of vehicle appearance and key component status, audio signals such as engine running sound, chassis vibration sound, and tire noise collected by microphones, and vibration signals of key parts such as the vehicle body and engine compartment recorded by vibration sensors, so as to achieve all-round and blind-angle perception of the vehicle's operating status.

[0031] The edge processing is handled by a lightweight AI chip built into the in-vehicle T-Box. This chip integrates a local lightweight AI model, enabling real-time processing of the collected raw data: First, a data cleaning algorithm removes invalid noise data caused by sensor malfunctions and signal interference, ensuring data accuracy. Then, large volumes of unstructured data, such as video and audio, undergo H.265 encoding and audio compression, significantly reducing data transmission bandwidth usage. Finally, a preliminary anomaly detection algorithm is run to quickly analyze the processed data, filtering out potentially anomalous feature data. Only encrypted key feature data is transmitted to the cloud platform via the vehicle network, ensuring data transmission security and preventing invalid data from consuming transmission resources.

[0032] Next, step S102 is executed: the cloud receives encrypted data from the vehicle, performs multimodal data fusion and feature extraction, performs anomaly detection and prediction based on the AI ​​model, and generates early warnings.

[0033] After receiving encrypted data uploaded from the vehicle via a secure communication protocol, the cloud platform first performs multimodal data fusion and feature extraction. Using a time alignment algorithm, data from different acquisition frequencies and sources, such as 100 parameter samples per second from the CAN bus and 15 frames per second from the camera, are precisely aligned on a unified timeline to eliminate timing discrepancies. Then, through a multimodal fusion model, the numerical features of structured data are deeply fused with the texture and frequency features of unstructured data. Abnormal frequency features are extracted from audio data, component appearance damage features are identified from video data, and abnormal vibration amplitude features are extracted from vibration data. Ultimately, a panoramic view comprehensively reflecting the vehicle's current operating status is formed.

[0034] Anomaly detection and prediction are performed by a core AI model deployed in the cloud. In this embodiment, a deep recurrent neural network or Transformer model is used to achieve anomaly identification through dual comparisons: First, the current vehicle status panoramic data is compared in real time with the health benchmark data generated by the vehicle's digital twin model. This digital twin model is built based on vehicle design parameters and factory test data and can accurately simulate the vehicle's healthy operating state. Second, it is statistically compared with the fleet's historical healthy operating data, using big data analysis to uncover vehicle operating patterns. When the model detects that the deviation between the current data and the health benchmark or historical data exceeds a preset threshold, it immediately generates early warning information, clarifying the warning level, the possible fault type, and the scope of risk.

[0035] The anomaly detection methods can be flexibly selected and combined according to the data type. Specifically, for complex multimodal fusion data, a reconstruction-based anomaly detection method is used. This involves training a massive amount of vehicle health operation data using an autoencoder or variational autoencoder to learn the compressed representation of normal data, i.e., latent space features. During the inference phase, if the model's reconstruction error for the current data is significantly higher than the normal threshold, it is judged as an anomaly. For structured data with strong temporal sequence characteristics, such as engine speed changes during continuous driving, a time-series prediction-based anomaly detection method is used. This involves training the model using LSTM or temporal convolutional networks to enable it to predict data for multiple future time steps based on sequence data from the past period. If the deviation between the actual data value and the predicted value exceeds a set range, it is judged as an anomaly.

[0036] For simple signals of a single type, such as vibration and audio, an ultra-lightweight dynamic threshold monitoring method is adopted. Time-domain features and frequency-domain features are extracted during the edge processing stage. The time-domain features include peak value, mean, and variance, while the frequency-domain features include peak frequency and frequency band energy. The cloud dynamically adjusts the monitoring threshold according to different vehicle operating states, including idling, constant speed, acceleration, and deceleration. When the feature value exceeds the threshold range of the corresponding state, an anomaly prompt is triggered.

[0037] Subsequently, step S103 is executed: the cloud controls the aggregation of local model parameters from multiple vehicles through federated learning technology, updates the global AI model, and sends it to the vehicles via OTA.

[0038] To enable continuous evolution of the diagnostic model while protecting individual vehicle data privacy and OEM core data, this embodiment employs federated learning technology to construct a global model optimization mechanism.

[0039] First, each vehicle connected to the system acts as a federated learning node. It performs several rounds of local training using newly generated runtime data and the current global AI model W_global distributed from the cloud. The model parameters are optimized through the backpropagation algorithm to obtain the local model update W_local^k, where k is the kth vehicle.

[0040] After training is completed, the vehicle only calculates the model parameter update difference ΔW^k=W_local^k-W_global, and encrypts the update difference using an asymmetric encryption algorithm. Then, the encrypted model parameter update difference is uploaded to the cloud aggregation server. No original data is transmitted throughout the process to ensure data privacy and security.

[0041] After receiving the encrypted model parameter update differences from all vehicles participating in this round of training, the cloud aggregation server can use a secure aggregation protocol to calculate the average value of all update differences, ΔW_avg=(1 / K)*Σ(ΔW^k), where K is the number of vehicles participating in this round of training, without being able to decrypt the update difference of a single vehicle model. Then, it updates the global AI model parameters according to the preset learning rate η to obtain the new global model W_global_new=W_global+η*ΔW_avg.

[0042] Finally, the cloud uses OTA (Over-The-Air) technology to distribute the updated global AI model to all vehicles in the fleet, replacing the original local model and completing a model evolution iteration. As the number of iterations increases, the diagnostic accuracy and generalization ability of the global model continue to improve, enabling it to adapt to more complex fault scenarios.

[0043] Next, step S104 is executed: the cloud control system generates the optimal maintenance strategy based on the fault diagnosis results and real-time information through the dynamic maintenance decision optimization engine.

[0044] After confirming the diagnostic results such as fault type and severity level through step S102, the cloud platform starts the dynamic maintenance decision optimization engine. This engine models the maintenance decision problem as a Markov decision process and generates the optimal maintenance strategy through deep reinforcement learning algorithms.

[0045] The Markov decision-making process includes multi-dimensional information, such as fault types including brake system wear, electronic control system failure, and chassis noise, and fault severity levels including minor warning, moderate fault, and emergency fault. It also includes the vehicle's current location coordinates, remaining warranty period, a list of nearby service centers and their real-time status, including technician availability, technician skill profile, target spare parts inventory quantity, work bay occupancy, and service center service rating.

[0046] The actions include a variety of optional strategies, such as recommending immediate repair, recommending repair within a specified time window, recommending a specific repair center, triggering OTA software remote repair, and arranging roadside assistance.

[0047] The reward function calculates the total reward based on the repair effect and the overall cost. The specific formula is r=R1-R2-R3-C, where R1 is the positive reward for successful fault repair, R2 is the negative reward for users giving bad reviews due to poor repair experience, such as excessive waiting time and poor repair effect, R3 is the high negative reward for vehicle breakdown and safety accidents caused by delayed repair, and C is the economic cost incurred in the repair process, including spare parts cost, technician labor cost, roadside assistance cost, etc.

[0048] The Dynamic Maintenance Decision Optimization Engine's DRL agent continuously learns from maintenance execution results and user feedback to optimize action selection strategies. During the decision-making process, the engine comprehensively analyzes all real-time information: for minor software faults, it prioritizes OTA (Over-The-Air) remote software repair, eliminating the need for the user to visit a repair center. For moderate faults requiring spare parts replacement, it selects the nearest repair center with the corresponding spare parts in stock and available skilled technicians, generating a reservation order. For emergency faults, it immediately recommends the nearest repair center and simultaneously notifies the roadside assistance team, while also proactively sending fault details to the repair center to ensure repair resources are ready in advance, achieving "knowing the problem before the car arrives."

[0049] Finally, step S105 is executed: the maintenance strategy is pushed to the user terminal and the server terminal, and the corresponding maintenance operation is performed.

[0050] After generating the optimal repair strategy in the cloud, it is simultaneously pushed to multiple terminals: On the user's end, the car owner's app receives clear and easy-to-understand warning notifications and repair suggestions. The notification includes the fault type, risk level, specific impact, recommended repair plan, and estimated cost. For example, if the vehicle's brake pads are worn close to their limit, it is considered a moderate fault; continuing to drive may affect braking performance, and it is recommended to visit the shop for replacement within the week. The shop currently has available technicians and matching parts. Users can directly schedule repairs, check repair progress, and communicate online through the app. The service-side repair workshop terminal includes technician workstations and AR smart glasses. The workshop terminal receives detailed fault diagnosis reports, including root cause analysis, detailed repair procedure guidance, a list of required tools, spare parts specifications, and disassembly diagrams of key components. Technicians can prepare relevant tools and spare parts in advance and plan the repair process. This invention not only outputs the cause of the fault but also provides remote service centers with repair plan suggestions, spare parts demand prediction, and technician dispatch optimization, improving the efficiency of the entire service chain.

[0051] After the maintenance operation is completed, the server will feed back the final maintenance results, including maintenance completion time, actual replacement parts, fault repair status, and user satisfaction rating, to the cloud platform. The cloud will then update the fault case, maintenance process data, and decision results to the fault case library and knowledge graph library, providing data support for subsequent model training and decision optimization. This forms a complete closed loop of "data collection - anomaly detection - model evolution - maintenance decision - result feedback," continuously improving the system's diagnostic accuracy and decision efficiency.

[0052] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0053] like Figure 3 The diagram shows a hardware structure of an electronic device according to the present invention, including a memory 302, a processor 301, and an electronic device program on the memory 302. The processor 301 executes the electronic device program to implement the steps of the vehicle remote intelligent diagnostic method of any of the above embodiments.

[0054] Figure 3 Take processor 301 as an example.

[0055] The electronic device may also include an input device 303 and a display device 304.

[0056] The processor 301, memory 302, input device 303 and display device 304 can be connected by a bus or other means. The figure shows an example of connection by bus.

[0057] The memory 302, as a non-volatile electronic device readable storage medium, can be used to store non-volatile software programs, non-volatile electronic device executable programs, and modules, such as the program instructions / modules corresponding to the vehicle remote intelligent diagnostic method in the embodiments of this application. The processor 301 executes various functional applications and data processing by running the non-volatile software programs, instructions, and modules stored in the memory 302, thereby realizing the vehicle remote intelligent diagnostic method in the above embodiments.

[0058] The memory 302 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the vehicle remote intelligent diagnostic method. Furthermore, the memory 302 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 302 may optionally include memory remotely located relative to the processor 301, and these remote memories can be connected via a network to the apparatus performing the vehicle remote intelligent diagnostic method. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0059] The input device 303 can receive user clicks and generate signal inputs related to user settings and function control of the vehicle remote intelligent diagnostic method. The display device 304 may include a display screen or other display equipment.

[0060] When one or more modules are stored in the memory 302, and are run by one or more processors 301, the vehicle remote intelligent diagnostic method in any of the above method embodiments is executed.

[0061] When the electronic device disclosed in this invention is running, it can execute all the steps of the above-mentioned vehicle remote intelligent diagnostic method. By collecting various types of data through multiple different sensors at the vehicle end, it can comprehensively grasp the vehicle status. In addition, edge processing is performed at the vehicle end first, so that not all raw data is transmitted to the cloud, which greatly reduces the pressure of data transmission. Then, anomaly detection and prediction are performed through AI model to generate early warnings, so as to predict vehicle risks in advance. The repair plan is pushed to both the car owner and the repair shop at the same time. Both parties can obtain information at the same time. The repair shop can prepare parts and work positions in advance, and the car owner can also arrange time in time without repeated communication.

[0062] An embodiment of the present invention provides an electronic device readable storage medium storing an electronic device program / instructions, which, when executed by a processor 301, implements all the steps of the vehicle remote intelligent diagnostic method described above.

[0063] In the context of this disclosure, a storage medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. The storage medium can be a machine-readable signal medium or a machine-readable storage medium. Optionally, the storage medium can be a non-transitory electronically readable storage medium, such as a ROM, random access memory (RAM), compact disc ROM (CD ROM), magnetic tape, floppy disk, and optical data storage device.

[0064] One embodiment of the present invention provides an electronic device program product, including an electronic device program / instruction, which, when executed by a processor, implements the steps of the vehicle remote intelligent diagnostic method as described above.

[0065] By running the aforementioned electronic device program, all steps of the vehicle remote intelligent diagnostic method described above can be executed. By collecting various types of data through multiple different sensors on the vehicle, the vehicle status can be fully grasped. In addition, edge processing is performed on the vehicle first, so that not all raw data is transmitted to the cloud, greatly reducing the pressure of data transmission. Then, anomaly detection and prediction are performed through AI models to generate early warnings, enabling early prediction of vehicle risks. Repair plans are simultaneously pushed to the car owner and the repair shop, and both parties can obtain information at the same time. The repair shop can prepare parts and work bays in advance, and the car owner can also arrange time in a timely manner without repeated communication.

[0066] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for remote intelligent diagnosis of vehicles, characterized in that, include: The vehicle control unit collects multimodal data of the vehicle through multiple source sensors and performs edge processing on the multimodal data; The cloud controls the reception of encrypted data from the vehicle, performs multimodal data fusion and feature extraction, and uses AI models to detect and predict anomalies, generating early warnings. The cloud-based control system aggregates local model parameters from multiple vehicles using federated learning technology, updates the global AI model, and sends the updated data to the vehicles via OTA. Based on fault diagnosis results and combined with real-time information, the cloud-based control system generates the optimal maintenance strategy through a dynamic maintenance decision optimization engine. The maintenance strategy is pushed to the user terminal and the server terminal, and the corresponding maintenance operations are executed.

2. The vehicle remote intelligent diagnostic method according to claim 1, characterized in that, The control vehicle terminal collects multimodal data of the vehicle through multi-source sensors and performs edge processing on the multimodal data, including: Collect structured and unstructured data. The structured data includes CAN bus data and GPS location data, while the unstructured data includes camera footage, microphone audio, and vibration signals. Using the lightweight AI chip built into the vehicle's T-Box, a local AI model is run to perform real-time cleaning, compression, and preliminary analysis of the raw data.

3. The vehicle remote intelligent diagnostic method according to claim 1, characterized in that, The process of multimodal data fusion and feature extraction in the cloud, and anomaly detection and prediction based on AI models, includes: By aligning data from different sources and frequencies on a timeline, a unified panoramic view of vehicle status is formed. Using deep recurrent neural networks or Transformer models, compare current data with health benchmarks generated by vehicle digital twin models, or compare them with fleet historical data, to identify deviations and issue early warnings; Among them, the anomaly detection methods include at least one of reconstruction-based anomaly detection, time-series prediction-based anomaly detection, or ultra-lightweight dynamic threshold monitoring.

4. The vehicle remote intelligent diagnostic method according to claim 3, characterized in that: The reconstruction-based anomaly detection includes: Use an autoencoder or variational autoencoder to learn a compressed representation of normal data; During inference, the reconstruction error is used to determine whether the data is abnormal; The anomaly detection based on time-series prediction includes: Use LSTM or temporal convolutional networks to predict future data based on past sequence data; Anomalies are determined by the deviation between the actual value and the predicted value.

5. The vehicle remote intelligent diagnostic method according to claim 1, characterized in that, The control cloud aggregates local model parameters from multiple vehicles using federated learning technology, updates the global AI model, and distributes it to the vehicles via OTA, including: Each vehicle trains a global model locally using the new data, and obtains local model parameter updates. The encrypted local model parameters are updated and uploaded to the cloud aggregation server; The cloud uses a secure aggregation protocol to calculate the average update and update the global model. The updated global model will be delivered to the vehicle via OTA.

6. The vehicle remote intelligent diagnostic method according to claim 5, characterized in that, The federated learning includes the following steps: When the k-th vehicle uses new data locally, the global model is trained to obtain a local update; The vehicle control unit uploads the encrypted model update to the cloud aggregation server; Control the cloud to use a secure aggregation protocol to calculate average updates and update the global model; The updated global model is then distributed to the fleet vehicles via OTA, completing one iteration.

7. The vehicle remote intelligent diagnostic method according to claim 1, characterized in that: In the process of generating the optimal maintenance strategy through the dynamic maintenance decision optimization engine, the model is a Markov decision process, and the state includes fault type, fault severity level, vehicle location, remaining warranty period, list of nearby service centers and their real-time status. Actions include recommending immediate repair, recommending repair within one week, recommending a repair center, and triggering an OTA software update; The reward function calculates the total reward based on successful repair, user feedback, repair delay costs, and repair economic costs.

8. An electronic device comprising a memory, a processor, and an electronic device program on the memory, characterized in that, The processor executes the electronic device program to implement the steps of the vehicle remote intelligent diagnostic method according to any one of claims 1-7.

9. An electronic device readable storage medium having an electronic device program / instructions stored thereon, characterized in that, When the electronic device program / instructions are executed by the processor, they implement the steps of the vehicle remote intelligent diagnostic method according to any one of claims 1-7.

10. An electronic device program product, comprising an electronic device program / instructions, characterized in that, When the electronic device program / instructions are executed by the processor, they implement the steps of the vehicle remote intelligent diagnostic method according to any one of claims 1-7.

Images