Motor failure fault-tolerant torque distribution system for distributed driving electric vehicle

Abstract

The invention discloses a motor failure fault-tolerant torque distribution system for a distributed driving electric vehicle, and belongs to the technical field of vehicle control. The invention aims to solve the technical problem of man-machine conflict caused by tension misoperation of a driver due to sudden change of dynamic response of a vehicle after a motor fault, and the system comprises a data acquisition and processing unit, a self-correction module, a driver intention and cognitive state cognition module, a predictive deduction module and a man-machine collaborative torque optimization module. The self-correction module identifies motor fault evolution characteristics on line and corrects a digital twinborn model in real time to reproduce fault dynamics, the predictive deduction module is based on a corrected pathological digital twinborn body and is coupled with a driver model, potential man-machine conflicts are predicted and restrained, the stability of a vehicle is guaranteed, and meanwhile the reliability of the vehicle is improved. Smooth and predictable fault response is achieved, and the cooperative driving safety of the vehicle under the fault working condition is fundamentally improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle control, in particular to a motor failure fault-tolerant torque distribution system applied to a distributed drive electric vehicle. BACKGROUND

[0002] The distributed drive electric vehicle is equipped with independent drive motors on each wheel or axle, which has the ability to independently adjust the driving or braking torque of each wheel with high precision and high dynamic response. This driving configuration not only greatly broadens the boundaries of vehicle dynamics control, but also provides an ideal execution platform for improving vehicle handling stability and driving safety through advanced torque vectoring control technology. When the vehicle is driving normally, these systems can work efficiently to optimize driving performance and energy consumption.

[0003] However, as the complexity of the system increases, the possibility of failure of the drive motor and its control unit also rises. The motor may lose part or all of its torque output ability due to overheating, loss of sensor signal, internal short circuit, and other reasons. This sudden power loss will directly disrupt the original force balance of the vehicle, especially in conditions such as turning or high-speed driving, which require high stability, and may cause the vehicle to produce unintended yaw or roll, posing a serious threat to driving safety. Therefore, researching and deploying effective motor failure fault-tolerant control technology is crucial to ensuring the safety of distributed drive electric vehicles.

[0004] Currently, fault-tolerant control technology for such failures has become a research hotspot in this field. The conventional technical route usually focuses on one core goal: after detecting a fault, quickly redistribute the output torque of the remaining healthy motors through control algorithms to compensate for the loss of driving force and yaw moment of the faulty motor at the physical level. These methods strive to restore the vehicle's overall motion state (such as longitudinal acceleration and yaw angular velocity) to the driver's desired value as quickly as possible, and the core of their design is to treat the vehicle as an independent physical object to be stabilized, with the goal of maintaining the mathematical stability of the vehicle dynamics model.

[0005] However, the prior art generally ignores a crucial dynamic factor, the driver in the closed-loop system, in the process of achieving the above-mentioned objectives. While the traditional fault-tolerant strategy can theoretically maintain the macro trajectory of the vehicle when performing aggressive torque compensation, it may cause dramatic and unnatural changes in the transient response characteristics of the vehicle. For example, to compensate for the torque loss of the outside wheels, the driving force of the inside wheels is instantaneously increased substantially, which makes the vehicle feel abrupt and "forced into" the curve. This nonlinear, contrary to the driver's daily experience, vehicle dynamic feedback is easily misinterpreted by the driver as a precursor to vehicle loss of control, triggering his instinctive and excessive corrective operation, such as unnecessary steering wheel counterpunching or emergency braking. This driver's antagonistic operation induced by the controller behavior forms a dangerous "man-machine conflict", which may escalate a controllable fault scenario into a real danger.

[0006] Therefore, the existing fault-tolerant control method, while solving the problem of physical stability of the vehicle to some extent, fails to fully consider the cooperative stability of man and machine as a whole system. They lack a forward-looking perspective and cannot foresee and actively avoid the negative impact of their control strategy on the driver's behavior. There is an urgent need in the art for a new fault-tolerant control paradigm that not only stabilizes the vehicle but also stabilizes the driver, achieving harmonious coexistence and cooperative safety of the man-vehicle system under fault conditions. SUMMARY

[0007] To solve the above technical problems, the present application provides a motor failure fault-tolerant torque distribution system for a distributed drive electric vehicle. The system predicts the future state of the "man-vehicle" closed-loop system including the driver by constructing a digital twin model that can dynamically self-correct, and decides the optimal torque distribution scheme based on the optimization goal of man-machine cooperation.

[0008] Specifically, the technical solution provided by the present application includes:

[0009] A self-correcting module, a predictive reasoning module, and a man-machine cooperative torque optimization module.

[0010] The function of the self-correcting module is to ensure that the vehicle model relied upon by the system can accurately reproduce the actual dynamic characteristics of the faulty vehicle. The module monitors the operating state of each drive motor of the vehicle in real time, detects faults by comparing the residual error between the motor command value and the sensor feedback value. Once a fault is detected, the module will immediately identify and parameterize the fault behavior, thereby obtaining the motor fault evolution characteristics that characterize the fault dynamic process. The motor fault evolution characteristics can be one or a combination of the following:

[0011] Torque output decay factor: Characterizes the degree of reduction in the maximum torque that a motor can output due to performance degradation compared to its normal state.

[0012] Torque response lag time: Characterizes the delay in the actual torque output of a faulty motor after receiving a torque command.

[0013] Output torque fluctuation intensity: Quantifies the irregular, non-commanded fluctuations or jitter characteristics in the output torque of a faulty motor.

[0014] After obtaining the above characteristics, the self-correction module injects and modifies these characteristic parameters into the model parameters of the corresponding faulty motor in a digital twin model that is running synchronously with the actual vehicle, thereby transforming the digital twin model from a healthy vehicle mirror to a dynamic model that can accurately simulate the "sick" working state.

[0015] The core function of the predictive reasoning module is to provide the decision-making system with the ability to predict future risks after a fault occurs. This module is configured to receive the digital twin model modified by the self-correction module and use it as the basis for reasoning. Instead of reasoning about a single strategy, it simulates the dynamic response of the vehicle when multiple candidate fault-tolerant torque distribution strategies are applied to the modified digital twin model in parallel. Crucially, the module introduces a pre-set driver model during simulation, which can predict the closed-loop operations (such as steering correction and pedal adjustment) that the driver may generate based on the simulated vehicle dynamic response (such as yaw and roll). In this way, the module ultimately outputs multiple possible evolution states of the human-vehicle system as a whole within a pre-set time window under different torque strategies.

[0016] The human-machine collaborative torque optimization module is the decision-making core of the system. This module receives and evaluates multiple human-vehicle system evolution states generated by the predictive reasoning module. The basis for evaluation is a specially designed collaborative safety cost function. Through optimization calculation, the module selects the optimal strategy that minimizes the cost function value and generates the optimal fault-tolerant torque distribution command accordingly. This command is a torque vector composed of torque values assigned to each remaining healthy drive motor, which is ultimately issued to the corresponding motor controller for execution.

[0017] To achieve human-machine collaboration, the collaborative safety cost function is specially configured to include a human-machine behavior conflict cost term, which aims to quantify and penalize fault-tolerant torque distribution strategies that, although physically stable the vehicle, may not conform to the driver's intuition and even induce the driver to produce intense antagonistic operations. Specifically, the human-machine behavior conflict cost J cmf can be calculated by the following formula:

[0018]

[0019] wherein M z,pred (t) is the predicted yaw moment generated by the system for stabilizing the vehicle, as derived by the predictive derivation module;

[0020] is the predicted steering wheel angular velocity of the driver, as derived by the driver model; and f ] is the preset time window. The physical meaning of this formula is that when the direction of the stabilizing moment generated by the system is opposite to that of the driver's correction operation and both are relatively intense, the cost value will significantly increase, thereby inhibiting such strategies in the optimization process.

[0021] Further, to achieve comprehensive safety control, the collaborative safety cost function can further include other cost terms, such as: a vehicle posture stability cost for punishing the deviation of the predicted vehicle yaw motion from the driver's intention; an intention following cost for punishing the deviation of the predicted vehicle trajectory from the driver's desired path; and an actuator limit cost for punishing the overload of the remaining healthy drive motor or the saturation of tire adhesion.

[0022] In a preferred embodiment, the system further includes a driver intention and cognitive state cognition module. On the one hand, the module infers the driver's driving intention (such as the desired path and the desired acceleration) based on the driver's real-time operation instructions (such as the steering wheel angle and the pedal opening), and provides the intention as a reference to the predictive derivation module. On the other hand, the module evaluates the real-time cognitive state of the driver by analyzing the dynamic characteristics of the driver's operation signals, for example, by calculating the cognitive load index I cog to quantify the tension of the driver:

[0023]

[0024] wherein P δ (ω) is the power spectral density of the steering wheel angle signal, [ω h,1 ,ω h,2 ] is the preset high-frequency operation frequency interval that can reflect the tension state of the driver, is the variance of the acceleration pedal depression rate, and c1 and c2 are preset weight coefficients. The cognitive state evaluation result can be used to dynamically adjust the weights of each cost term in the collaborative safety cost function, thereby making the fault-tolerant strategy more adaptive.

[0025] The application provides a motor failure fault-tolerant torque distribution system for a distributed drive electric vehicle. The application has the following advantages:

[0026] 1、The invention is no longer limited to the traditional fault-tolerant control of the "lost sheep to repair the loss" type of thinking, that is, only in the physical level to stabilize the vehicle. By introducing the predictive reasoning module, the invention can foresee the impact of different compensation strategies on the driver's psychology and behavior, thereby prospectively eliminating the potential vicious cycle of "machine mutation - human panic - man-machine confrontation". This method improves safety from the root, and upgrades the control target from single vehicle physical stability to the harmony and stability of the whole "man-vehicle" closed-loop system.

[0027] 2、Unlike traditional methods that rely on static and healthy models, the self-correcting module of the invention can capture and quantify the subtle torque output decay and response lag of the motor after failure occurs. It no longer creates a general and idealized vehicle model, but a "pathological" digital twin that can reflect the current fault characteristics in real time. It is this highly realistic dynamic model that makes subsequent future state reasoning of human-vehicle interaction have realistic significance and sufficient accuracy, ensuring the reliability of the final decision.

[0028] 3、The invention goes beyond the traditional cognition of regarding simple driver operation as the control target, and draws a dynamic "cognitive portrait" for the driver by analyzing the slight shaking on the steering wheel or the disorder of the acceleration rhythm. The system can determine whether the driver is in a calm state or a tense state, and use this information to dynamically adjust the decision weight in the human-machine collaborative torque optimization module. This makes the intervention scale of the fault-tolerant strategy adaptively match the driver's psychological tolerance, providing appropriate assistance rather than surprise at critical moments.

[0029] 4、When making an optimal decision, the invention introduces a unique value scale - the human-machine behavior conflict cost function. It enables the system to not only measure the efficiency of a fault-tolerant strategy in stabilizing the vehicle, but also quantifies the risk of the strategy inducing the driver to produce antagonistic correction operations. Those compensation actions that are stable in mathematics but abrupt and counterintuitive in body sense will have a high "conflict cost" and thus be suppressed in decision-making. This ensures that the system ultimately selects a solution that is physically effective and psychologically most acceptable to the driver.

[0030] 5、The technical solution of the invention is significantly improved in terms of the driver's actual experience. When a fault occurs, the driver no longer feels the abrupt and shocking "pulling sensation" or "loss of control", but a smooth and perceptible transition after the system intervention. This smooth dynamic response can greatly alleviate the driver's panic, allowing him to remain calm during the critical window period after the failure, thus retaining confidence and ultimate control over the vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1A system function structure block diagram of the present application;

[0032] Figure 2 A method overall work flow chart of the present application;

[0033] Figure 3 A principle schematic diagram of the evolution characteristic of the present application;

[0034] Figure 4 A comparative schematic diagram of the digital twin model self-correction effect of the present application;

[0035] Figure 5 A principle schematic diagram of the driver cognitive state evaluation of the present application. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be described in detail below. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0037] Referring to the drawings Figure 1 The distributed drive electric vehicle motor failure fault-tolerant torque distribution system can include a vehicle-mounted high-performance computing platform, various vehicle-mounted sensors and vehicle actuator controllers in hardware implementation. The system is integrated as a whole for cooperative work in software layer, which can include the following modules: a data acquisition and processing unit a, a self-correction module b, a driver intention and cognitive state cognition module c, a predictive reasoning module d, and a man-machine cooperative torque optimization module d.

[0038] The overall working principle of the system is that the real-time state information of the vehicle and the driver is obtained through the data acquisition and processing unit a. During normal driving of the vehicle, the system background continuously runs a digital twin model highly synchronized with the real vehicle state. When detecting that any drive motor fails, the self-correction module b is activated. Instead of regarding the failure as a simple binary event, the self-correction module b immediately quantitatively characterizes the dynamic evolution process of the failure, and modifies the internal parameters of the digital twin model in real time according to the characterization result, so that it changes from a healthy vehicle mirror image to a "pathological mirror image" that can accurately reproduce the failure characteristics.

[0039] Meanwhile, the driver intention and cognitive state cognitive module c continuously analyzes the driver's actions to infer their true driving intentions and current level of tension. Subsequently, the predictive inference module d, based on the corrected "pathological mirror image," infers multiple different torque compensation strategies in parallel, coupling the driver model during the inference process to predict the evolution trend of the entire "human-vehicle" system in the next few seconds after the driver interacts with the "sick" vehicle under each strategy.

[0040] Ultimately, the human-machine collaborative torque optimization module d comprehensively evaluates these predicted future evolution trends. By solving a collaborative safety cost function that includes the cost of human-machine interaction conflicts, it selects an optimal torque distribution scheme that not only stabilizes the vehicle but also guides the driver to perform smooth and correct operations. The instructions generated by this scheme are then sent to the vehicle's motor controller for execution. This process repeats cyclically within a high-frequency control cycle, thus forming a forward-looking closed-loop control system capable of adapting to fault evolution and driver dynamics.

[0041] Reference Figure 1 , Figure 3 , Figure 4 and Figure 5 In one specific embodiment of the present invention, the core of the digital twin model is based on a vehicle dynamics model capable of describing key vehicle dynamics. For example, a three-degree-of-freedom model, mathematically expressed as follows:

[0042] Longitudinal motion equation:

[0043]

[0044] Lateral motion equations:

[0045]

[0046] The yaw motion equation is described as the sum of the moments of all external forces acting on the vehicle about its Z-axis, specifically expressed as the sum of the yaw moment generated by the longitudinal force and the yaw moment generated by the lateral force:

[0047]

[0048] Among them, the yaw moment M generated by the longitudinal force z,x for:

[0049]

[0050] The yaw moment M generated by the lateral force z,y for:

[0051] M z,y =L f (F y,1 +Fy,2 )-L r (F y,3 +F y,4 );

[0052] In the above formula, the definitions of the symbols are as follows: m represents the mass of the whole vehicle; v x and v y represent the longitudinal and lateral vehicle speeds in the vehicle coordinate system, respectively; γ represents the yaw rate of the vehicle; I z represents the moment of inertia of the vehicle around the Z axis; F x,i and F y,i represent the longitudinal force and lateral force on the wheel numbered i (1-left front, 2-right front, 3-left rear, and 4-right rear); F wind represents the air resistance; L f and L r represent the distances from the vehicle mass center to the front axle and rear axle, respectively; W f and W r represent the front wheel track and rear wheel track, respectively. This dynamic model constitutes the basis for all subsequent predictive derivations.

[0053] In the specific implementation of the present application, the cooperative work of the core modules of the system is the key to realizing the technical effects of the present application. The specific implementation methods of the data acquisition and processing unit a, the self-correction module b, the driver intention and cognitive state cognition module c, the predictive derivation module d, and the human-machine cooperative torque optimization module d will be described in detail below.

[0054] The data acquisition and processing unit a is the information input end of the whole system. It acquires the data stream of the vehicle bottom layer sensors and controllers through the vehicle controller area network (CAN) bus at a high frequency. These data include but are not limited to: the wheel speed sensor signals of the four wheels, the angle signals of the steering wheel angle sensor and the torque signals of the torque sensor, the three-axis acceleration and three-axis angular velocity output by the inertial measurement unit (IMU), the opening signals of the acceleration and braking pedals, and the actual speed, current and temperature information feedback by the drive motor controllers. The acquired raw data stream is preprocessed by this unit, including necessary low-pass filtering to eliminate high-frequency noise, coordinate system transformation to unify the data reference, and fusion estimation of key vehicle states (such as longitudinal vehicle speed v x , lateral vehicle speed v y ) through state estimation algorithms (such as Kalman filter) to provide stable and reliable data input to the subsequent modules.

[0055] The preprocessed data stream is delivered to the self-correction module b, which is the core of ensuring the consistency of the digital twin model and the dynamic of the faulty real vehicle. This module first performs fault detection. Specifically, it compares the command torque T cmd,iThe actual output torque T est,i estimated based on the motor feedback current and speed i is calculated to find the residual e cmd,i = T est,i When the absolute value of the residual exceeds a preset dynamic threshold that can tolerate normal working errors, the system determines that the motor i has failed.

[0056] Once the failure is confirmed, the self-repairing module b immediately starts an online parameter identification procedure to quantitatively characterize the behavior of the failure, i.e., to extract the motor failure evolution features. In this embodiment, the identification process can use a recursive least squares method (RLS) with a forgetting factor to identify the following set of parameters that can describe the failure dynamics in an online and real-time manner: a torque output decay factor λ T,i with a value range of 0 to 1, representing the proportion of the motor output capacity compared to the healthy state; a torque response lag time τ d,i to quantify the degree of delay of the actual torque response to the command; and an output torque fluctuation intensity σ n,i representing the standard deviation of the non-command random fluctuations in the output torque of the failed motor.

[0057] Subsequently, the identified set of failure feature parameters is used to modify the mathematical model of the corresponding failed motor i in the digital twin model in real time. If the healthy motor model is a standard first-order inertial link, the modified failure motor model can be expressed as:

[0058]

[0059] where T m,i is the actual output torque of the motor i, T cmd,i is the control command torque, τ m,i is the inherent time constant of the motor, λ T,i and τ d,i are the online identified failure feature parameters. w i (t) is a Gaussian white noise process, and its statistical characteristics (such as standard deviation) are determined by the identified output torque fluctuation intensity σ n,i Through this mechanism, the digital twin model completes the self-repairing from the healthy state to the "sick" state.

[0060] At the same time that the self-repairing module b is working, the driver intention and cognitive state recognition module c analyzes the driver behavior in parallel. This module first calculates the driver's expected yaw rate γ sw based on the driver's steering wheel angle δ x and the current vehicle speed v des :

[0061]

[0062] where L is the wheelbase of the vehicle, N gear is the steering ratio, and K is the understeer gradient. Meanwhile, this module maps out the longitudinal acceleration a acc desired by the driver based on the accelerator pedal opening a brk and brake pedal pressure P brk . x,des .

[0063] More importantly, this module assesses the current cognitive state of the driver by analyzing the dynamic characteristics of the driver's operation signals. Specifically, a cognitive load index I cog is calculated to quantify the degree of tension or overreaction of the driver:

[0064]

[0065] In this formula, P δ (ω) is the power spectral density of the steering wheel angle signal, and the integration interval [ω h,1 ,ω h,2 ] is set to cover the frequency range of typical high-frequency correction operations by human drivers in a state of tension; is the variance of the accelerator pedal depression rate; and c1 and c2 are preset weight coefficients. The result of the calculation of this cognitive load index will be used in subsequent optimization decisions.

[0066] The self-correction module b and the output of the driver intention and cognitive state recognition module c are jointly fed into the predictive reasoning module d. The core task of this module is to simulate the future. It first generates a series of candidate fault-tolerant torque distribution strategies, and then for each candidate strategy, it performs forward simulation based on the corrected digital twin model within a preset future time window. At each time step of the simulation, the system not only calculates the dynamic response of the vehicle, but also predicts the feedback operation that the driver might make to this response using the driver model, and uses this operation as the input to the next time step, thereby realizing complete reasoning of the "man-vehicle" closed-loop interactive system, and finally generating multiple future state evolution trajectories.

[0067] These future state trajectories generated by the predictive reasoning module d form the basis for decision-making by the man-machine collaborative torque optimization module d. This module sorts all candidate strategies by solving a multi-objective collaborative safety cost function J. This cost function J is constructed as a weighted sum of multiple sub-cost terms:

[0068] J = w stab J stab + w path J path + w conf J conf + wlim J lim

[0069] where w stab ,w path ,w conf ,w lim are weight coefficients of each term, whose values can be dynamically adjusted according to the driver's intention and the cognitive load index output by the cognitive state recognition module c. Each sub-cost term defines the requirements for vehicle posture stability (J stab ), the degree of compliance with the driver's original path intention (J path ), the suppression of human-machine behavior conflicts (J conf ), and the respect for the physical limits of the remaining healthy actuators (J lim ).

[0070] In the optimization solving process, any candidate torque distribution scheme must satisfy the physical constraints of the vehicle, mainly including the peak torque constraint of each remaining healthy drive motor, and the tire-road adhesion force of each wheel cannot exceed its adhesion limit, i.e. satisfying the tire adhesion ellipse constraint. Finally, by using efficient numerical optimization algorithms such as sequential quadratic programming (SQP), the optimal fault-tolerant torque distribution instruction that minimizes the total cost function J is solved under the premise of satisfying all constraints, and is executed.

[0071] Referring Figure 2 , in a preferred embodiment of the present application, the overall workflow performed by the motor failure fault-tolerant torque distribution system of the distributed drive electric vehicle is a closed-loop process that is repeated in a high-frequency control cycle. The following will be described in detail through a series of continuous steps.

[0072] Step S101: System initialization and real-time state synchronization.

[0073] After the vehicle is powered on, the system is first initialized, and the baseline digital twin model is loaded, which contains the vehicle dynamics parameters, tire model parameters, and performance model of each drive motor in the healthy state strictly consistent with the physical parameters of the real vehicle. After initialization, the system enters a continuous running state, continuously acquires real-time data of various sensors from the vehicle bus through the data acquisition and processing unit a, and drives the digital twin model to run, so that its internal state, such as vehicle speed, yaw rate, wheel speed, etc. is highly synchronized with the state of the real vehicle.

[0074] Step S102: Fault online detection and process activation judgment.

[0075] In each control cycle, the system performs a routine health check on all the driving motors. Specifically, the self-revision module b calculates the residual error between the command torque and the actual output torque of each motor. The system compares the residual error with a preset dynamic threshold that can accommodate the model error and measurement noise under normal conditions. If the residual error of all motors is within the threshold range, the system determines that the vehicle is in a healthy state, the fault-tolerant control process is not activated, and the system directly returns to step S101 to continue the state synchronization. Otherwise, if the residual error of any motor continuously exceeds the threshold, the system determines that the motor has failed and immediately activates the subsequent fault-tolerant control process, entering step S103.

[0076] Step S103: Fault characterization and driver state co-cognition.

[0077] Once the fault-tolerant process is activated, the system will start two key analysis modules in parallel. On the one hand, the self-revision module b immediately starts the online parameter identification algorithm for the confirmed faulty motor, quantifies its motor fault evolution characteristics, obtains key parameters such as torque output attenuation factor λ T,i , torque response lag time τ d,i , and completes the self-revision of the digital twin model based on these parameters. On the other hand, the driver intent and cognitive state cognition module c analyzes the real-time operation behavior of the driver at the same time, infers the expected driving trajectory of the driver, and evaluates the current cognitive load index I cog .

[0078] Step S104: Future state deduction of the man-vehicle system based on the revised twin. This step is the core of the forward-looking control of the present application. The predictive deduction module d receives two key inputs from step S103: the revised "pathological" digital twin model that can reproduce the fault characteristics, and the driver model that contains the current intent and state of the driver. Based on these two, the module performs super-real-time simulation on multiple candidate fault-tolerant torque allocation strategies within a preset future time window (e.g., 2-3 seconds) in parallel. In the simulation, it not only simulates the dynamics of the "sick" vehicle under different strategies, but more importantly, it couples the driver model into the simulation loop to predict the instinctive closed-loop correction operation of the driver on the abnormal dynamics of the vehicle. Finally, the module outputs multiple sets of future state trajectory data that can fully exhibit the interactive evolution process of the "man-vehicle" system.

[0079] Step S105: Human-machine collaborative torque optimization decision.

[0080] The human-machine collaborative torque optimization module d comprehensively evaluates and decides the multiple sets of future state trajectories output by step S104. It quantitatively scores each future trajectory by using a collaborative safety cost function J containing multiple sub-items as the evaluation standard. The design of the cost function aims to seek the optimal balance between vehicle stability, driver intention following, human-machine operation coordination, and system execution capability. The module solves the set of fault-tolerant torque distribution strategies that can minimize the total cost function J by executing an efficient numerical optimization algorithm under the premise of meeting all physical constraints.

[0081] Step S106: Optimal instruction issuance and closed-loop cycle. After solving the optimal fault-tolerant torque distribution scheme, the system converts it into a set of explicit torque instructions, i.e., a torque vector containing specific torque values assigned to each remaining healthy drive motor. The instructions are issued to the corresponding motor controllers through the vehicle bus for accurate execution. At this point, a complete fault-tolerant control cycle is completed. Subsequently, the system seamlessly returns to step S101 to start a new round of state synchronization, fault evolution monitoring, prediction, and optimization decision-making using the latest sensor data, thereby forming a closed-loop control system that can dynamically adapt to fault development and driver state changes.

[0082] To more specifically illustrate the implementation process of the technical solutions provided by the present application and the beneficial effects brought by them, the application of the present application will be exemplified by a typical fault scenario. It should be understood that this scenario is only one example of the many possible applications of the present application, and its purpose is to illustrate how the present application solves specific technical problems, rather than any form of limitation on the scope of protection of the present application.

[0083] In this embodiment, assume that a distributed drive electric vehicle equipped with the system of the present application is making a smooth left turn at a speed of 60 km / h on a wet road. During this process, the right front wheel drive motor located on the outside of the turn begins to gradually degrade in torque output capability due to initial insulation deterioration of the internal winding.

[0084] At the beginning of system operation, the self-correction module b continuously monitors the residual error e between the command torque and the estimated output torque of the right front motor i is always within the normal threshold. When the motor fault begins to appear, the residual error e i begins to steadily exceed the threshold, and the self-correction module b immediately confirms the fault and starts online parameter identification. The module identifies that the main manifestation of the motor fault evolution characteristic of this motor is the continuous decline of the torque output attenuation factor λ T,i , for example, slowly decreasing from 1.0 to 0.7 within a few seconds, while other fault characteristic parameters do not change significantly. The module immediately converts this dynamically changing λ T,iThe value is updated in real time into the performance model of the right front motor in the digital twin model.

[0085] At this moment, the driver, in order to maintain the intended steering trajectory, applies a steering angle δ sw to the steering wheel, which remains substantially constant. From this, the driver intent and cognitive state cognitive module c infers the driver's desired yaw rate γ des and uses this as an important benchmark for subsequent optimization.

[0086] Subsequently, the predictive reasoning module d, based on the revised "sick" digital twin model with the right front motor capability decayed to 70%, starts to reason over a variety of candidate fault-tolerant torque allocation strategies. For the sake of illustration, two representative strategies are listed here:

[0087] Strategy A (locally aggressive compensation strategy): The goal of this strategy is to quickly and completely compensate for the lost yaw moment on the front axle of the vehicle. The specific approach is to add the full drive torque lost by the right front wheel to the left front wheel in equal amount, in an attempt to mathematically instantaneously balance the yaw moment of the vehicle.

[0088] Strategy B (human-machine collaborative global compensation strategy): This strategy is the one preferred by the invention. It dynamically and proportionally allocates the drive torque lost by the right front wheel to the remaining three healthy motors (left front, left rear, right rear) according to an optimization principle that takes into account the smoothness of the vehicle's dynamic response and the tire load, rather than just the left front motor.

[0089] In the simulation of the predictive reasoning module d, the two strategies exhibit drastically different future evolution trends of the "human-vehicle" system. For strategy A, since the compensation moment is completely provided by the left front wheel, the drive force difference between the left and right wheels of the front axle increases dramatically, and the vehicle yaw response predicted by the digital twin model is abnormally sensitive, showing an "oversteer" phenomenon that exceeds the driver's expectation. Coupled with the driver model, the driver is predicted to feel uneasy due to this abrupt, nonlinear vehicle dynamics and will instinctively and quickly correct the steering wheel to counteract this trend.

[0090] For strategy B, since the compensation moment is more smoothly distributed throughout the vehicle, although the vehicle yaw rate predicted by the digital twin model can also closely follow the driver's desired value γ des , its response process is more linear and gentle. Therefore, the driver model predicts that the driver will not have a strong counter-corrective intent under this smooth dynamic response, and the steering wheel operation will remain stable. Finally, the human-machine collaborative torque optimization module d evaluates the cost of the above two future trajectories. In the evaluation of strategy A, although its vehicle posture stability cost J stabThe value of J conf may be very high due to the predicted sharp steering wheel kickback operation by the driver. In contrast, strategy B has a very low value of J conf due to the predicted smooth driver behavior, and its vehicle pose stability and intent following cost are also at an excellent level. After the comprehensive calculation, the total cost function J of strategy B is significantly lower than that of strategy A.

[0091] Therefore, the system finally selects and executes strategy B. The output optimal torque distribution instruction of strategy B is issued to each motor controller. In the actual vehicle, the driver feels that the vehicle still maintains stable and predictable steering characteristics after the failure occurs, thereby avoiding the panic-induced misoperation, and finally safely maintains the human-vehicle system in a controllable state. This implementation scenario clearly shows how the present application fundamentally improves the actual driving safety of the vehicle under failure by predicting and avoiding human-machine conflicts.

[0092] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A motor failure tolerant torque distribution system for a distributed drive electric vehicle, characterized by, The method comprises the following steps: a self-repairing module is used to monitor the motor fault evolution characteristics of the faulty driving motor in real time, and dynamically correct a digital twin model running synchronously with the actual vehicle based on the motor fault evolution characteristics; a predictive reasoning module is configured to reason the evolution state of the human-vehicle system within a preset time window in the future under a plurality of candidate fault-tolerant torque distribution strategies based on the digital twin model corrected by the self-repairing module and combined with a preset driver model; a human-machine collaborative torque optimization module is used to evaluate the evolution state of the human-vehicle system under the plurality of candidate fault-tolerant torque distribution strategies by optimizing a collaborative safety cost function, and select the optimal strategy from among them to generate optimal fault-tolerant torque distribution instructions and issue them to the remaining healthy driving motor.

2. The electric motor failure tolerant torque distribution system for distributed drive electric vehicles of claim 1, wherein, The self-repairing module is specifically used for: detecting faults by comparing the residual error between the motor instruction value and the feedback value; after detecting the fault, online identifying and quantifying the motor fault evolution characteristics; injecting the identified motor fault evolution characteristics into and modifying the model parameters of the corresponding faulty motor in the digital twin model to complete the correction of the digital twin model.

3. The electric motor failure tolerant torque distribution system for distributed drive electric vehicles of claim 2, wherein, The motor fault evolution characteristics include at least one of the following: a torque output attenuation factor for representing the degree of decline in the maximum output torque capability of the motor; a torque response lag time for representing the delay characteristics of the actual torque response to the instruction torque of the motor; an output torque fluctuation intensity for representing the irregular fluctuation characteristics of the output torque of the faulty motor.

4. The electric motor failure tolerant torque distribution system for distributed drive electric vehicles of claim 1, wherein, The predictive reasoning module is specifically used for: receiving the corrected digital twin model as the basic model of vehicle state evolution; simulating a plurality of candidate fault-tolerant torque distribution strategies applied to the corrected digital twin model in parallel; in the simulation process, introducing the driver model to predict the closed-loop operation of the driver on the vehicle dynamic response to generate the future evolution state of the human-vehicle system.

5. The electric motor failure tolerant torque distribution system for distributed drive electric vehicles of claim 1, wherein, The collaborative safety cost function optimized by the human-machine collaborative torque optimization module is configured to include a human-machine behavior conflict cost term to punish the fault-tolerant torque distribution strategy that may induce the driver to produce violent antagonistic operation.

6. The electric motor failure tolerant torque distribution system for distributed drive electric vehicles of claim 5, wherein, The human-machine behavior conflict cost J conf The calculation method of the motor failure fault-tolerant torque distribution system of the distributed drive electric vehicle is defined by the following formula: wherein M z,pred (t) is a predicted yaw moment generated by the system for stabilizing the vehicle, as derived by the predictive derivation module, is the driver steering wheel angular velocity, as predicted by the driver model, [t0, t f ] is the preset time window.

7. The electric motor failure tolerant torque distribution system for distributed drive electric vehicles of claim 5, wherein, The collaborative safety cost function further includes at least one of the following: a vehicle attitude stability cost for punishing the deviation of the predicted vehicle yaw motion from the driver's intention; an intention following cost for punishing the deviation of the predicted vehicle driving trajectory from the driver's expected path; an actuator limit cost for punishing the torque distribution that causes the remaining healthy driving motor to be overloaded or the tire adhesion to be saturated.

8. The electric motor failure tolerant torque distribution system for distributed drive electric vehicles of claim 1, wherein, Further comprising a driver intention and cognitive state cognition module for: inferring the driving intention of the driver based on the real-time operation instruction of the driver and providing the intention to the predictive reasoning module; evaluating the real-time cognitive state of the driver by analyzing the dynamic characteristics of the driver's operation signal, and using the state to dynamically adjust the weight in the collaborative safety cost function.

9. The electric motor failure-tolerant torque distribution system for distributed- drive electric vehicles of claim 1, wherein, The driver's intention and cognitive state cognitive module evaluates the cognitive state by calculating a cognitive load index I cog defined by the following formula: wherein P δ (ω) is the power spectral density of the steering wheel angle signal, [ω h,1 ,ω h,2 ] is the preset high-frequency operating frequency range, a motor failure fault-tolerant torque distribution system for a distributed drive electric vehicle is the variance of the accelerator pedal depression rate, and c1 and c2 are preset weight coefficients.

10. The motor fault-tolerant torque distribution system for distributed drive electric vehicles of claim 1, wherein, The optimal fault-tolerant torque distribution instruction output by the human-machine collaborative torque optimization module is a torque vector composed of torque values allocated to each remaining healthy driving motor.

Images