Wheel motor fault control method and system for unmanned ground vehicle

Abstract

This invention relates to a control method and system for unmanned ground vehicles experiencing hub motor failure. The method includes: S1, collecting operational status data and hub motor operational data of the unmanned ground vehicle; S2, establishing a nominal prediction model for the vehicle based on the operational status data; S3, constructing a residual sequence based on the prediction results of the nominal prediction model and the actual measurement results of the vehicle, and performing statistical analysis on the residual sequence based on a dynamic sliding window adaptive residual analysis mechanism to output faulty wheel position information, fault indication quantity, and health factor; S4, driving a residual learning model with historical vehicle operational data, current vehicle status, control input, the residual sequence, and the health factor input data. In this invention, the nominal model provides a prediction framework that conforms to physical laws and the ability to handle explicit constraints, ensuring the interpretability and safety of the control method.

Description

Technical Field

[0001] This invention relates to the field of unmanned ground vehicles, and in particular to a control method and system for unmanned ground vehicles in the event of hub motor failure. Background Technology

[0002] Four-wheel hub motor-driven unmanned ground vehicles possess advantages such as independent drive, high degree of control freedom, and strong maneuverability, making them promising for unmanned inspection, unmanned transportation, and complex operation scenarios. These vehicles can achieve trajectory tracking and stability control by independently adjusting the drive torque of each wheel. However, hub motors are susceptible to factors such as component aging, road impacts, thermal loads, electromagnetic interference, and sensor malfunctions during actual operation, leading to faults such as output attenuation, response hysteresis, torque bias, and even failure. Hub motor failures alter the vehicle's drive force distribution, causing a mismatch between the vehicle's actual dynamic characteristics and the control model, thus affecting the vehicle's trajectory tracking accuracy and driving stability.

[0003] In the existing technology, control methods based on vehicle dynamics models have good physical interpretability, but they are sensitive to model accuracy under fault conditions and have difficulty adapting to changes in the actuation capability of hub motors. Although data-driven control or compensation methods have certain nonlinear fitting capabilities, they usually have problems such as difficulty in explicitly handling constraints and insufficient engineering interpretability.

[0004] Therefore, there is an urgent need for a fault-tolerant control method for hub motors that can balance model constraint expression capability and online adaptive capability, so as to improve the control performance and operational safety of unmanned ground vehicles under fault conditions. Summary of the Invention

[0005] In view of this, the present invention aims to propose a control method and system for unmanned ground vehicles in the event of hub motor failure, so as to solve the problems in the prior art.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows: This invention discloses a control method for unmanned ground vehicles in the event of hub motor failure, comprising the following steps: S1 collects operational status data and hub motor operation data of unmanned ground vehicles; S2, Establish a vehicle nominal prediction model based on the operating status data; S3, construct a residual sequence based on the prediction results of the vehicle nominal prediction model and the actual measurement results of the vehicle, and perform statistical analysis on the residual sequence based on the dynamic sliding window adaptive residual analysis mechanism to output fault wheel position information, fault indication quantity and health factor. S4, drive the residual learning model with the vehicle's historical operating data, current vehicle status, control input, residual sequence, and health factor input data to obtain the model mismatch compensation amount and its uncertainty; S5, inject the model mismatch compensation amount into the vehicle nominal prediction model to obtain the data model collaborative prediction model; S6. Based on the data model and collaborative prediction model, construct a fault-tolerant model predictive control to solve for the total longitudinal force of the vehicle target and the target additional yaw moment. S7. Based on the health factors, the upper and lower limits of the available torque of each wheel hub motor are dynamically updated, and the wheel torque reconstruction and distribution problem is constructed based on the total longitudinal force of the whole vehicle and the target additional yaw moment. The torque command of each wheel hub motor is obtained by solving the problem.

[0007] Furthermore, the vehicle nominal prediction model established in S2 has a state vector that includes the vehicle's longitudinal velocity, lateral velocity, yaw rate, angular velocity of each wheel, position coordinates of the vehicle in the inertial coordinate system, and heading angle.

[0008] Furthermore, the residual sequence constructed in S3 is generated based on the difference between the reference angular velocity and the actual angular velocity of each wheel; the length of the dynamic sliding window is adaptively adjusted according to the vehicle speed fluctuation, yaw rate and lateral acceleration.

[0009] Furthermore, the health factor is a continuous variable with a value range between 0 and 1. The smaller the value, the more severe the corresponding hub motor failure. It is calculated by the ratio of the comprehensive fault evaluation quantity to the dynamic threshold. The fault indication quantity is generated based on the comparison result between the health factor and the preset threshold.

[0010] Furthermore, the data-driven residual learning model in S4 is a Bayesian Transformer model, whose input features include vehicle state vector, control input vector, health factor vector, and residual vector; its output includes the predicted compensation mean as the model mismatch compensation amount and the predicted uncertainty used to quantify the compensation uncertainty.

[0011] Furthermore, the predicted compensation mean is directly injected into the vehicle's nominal prediction model as a state compensation term to correct the state prediction equation; the prediction uncertainty is used to dynamically adjust the cost function weights or safety margins of the constraints in the fault-tolerant model's predictive control problem.

[0012] Furthermore, the fault-tolerant model predictive control constructed in S6 has optimization objectives including at least trajectory tracking error, control increment, slack variables, and prediction uncertainty from the data-driven residual learning model; and its constraints include at least vehicle yaw stability constraints, tire adhesion constraints, and actuator saturation constraints.

[0013] Furthermore, the wheel torque reconstruction and distribution in S7 has optimization objectives including at least an error term for tracking the total longitudinal force and additional yaw moment of the vehicle, a torque change smoothing term, and a slip energy consumption suppression term.

[0014] This invention also discloses an unmanned ground vehicle control system for implementing the above method in the case of hub motor failure, comprising: The information acquisition module is used to collect vehicle operating status information and hub motor operating information; The nominal model module is used to build and output state prediction results based on the vehicle's nominal prediction model. The fault perception module is used to analyze the residual between the prediction results and the actual measurement results based on the dynamic sliding window adaptive residual analysis mechanism, and output the fault wheel position information, fault indication quantity and health factor. The data model collaborative prediction module is used to output the model compensation amount and compensation uncertainty through a data-driven residual learning model based on the vehicle's historical operating data, current status, control input, fault perception information and residual sequence, and to construct a data model collaborative prediction model. The fault-tolerant control module is used to solve the fault-tolerant model predictive control problem based on the data model and the collaborative prediction model, and output the total longitudinal force of the vehicle target and the additional yaw moment of the target. The torque reconstruction and allocation module is used to update the available torque constraints of each wheel hub motor according to the health factors, and solve for the torque command of each wheel hub motor.

[0015] Compared with the prior art, the present invention has the following advantages: In this invention, the nominal model provides a predictive framework that conforms to physical laws and the ability to handle explicit constraints, ensuring the interpretability and safety of the control method; while the data-driven model learns online and compensates for model mismatch errors caused by hub motor failures, enabling the control system to adaptively cope with unknown fault types and time-varying fault degrees, overcoming the shortcomings of pure model methods in terms of accuracy degradation under fault conditions and the difficulty of constraint handling in pure data methods. Attached Figure Description

[0016] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram showing the kinematic relationship and tire forces of the four-wheel hub motor independently driven unmanned ground vehicle according to the present invention. Figure 2 This is a schematic diagram of the overall architecture of the control system of the present invention; Figure 3 This is a flowchart of the control method of the present invention; Figure 4 This is a comparison chart of the trajectory tracking performance of the present invention under single-wheel failure conditions; Figure 5 This is a comparison chart of the trajectory tracking performance of the present invention under dual-wheel failure conditions. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0018] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "back," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0019] Furthermore, in the description of this invention, unless otherwise explicitly defined, the terms "installation," "connection," "linking," and "connector" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention in light of the specific circumstances.

[0020] The following will refer to the appendix. Figures 1 to 5 The present invention will be described in detail with reference to the embodiments.

[0021] Example 1 Overall, this invention discloses a control method for unmanned ground vehicles in the event of hub motor failure, comprising the following steps: S1 collects operational status data and hub motor operation data of unmanned ground vehicles; S2, Establish a vehicle nominal prediction model based on the operating status data; S3, construct a residual sequence based on the prediction results of the vehicle nominal prediction model and the actual measurement results of the vehicle, and perform statistical analysis on the residual sequence based on the dynamic sliding window adaptive residual analysis mechanism to output fault wheel position information, fault indication quantity and health factor. S4, drive the residual learning model with the vehicle's historical operating data, current vehicle status, control input, the residual sequence, and health factor input data to obtain the model mismatch compensation amount and its uncertainty; S5. Inject the model mismatch compensation amount into the vehicle nominal prediction model to obtain the data model collaborative prediction model. S6, based on the data model and collaborative prediction model, construct a fault-tolerant model to predict the control problem, and solve the total longitudinal force of the vehicle target and the target additional yaw moment; S7 dynamically updates the upper and lower limits of the available torque of each wheel hub motor based on health factors, and constructs a wheel torque reconstruction and distribution problem based on the total longitudinal force of the vehicle target and the target additional yaw moment, and solves the torque command of each wheel hub motor.

[0022] In this embodiment, the nominal vehicle prediction model established in S2 has a state vector that includes the vehicle's longitudinal velocity, lateral velocity, yaw rate, angular velocity of each wheel, position coordinates of the vehicle in the inertial coordinate system, and heading angle.

[0023] The signal data collected in the state vector is the vehicle's longitudinal velocity. lateral velocity yaw rate Heading angle Location coordinates and angular velocity of the four wheels , , , Front wheel steering angle Output torque of each wheel hub motor , , , and control cycle number .

[0024] The vehicle coordinate system includes the vehicle body coordinate system. and inertial coordinate system The origin of the vehicle coordinate system is set at the vehicle's center of mass. The axis points in the direction the vehicle is moving. The axis points to the left side of the vehicle; the inertial coordinate system is used to characterize the vehicle's position and heading in the global plane.

[0025] The relationship of the vehicle's position change in the inertial coordinate system is as follows: In the formula, and These represent the position coordinates of the vehicle's center of mass in the inertial coordinate system; and These represent the first derivatives of the position coordinates with respect to time; Indicates the vehicle's heading angle; This represents the longitudinal velocity of the vehicle in the vehicle coordinate system; This represents the lateral velocity of the vehicle in the vehicle coordinate system.

[0026] Define the vehicle state vector as follows: In the formula, Indicates time The vehicle state vector, Indicates the yaw rate of the vehicle. , , , These represent the angular velocities of the left front wheel, right front wheel, left rear wheel, and right rear wheel, respectively.

[0027] Define the vehicle-level control input vector as: In the formula, Indicates time The total longitudinal force of the entire vehicle; Indicates time The target is to add yaw moment.

[0028] When describing using the target longitudinal acceleration: In the formula, Indicates the overall vehicle weight. This indicates the longitudinal acceleration of the target.

[0029] The vehicle nominal value prediction model is established as follows: It should be further explained that the vehicle's planar dynamics in the vehicle coordinate system satisfy: In the formula, Indicates the overall vehicle weight. Represents a set of wheels. and They represent the first The longitudinal and lateral tire forces of each wheel in the vehicle's coordinate system. This represents the yaw moment of inertia of the vehicle about its center of mass. and They represent the first The position coordinates of each wheel relative to the vehicle's center of mass in the vehicle coordinate system. This indicates an additional yaw moment.

[0030] It should be further explained that the rotational dynamics of a wheel can be expressed as: In the formula, This represents the equivalent moment of inertia of the wheel. Indicates the first The angular velocity of each wheel; Indicates the first Each hub motor outputs torque; Indicates the effective rolling radius of the wheel; Indicates the rolling resistance coefficient; Indicates the first Normal load on each wheel.

[0031] The force transformation relationship between the tire coordinate system and the vehicle coordinate system is as follows: In the formula, Indicates the first The steering angle of each wheel; for front-wheel steering vehicles, the front wheel steering angle can be taken as... The rear wheel steering angle is zero.

[0032] No. The speed of the wheel center can be expressed as: In the formula, and They represent the first The longitudinal and lateral velocities of each wheel center in the vehicle coordinate system.

[0033] Accordingly, the first The slip ratio and sideslip angle of each wheel are: In the formula, Indicates the first Wheel slip ratio, Indicates the first Each wheel slip angle and Both represent extremely small positive numbers to prevent the denominator from being zero.

[0034] Based on the above settings, as a preferred embodiment, the longitudinal and lateral forces of the tire are represented using the MagicFormula model as follows: In the formula, , , , These represent the longitudinal peak factor, shape factor, stiffness factor, and curvature factor, respectively. , , , These represent the lateral peak factor, shape factor, stiffness factor, and curvature factor, respectively.

[0035] After linearizing and discretizing the above continuous model at the current operating point, the nominal prediction model is obtained: In the formula, Indicates time The nominal predicted state, Represents the discrete state matrix. This represents a discrete input matrix.

[0036] The residual sequence constructed in S3 is generated based on the difference between the reference angular velocity and the actual angular velocity of each wheel; the residual sequence is constructed as follows: In the formula, Indicates time No. The residual of each wheel Indicates the first Reference angular velocity of each wheel Indicates the first The actual angular velocity of each wheel.

[0037] The length of the dynamic sliding window is adaptively adjusted based on vehicle speed fluctuations, yaw rate, and lateral acceleration. It is represented as: In the formula, Indicates the length of the dynamic sliding window. and These represent the lower and upper bounds of the window length, respectively. Indicates the base window length. , , These represent the weighting coefficients corresponding to velocity fluctuation, yaw rate, and lateral acceleration, respectively. This represents the standard deviation of vehicle speed fluctuation. Indicates the lateral acceleration of the vehicle. ω represents the yaw rate.

[0038] Inside the window, the first The mean and standard deviation of the residuals for each wheel are as follows: In the formula, This represents the mean of the residuals within the window. This represents the standard deviation of the residuals within the window.

[0039] The dynamic threshold is expressed as: in: In the formula, Indicates the first Dynamic fault threshold for each wheel Indicates the dynamic threshold adjustment factor. Represents the basic threshold coefficient. This represents the weighting coefficient for the shift in direction. This represents the average steering angular velocity of the front wheels.

[0040] Based on the above settings, the health factor is a continuous variable with a value range between 0 and 1. The smaller the value, the more severe the corresponding hub motor failure. It is calculated by the ratio of the comprehensive failure evaluation quantity to the dynamic threshold.

[0041] It should be noted that the comprehensive fault evaluation metric is constructed as follows: In the formula, Indicates the first The comprehensive fault evaluation quantity of each wheel, This represents the weighting coefficient between the current residual and the statistical residual.

[0042] Health factors are defined as: In the formula, Indicates time No. Health factors of each wheel This represents a very small positive number to prevent the denominator from being zero. The closer the health factor is to 1, the closer the execution capability of this round is to a normal state; the closer the health factor is to 0, the more severe the failure of this round.

[0043] In this embodiment, the fault indication is generated based on the comparison result between the health factor and a preset threshold.

[0044] In the formula, Indicates the first Fault indication of each wheel, This indicates the threshold for determining health factors.

[0045] In this embodiment, since there is a deviation between the actual vehicle dynamics and the nominal model under fault conditions, the data-driven residual learning model in S4 is preferably a Bayesian Transformer model to compensate for the prediction error online. Its input features include the vehicle state vector, control input vector, health factor vector, and residual vector; its output includes the mean of the prediction compensation as the model mismatch compensation amount and the prediction uncertainty used to quantify the compensation uncertainty.

[0046] In detail, the input feature vector for constructing the model is: In the formula, Indicates time The joint input feature vector, Represents the vehicle state vector. Represents the control input vector. Represents a health factor vector. This represents the residual vector.

[0047] The timing input, after embedding, is represented as: In the formula, Indicates the first Embedded representation of a time-series input Represents the input embedding matrix. This represents the positional encoding vector.

[0048] As a further improvement to the control method of this invention, the Bayesian Transformer is trained using variational inference, and its objective function is: In the formula, Describe the objective function representing the lower bound of the evidence. Represents network parameters, Represents the training dataset. Represents the posterior distribution of the parameters. This represents the prior distribution of the parameters.

[0049] As a further improvement to the control method of the present invention, in the online inference stage, the first The compensation output of the second sampling is expressed as: In the formula, Indicates the first The compensation output obtained from the second sampling. This represents the Bayesian Transformer mapping function. Indicates the length of the timing window.

[0050] As a further improvement to the control method of the present invention, the model compensation mean and compensation uncertainty are expressed as follows: In the formula, Indicates time The average compensation, Indicates time The compensated covariance matrix, Indicates the number of samples.

[0051] After injecting the compensation mean into the nominal prediction model, the resulting data model collaborative prediction model is as follows: In the formula, This indicates the predicted state after compensation.

[0052] The predicted compensation mean is directly injected into the vehicle's nominal prediction model as a state compensation term to correct the state prediction equation; the prediction uncertainty is used to dynamically adjust the cost function weights or safety margins of the constraints in the fault-tolerant model's predictive control problem.

[0053] Based on the above settings, the fault-tolerant model predictive control constructed in S6 has optimization objectives that include at least trajectory tracking error, control increment, slack variables, and prediction uncertainty from the data-driven residual learning model.

[0054] Based on the current vehicle status Reference trajectory Health factors Together with a collaborative prediction model, we construct model predictive control. The cost function is defined as: In the formula, This represents the model's predictive control cost function. Indicates the length of the prediction time domain. Indicates the length of the control time domain. Indicates time For time The predicted output, Indicates reference output, This represents the output error weight matrix. This indicates the control increment. This represents the control increment weight matrix. This represents the penalty coefficient for slack variables. Represents slack variables. This represents the uncertainty penalty weight.

[0055] Its constraints include at least vehicle yaw stability constraints, tire adhesion constraints, and actuator saturation constraints, as shown below: In the formula, and These represent the state bound and the upper bound, respectively. and These represent the upper and lower bounds of the control input after contraction based on health factors. This indicates the upper limit of the permissible yaw rate. This indicates the sideslip angle of the vehicle's center of gravity at the predicted time. This indicates the upper limit of the allowable sideslip angle. This represents the road surface adhesion coefficient.

[0056] By solving the above optimization problem, the target control quantity for the entire vehicle is obtained: That is, the total longitudinal force of the target Additional yaw moment to the target After obtaining the total longitudinal force and additional yaw moment of the vehicle, the vehicle-level targets are converted into torque commands for the four-wheel hub motors. Furthermore, the wheel torque reconstruction and distribution in S7 optimizes at least the following objectives: an error term for tracking the total longitudinal force and additional yaw moment of the vehicle, a torque change smoothing term, and a slip energy consumption suppression term.

[0057] The vehicle-level target vector is represented as: In the formula, This represents the target vector at the vehicle level.

[0058] The torque vector of the four wheels is represented as follows: In the formula, This represents the torque vector of the four-wheel hub motor.

[0059] The mapping relationship between the vehicle-level target and the wheel torque is represented by the allocation matrix as follows: In the formula, Represents the wheel torque distribution matrix. Indicates the wheel track of the vehicle.

[0060] The objective function for torque reconstruction and distribution is expressed as: In the formula, This represents the torque distribution cost function. This represents the weight matrix for vehicle target tracking. This represents the torque smoothing weighting coefficient. This represents the weighting coefficient for slip energy consumption penalty.

[0061] As a further improvement to the control method of the present invention, the torque constraints of each wheel are expressed as follows: when At that time, take: In the formula, and They represent the first The minimum and maximum torque that each wheel is allowed to output under healthy conditions.

[0062] In summary, the nominal model provides a predictive framework that conforms to physical laws and the ability to handle explicit constraints, ensuring the interpretability and safety of the control method; while the data-driven model learns online and compensates for model mismatch errors caused by hub motor failures, enabling the control system to adaptively cope with unknown fault types and time-varying fault degrees, overcoming the shortcomings of pure model methods in terms of accuracy degradation under fault conditions and the difficulty of constraint handling in pure data methods.

[0063] Example 2 This embodiment discloses a control system for an unmanned ground vehicle experiencing a hub motor failure, which is used to implement the method in embodiment one above, including: The information acquisition module is used to collect vehicle operating status information and hub motor operating information; The nominal model module is used to build and output state prediction results based on the vehicle's nominal prediction model. The fault perception module is used to analyze the residual between the prediction results and the actual measurement results based on the dynamic sliding window adaptive residual analysis mechanism, and output the fault wheel position information, fault indication quantity and health factor. The data model collaborative prediction module is used to output the model compensation amount and compensation uncertainty through a data-driven residual learning model based on the vehicle's historical operating data, current status, control input, fault perception information and residual sequence, and to construct a data model collaborative prediction model. The fault-tolerant control module is used to solve the fault-tolerant model predictive control problem based on the data model and the collaborative prediction model, and output the total longitudinal force of the vehicle target and the additional yaw moment of the target. The torque reconstruction and allocation module is used to update the available torque constraints of each wheel hub motor according to the health factors, and solve for the torque command of each wheel hub motor.

[0064] Example 3 This embodiment illustrates the trajectory tracking control effect of the present invention under single-wheel hub motor failure conditions. The test subject is an unmanned ground vehicle independently driven by four-wheel hub motors, with the vehicle traveling at a target speed. The vehicle travels along the center line of a pre-set S-shaped road. During the test, after the vehicle reaches the predetermined position, a fault is applied to the left front wheel hub motor.

[0065] In this embodiment, the on-board controller first constructs a nominal prediction model based on the real-time vehicle state. The vehicle state vector can be represented as: In the formula, Indicates the longitudinal speed of the vehicle. Indicates the lateral speed of the vehicle. Indicates yaw rate. , , , These represent the angular velocities of the left front wheel, right front wheel, left rear wheel, and right rear wheel, respectively. , Indicates the vehicle's position coordinates. Indicates the vehicle's heading angle.

[0066] The fault detection module constructs a residual based on the difference between the reference angular velocity and the actual angular velocity of the left front wheel: In the formula, Indicates the residual weight of the left front wheel. This indicates the reference angular velocity of the left front wheel. This indicates the actual angular velocity of the left front wheel.

[0067] After a fault is detected, a dynamic sliding window residual analysis mechanism is used to calculate the health factor: In the formula, Indicates the health factor of the left front wheel. This indicates the overall fault assessment score for the left front wheel. Indicates the dynamic threshold of the left front wheel. This represents the smallest positive number that prevents the denominator from being zero.

[0068] Once a fault in the left front wheel is detected, the data-model co-prediction module calculates the compensation amount based on the state vector, control input, health factor, and residual vector, and then injects it into the prediction model. In the formula, The discrete state matrix, For discrete input matrices, To control the input vector, This is the compensated mean output by the Bayesian Transformer.

[0069] The fault-tolerant model predictive controller solves for the total longitudinal force and additional yaw moment of the vehicle target: Furthermore, the vehicle-level control objective is mapped to the torque of the four wheels, and the torque distribution optimization model is as follows: In the formula, This represents the torque vector of the four wheels. This represents the vehicle-level target vector. Represents the torque distribution matrix. This represents the target tracking weight matrix. This indicates the torque smoothing weight.

[0070] The torque constraint on the left front wheel is reduced based on health factors: In the formula, This indicates the torque command for the left front wheel. and These represent the minimum and maximum torque of the left front wheel when it is in good working order.

[0071] Experimental results show that after a left front wheel failure, an uncontrolled vehicle rapidly deviates from the road boundary. While traditional MPC can still track the reference path, the lateral deviation increases significantly after the failure. In contrast, the BT-MPC method described in this invention maintains the trajectory closer to the reference centerline, exhibiting higher stability and fault tolerance. Statistical results show that under this single-wheel failure condition, the mean lateral error of the method described in this invention is [value missing]. Lower than traditional MPC ; the lateral error variance is Lower than traditional MPC Meanwhile, the mean longitudinal error of the method of this invention is Lower than traditional MPC The longitudinal error variance is Lower than traditional MPC This demonstrates that the present invention can significantly improve trajectory tracking accuracy and reduce error fluctuations under single-wheel failure conditions.

[0072] In addition, after the left front wheel failed in the experiment, its steering angle and driving torque changed abruptly. The other three wheels quickly compensated and adjusted under the action of the controller. The steering action of the right front wheel increased, and the rear wheels made coordinated adjustments in amplitude and phase. At the same time, the torque of the remaining healthy wheels increased to compensate for the power loss of the failed wheel, thereby maintaining the vehicle's stable driving along the S-shaped path.

[0073] Example 4 This embodiment illustrates the robust fault-tolerant control effect of the present invention under dual-wheel hub motor failure conditions. The test object remains an unmanned ground vehicle independently driven by four-wheel hub motors, with the vehicle traveling at a target speed. The vehicle travels along the center line of a pre-set circular road. During the test, a fault is applied to both the left front wheel and the right rear wheel simultaneously while the vehicle is in motion.

[0074] In this embodiment, the fault detection module constructs residuals for the left front wheel and the right rear wheel respectively: In the formula, and These represent the angular velocity residuals of the left front wheel and the right rear wheel, respectively.

[0075] Furthermore, the health factor vector under dual-fault conditions is obtained: In the formula, and It decreases significantly after a failure occurs, and is used to characterize the degree of degradation in the corresponding wheel's actuation capability.

[0076] The data-model co-prediction module uses the following input feature vectors for online compensation estimation: In the formula, Represents the four-round residual vector. This represents the joint feature vector input to the Bayesian Transformer.

[0077] Bayesian Transformer output model compensation mean and compensation uncertainty The nominal model is modified through the following relationship: The fault-tolerant model predictive controller solves the control problem based on the modified predictive model, and its cost function can be expressed as: In the formula, Indicates the length of the prediction time domain. Indicates the length of the control time domain. Indicates the predicted output. Indicates reference output, and These represent the weight matrices, This represents the uncertainty penalty weight.

[0078] During torque distribution, the torque boundaries of the left front wheel and the right rear wheel contract according to their respective health factors: When the health factor drops to zero, the torque of the corresponding faulty wheel is set to zero, and the remaining healthy wheels take on the compensation task.

[0079] Experimental results show that when both the left front wheel and right rear wheel fail simultaneously, the uncontrolled vehicle quickly loses its trajectory tracking ability and deviates significantly during turns. While traditional MPC still retains some tracking capability, its lateral and longitudinal errors increase significantly in circular curves. In contrast, the BT-MPC method described in this invention maintains a better fit to the reference centerline, demonstrating higher control accuracy and robustness. Statistical results show that under dual-wheel failure conditions, the mean lateral error of the method described in this invention is [value missing]. Significantly lower than traditional MPC ; the lateral error variance is Significantly lower than traditional MPC Meanwhile, the mean longitudinal error of the method of this invention is Lower than traditional MPC The longitudinal error variance is Lower than traditional MPC This demonstrates that the present invention can effectively suppress average deviation and extreme value fluctuations, maintaining stable vehicle tracking even under conditions of dual-wheel failure.

[0080] Furthermore, after a dual-wheel failure, the steering angle response of the left front wheel and right rear wheel rapidly decreases, while the remaining healthy wheels, namely the right front wheel and left rear wheel, exhibit significant compensatory changes in amplitude and phase. The corresponding drive torque is also rapidly reconstructed, with a significant increase in torque for the right front wheel and left rear wheel. On curved sections, frequent adjustments and even short-term reverse torque occur to generate sufficient lateral force to ensure stable vehicle cornering. Therefore, the method described in this invention can achieve coordinated reconstruction of steering and drive torque under multiple failure conditions, demonstrating strong engineering applicability.

[0081] This patent addresses the issues of model mismatch, increased trajectory deviation, and decreased stability that easily occur in four-wheel hub motor-driven unmanned ground vehicles when the hub motors fail. It proposes a data-model collaborative fault-tolerant control method and system. The core idea is as follows: first, online fault detection and health factor assessment are achieved through dynamic sliding window residual analysis; then, Bayesian Transformer is used to compensate for model errors under fault conditions online; based on this, model predictive control is combined to solve for the total longitudinal force and additional yaw moment of the entire vehicle; finally, wheel torque reconstruction and distribution are completed based on health factors. The patent's key feature is the organic integration of fault perception, model correction, fault-tolerant control, and torque reconstruction. It can maintain good trajectory tracking accuracy and driving stability even under single-wheel or multi-wheel hub motor failures, demonstrating strong engineering application value.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A control method for an unmanned ground vehicle in case of hub motor failure, characterized in that, Includes the following steps: S1 collects operational status data and hub motor operation data of unmanned ground vehicles; S2, Establish a vehicle nominal prediction model based on the operating status data; S3, construct a residual sequence based on the prediction results of the vehicle nominal prediction model and the actual measurement results of the vehicle, and perform statistical analysis on the residual sequence based on the dynamic sliding window adaptive residual analysis mechanism to output fault wheel position information, fault indication quantity and health factor. S4, drive the residual learning model with the vehicle's historical operating data, current vehicle status, control input, residual sequence, and health factor input data to obtain the model mismatch compensation amount and its uncertainty; S5, inject the model mismatch compensation amount into the vehicle nominal prediction model to obtain the data model collaborative prediction model; S6. Based on the data model and collaborative prediction model, construct a fault-tolerant model predictive control to solve for the total longitudinal force of the vehicle target and the target additional yaw moment. S7. Based on the health factors, the upper and lower limits of the available torque of each wheel hub motor are dynamically updated, and the wheel torque reconstruction and distribution problem is constructed based on the total longitudinal force of the whole vehicle and the target additional yaw moment. The torque command of each wheel hub motor is obtained by solving the problem.

2. The method for controlling an unmanned ground vehicle in case of hub motor failure according to claim 1, characterized in that: The nominal vehicle prediction model established in S2 has a state vector that includes the vehicle's longitudinal velocity, lateral velocity, yaw rate, angular velocity of each wheel, position coordinates of the vehicle in the inertial coordinate system, and heading angle.

3. The unmanned ground vehicle control method for hub motor failure according to claim 1, characterized in that: The residual sequence constructed in S3 is generated based on the difference between the reference angular velocity and the actual angular velocity of each wheel; the length of the dynamic sliding window is adaptively adjusted according to the vehicle speed fluctuation, yaw rate and lateral acceleration.

4. The unmanned ground vehicle control method for hub motor failure according to claim 1, characterized in that: The health factor is a continuous variable with a value between 0 and 1. The smaller the value, the more severe the corresponding hub motor failure. It is calculated by the ratio of the comprehensive fault evaluation quantity to the dynamic threshold. The fault indication quantity is generated based on the comparison result of the health factor and the preset threshold.

5. The unmanned ground vehicle control method for hub motor failure according to claim 1, characterized in that: The data-driven residual learning model in S4 is a Bayesian Transformer model. Its input features include vehicle state vector, control input vector, health factor vector, and residual vector. Its output includes the mean of the predicted compensation as the model's mismatch compensation amount and the predicted uncertainty used to quantify the compensation uncertainty.

6. The unmanned ground vehicle control method for hub motor failure according to claim 5, characterized in that: The predicted compensation mean is directly injected into the vehicle's nominal prediction model as a state compensation term to correct the state prediction equation; the prediction uncertainty is used to dynamically adjust the cost function weights or safety margins of the constraints in the fault-tolerant model's predictive control problem.

7. The unmanned ground vehicle control method for hub motor failure according to claim 1, characterized in that: The fault-tolerant model predictive control constructed in S6 has optimization objectives including at least trajectory tracking error, control increment, slack variables, and prediction uncertainty from the data-driven residual learning model; its constraints include at least vehicle yaw stability constraints, tire adhesion constraints, and actuator saturation constraints.

8. The unmanned ground vehicle control method for hub motor failure according to claim 1, characterized in that: The wheel torque reconstruction and distribution in S7 has optimization objectives including at least an error term for tracking the total longitudinal force and additional yaw moment of the vehicle, a torque change smoothing term, and a slip energy consumption suppression term.

9. An unmanned ground vehicle control system for implementing the method according to any one of claims 1 to 8 in the case of a hub motor failure, characterized in that, include: The information acquisition module is used to collect vehicle operating status information and hub motor operating information; The nominal model module is used to build and output state prediction results based on the vehicle's nominal prediction model. The fault perception module is used to analyze the residual between the prediction results and the actual measurement results based on the dynamic sliding window adaptive residual analysis mechanism, and output the fault wheel position information, fault indication quantity and health factor. The data model collaborative prediction module is used to output the model compensation amount and compensation uncertainty through a data-driven residual learning model based on the vehicle's historical operating data, current status, control input, fault perception information and residual sequence, and to construct a data model collaborative prediction model. The fault-tolerant control module is used to solve the fault-tolerant model predictive control problem based on the data model and the collaborative prediction model, and output the total longitudinal force of the vehicle target and the additional yaw moment of the target. The torque reconstruction and allocation module is used to update the available torque constraints of each wheel hub motor according to the health factors, and solve for the torque command of each wheel hub motor.

Images