New energy automobile motor control hardware fault diagnosis method and system based on AI edge calculation

Abstract

The invention discloses a new energy automobile motor control hardware fault diagnosis method and system based on AI edge calculation, and aims to solve the technical problems of poor real-time performance, strong network dependence, data potential safety hazards and single function of existing pure cloud diagnosis. The system adopts a double-layer architecture of edge computing and cloud collaboration, an edge computing intelligent unit collects multi-source time sequence data of motor control hardware at a high frequency not lower than 100 kHz, after filtering and normalization preprocessing, fault diagnosis is completed through fusion of a 1D-CNN + LSTM lightweight AI hybrid model and an expert rule base, and the fault diagnosis efficiency is improved. Generating a hierarchical control instruction of'warning-power reduction operation-immediate shutdown '; and the cloud end only aggregates the desensitization diagnosis abstract data optimization model and reversely updates the edge end. According to the invention, millisecond-level fault response is realized, network dependence and data transmission quantity are reduced, diagnosis precision and system adaptability are improved, vehicle maintenance cost is reduced, and the method is suitable for a new energy vehicle motor control hardware full-condition fault diagnosis scene.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for new energy vehicles, specifically to a fault diagnosis method and system for new energy vehicle motor control hardware based on AI edge computing. Background Technology

[0002] The motor control hardware of new energy vehicles is a core component of the vehicle's power system, and its operating status directly affects the vehicle's driving safety and reliability. Existing motor control hardware fault diagnosis technologies mostly adopt a single architecture of "terminal acquisition - cloud diagnosis," which transmits all motor operating information to a cloud computing platform via a GPS vehicle terminal, where fault identification and data storage are completed by the cloud.

[0003] However, this type of technology has significant drawbacks: First, it lacks real-time performance. The combined delay of wireless communication transmission and cloud processing time cannot meet the millisecond-level response requirements for transient faults such as open circuits and short circuits in power modules, which can easily lead to the expansion of faults. Second, it is highly dependent on the network. In scenarios with weak or interrupted signals, such as tunnels and remote areas, the diagnostic function will directly fail. Third, it has low data processing efficiency and security. Uploading all data consumes a large amount of network bandwidth and lacks desensitization processing, which poses a risk of leakage of sensitive vehicle information. Fourth, it lacks a functional closed loop. It only outputs diagnostic results without an active protection mechanism, which cannot promptly curb the damage of faults to hardware.

[0004] The root cause of the above problems lies in the limitations of the early technology roadmap and the bottleneck of transmission and processing in the industry: insufficient computing power of edge computing chips leads to the reliance on the cloud for core diagnostic functions, there is an inherent contradiction between wireless communication latency and fault response requirements, and the full data transmission does not take into account the load of the vehicle network and data security requirements, ultimately making it difficult to adapt to the comprehensive requirements of new energy vehicles for real-time, reliability and security of fault diagnosis. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for fault diagnosis of motor control hardware in new energy vehicles based on AI edge computing. It aims to solve the technical problems existing in current fault diagnosis technologies for motor control hardware in new energy vehicles, such as poor real-time performance, strong network dependence, data security risks, narrow diagnostic scope, and lack of proactive protection and self-evolution capabilities.

[0006] A fault diagnosis system for motor control hardware of new energy vehicles based on AI edge computing includes an edge computing intelligent unit and a cloud collaborative platform, wherein the edge computing intelligent unit is communicatively connected to the cloud collaborative platform;

[0007] The edge computing intelligent unit includes a data acquisition module, a data preprocessing module, an AI diagnostic model module, a fault decision module, and a local communication module;

[0008] The data acquisition module is used to acquire multi-source timing data of the new energy vehicle motor control hardware at a sampling frequency of not less than 100kHz. The multi-source timing data includes three-phase output current, three-phase output voltage, DC bus current, DC bus voltage, power module junction temperature, power module case temperature, gate drive voltage waveform, and motor resolver sensor signal.

[0009] The data preprocessing module is used to perform digital filtering and energy normalization on the multi-source time series data. The energy normalization process maps the data to the interval [-1, 1].

[0010] The AI ​​diagnostic model module is a lightweight hybrid model of 1D-CNN and LSTM cascaded, used to extract local fault features and dynamic time-dimensional dependencies of multi-source time series data, and output fault type and fault confidence.

[0011] The fault decision module has a built-in expert rule base, which is used to calculate the comprehensive confidence level by combining the fault type and confidence level output by the AI ​​diagnostic model, classify the fault level through dynamic thresholds and generate graded control instructions. The fault level includes warning, power reduction operation and immediate shutdown.

[0012] The local communication module is used to send control commands to the vehicle controller via the CAN bus and upload de-identified diagnostic summary data to the cloud collaborative platform.

[0013] The cloud-based collaborative platform is used to aggregate de-identified diagnostic summary data from multiple vehicles, optimize AI diagnostic model parameters, and update edge computing intelligent units in reverse.

[0014] Specifically, the data acquisition module has a built-in phase-locked loop synchronization mechanism to achieve time alignment of multiple sampling channels, and uses transient energy function and multi-scale sliding integral method to enhance data features.

[0015] Specifically, the digital filtering in the data preprocessing module adopts an adaptive weighting method based on the transient energy function, and the filtering formula is as follows:

[0016] ;

[0017] in Let M be the original signal from the i-th channel, and M be the length of the sliding window. Let be the transient energy function. This is the filter adjustment factor.

[0018] Specifically, the expert rule base contains logical rules based on the mechanism of the motor control system, and the comprehensive confidence level is determined by the formula... Calculate, where is the probability predicted by the AI ​​model. β is the rule logic judgment value, and β is the adjustment factor.

[0019] Specifically, the dynamic threshold is adaptively adjusted using the formula, wherein... Based on the threshold, For multi-scale energy integrals, This is the sensitivity coefficient.

[0020] Specifically, in the graded control command, the upper limit of the motor output power for reduced power operation is determined by the formula... Calculation, where This refers to the rated power of the motor. For warning thresholds, This is the threshold for immediate shutdown.

[0021] A method for fault diagnosis of motor control hardware in new energy vehicles based on AI edge computing includes the following steps:

[0022] Step 1: Data acquisition. The data acquisition module acquires multi-source timing data of the motor control hardware at a frequency of not less than 100kHz, and the time alignment of multiple signals is achieved through the phase-locked loop synchronization mechanism.

[0023] Step 2: Data preprocessing. The data preprocessing module performs adaptive filtering and energy normalization on the collected raw data to preserve the temporal characteristics of the data.

[0024] Step 3: AI model inference. Input the preprocessed data into the 1D-CNN+LSTM hybrid model to extract local fault features and their time dimension dependencies, and output the fault type and confidence level.

[0025] Step 4: Fault decision-making. Combine the AI ​​model output with the expert rule base to calculate the comprehensive confidence level, classify the fault level through dynamic thresholds and generate hierarchical control instructions.

[0026] Step 5: Local execution and communication, sending control commands via CAN bus, and simultaneously uploading de-identified diagnostic data after caching;

[0027] Step 6: Cloud Collaboration and Model Evolution. The cloud aggregates data from multiple vehicles to optimize model parameters, and distributes them to the edge computing intelligent unit through incremental or full updates.

[0028] Specifically, the transient energy function formula in step 1 of a fault diagnosis method for new energy vehicle motor control hardware based on AI edge computing is as follows: ,in For the i-th sampled value, The average value under the baseline operating conditions. is the sensor weighting coefficient, and N is the number of acquisition channels.

[0029] According to the above-mentioned method for fault diagnosis of new energy vehicle motor control hardware based on AI edge computing, the model parameter optimization in step 6 adopts a weighted gradient fusion strategy, and the update formula is as follows:

[0030] ,

[0031] in The parameters are the old ones, and η is the learning rate. Let v be the loss function for the data of vehicle v. V represents the data weight, and V represents the total number of aggregated vehicles.

[0032] According to the above-mentioned method for diagnosing hardware faults in the motor control of new energy vehicles based on AI edge computing, in step 4, when the comprehensive confidence level... < When, issue a warning and log it; when ≤ < When, power reduction operation is performed; when ≥ If this occurs, immediately shut down the system and activate the security protection mechanism.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] This invention performs core diagnostics locally at the edge, achieving millisecond-level fault identification. It can promptly capture transient faults, preventing fault escalation and ensuring vehicle safety. In scenarios with weak or interrupted signals, the edge unit can independently perform diagnosis and control without affecting core functions. Data transmission volume is reduced by more than 80% compared to existing technologies, lowering the overall vehicle network load; desensitization processing avoids the leakage of sensitive information, complying with data security regulations. It covers both sudden and progressive faults in multiple core components, improving diagnostic accuracy by more than 30% compared to traditional methods, reducing false positives and false negatives. Hierarchical control commands proactively protect hardware, and the model self-evolves to adapt to multiple vehicle models and operating conditions, reducing subsequent maintenance costs by 40%. Attached Figure Description

[0035] Figure 1 This is a system architecture diagram illustrating the present invention;

[0036] Figure 2 This is a flowchart illustrating the data acquisition and preprocessing process of this invention.

[0037] Figure 3 This is a structural diagram of the AI ​​model shown in this invention;

[0038] Figure 4 This is a flowchart illustrating the fault decision-making and hierarchical control process of the present invention.

[0039] Figure 5This is a flowchart illustrating the cloud collaboration and model evolution process of this invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] refer to Figures 1-5 This invention provides a method and system for diagnosing hardware faults in the motor control of new energy vehicles based on AI edge computing. The system hardware configuration is as follows:

[0042] The edge computing intelligent unit adopts an automotive-grade high-performance chip, integrating a multi-channel high-frequency sampling module (sampling rate adjustable from 100kHz to 200kHz), a CAN communication interface, and a wireless communication module; the data acquisition module includes a current sensor, a voltage sensor, a temperature sensor, a waveform acquisition device, and a resolver sensor, which are deployed at the power module, gate drive circuit, and motor interface of the motor controller, respectively; the cloud collaboration platform adopts a distributed server cluster, which has the functions of big data storage, model training, and parameter distribution.

[0043] Data acquisition and preprocessing implementation

[0044] The data acquisition module uses a phase-locked loop to synchronize the clocks of multiple sensors, ensuring a one-to-one correspondence between sampling points on different channels. In the transient energy function calculation, sensor weighting coefficients are used. Based on component importance, the weights are set as follows: power module-related sensors are assigned a weight of 0.3, gate drive-related sensors are assigned a weight of 0.25, resolver sensors are assigned a weight of 0.2, and all other sensors are assigned a weight of 0.25; the baseline operating condition average is... Obtained through calibration tests conducted before the vehicle leaves the factory.

[0045] In adaptive filtering, the sliding window length M is set to 50, and the filter adjustment factor α is set to 0.6; in energy normalization, the mean... With variance Based on historical normal operating data, the normalization adjustment coefficient λ was set to 0.1, and the sampling duration T was set to 1s to ensure that the data was mapped to the [-1,1] interval and that fault characteristics were preserved.

[0046] AI Model Training and Inference

[0047] The 1D-CNN layer uses three convolutional blocks with kernel lengths K of 3, 5, and 7, and output channels of 16, 32, and 64 respectively. The LSTM layer has two hidden layers, each containing 64 neurons. The output layer uses the Softmax activation function to output confidence scores for 12 fault types (including power module open circuit, short circuit, temperature drift, gate drive abnormality, etc.). The model is trained using the Adam optimizer with a learning rate of 0.001. The training dataset contains fault simulation data and real-world vehicle operation data from over 1000 vehicles.

[0048] Fault Decision Implementation

[0049] The expert rule base contains 20+ logical rules, such as "AI model output power module open circuit fault ( If the DC bus current is ≤0.7A and the DC bus current is ≤0.5A, then =1"AI model outputs temperature drift fault ( If the junction temperature of the power module is ≥0.6 and the continuous rise rate of the junction temperature is ≥2℃ / s, then =0.9". The overall confidence adjustment factor β is set to 0.6 to balance the weights of AI model prediction and rule logic.

[0050] In dynamic thresholding, the basic threshold Setting: Warning Threshold =0.5, immediate shutdown threshold =0.8; the sensitivity coefficient γ is set to 0.3, and the threshold is adaptively adjusted according to the system energy state to avoid misjudgment of fluctuations under normal operating conditions.

[0051] (v) Hierarchical control implementation

[0052] When the overall confidence level When 0.5 < 0.5, the system is considered fault-free and only records operational data; when 0.5 ≤ When the value is less than 0.8, a warning command is generated, which is displayed on the instrument panel to alert the driver, and a fault log is recorded simultaneously; when... When the value is ≥0.8, the fault type is determined as follows: for non-fatal faults, a power reduction operation command is generated, and the motor output power is gradually limited according to the formula, with the power attenuation rate set to 5kW / s; for fatal faults, an immediate stop command is generated, and the power module power supply is cut off within 5ms through a transient power-off control algorithm, activating the vehicle's emergency braking system.

[0053] (vi) Cloud Collaboration and Model Evolution

[0054] The edge computing intelligent unit uploads anonymized diagnostic summary data (including fault type, occurrence time, overall confidence level, and vehicle driving condition) every 24 hours. The cloud platform aggregates multi-vehicle data by vehicle type and driving condition. In model optimization, the learning rate η is set to 0.0005, and the vehicle data weights are... The weighting is set based on data quality and vehicle mileage (data with a mileage ≥ 50,000 km is weighted at 1.2, and others at 1.0). Edge model updates use incremental loading, prioritizing the updating of key layer parameters (LSTM hidden layer weights), and performing a full refresh every 7 days to ensure continuous improvement in model stability and diagnostic accuracy.

[0055] (vii) System test results

[0056] In real-vehicle testing, the system's response time for identifying power module short-circuit faults is ≤3ms, and its accuracy in identifying temperature drift faults is ≥95%. In network interruption scenarios, the independent diagnostic accuracy of edge units remains ≥90%. The amount of data transmitted is reduced by 85% compared to existing pure cloud solutions, and the network bandwidth usage of the entire vehicle is reduced by 60%. After iterative optimization on more than 100 vehicles, the diagnostic accuracy is improved by 38% compared to the initial version, and the maintenance cost is reduced by 42% compared to traditional solutions.

[0057] This invention comprehensively addresses the core deficiencies of existing technologies through an "edge computing + cloud collaboration" architecture, the fusion of AI models and expert rule bases, hierarchical control, and self-evolution mechanism. It provides an efficient, reliable, and safe fault diagnosis solution for motor control hardware in new energy vehicles and can be widely applied to motor control systems of pure electric and hybrid vehicles.

[0058] The above embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

Claims

1. A hardware fault diagnosis system for new energy vehicle motor control based on AI edge computing, characterized in that, It includes an edge computing intelligent unit and a cloud collaboration platform, wherein the edge computing intelligent unit is communicatively connected to the cloud collaboration platform; The edge computing intelligent unit includes a data acquisition module, a data preprocessing module, an AI diagnostic model module, a fault decision module, and a local communication module; The data acquisition module is used to acquire multi-source timing data of the new energy vehicle motor control hardware at a sampling frequency of not less than 100kHz. The multi-source timing data includes three-phase output current, three-phase output voltage, DC bus current, DC bus voltage, power module junction temperature, power module case temperature, gate drive voltage waveform, and motor resolver sensor signal. The data preprocessing module is used to perform digital filtering and energy normalization on the multi-source time series data. The energy normalization process maps the data to the interval [-1, 1]. The AI ​​diagnostic model module is a lightweight hybrid model of 1D-CNN and LSTM cascaded, used to extract local fault features and dynamic time-dimensional dependencies of multi-source time series data, and output fault type and fault confidence. The fault decision module has a built-in expert rule base, which is used to calculate the comprehensive confidence level by combining the fault type and confidence level output by the AI ​​diagnostic model, classify the fault level through dynamic thresholds and generate graded control instructions. The fault level includes warning, power reduction operation and immediate shutdown. The local communication module is used to send control commands to the vehicle controller via the CAN bus and upload de-identified diagnostic summary data to the cloud collaborative platform. The cloud-based collaborative platform is used to aggregate de-identified diagnostic summary data from multiple vehicles, optimize AI diagnostic model parameters, and update edge computing intelligent units in reverse.

2. The system according to claim 1, characterized in that, The data acquisition module has a built-in phase-locked loop synchronization mechanism to achieve time alignment of multiple sampling channels, and enhances data features through transient energy functions and multi-scale sliding integral methods.

3. The system according to claim 1, characterized in that, The digital filtering in the data preprocessing module employs an adaptive weighting method based on the transient energy function, and the filtering formula is as follows: ； in Let M be the original signal from the i-th channel, and M be the length of the sliding window. Let be the transient energy function. This is the filter adjustment factor.

4. The system according to claim 1, characterized in that, The expert rule base contains logical rules based on the mechanism of the motor control system, and the comprehensive confidence level is expressed by the formula: Calculation, where Predicting probabilities for AI models β is the rule logic judgment value, and β is the adjustment factor.

5. The system according to claim 1, characterized in that, The dynamic threshold is expressed by the formula ; Adaptive adjustment, in which Based on the threshold, For multi-scale energy integrals, This is the sensitivity coefficient.

6. The system according to claim 1, characterized in that, In the graded control command, the upper limit of the motor output power during reduced power operation is determined by the formula. Calculation, where This refers to the rated power of the motor. For warning thresholds, This is the threshold for immediate shutdown.

7. A method for diagnosing hardware faults in the motor control of new energy vehicles based on AI edge computing, characterized in that, Includes the following steps: Step 1: Data acquisition. The data acquisition module acquires multi-source timing data of the motor control hardware at a frequency of not less than 100kHz, and the time alignment of multiple signals is achieved through the phase-locked loop synchronization mechanism. Step 2: Data preprocessing. The data preprocessing module performs adaptive filtering and energy normalization on the collected raw data to preserve the temporal characteristics of the data. Step 3: AI model inference. Input the preprocessed data into the 1D-CNN+LSTM hybrid model to extract local fault features and their time dimension dependencies, and output the fault type and confidence level. Step 4: Fault decision-making. Combine the AI ​​model output with the expert rule base to calculate the comprehensive confidence level, classify the fault level through dynamic thresholds and generate hierarchical control instructions. Step 5: Local execution and communication, sending control commands via CAN bus, and simultaneously uploading de-identified diagnostic data after caching; Step 6: Cloud Collaboration and Model Evolution. The cloud aggregates data from multiple vehicles to optimize model parameters, and distributes them to the edge computing intelligent unit through incremental or full updates.

8. The method according to claim 7, characterized in that, The transient energy function formula mentioned in step 1 is: ,in For the i-th sampled value, The average value under the baseline operating conditions. is the sensor weighting coefficient, and N is the number of acquisition channels.

9. The method according to claim 7, characterized in that, The model parameter optimization in step 6 employs a weighted gradient fusion strategy, and the update formula is as follows: ； in The parameters are the old ones, and η is the learning rate. Let v be the loss function for the data of vehicle v. V represents the data weight, and V represents the total number of aggregated vehicles.

10. The method according to claim 7, characterized in that, In step 4, when the overall confidence level < When, issue a warning and log it; when ≤ < When, power reduction operation is performed; when ≥ If this occurs, immediately shut down the system and activate the security protection mechanism.

Images