Fault fast perception and direct compensation control method for electric vehicle wheel hub motor with stability constraint

Abstract

The application discloses a kind of with stability restraint electric vehicle wheel hub motor fault fast perception and direct compensation control method, belong to electric vehicle active safety control technical field.The application adopts FFT-WMFCC high-frequency fault feature extraction to realize early fault accurate detection;PID+PPO direct compensation loose coupling layered control architecture is constructed, and millisecond level fault response is realized by discarding parameter iteration;Based on state deviation, Lyapunov function is constructed to complete closed-loop system stability proof and control weight dynamic scheduling;Design fusion tracking accuracy, stability, control smoothness and motor physical constraint multi-objective reward function.The fault detection F1 score of the application is 0.92, fault response time is less than or equal to 55ms, trajectory tracking root mean square error is less than or equal to 0.08m, side angle peak value is 0 rad, and motor maximum temperature rise is less than or equal to 18.5 DEG C, which can significantly improve the trajectory tracking accuracy, lateral stability and driving safety of vehicle under fault condition.

Description

Technical Field

[0001] This invention belongs to the field of active safety control technology for electric vehicles, specifically relating to a method for rapid detection and direct compensation control of hub motor faults in electric vehicles with stability constraints. Background Technology

[0002] Driven by the "dual-carbon" strategy, four-wheel independent drive electric vehicles, with their advantage of independently adjustable torque for each of the four wheels, possess significant competitiveness in handling stability, active safety, and energy conservation, making them an important technological route for high-end electrified vehicles. In-wheel motors operate under conditions of vibration, high temperature, and strong coupling, making them prone to failures such as bearing wear, stator winding aging, and reducer jamming, ultimately manifesting as a decrease in output torque. Actual test data shows that the failure rate of in-wheel motors is approximately 2.3 times per 100,000 kilometers, with torque attenuation failures accounting for 67%. This directly disrupts the vehicle's dynamic balance, leading to yaw instability, trajectory deviation, and even traffic accidents. Therefore, high-performance actuator fault-tolerant control has significant engineering value. The existing fault-tolerant control methods have the following obvious defects: (1) Model-driven methods rely on accurate vehicle and fault models. Their performance drops significantly when road surface adhesion changes, load migration occurs, and model mismatch occurs. The computational load is large under multi-motor faults, making it difficult to meet the 100Hz real-time requirements of vehicle-mounted systems. (2) Traditional reinforcement learning fault-tolerant control often adopts an indirect structure of "RL adjusting PID parameters". After a fault occurs, multiple iterations are required to respond. The fault recovery delay exceeds 3 seconds, resulting in poor adaptability to dynamic operating conditions. (3) Most schemes lack rigorous closed-loop stability proofs, which is a fatal flaw in the automotive field that emphasizes functional safety. At the same time, existing schemes do not adequately consider physical constraints such as motor torque and temperature rise, which may lead to over-control, causing overcompensation and secondary damage to actuators. (4) The lack of rigorous closed-loop stability proofs makes it impossible to meet the functional safety requirements of automobiles. (5) Traditional time-domain / frequency-domain fault features are not sensitive to early weak faults, resulting in low detection accuracy, high false alarm rate, and missed optimal compensation opportunities. Summary of the Invention

[0003] This invention aims to overcome the shortcomings of existing technologies by providing a fast detection and direct compensation control method for hub motor faults in electric vehicles with stability constraints. This method addresses the problems of perception lag, slow response, instability risk, and lack of constraints in traditional fault-tolerant control methods, thereby achieving accurate early fault detection, millisecond-level direct compensation, global stability assurance, and actuator safety constraints.

[0004] To solve the above-mentioned technical problems, the present invention is implemented as follows:

[0005] A method for rapid fault detection and direct compensation control of hub motor in electric vehicles with stability constraints includes the following steps:

[0006] (1) A vehicle dynamics model characterizing the torque decay fault of the hub motor is established based on Newton-Euler equations;

[0007] (2) The vehicle status signal is sampled at high frequency, and the fault features for detecting early torque decay faults are extracted using the fast FFT-WMFCC method.

[0008] (3) Construct a parallel hierarchical fault-tolerant control architecture; the architecture includes an independently operating basic PID control loop and a PPO reinforcement learning compensation loop; wherein, the total control input is obtained by weighted fusion of the torque commands output by the above two loops;

[0009] (4) Construct a Lyapunov function based on the deviation between the actual state of the vehicle and the state of the health model. Based on the energy value of the function, dynamically schedule the weights allocated to the PPO compensation loop in the weighted fusion step (3). The Lyapunov function is used to ensure that the closed-loop system is globally asymptotically stable under preset conditions or consistently eventually bounded under bounded disturbances.

[0010] (5) Design a multi-objective reward function to optimize the strategy of PPO compensation loop; the reward function includes penalty terms related to trajectory tracking error, convergence of the Lyapunov function, smoothness of control action and actuator constraints.

[0011] (6) Closed-loop fault-tolerant control is achieved by using EKF state observation and torque limiting to output the final control command.

[0012] Furthermore, the torque attenuation fault model described in step (1) is as follows: ;in, For a healthy motor, the desired torque, This represents the actual output torque after the fault. Let be the fault coefficient of the i-th hub motor (i=1, 2, 3, 4), and let be the fault coefficient. It covers constant faults, gradual faults, and intermittent faults.

[0013] Further, the FFT-WMFCC method in step (2) specifically includes: synchronously sampling multiple signals including vehicle speed, lateral speed, yaw rate, roll angle, pitch angle and torque of each wheel; windowing the sampled signals and performing fast Fourier transform to extract the energy of the frequency bands related to mechanical faults, electrical faults and power imbalances as frequency domain features; fusing the frequency domain features and time domain features according to a preset weight ratio, and then generating the fault features through weighted Mel-Cepstral Transform.

[0014] Furthermore, in the parallel hierarchical fault-tolerant control architecture described in step (3), the total control input is represented as:

[0015]

[0016] in, This is the overall control input vector for the vehicle. The four-wheel basic control vectors output by the basic PID controller. The four-wheel compensation control vector output by the PPO compensation controller. The weighting coefficients for basic PID control. The dynamic weights of the PPO compensation control weight coefficients satisfy... , 0≤ ≤1, 0≤ ≤1.

[0017] Furthermore, the basic PID control loop in step (3) is a cascade structure and operates at the first control frequency; the PPO compensation loop directly generates independent compensation torque for the four wheels and operates at a second control frequency lower than the first control frequency.

[0018] Furthermore, the dynamic scheduling in step (4) specifically means: when the energy value of the Lyapunov function represents the degree to which the system deviates from the healthy state, the weight coefficient corresponding to the PPO compensation loop is dynamically increased.

[0019] Furthermore, the Lyapunov function mentioned in step (4) is:

[0020]

[0021] In the formula This represents the deviation vector between the actual vehicle state and the health model state. It is a positive definite weighted matrix; the degree to which the system deviates from a healthy state is caused by the torque attenuation fault of the hub motor.

[0022] Furthermore, the multi-objective reward function described in step (5) is designed as follows:

[0023]

[0024] For trajectory tracking error term, Let Lyapunov function values ​​represent the stability convergence term. The control smoothing term characterizes the magnitude of torque command changes between adjacent time points. Actuator constraint penalty applied to actions exceeding preset torque and / or temperature rise thresholds.

[0025] Furthermore, the torque limit in step (6) ranges from 0 to the maximum rated torque of the motor.

[0026] The present invention also provides a four-wheel independent drive electric vehicle controller, which is configured to execute the above-described method for rapid detection and direct compensation control of electric vehicle hub motor faults with stability constraints.

[0027] This invention proposes a hierarchical fault-tolerant control scheme centered on stability. It primarily employs a design that constructs a basic steady-state control layer and a safety compensation control layer, decoupling accuracy and robustness. Safety compensation is triggered by the perception of anticipated instability characteristics. Furthermore, the stability of the control system is transformed from a target to a design constraint. First, this invention achieves early fault perception through the FFT-WMFCC method using high-frequency multi-source signal fusion. Based on this, a loosely coupled architecture of basic PID + PPO direct compensation is constructed, allowing reinforcement learning to shift from a decision tuner to a decision executor, achieving millisecond-level direct response from the root cause. Further, Lyapunov theory provides a global stability guarantee for the entire closed-loop system, and this stability constraint, along with actuator physical constraints, is incorporated into the multi-objective reward optimization of PPO, ensuring a safe, smooth, and reliable rapid response process. Ultimately, this forms an advanced fault-tolerant control system that meets automotive functional safety standards.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] 1. This invention uses the FFT-WMFCC method to weighted fuse time-frequency domain features, which can still reliably identify 10-20% of weak faults, and the fault detection F1 score can be significantly improved from the conventional 0.83 to 0.92.

[0030] 2. Thanks to the innovative loosely coupled direct compensation architecture, the PPO of this invention directly outputs the compensation torque, avoiding parameter iteration delay and compressing the fault response time from more than 3 seconds in traditional reinforcement learning methods to milliseconds of less than or equal to 55ms.

[0031] 3. Global stability can be proven: Based on Lyapunov theory, the global asymptotic stability of the closed-loop system under ideal conditions and the uniform final boundedness under bounded perturbations are mathematically rigorously guaranteed, thus meeting the core safety requirements of automobiles.

[0032] 4. Under high dynamic conditions, the root mean square error of trajectory tracking is ≤0.08m, which is 61.9% lower than that of the traditional RL method. It has been rigorously proven by Lyapunov that the peak side slip angle is effectively clamped at 0rad, and the lateral stability is excellent.

[0033] 5. By explicitly incorporating physical constraints such as torque and temperature rise into the reward function and the final limiting stage, this invention ensures that the motor operates within a safe range with a maximum temperature rise of ≤18.5℃, avoiding secondary damage and improving system lifespan.

[0034] 6. This invention exhibits excellent robustness to uncertainties such as sensor noise and changes in road surface adhesion coefficient. It can withstand fluctuations of ±20% in sensor noise and road surface adhesion, making it suitable for complex working conditions. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of a hierarchical fault-tolerant control framework;

[0036] Figure 2 A schematic diagram of the vehicle coordinate system and motor numbering;

[0037] Figure 3 Flowchart for FFT-WMFCC fault feature extraction;

[0038] Figure 4 A network structure diagram of PPO actors-critics;

[0039] Figure 5 This is a basic cascaded PID control structure diagram;

[0040] Figure 6 A graph showing the change in rewards during training rounds;

[0041] Figure 7 This is a comparison chart of trajectories under dual-track shifting conditions;

[0042] Figure 8 Comparison chart of trajectories under circular acceleration conditions;

[0043] Figure 9 A comparison diagram of sideslip angles under dual-track shifting conditions;

[0044] Figure 10 A comparison chart of the sideslip angle under circular acceleration conditions;

[0045] Figure 11 A comparison chart of core performance indicators;

[0046] Figure 12 This is a comparison chart of three fault types. Detailed Implementation

[0047] The present invention will now be described in detail through specific embodiments. These embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art. As used throughout the specification and claims, the terms "comprising" or "including" are open-ended and are interpreted as "comprising but not limited to". The following descriptions are preferred embodiments for carrying out the invention; however, these descriptions are intended to illustrate the general principles of the specification and are not intended to limit the scope of the invention. The scope of protection of the present invention is determined by the appended claims.

[0048] Example 1

[0049] This embodiment provides a method for rapid detection and direct compensation control of hub motor faults in electric vehicles with stability constraints, which specifically includes:

[0050] 1. System and Parameter Settings

[0051] The simulation platform was built using CarSim 2024.0 and Simulink R2024b. The simulated vehicle had a total mass of 1500 kg, a wheelbase of 1.6 m, a front and rear wheelbase of 1.0 m and 1.2 m respectively, and a wheel radius of 0.3 m. The vehicle was equipped with four independent hub motors, each with a maximum output torque of 305 N·m. The control system was configured as follows: a basic PID controller with a control frequency of 100 Hz and a PPO compensator with a control frequency of 20 Hz. An extended Kalman filter (EKF) was used as the state observer.

[0052] 2. Establishment of dynamic and fault models

[0053] Reference Figure 1 The vehicle coordinate system and motor numbers shown are used to establish a dynamic model of a four-wheel independently driven electric vehicle based on Newton's and Euler's equations. The model covers the vehicle's longitudinal, lateral, and yaw motions. During the modeling process, it is assumed that the tires operate within a linear lateral slip range. A mapping relationship is established between the motor output torque and the wheel driving force, taking into account transmission efficiency, transmission ratio, and wheel rolling radius. To simulate potential faults during actual operation, a fault factor is introduced. To characterize the degree of torque attenuation, the following fault model is established:

[0054] ,

[0055] 1. For health, <1 indicates torque decay, covering constant, gradual, and intermittent faults. Let be the expected output torque of the i-th healthy motor under the current operating conditions.

[0056] This represents the actual output torque of the motor under fault conditions. Fault coefficient. The value range is set to .when When 1 indicates that the motor is in a fully healthy state; when A value less than 1 indicates a torque decay fault in the motor. This model can cover three typical fault modes: constant fault, gradual fault, and intermittent fault. Meanwhile, it assumes no fault (…). The system response trajectory at time 1) is used as the baseline for the health model and is used for subsequent calculation of state deviation.

[0057] 3. Fault Feature Extraction

[0058] The six status signals used in this invention all come from the vehicle's existing CAN bus sensors, requiring no additional dedicated sensors, resulting in zero hardware modifications and strong engineering applicability. The specific signal sources are as follows:

[0059] (1) Vehicle speed and lateral speed: output by the vehicle inertial measurement unit (IMU);

[0060] (2) Yaw rate: directly acquired by the IMU gyroscope;

[0061] (3) Roll angle and pitch angle: output from IMU attitude calculation;

[0062] (4) Four-wheel torque signal: reported in real time by the controller of each wheel hub motor via CAN bus.

[0063] All signals are synchronously transmitted to the controller via the vehicle's Ethernet / CAN bus, requiring no additional hardware.

[0064] This embodiment achieves feature extraction for early torque attenuation faults through the following steps, referring to... Figure 3 The process is shown below.

[0065] First, high-frequency synchronous sampling is performed. Six signals—vehicle speed, lateral speed, yaw rate, roll angle, pitch angle, and torque of all four wheels—are independently sampled at a frequency of 1000Hz. A 10-point buffer is configured to enable synchronous data updates at 0.01s. Next, a Hanning window is applied to each sampled signal to suppress spectral leakage. Then, a radix-2 Fast Fourier Transform (FFT) algorithm is used to convert the time-domain signal into a frequency-domain amplitude spectrum. In the frequency domain, energy features are extracted based on preset fault-related frequency bands. Specifically, energy in the 20Hz-80Hz band is extracted as a mechanical fault-sensitive feature, energy in the 50Hz-100Hz band as an electrical fault-sensitive feature, and energy in the 10Hz-30Hz band as a power imbalance-sensitive feature. Finally, the extracted frequency-domain features are fused with the original time-domain features at a weighted ratio of 4:6, and a weighted Mel-frequency cepstral transform is performed to generate a WMFCC fault feature vector. The experimental results show that the F1 score for fault detection can reach 0.92 using the above-mentioned FFT-WMFCC method.

[0066] 4. Layered fault-tolerant control architecture and instruction generation

[0067] See Figure 1 and Figure 5This invention constructs a loosely coupled hierarchical control architecture combining a basic PID controller and a PPO reinforcement learning compensator. The basic PID controller adopts a cascade structure, with its outer loop responsible for speed tracking and its inner loop responsible for distributing the total required torque to the four wheels. This controller serves as a steady-state base, stably outputting the basic torque vector at a control frequency of 100Hz. The PPO compensator, as a safety compensation unit, is configured as follows: Figure 4 The actor-critic network structure is shown. The compensator uses the state deviation output from the state observer and the basic control quantity as network inputs, and directly outputs the four-wheel compensation torque vector at a control frequency of 20Hz. This compensator only generates the superimposed torque value and does not participate in the adjustment or iteration of any parameters within the PID controller. Ultimately, the total control input is obtained by weighted fusion of the torque commands output from the two loops mentioned above, with the formula: Total Control Input: The dynamic weighting coefficients satisfy Both values ​​range from 0 to 1. Based on the PID control vector, This is the PPO compensation control vector. , These are dynamic weighting coefficients.

[0068] The overall control command output by this invention The torque command is for four-wheel independent drive, which is applied independently to the four hub motors at the front left, front right, rear left, and rear right, forming a differential drive and independent compensation control mode.

[0069] The PPO compensator outputs differentiated compensation amounts for the faulty motor and the healthy motor, achieving single-wheel failure, four-wheel coordination, independent adjustment, and differential balancing. From a dynamic perspective, it quickly counteracts yaw interference caused by torque imbalance, ensuring the vehicle maintains high-precision trajectory tracking and lateral stability even under fault conditions. Combined with... Figure 4 The PPO actor-critic network structure shown in this embodiment uses a three-layer fully connected network for both the Actor and Critic: the input layer is the state bias vector. With basic control vector The hidden layers are two fully connected layers, both using ReLU activation functions; the Actor output layer provides independent torque compensation for four wheels. The activation function used is Tanh; the Critic output layer uses a state-value function with a linear activation function. The network employs gradient descent for synchronous optimization, resulting in smooth policy updates and high training efficiency.

[0070] 5. Stability Constraints and Dynamic Weight Scheduling

[0071] To ensure the stability of the control system, this embodiment uses the deviation vector between the actual vehicle state and the health model state described in step 2. As the independent variable, construct a quadratic Lyapunov function, whose expression is:

[0072]

[0073] in, It is a 9×9 positive definite weighting matrix, and the specific values ​​in the embodiment are... = This embodiment uses the energy value calculated by the Lyapunov function to schedule weight coefficients in real time. The magnitude of the energy value directly reflects the degree to which the current system state deviates from the healthy baseline. The scheduling logic is as follows: when the system is fault-free and the energy value is low, Dominantly relying on basic PID control to ensure steady-state accuracy; when a fault is detected, or the system deviation worsens or the energy value increases, the weighting coefficient of the PPO compensation loop is dynamically increased. The more severe the fault, the higher the compensation weight, in order to achieve rapid correction.

[0074] Lyapunov function Taking the first-order time derivative along the system trajectory, we get:

[0075]

[0076] Under ideal assumptions:

[0077] (1) The system is free from external disturbances;

[0078] (2) Failure coefficient Constant or slowly changing;

[0079] (3) Control input satisfies , and satisfy ;

[0080] (4) The PPO compensation term reduces the deviation. Monotonic convergence.

[0081] Substituting the closed-loop dynamics, we get:

[0082]

[0083] Under the design conditions where Lyapunov conditions are satisfied and control weights dynamically increase with deviation, it can be guaranteed that:

[0084]

[0085] Where α>0 is a constant.

[0086] According to Lyapunov's theorem, a closed-loop system is globally asymptotically stable under ideal conditions; under bounded perturbations, the system is uniformly and eventually bounded.

[0087] 6. Optimization of Multi-Objective Reward Function

[0088] To train and optimize the policy network in the PPO compensation loop, a reward function that integrates multiple performance metrics and security constraints is designed. This embodiment uses a weighted sum implementation as follows:

[0089]

[0090] The engineering roles of each weighting coefficient and penalty term are as follows:

[0091] =5.0 is the lateral trajectory error penalty coefficient, used to ensure trajectory tracking accuracy;

[0092] =3.0 is the yaw rate error penalty coefficient, used to ensure the lateral stability of the vehicle;

[0093] =2.0 is the Lyapunov energy penalty coefficient, used to force the system to asymptotically stable convergence;

[0094] =0.1 is the control increment penalty coefficient, used to suppress drastic jitter in the output command and improve control smoothness;

[0095] =10.0 is the actuator constraint penalty coefficient, used to limit motor torque and temperature rise, and avoid hardware overload and secondary damage.

[0096] Through the aforementioned multi-objective weighted reward function, the PPO agent can simultaneously optimize tracking accuracy, system stability, control smoothness, and actuator safety during training.

[0097] 7. Closed-loop safety control output and verification

[0098] Within each control cycle, the actual vehicle state, health state, and the deviation vector between them are output and filtered by the EKF state observer. The weights calculated in step 4 and the torque commands calculated in step 3 are fused together to generate the total control quantity. Finally, a strict torque limiting constraint is applied to the total control quantity, restricting it to a safe range of 0 to 305 N·m. The limited command is then sent to the drivers of the four hub motors to form a complete closed-loop safety fault-tolerant control.

[0099] To verify the effectiveness of the method in this embodiment, two sets of typical test conditions were set up:

[0100] Condition 1: 100m circular acceleration condition. At the 20th second of the simulation, a fault with a 30% torque reduction is injected into motor 1 (left front wheel).

[0101] Condition 2: Double lane change steering condition. At the 3rd second of the simulation, a fault with a 30% torque reduction is injected into motor 3 (right rear wheel).

[0102] Meanwhile, three schemes were set up for effect comparison: a scheme without any fault-tolerant control (no fault tolerance), a fault-tolerant scheme based on traditional reinforcement learning to adjust PID parameters (traditional RL fault tolerance), and the scheme proposed in the embodiments of this invention.

[0103] Figure 6 The curves showing the gradual convergence of reward values ​​during the training process of the method of this invention are presented. The training tends to stabilize and converge after about 4000 rounds. After convergence, the average reward fluctuation is less than 5%, with no violent oscillations, indicating that the policy learning efficiency is high and the training process is stable and reliable. Figure 7 and Figure 8 The comparison of vehicle trajectories under dual lane change and circular acceleration conditions is shown respectively. It can be seen intuitively that the actual trajectory of the method of the present invention almost coincides with the expected trajectory, and the root mean square error of trajectory tracking does not exceed 0.08m, which is 61.9% lower than the traditional RL fault-tolerant scheme. Figure 9 and Figure 10 By comparing the changes in vehicle sideslip angle under two operating conditions, the method of this invention can effectively clamp the peak sideslip angle to 0 rad, demonstrating excellent lateral stability. Figure 11 A comprehensive comparison of multiple core performance indicators shows that the method in this embodiment is significantly better than the comparative scheme in terms of fault detection F1 score (0.92), fault response time (≤55ms), and maximum motor temperature rise (≤18.5℃), thus fully verifying its effectiveness.

[0104] To verify the independent contributions of the FFT-WMFCC fault detection and PPO direct compensation architectures of this invention, two sets of comparative schemes were added: Comparative scheme A uses traditional frequency domain features (FFT only) + PPO direct compensation; Comparative scheme B uses FFT-WMFCC features + RL-tuned PID indirect compensation. The results show that: Comparative scheme A has a fault detection F1 score of only 0.83, failing to identify early, weak faults; Comparative scheme B has a fault response delay ≥2.8s, resulting in poor dynamic tracking performance. This invention simultaneously optimizes fault detection accuracy and response speed, achieving unexpected technical effects.

[0105] To further verify its adaptability to different failure modes, quantitative tests were conducted under constant failure, gradual failure, and intermittent failure conditions, such as... Figure 12As shown: Under constant fault conditions, the detection time of this invention is ≤55ms, F1=0.92, and tracking error is ≤0.08m; under gradual fault conditions, this invention can detect the fault trend 0.6s to 1.2s in advance, F1=0.90, and tracking error is ≤0.09m; under intermittent fault conditions, F1=0.91, and error is ≤0.10m. Traditional methods have a detection delay of ≥1.5s for gradual faults, making early prediction impossible. Experiments demonstrate that this invention has a significant advance sensing capability for gradual faults.

[0106] 8. Control Process

[0107] Based on the above steps, the complete execution flow of this embodiment is as follows:

[0108] (1) Power on the system and initialize the vehicle model parameters, Lyapunov weight matrix, PPO algorithm hyperparameters and EKF filter parameters.

[0109] (2) The six vehicle status signals are sampled cyclically at a frequency of 1000Hz and the data is synchronized through a 10-point buffer.

[0110] (3) Perform the FFT-WMFCC fault feature extraction process on the cached data to determine whether a torque attenuation fault has occurred.

[0111] (4) The EKF state observer is based on the measurement signal and the dynamic model, and outputs the filtered actual state, health state reference value and current state deviation vector $\mathbf{e}$.

[0112] (5) Based on the state deviation vector e, calculate the value of the Lyapunov function and its derivative, determine the stability margin of the system, and calculate and output the dynamic weight coefficients in real time. and .

[0113] (6) Basic PID controller (100Hz) outputs basic torque vector The PPO compensator (20Hz) outputs a compensation torque vector. The final control command is obtained after weighted fusion.

[0114] (7) The final control command is output to each wheel hub motor after being limited by torque from 0 to 305 N·m.

[0115] (8) Return to step 2 and repeat until the vehicle is powered off.

[0116] Through the complete closed-loop control process described above, this embodiment successfully achieves accurate detection of early faults in the hub motor, millisecond-level direct compensation response, global closed-loop stability assurance, and safe operation throughout the actuator's entire life cycle.

[0117] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for rapid detection and direct compensation control of hub motor faults in electric vehicles with stability constraints, characterized in that, Includes the following steps: (1) A vehicle dynamics model characterizing the torque decay fault of the hub motor is established based on Newton-Euler equations; (2) The vehicle status signal is sampled at high frequency, and the fault features for detecting early torque decay faults are extracted using the fast FFT-WMFCC method. (3) Construct a parallel hierarchical fault-tolerant control architecture; the architecture includes an independently operating basic PID control loop and a PPO reinforcement learning compensation loop; wherein, the total control input is obtained by weighted fusion of the torque commands output by the above two loops; (4) Construct a Lyapunov function based on the deviation between the actual state of the vehicle and the state of the health model. Based on the energy value of the function, dynamically schedule the weights allocated to the PPO compensation loop in the weighted fusion step (3). The Lyapunov function is used to ensure that the closed-loop system is globally asymptotically stable under preset conditions or consistently eventually bounded under bounded disturbances. (5) Design a multi-objective reward function to optimize the strategy of PPO compensation loop; the reward function includes penalty terms related to trajectory tracking error, convergence of the Lyapunov function, smoothness of control action and actuator constraints. (6) Closed-loop fault-tolerant control is achieved by using EKF state observation and torque limiting to output the final control command.

2. The method for rapid detection and direct compensation control of electric vehicle hub motor faults with stability constraints according to claim 1, characterized in that, The hub motor torque attenuation fault model mentioned in step (1) is as follows: ;in, For a healthy motor, the desired torque, This represents the actual output torque after the fault. Let be the fault coefficient of the i-th hub motor (i=1, 2, 3, 4). It covers constant faults, gradual faults, and intermittent faults.

3. The method for rapid detection and direct compensation control of electric vehicle hub motor faults with stability constraints according to claim 2, characterized in that, The FFT-WMFCC method described in step (2) specifically includes: synchronously sampling multiple signals including vehicle speed, lateral speed, yaw rate, roll angle, pitch angle, and torque of each wheel; windowing the sampled signals and performing fast Fourier transform to extract the energy of the frequency bands related to mechanical faults, electrical faults, and power imbalances as frequency domain features; fusing the frequency domain features with the time domain features according to a preset weight ratio, and then generating the fault features through weighted Mel-Cepstral Transform.

4. The method for rapid detection and direct compensation control of electric vehicle hub motor faults with stability constraints according to claim 3, characterized in that, In the parallel hierarchical fault-tolerant control architecture described in step (3), the total control input is represented as: ； in, This is the overall control input vector for the vehicle. The four-wheel basic control vectors output by the basic PID controller. The four-wheel compensation control vector output by the PPO compensation controller. The weighting coefficients for basic PID control. The dynamic weights of the PPO compensation control weight coefficients satisfy... , 0≤ ≤1, 0≤ ≤1.

5. The method for rapid detection and direct compensation control of electric vehicle hub motor faults with stability constraints according to claim 4, characterized in that, The basic PID control loop in step (3) is a cascade structure and operates at the first control frequency; the PPO compensation loop directly generates independent compensation torque for the four wheels and operates at a second control frequency lower than the first control frequency.

6. The method for rapid detection and direct compensation control of electric vehicle hub motor faults with stability constraints according to claim 1, characterized in that, The dynamic scheduling in step (4) specifically means: when the energy value of the Lyapunov function represents the degree to which the system deviates from the healthy state, the weight coefficient corresponding to the PPO compensation loop is dynamically increased.

7. The method for rapid detection and direct compensation control of electric vehicle hub motor faults with stability constraints according to claim 6, characterized in that, The Lyapunov function mentioned in step (4) is: ； In the formula This represents the deviation vector between the actual vehicle state and the health model state. It is a positive definite weighted matrix; The degree to which the system deviates from its healthy state is caused by a torque attenuation fault in the hub motor.

8. The method for rapid detection and direct compensation control of electric vehicle hub motor faults with stability constraints according to claim 1, characterized in that, The multi-objective reward function described in step (5) Designed as follows: ； For trajectory tracking error term, Let Lyapunov function values ​​represent the stability convergence term. The control smoothing term characterizes the magnitude of torque command changes between adjacent time points. Actuator constraint penalty applied to actions exceeding preset torque and / or temperature rise thresholds.

9. The method for rapid detection and direct compensation control of electric vehicle hub motor faults with stability constraints according to claim 8, characterized in that, The torque limit range mentioned in step (6) is from 0 to the maximum rated torque of the motor.

10. A four-wheel independent drive electric vehicle controller, characterized in that, The controller is configured to perform the method according to any one of claims 1 to 9.

Images