Self-adaptive control method for steering stability of chassis of light passenger car

Abstract

The invention discloses a light passenger car chassis steering stability self-adaptive control method, which comprises the following steps of: collecting multi-dimensional working condition parameters in real time, and covering 12 types of key parameters such as a steering wheel rotating angle and a wheel speed; after preprocessing and abnormity diagnosis, double models are adopted to estimate the dynamic state of the vehicle in parallel; dynamically setting a control target, and realizing personalized control in combination with driver style recognition; the control quantity is optimized through multi-mode strategy dynamic switching and steering and braking system cooperative control; and actuator dynamic compensation, scenarized self-learning and multi-stage fault handling mechanisms are introduced. According to the method, the complex working condition adaptability is improved, the response speed is controlled, the subsystem cooperation efficiency is improved, the system robustness and the long-term use performance are enhanced, and the driving safety of the light bus is effectively guaranteed.

Description

Technical Field

[0001] This invention relates to the field of automotive control methods, specifically to an adaptive control method for steering stability of a light passenger vehicle chassis. Background Technology

[0002] In the driving safety performance system of light commercial passenger vehicles, chassis steering stability is the core indicator determining vehicle handling safety. It directly affects the vehicle's attitude control ability in complex road conditions (such as congested urban roads, mountain curves, and icy roads), different load states (such as empty, half-loaded, and fully loaded with cargo or passengers), and dynamic driving processes (such as rapid acceleration, rapid deceleration, and emergency steering). Currently, chassis steering stability control of light commercial passenger vehicles mainly relies on the synergistic effect of three types of technologies: traditional hydraulic power steering (HPS), electric power steering (EPS), and electronic stability program (ESP).

[0003] Traditional hydraulic power steering (HPS) systems provide steering assistance via an engine-driven hydraulic pump, with its assistance characteristics fixed by mechanical parameters. The limitation of this design is its inability to dynamically adjust the amount of assistance based on the vehicle's real-time driving conditions. In low-speed steering scenarios (such as parking or U-turns in narrow urban areas), insufficient assistance may lead to increased driver fatigue and difficulty in maneuvering. Conversely, at high speeds (such as on highways), excessive assistance may result in oversteering, significantly reducing vehicle stability.

[0004] Electric power steering (EPS) provides assistance via an electric motor, offering a degree of adjustable power steering characteristics compared to HPS. Its control logic is typically based on a preset linear or piecewise linear mapping relationship, allowing for adjustments to the power steering output based on limited operating parameters such as vehicle speed and steering wheel torque. However, this adjustment method still has significant limitations: the preset nature of the mapping relationship makes it difficult to cover complex and changing real-world driving environments. For example, when the load changes abruptly (such as suddenly loading cargo) or the road surface adhesion coefficient changes dynamically (such as alternating between dry and wet roads), the accuracy of the power steering adjustment drops significantly, failing to match the real-time changes in vehicle dynamics.

[0005] Electronic Stability Program (ESP) corrects oversteer or understeer by applying braking intervention to individual wheels. Its core control strategy is based on a threshold triggering mechanism: the system only initiates braking intervention when parameters such as the vehicle's slip angle and yaw rate exceed set thresholds. This passive intervention mode has two major problems: first, response lag; when the vehicle shows a slight tendency to become unstable or is in a dynamic transitional situation (such as the instant of entering a curve from straight driving), the intervention is often not timely, resulting in poor attitude adjustment; second, the intervention force is difficult to control precisely, and excessive or insufficient braking may exacerbate vehicle instability.

[0006] The core problems of existing technologies focus on two aspects:

[0007] Insufficient adaptability to operating conditions: Existing control methods rely on preset control parameters or fixed logic, which cannot perceive and respond to dynamic changes in the vehicle's critical state in real time. Specifically, changes in vehicle load (such as increases or decreases in the number of passengers, differences in cargo weight and distribution) will cause a shift in the center of gravity, directly altering steering characteristics (e.g., a rearward shift of the center of gravity when fully loaded can easily lead to understeer), but existing systems struggle to adaptively adjust their control strategies. Sudden changes in road surface adhesion coefficient (such as transitioning from dry asphalt to icy or snowy roads) will significantly reduce tire grip, and if the ESP braking intervention threshold is not adjusted synchronously, it can easily lead to vehicle sideslip risk. Furthermore, the impact of factors such as tire wear and tire pressure changes on steering stability has not been effectively incorporated into the control logic of existing systems, resulting in a significant decrease in control accuracy under complex operating conditions.

[0008] Lack of control coordination: The control processes of the steering system (HPS / EPS) and the ESP system are independent, lacking a dynamic coordination mechanism. Their control actions may couple and interfere with each other, weakening the overall control effect. For example, when ESP applies braking to the inner wheels to correct oversteer, if EPS continues to provide positive assistance according to normal logic, it will further exacerbate the vehicle's rotational tendency, leading to greater instability. Conversely, if the timing of EPS assist adjustment and ESP braking intervention is misaligned, it will also cause the optimization of vehicle dynamics to fail. This mode of independent subsystem operation cannot achieve coordinated optimization of the vehicle's global dynamics, making it difficult to meet the high stability requirements of light commercial vehicles in complex scenarios.

[0009] Therefore, a solution is needed. Summary of the Invention

[0010] The purpose of this invention is to provide an adaptive control method for steering stability of a light passenger vehicle chassis to solve the problems mentioned in the background art.

[0011] To achieve the above objectives, the present invention provides the following technical solution:

[0012] An adaptive control method for steering stability of a light passenger vehicle chassis, the control method being as follows:

[0013] Step 1: Real-time acquisition of multi-dimensional operating parameters

[0014] A cluster of sensors installed at various locations on the vehicle collects key parameters affecting steering stability in real time. These parameters include: using a steering wheel angle sensor to collect the steering wheel angle and rotation speed; using wheel speed sensors to obtain the real-time rotational speed of each wheel; using an acceleration sensor to measure the vehicle's longitudinal and lateral acceleration; using a sideslip angle sensor to collect the instantaneous sideslip angle of the vehicle body; using a yaw rate sensor to record the vehicle's rotational speed around its vertical axis; using a load sensor to detect the total mass and center of gravity distribution of the vehicle body; using road surface recognition sensors (such as LiDAR and cameras) to identify road surface types (dry, wet, snowy, etc.) and estimate the road adhesion coefficient; using a tire pressure monitoring system (TPMS) to obtain real-time tire pressure and wear parameters; and using force sensors on the suspension springs and shock absorbers to collect the vertical load on the suspension, and displacement sensors to collect the suspension compression. All collected parameters are sampled at a frequency of 100Hz to ensure real-time data accuracy.

[0015] Step 2: Preprocessing of operating parameters and diagnosis of anomalies

[0016] The collected multi-dimensional parameters are preprocessed, including data filtering, outlier removal, and time synchronization. A Kalman filter algorithm is used to filter noise from dynamic parameters such as sideslip angle and yaw rate, improving measurement accuracy. Anomalies caused by sensor faults are identified and removed using the 3σ criterion combined with a deep learning anomaly detection model; the deep learning model can detect gradual sensor faults 0.5-1 second in advance. Data from different sensors is synchronized and aligned based on timestamps. Redundant sensor data (such as wheel speed sensor and GPS speed signals) are weighted and fused, with the fusion weights dynamically adjusted based on the historical errors of each sensor. A sensor health assessment model is constructed to diagnose faults by comparing the deviations of redundant sensor data; if an anomaly is detected, the system automatically switches to backup sensor data or activates the parameter estimation model.

[0017] Step 3: Real-time estimation of vehicle dynamics

[0018] Based on the preprocessed operating parameters, a dual-model parallel estimation strategy (a simplified two-degree-of-freedom model and a refined multibody dynamics model) is adopted, combined with the Extended Kalman Filter (EKF) algorithm to construct a vehicle dynamics state estimation model. The two-degree-of-freedom model is used for fast calculation (cycle ≤ 10ms), while the multibody dynamics model includes detailed characteristics of the suspension and tires (cycle 50ms), and the output is fused through Kalman filtering. Specifically, this includes: estimating the actual vehicle speed based on wheel speed, longitudinal acceleration, and road adhesion coefficient; estimating the ideal yaw rate using the two-degree-of-freedom vehicle model by combining steering wheel angle, vehicle speed, and center of gravity position; calculating the instability tendency index based on the deviation between the actual and ideal yaw rates and the slip angle data; and correcting the tire lateral stiffness based on tire condition and road adhesion coefficient.

[0019] Step 4: Dynamically setting stability control targets

[0020] The control objectives are dynamically set based on the vehicle's current state: at low speeds (≤30km / h) and large turns, ease of use is prioritized, with a small steering resistance coefficient; at medium speeds (30-80km / h), a balance between ease of use and stability is achieved, with a medium steering gain coefficient; and at high speeds (≥80km / h), stability is the core focus, with a large steering damping coefficient. A driver style recognition module is introduced, which analyzes historical data such as steering operation and pedal intensity to classify drivers into aggressive, moderate, and cautious types, and adjusts the yaw rate threshold for different types. The system also supports manual selection of control modes by the driver for personalized settings.

[0021] Step 5: Dynamic switching of multimodal control strategies

[0022] Based on the dynamic state estimation results and control objectives, a multimodal control strategy library (normal stable mode, mildly corrected mode, and emergency intervention mode) is constructed and dynamically switched. In the normal mode, only EPS assist is adjusted. In the mildly corrected mode, EPS assist characteristics are adjusted, and ESP applies slight braking (≤5 bar) to a single wheel. In the emergency intervention mode, EPS provides reverse assist, and ESP applies coordinated braking (5-15 bar) to multiple wheels while reducing engine output torque. A transition condition buffer mechanism is added, and a smooth switching is achieved through a control variable ramp transition algorithm (brake pressure change rate ≤2 bar / ms, EPS assist adopts an exponential transition curve). A mode switching hysteresis threshold is established to avoid frequent jitter.

[0023] Step 6: Calculation of coordinated control quantities between steering and braking systems

[0024] For the current control mode, the coordinated control quantities of EPS and ESP are calculated; the EPS system calculates the target assist torque through a fuzzy control algorithm (fuzzy rules optimized by a genetic algorithm) to achieve nonlinear dynamic adjustment; the ESP system calculates the target braking pressure of each wheel based on the yaw rate deviation and sideslip angle using a model predictive control (MPC) algorithm; an energy optimization objective function is introduced to dynamically adjust the MPC weight coefficients, prioritizing braking wheels and assist curves with lower energy consumption; a coupling compensation model of EPS and ESP control quantities is established to correct the assist torque in real time to avoid the control effects canceling each other out.

[0025] Step 7: Actuator Dynamic Response Compensation

[0026] Considering the dynamic response delays of the EPS motor (approximately 50ms delay) and the ESP brake actuator (approximately 100ms delay), a dynamic response model for the actuators is constructed to provide advance compensation for the control input. By collecting signals of motor current, temperature, and brake line pressure fluctuations, an actuator aging model is established to estimate changes in response characteristics (such as increased delay) in real time and dynamically adjust the compensation time (0.1-0.2 seconds). The vehicle state 0.1-0.2 seconds later is predicted, and the control input is corrected to ensure that the actuator action is synchronized with the vehicle's dynamic requirements.

[0027] Step 8: Real-time feedback on control effects and parameter self-learning

[0028] Real-time acquisition of feedback parameters such as yaw rate and sideslip angle, comparison with control targets to calculate errors, and closed-loop adjustment of EPS assist torque and ESP braking pressure through PID controller; construction of a scenario-based self-learning model, dividing operating conditions into typical scenarios such as urban roads and highways, identifying scenarios through GPS and map data and calling corresponding optimization parameters; optimization of fuzzy control rules and MPC weight coefficients through reinforcement learning algorithms to adapt to vehicle aging (such as tire wear) and driving habits; support for updating the basic parameters of the self-learning model via OTA.

[0029] Step 9: Switch to Fail-Safe Mode

[0030] The system monitors the status of sensors, EPS, and ESP actuators in real time. When a critical component failure is detected, a fail-safe mode is activated. The system responds according to the severity of the failure: for minor failures (such as failure of non-critical sensors), a partial redundancy scheme is activated; for moderate failures (such as partial limitation of EPS functions), a degraded control mode is activated to retain core functions; for severe failures (such as complete failure of ESP), the system switches to mechanical backup mode, activates hazard warning lights, sends fault information and location via the vehicle network, limits the maximum vehicle speed (≤40km / h), and guides the vehicle to a safe stop.

[0031] In a preferred embodiment of the present invention, in step 1, a dynamic estimation model of suspension stiffness is established by combining suspension vertical load and compression data, and the vehicle roll stiffness parameters are corrected in real time to improve the accuracy of dynamic state estimation during sharp turns, road undulations and load switching.

[0032] In a preferred embodiment of the present invention, in step 2, the parameter fusion algorithm reduces the weight of GPS when the GPS signal is weak (such as in a tunnel environment) and increases the GPS weight when the wheel speed sensor is abnormal due to slippage; the deep learning anomaly detection model improves the ability to identify gradual faults such as slow sensor drift by training a large number of normal and abnormal working condition parameter features.

[0033] In a preferred embodiment of the present invention, in step 3, the dual-model parallel estimation strategy mainly uses the results of the two-degree-of-freedom model under simple working conditions such as flat roads, and increases the weight of the multibody dynamics model under complex working conditions such as bumpy roads and continuous curves to improve the robustness of state estimation.

[0034] In a preferred embodiment of the present invention, in step 4, the driver style recognition module increases the allowable threshold for yaw rate for aggressive drivers to reduce intervention, and lowers the threshold for cautious drivers to intervene earlier, thus balancing the needs for driving control and stability.

[0035] In a preferred embodiment of the present invention, in step 5, the transition condition buffer mechanism avoids vehicle body posture fluctuations during mode switching by limiting the maximum rate of change of braking pressure and the EPS assist transition curve, and the hysteresis threshold prevents frequent mode switching jitter under critical conditions.

[0036] In a preferred embodiment of the present invention, in step 6, the energy optimization objective function prioritizes braking of the inner rear wheel to reduce kinetic energy loss under the mildly corrected mode, considers the motor efficiency characteristics when calculating the EPS assist torque, and adjusts the ESP braking distribution strategy under the braking energy recovery condition to balance recovery efficiency and stability.

[0037] In a preferred embodiment of the present invention, in step 7, when the brake actuator response delay increases from 100ms to 150ms, the actuator health status monitoring automatically extends the compensation time from 0.1 seconds to 0.15 seconds, and triggers a maintenance reminder when the performance degradation exceeds the threshold.

[0038] In a preferred embodiment of the present invention, in step 8, the scenario-based learning mechanism focuses on optimizing the power assist characteristics and braking intervention timing during continuous steering in mountainous curve scenarios, and the learning data of different scenarios are accumulated independently to avoid interference.

[0039] In a preferred embodiment of the present invention, in step 9, the multi-level fault response strategy, in addition to limiting vehicle speed, pushes fault information in real time through the vehicle network backend for serious faults and guides the driver to the nearest repair station in combination with map data.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] 1. Significantly improves adaptability to operating conditions

[0042] The solution acquires multi-dimensional operating parameters in real time, covering key information such as vehicle steering state, wheel motion, load distribution, road surface type, and tire condition, with a sampling frequency of 100Hz to ensure real-time perception of dynamic operating conditions. Combined with a dual-model parallel estimation strategy, a two-degree-of-freedom model is used for rapid calculation in simple operating conditions (such as flat roads), while a multi-body dynamics model is weighted for complex operating conditions (such as bumpy roads and continuous curves), significantly improving the accuracy of vehicle dynamic state estimation. This design enables the system to respond in real time to factors such as load changes (such as center of gravity shift), road adhesion coefficient fluctuations (such as icy roads), and tire aging, dynamically adjusting the control strategy and completely solving the problem of insufficient adaptability of existing technologies to complex operating conditions.

[0043] 2. Achieve deep coordination between steering and braking systems

[0044] The solution breaks down the independent operating modes of the EPS and ESP systems by dynamically switching multi-modal control strategies and calculating collaborative control quantities. In the normal stable mode, the EPS only adjusts the assist; in the mild correction mode, the EPS and ESP work together to provide slight intervention; and in the emergency intervention mode, the EPS provides reverse assist, the ESP applies multi-wheel braking, and the engine torque limits, forming a progressive collaborative control logic. Simultaneously, a coupled compensation model corrects the control quantities in real time to prevent the EPS and ESP actions from canceling each other out. This collaborative mechanism eliminates coupling interference between subsystems and achieves global optimization of vehicle dynamics.

[0045] 3. Improve control response speed and accuracy

[0046] To address the actuator delay issue (EPS motor delay of approximately 50ms, ESP brake actuator delay of approximately 100ms), the solution constructs an advance compensation model through actuator dynamic response compensation (step 7), and adjusts the compensation time in real time (0.1-0.2 seconds) in conjunction with the actuator aging state to ensure that the control action is synchronized with the vehicle's dynamic needs. At the same time, the solution limits the brake pressure change rate (≤2bar / ms) and EPS assist transition curve through the control quantity ramp transition algorithm, avoiding vehicle attitude fluctuations during mode switching, thus solving the lag problem of existing ESP passive intervention.

[0047] 4. Supports personalized and adaptive control

[0048] A driver style recognition module is introduced to classify drivers into three types (aggressive, moderate, and cautious) by analyzing data such as steering operation and pedal intensity. The yaw rate threshold is adjusted accordingly to balance control and stability. Combined with a scenario-based self-learning model, the system identifies scenarios such as urban roads and mountain curves using GPS and map data, calls up corresponding optimization parameters, and continuously optimizes control rules through reinforcement learning to adapt to vehicle aging (such as tire wear) and changes in driving habits, achieving personalized control for each driver.

[0049] 5. Enhance system security and reliability

[0050] The multi-level fault-safe mode responds according to the severity of the fault: minor faults activate local redundancy, moderate faults initiate degraded control, and severe faults switch to mechanical backup and limit vehicle speed (≤40km / h). Simultaneously, fault information is pushed via the vehicle-to-everything (V2X) network and the system guides the user to a repair station. This design significantly improves the system's fault tolerance under abnormal conditions, ensuring vehicle driving safety.

[0051] In summary, this solution comprehensively addresses the shortcomings of existing technologies in terms of adaptability to operating conditions and control coordination through dynamic perception, collaborative control, precise compensation, and self-learning mechanisms, significantly improving the steering stability and driving safety of light commercial vehicles in complex scenarios. Attached Figure Description

[0052] Figure 1 This is a flowchart of the control method of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] This solution discloses an adaptive control method for steering stability of a light commercial vehicle chassis. Through multi-dimensional parameter sensing, dynamic state estimation, collaborative control strategies, and a self-learning mechanism, it achieves precise control of steering stability under complex operating conditions. The implementation process is detailed below with specific steps:

[0055] Step 1: Real-time acquisition of multi-dimensional operating parameters

[0056] Technical Content: A cluster of sensors installed at various locations on the vehicle collects key parameters affecting steering stability in real time, including steering wheel angle and rotation speed (steering wheel angle sensor), real-time wheel speed (wheel speed sensor), longitudinal and lateral acceleration (acceleration sensor), instantaneous body slip angle (slip angle sensor), rotational speed around the vertical axis (yaw rate sensor), total vehicle mass and center of gravity distribution (load sensor), road surface type and adhesion coefficient (road surface recognition sensors such as LiDAR and cameras), real-time tire pressure and wear status (TPMS), and suspension vertical load (force sensors on suspension springs and shock absorbers) and compression (displacement sensor). All parameters are sampled at a frequency of 100Hz. Simultaneously, by combining the suspension vertical load and compression data, a dynamic estimation model for suspension stiffness is established to correct the vehicle's roll stiffness parameters in real time.

[0057] Solution Analysis: The core of vehicle steering stability lies in the accurate perception of dynamic states. As a key component connecting the vehicle body and wheels, the stiffness characteristics of the suspension system directly affect the vehicle's roll response (such as the degree of body tilt during sharp turns). In traditional control methods, suspension stiffness is mostly a fixed parameter, which cannot adapt to dynamic conditions such as road undulations and load changes, leading to deviations in roll stiffness estimation, which in turn affects the accuracy of core state parameters such as yaw rate and slip angle.

[0058] This step uses force and displacement sensors to collect the vertical load and compression of the suspension. Based on Hooke's Law and the principles of mechanics of materials, a dynamic estimation model for suspension stiffness is constructed: Suspension Stiffness = Vertical Load / Compression. This model can capture the changes in suspension stiffness in real time under different loads (such as switching between full load and no load) and different road surfaces (such as sudden changes in compression caused by potholes), and feed this data back to the roll stiffness parameter correction. Roll stiffness is a key indicator determining vehicle roll stability. Its dynamic correction allows the system to more accurately predict the vehicle's tilt trend during sharp turns, adjust the assessment of tire grip in a timely manner when the road surface is uneven, and quickly adapt to changes in steering characteristics caused by center of gravity shift during load switching.

[0059] In addition, the multi-dimensional parameter acquisition covers the entire chain of information from driver operation (steering wheel angle) to vehicle status (wheel speed, acceleration) and environmental factors (road surface type). The 100Hz sampling frequency ensures real-time capture of transient conditions (such as emergency steering), providing a high-fidelity data foundation for subsequent state estimation and control strategies.

[0060] Technical Results: By comprehensively collecting multi-dimensional parameters and dynamically correcting suspension stiffness, the system overcomes the limitations of "fixed parameters" in traditional methods, achieving a deep perception of vehicle dynamic characteristics. In sharp turning scenarios, real-time corrected roll stiffness parameters enable the system to more accurately identify vehicle roll risks, providing an accurate basis for subsequent braking or steering intervention. When the road surface is uneven, the coordinated collection of suspension compression and vertical load can quickly reflect the impact of road impacts on vehicle attitude, avoiding control lag caused by suspension stiffness estimation errors. When load changes (such as sudden loading of cargo), the coordinated analysis of center of gravity distribution and suspension stiffness can adapt to changes in steering characteristics in a timely manner, preventing understeer or oversteer caused by center of gravity shift. Overall, this step lays a high-accuracy, high-real-time data foundation for the entire control method, significantly improving the system's ability to perceive complex operating conditions.

[0061] Step 2: Preprocessing of operating parameters and diagnosis of anomalies

[0062] Technical Content: Preprocessing of collected multi-dimensional parameters includes data filtering, outlier removal, and time synchronization. A Kalman filter algorithm is used to filter noise from dynamic parameters such as sideslip angle and yaw rate. Anomaly detection models (trained with extensive normal and abnormal operating condition parameter features) are used in conjunction with the 3σ criterion to identify and remove abnormal data, enabling the detection of gradual sensor faults 0.5-1 seconds in advance. Data from different sensors is synchronized and aligned based on timestamps. Redundant sensor data (such as wheel speed sensor and GPS speed signals) are weighted and fused, with the fusion weights dynamically adjusted based on the historical errors of each sensor (e.g., reducing the weight of GPS signals when they are weak in a tunnel environment, and increasing the weight of GPS when wheel speed sensors slip). A sensor health assessment model is constructed to diagnose faults by comparing the deviations of redundant data, automatically switching to backup data or activating the parameter estimation model.

[0063] Solution Analysis: The reliability of sensor data directly determines the effectiveness of the control strategy. However, in actual driving, sensors are susceptible to noise (such as yaw rate fluctuations caused by electromagnetic interference), environmental interference (such as GPS signal attenuation in tunnels), and hardware failures (such as wheel speed sensor slippage), leading to data distortion. In this step, the Kalman filter algorithm effectively filters out high-frequency noise in dynamic parameters such as sideslip angle and yaw rate by establishing a dynamic system model and combining the weighted fusion of predicted and measured values. Its core principle is to minimize the estimation error variance and improve data smoothness by using iterative updates of the state equation and observation equation. The 3σ criterion, as a classic statistical method, can quickly identify abrupt outliers that deviate from the mean by three times the standard deviation. Meanwhile, deep learning anomaly detection models (such as LSTM-based time series models) can capture gradual faults (such as signal drift caused by a slow decrease in sensor sensitivity) by learning the normal time series characteristics of sensor data, achieving early warning. The combination of the two forms a full-scene anomaly detection capability of "abrupt + gradual".

[0064] Redundant data weighted fusion embodies the concept of "dynamic adaptation": wheel speed sensors are susceptible to the relative slippage between the tires and the ground when the vehicle skids, leading to speed measurement deviations. In this case, increasing the GPS weight can utilize its absolute position and speed correction. Conversely, GPS signals are attenuated due to obstruction in tunnels; reducing its weight and relying on wheel speed sensors (which are unaffected by obstruction) ensures reliable speed data. Dynamic weight adjustment is based on historical error statistics (e.g., calculating the variance of the deviation between the sensor's past 100 measurements and the actual value), ensuring the fusion result always favors the sensor with the smaller error. The sensor health assessment model uses redundant data cross-validation (e.g., if the deviation between wheel speed and GPS speed exceeds a threshold, one is deemed abnormal), achieving fault self-diagnosis and redundancy switching, avoiding system failure caused by a single sensor malfunction.

[0065] Technical Effects: This step significantly improves the quality and reliability of input data through multi-level data preprocessing and anomaly diagnosis. Kalman filtering effectively reduces measurement noise of dynamic parameters, greatly improving the estimation accuracy of core state parameters such as sideslip angle and yaw rate, providing a more stable foundation for subsequent dynamic state analysis. The combination of the 3σ criterion and deep learning model enables comprehensive and proactive identification of sensor faults, avoiding control deviations caused by the accumulation of gradual faults (such as sensor drift) while reducing the impact of sudden faults (such as instantaneous sensor failure) on the system.

[0066] The redundant data weighted fusion mechanism enables the system to maintain the measurement accuracy of key parameters (such as vehicle speed) even in complex environments (such as tunnels and slippery roads), solving the problem of poor environmental adaptability of single sensors. Sensor health assessment and redundancy switching enhance the system's fault tolerance; even if some sensors fail, core functions can still be maintained through backup data or estimation models. Overall, this step provides high-fidelity and high-reliability data support for the formulation of subsequent control strategies, reducing the risk of control errors caused by data distortion.

[0067] Step 3: Real-time estimation of vehicle dynamics

[0068] Technical Content: Based on preprocessed parameters, a dual-model parallel estimation strategy (a simplified two-degree-of-freedom model and a refined multibody dynamics model) is employed, combined with the Extended Kalman Filter (EKF) algorithm to construct a vehicle dynamics state estimation model. The two-degree-of-freedom model is used for fast calculation (cycle ≤ 10ms), while the multibody dynamics model includes detailed characteristics of the suspension and tires (cycle 50ms), and the output is fused through Kalman filtering. Specifically, this includes: estimating the actual driving speed based on wheel speed, longitudinal acceleration, and road adhesion coefficient; estimating the ideal yaw rate using the two-degree-of-freedom model by combining steering wheel angle, vehicle speed, and center of gravity position; calculating the instability tendency index based on the deviation between the actual and ideal yaw rates and the sideslip angle; and correcting the tire side stiffness based on tire condition and road adhesion coefficient. The dual-model parallel estimation strategy primarily uses the two-degree-of-freedom model results under simple conditions such as flat roads, while increasing the weight of the multibody dynamics model under complex conditions such as bumpy roads and continuous curves.

[0069] Solution Analysis: Vehicle dynamics (such as actual speed, yaw rate, and instability tendency) is the core basis for control strategies, but it cannot be directly measured and needs to be estimated through models. Traditional methods often use a single model (such as a simplified two-degree-of-freedom model), which makes it difficult to balance "computation speed" and "estimation accuracy": simplified models are fast to calculate but ignore fine characteristics such as suspension and tires, while complex models have high accuracy but are computationally time-consuming, failing to meet the requirements of real-time control.

[0070] The dual-model parallel strategy in this step is based on the principle of "scenario adaptation": the two-degree-of-freedom model simplifies the vehicle into a rigid body moving in the longitudinal and yaw directions, ignoring details such as suspension deformation and tire yaw characteristics. It quickly calculates the ideal yaw rate (cycle ≤10ms) using parameters such as steering wheel angle and vehicle speed, making it suitable for simple conditions such as flat roads and straight-line driving (where the vehicle's dynamic characteristics are relatively stable and the simplified model's error is acceptable). The multibody dynamics model, on the other hand, includes fine characteristics such as suspension elastic deformation and nonlinear changes in tire yaw stiffness. It reconstructs the interaction between the vehicle body, suspension, and tires through finite element analysis, accurately capturing the dynamic response under complex conditions such as bumpy roads (frequent suspension compression / extension) and continuous curves (dynamic changes in tire yaw force), but its calculation cycle is longer (50ms).

[0071] Extended Kalman Filter (EKF) resolves the conflict between speed and accuracy by fusing the outputs of two models: under simple conditions, it assigns higher weights (e.g., 0.8) to the two-degree-of-freedom model to ensure real-time performance; under complex conditions, it increases the weights of the multibody dynamics model (e.g., 0.6) to improve estimation accuracy by leveraging its fine-grained characteristics. Simultaneously, tire lateral stiffness correction is based on tire pressure, wear condition, and road surface adhesion coefficient (e.g., low adhesion coefficient on wet roads requires lower lateral stiffness), making the model output more closely reflect actual physical characteristics. The instability tendency index is calculated by combining the deviation between actual and ideal yaw rate (the larger the deviation, the higher the risk of instability) and lateral slip angle (excessive lateral slip angle increases the risk of sideslip), providing a quantitative basis for subsequent control interventions.

[0072] Technical Effects: The dual-model parallel estimation and EKF fusion mechanism enable the system to maintain high accuracy and real-time performance under different operating conditions. In simple scenarios such as flat roads, the rapid calculation of the two-degree-of-freedom model ensures the real-time response of the control strategy, avoiding intervention lag caused by computational delays. In complex scenarios such as bumpy roads or continuous curves, the refined characteristics of the multibody dynamics model significantly reduce the estimation errors of parameters such as actual vehicle speed and yaw rate, providing a more reliable basis for judging instability trends.

[0073] Dynamic correction of tire lateral stiffness resolves the estimation bias caused by the "fixed lateral stiffness" in traditional models (e.g., the lateral stiffness of worn tires decreases, and failure to correct this will underestimate the risk of sideslip), making the calculation of the instability tendency index more closely reflect the actual vehicle condition. Overall, this step provides accurate dynamic state support for subsequent control target setting and strategy switching, reduces misjudgments in control strategies due to inaccurate state estimation, and improves the system's adaptability to complex operating conditions.

[0074] Step 4: Dynamically setting stability control targets

[0075] Technical Content: The system dynamically sets control targets based on the vehicle's current state: at low speeds (≤30km / h) with large turns, ease of use is prioritized, with a small steering resistance coefficient; at medium speeds (30-80km / h), a balance between ease of use and stability is achieved, with a medium steering gain coefficient; and at high speeds (≥80km / h), stability is the core focus, with a large steering damping coefficient. A driver style recognition module is introduced. By analyzing historical data such as steering operation (e.g., steering speed, steering angle) and pedal intensity (accelerator / brake pedal force), drivers are categorized as aggressive, conservative, or cautious. The yaw rate threshold is adjusted for each type (increasing the threshold for aggressive drivers to reduce intervention, and decreasing the threshold for cautious drivers to intervene earlier). Drivers can manually select control modes for personalized settings.

[0076] Solution Analysis: The rationality of the control objective directly affects the driving experience and safety. Different vehicle speeds and different drivers have significantly different needs for "ease of steering" and "stability": at low speeds (such as parking and U-turns), drivers need easy steering to reduce the burden of operation; at high speeds (such as highways), they need stable steering to avoid oversteering; and the differences in driver style (such as aggressive drivers who prefer agile handling and cautious drivers who focus on stability) further require personalized adaptation of the control objective.

[0077] This step is based on the "vehicle speed-demand" matching principle: In low-speed, high-steering scenarios, a small steering resistance coefficient can reduce the driver's steering effort, and EPS provides greater assistance to achieve ease of use; in medium-speed scenarios, a medium steering gain coefficient balances steering sensitivity and stability, ensuring both handling agility and avoiding attitude fluctuations caused by excessive sensitivity; in high-speed scenarios, a large steering damping coefficient increases steering resistance, making steering more stable and reducing the impact of subtle operations on vehicle attitude. The core is to adapt the force feedback characteristics of the steering system to the safety requirements corresponding to the vehicle speed.

[0078] The driver style recognition module is based on behavioral psychology and machine learning principles: Aggressive drivers steer faster, at larger angles, and with heavier pedal pressure, and are more tolerant of vehicle dynamic responses. Therefore, increasing the yaw rate threshold (allowing for greater yaw fluctuations) reduces system intervention and preserves driving control. Conversely, cautious drivers operate more smoothly and are more sensitive to vehicle body fluctuations. Lowering the threshold allows the system to intervene earlier (such as with slight braking correction), enhancing their driving confidence. The manual selection mode further allows drivers to adjust the control logic according to their preferences (such as forcibly selecting "stability priority" mode in rain or snow), achieving "machine adapting to human" rather than "human adapting to machine."

[0079] Technical Effects: Dynamic control target setting breaks through the limitations of traditional "fixed parameter" control, achieving three-dimensional adaptation of "vehicle speed-demand-style". In low-speed scenarios, the improved steering ease reduces the driver's workload, especially in scenarios such as U-turns in narrow urban areas and maneuvering in parking lots, significantly reducing driver fatigue; in medium-speed scenarios, the balance strategy allows the vehicle to maintain handling agility while avoiding instability caused by over-steering, improving driving safety in suburban roads and national highways; in high-speed scenarios, the stability-priority setting increases steering damping, making the vehicle more stable when changing lanes or making evasive maneuvers at high speeds, reducing the risk of sideslip at high speeds.

[0080] The combination of driver style recognition and manual mode enables personalized control: aggressive drivers enjoy a more flexible driving experience as the system reduces unnecessary intervention; cautious drivers benefit from more proactive safety features as the system intervenes early to correct posture; and manual selection mode further meets the personalized needs of special scenarios (such as inclement weather). Overall, this approach not only enhances the driving experience in different scenarios but also strengthens the vehicle's overall safety by precisely matching needs, achieving a dynamic balance between "safety" and "handling."

[0081] Step 5: Dynamic switching of multimodal control strategies

[0082] Technical Content: Based on dynamic state estimation results and control objectives, a multi-modal control strategy library (normal stable mode, mildly corrected mode, and emergency intervention mode) is constructed and dynamically switched. In the normal stable mode, only EPS assist is adjusted; in the mildly corrected mode, EPS assist characteristics are adjusted, and ESP applies slight braking to a single wheel; in the emergency intervention mode, EPS provides reverse assist, ESP applies coordinated braking to multiple wheels, and simultaneously reduces engine output torque. A transition condition buffer mechanism is added, using a control variable ramp transition algorithm (limiting the maximum rate of change of braking pressure, and using an exponential transition curve for EPS assist) to achieve smooth switching, and establishing a mode switching hysteresis threshold to avoid frequent jitter under critical conditions.

[0083] Solution Analysis: The dynamic changes in vehicle driving status (such as from stable driving to mild instability, and then to emergency instability) require the intensity of the control strategy to increase accordingly, and a single control mode cannot cover the needs of all scenarios. The "threshold trigger" mode of the traditional ESP system (intervening only after instability) has a response lag, while the independent operation of the steering and braking systems may lead to action conflicts (such as EPS positive assist and ESP braking correction canceling each other out).

[0084] The multimodal strategy in this step is based on the principle of "progressive intervention": the normal stability mode is suitable for vehicles in good condition (such as driving at a constant speed in a straight line), and only adjusts the power assist through EPS (such as increasing damping at high speeds) to avoid unnecessary intervention; the mild correction mode is for minor instability trends (such as small deviations in yaw rate), and achieves "gentle correction" by adjusting the power assist characteristics of EPS (such as increasing steering resistance to suppress oversteer) and using ESP single-wheel mild braking (such as generating a reverse torque on the inner wheel); the emergency intervention mode is used for severe instability risks (such as excessive sideslip angle), and forms a "strong intervention" combination punch by using EPS reverse power assist (opposite to the driver's steering direction to suppress oversteer), ESP multi-wheel coordinated braking (differential distribution of braking pressure to different wheels to generate a torque to correct yaw) and engine torque reduction (reducing sideslip caused by driving force) to quickly stabilize the vehicle's attitude.

[0085] The transition buffer mechanism solves the "impact problem" of mode switching: the limit on the rate of change of braking pressure (such as a slow increase rather than a sudden increase) and the EPS exponential transition curve (smooth increase and decrease of assist) can avoid body posture fluctuations caused by sudden changes in control quantity (such as body tilt caused by sudden braking); the hysteresis threshold (such as setting the yaw rate deviation to exceed the threshold for 0.3 seconds before switching modes) can prevent frequent mode switching under critical conditions (such as instantaneous deviation caused by slight road bumps), and avoid body "shaking" affecting the driving experience.

[0086] Technical Effects: Multimodal dynamic switching enables precise matching of control strategies and vehicle states, upgrading from "passive response" to "active intervention." Lightweight control in the normal stable mode reduces energy consumption and system wear while avoiding the impact of excessive intervention on the driving experience; the "gentle intervention" of the mild correction mode promptly corrects the vehicle when it shows a slight tendency to become unstable, eliminating the risk of instability in its infancy, especially in scenarios such as slight sideslip on rainy roads or small attitude deviations in continuous curves, significantly improving driving smoothness; the "powerful combination punch" of the emergency intervention mode quickly stabilizes the vehicle body in emergency situations (such as sharp steering caused by avoiding obstacles), reducing the risk of serious accidents such as rollovers and fishtailing.

[0087] The transition buffer mechanism effectively solves the "shock and vibration" problem of mode switching: the smooth transition of braking pressure and power assist makes the driver almost unaware of the mode switching, improving driving comfort; the hysteresis threshold avoids frequent switching under critical conditions, preventing instability caused by repeated adjustments to the vehicle's posture, especially in scenarios such as continuously undulating roads and curve entrances, ensuring the continuity of the driving experience. Overall, this step achieves a balance between "safety" and "comfort" through progressive intervention and smooth switching, significantly improving the vehicle's control performance under complex dynamic conditions.

[0088] Step 6: Calculation of coordinated control quantities between steering and braking systems

[0089] Technical Content: For the current control mode, the coordinated control quantities of EPS and ESP are calculated. The EPS system calculates the target assist torque using a fuzzy control algorithm (fuzzy rules optimized by a genetic algorithm) to achieve nonlinear dynamic adjustment. The ESP system calculates the target braking pressure for each wheel based on yaw rate deviation and sideslip angle using a model predictive control (MPC) algorithm. An energy optimization objective function is introduced to dynamically adjust the MPC weight coefficients, prioritizing braking wheels and assist curves with lower energy consumption (e.g., prioritizing the inner rear wheel braking in a lightly corrected mode). Motor efficiency characteristics are considered when calculating the EPS assist torque. The ESP braking distribution strategy is adjusted under regenerative braking conditions to balance recovery efficiency and stability. A coupling compensation model for EPS and ESP control quantities is established to correct the assist torque in real time and prevent control effects from canceling each other out.

[0090] Solution Analysis: The coordination between the steering (EPS) and braking (ESP) systems is the core of improving control performance. Traditional independent control modes may lead to "action cancellation" (such as EPS positive assist exacerbating the oversteering that ESP is trying to correct), while energy consumption issues (such as unnecessary braking or assist consuming energy) also need to be optimized.

[0091] The collaborative control in this step is based on the principle of "global optimization": the fuzzy control algorithm optimizes the rule base through a genetic algorithm, which can handle the nonlinear characteristics of EPS assist torque (such as nonlinear assist demand under different vehicle speeds and steering angles) and achieve precise matching of assist torque; the MPC algorithm calculates the optimal distribution of braking pressure on each wheel by predicting the vehicle state in the short term (such as 0.5 seconds) (such as applying greater braking pressure to the outer wheel to correct understeer), and its core is to minimize the control quantity while satisfying stability constraints.

[0092] The energy optimization objective function achieves energy savings by adjusting the MPC weights (e.g., increasing the weight of the "energy consumption term"): During slight correction, the inner rear wheel brake experiences less braking energy loss due to its lower vertical load (caused by vehicle tilt); the EPS assist torque calculation considers motor efficiency characteristics (e.g., avoiding inefficient speed ranges), reducing motor energy consumption; under regenerative braking conditions, ESP prioritizes rear wheel braking (traditional regeneration systems often rely on the rear wheels), improving energy recovery rate while ensuring braking stability. The coupling compensation model calculates the real-time interaction between EPS and ESP (e.g., the yaw moment generated by ESP braking changes steering resistance, requiring EPS assist correction) to prevent their actions from canceling each other out, ensuring a synergistic effect of 1+1>2.

[0093] Technical benefits: Collaborative control calculation overcomes the drawbacks of traditional systems that operate independently, achieving global optimization of the steering and braking systems. The combination of fuzzy control and MPC algorithm makes the calculation of EPS assist torque and ESP braking pressure more accurate, especially under nonlinear conditions (such as large-angle steering and slippery roads), significantly improving the matching degree between control quantity and vehicle demand, and reducing attitude fluctuations caused by insufficient or excessive control quantity.

[0094] The energy optimization objective function reduces energy consumption while ensuring stability: prioritizing braking of the inner rear wheels reduces kinetic energy loss during minor corrections; consideration of motor efficiency characteristics lowers EPS energy consumption; and strategy adjustments under regenerative braking conditions improve energy recovery rate without affecting stability, especially for new energy light commercial vehicles, extending driving range. The coupled compensation model avoids control effects canceling each other out, ensuring that the actions of EPS and ESP work in tandem, significantly improving control efficiency under complex conditions (such as during emergency steering, where reverse assist and multi-wheel braking work together to stabilize the vehicle faster). Overall, this process improves stability while achieving energy conservation, balancing the dual goals of "safety" and "economy."

[0095] Step 7: Actuator Dynamic Response Compensation

[0096] Technical Content: Considering the dynamic response delays of the EPS motor (approximately 50ms delay) and the ESP brake actuator (approximately 100ms delay), a dynamic response model for the actuators is constructed to provide anticipatory compensation for the control inputs. By collecting signals of motor current, temperature, and brake line pressure fluctuations, an actuator aging model is established to estimate changes in response characteristics (such as increased delay) in real time and dynamically adjust the compensation time (0.1-0.2 seconds). The vehicle state 0.1-0.2 seconds later is predicted, and the control inputs are corrected to ensure that the actuator actions are synchronized with the vehicle's dynamic requirements. When the brake actuator response delay increases from 100ms to 150ms, the compensation time is automatically extended from 0.1 seconds to 0.15 seconds; a maintenance reminder is triggered when performance degradation exceeds a threshold.

[0097] Solution Analysis: Physical delays in actuators (EPS motor, ESP brake pump) (such as motor start-up and hydraulic line pressure build-up time) will cause control actions to lag behind the vehicle's dynamic needs (such as when emergency braking is required, a 100ms delay in actuators may miss the best intervention opportunity), and aging (such as motor wear and brake line aging) will further aggravate the delay and reduce control accuracy.

[0098] The dynamic compensation in this step is based on the principle of "advanced prediction": the actuator dynamic response model fits the delay characteristics (such as the current-torque response curve of the EPS motor and the ESP braking pressure-time curve) through experimental data, calculates the time difference between the control quantity and the actual output, and then outputs the control command in advance (e.g., if the braking pressure at time t is required, the command is issued at t-delay time) to offset the physical delay. The aging model monitors the motor current (increased current may lead to increased resistance due to bearing wear), temperature (overheating may reduce response speed), and brake line pressure fluctuations (increased fluctuations may lead to leakage due to aging of seals), estimates the delay changes in real time (e.g., from 100ms to 150ms), and adjusts the compensation time synchronously (from 0.1 seconds to 0.15 seconds) to ensure compensation accuracy.

[0099] Vehicle state prediction is based on a dynamic model (e.g., predicting the state 0.1-0.2 seconds later based on the current yaw rate and sideslip angle), and corrects the control inputs (e.g., appropriately increasing braking pressure if an increase in sideslip angle is predicted). This ensures that actuator actions not only offset the delay but also adapt to future states, achieving "precise synchronization." Maintenance reminders are triggered when performance degradation exceeds a threshold (e.g., a 50% increase in delay), preventing control failure due to excessive actuator aging.

[0100] Technical Effects: Dynamic response compensation solves the control lag problem caused by actuator "delay and aging," significantly improving the timeliness and accuracy of control actions. The proactive compensation mechanism ensures that the actual output time of EPS power assist and ESP braking is completely synchronized with the vehicle's dynamic needs. In emergency scenarios (such as sudden sideslip), it avoids missed intervention opportunities due to actuator delay, greatly improving the attitude correction effect. The dynamic adjustment of the aging model ensures that the system can maintain compensation accuracy even after long-term use of the actuators, solving the problem of the decline in effectiveness of traditional "fixed compensation" after aging and extending the effective working cycle of the system.

[0101] Vehicle state prediction further enhances the forward-looking nature of control, enabling actuator actions not only to respond to the current state but also to adapt to future changes in advance (such as predicting impending oversteering and adding reverse assistance in advance), significantly improving control accuracy. Maintenance reminders, by providing timely warnings of actuator aging, prevent control failures caused by component failures, ensuring long-term system reliability. Overall, this step upgrades the control strategy from "passive following" to "active prediction," significantly improving control effectiveness and system durability under complex dynamic conditions.

[0102] Step 8: Real-time feedback on control effects and parameter self-learning (integrated with claim 9)

[0103] Technical Content: Real-time acquisition of feedback parameters such as yaw rate and sideslip angle, comparison with control targets to calculate errors, and closed-loop adjustment of EPS assist torque and ESP braking pressure through a PID controller; construction of a scenario-based self-learning model, dividing operating conditions into typical scenarios such as urban roads, highways, and mountain curves, identifying scenarios through GPS and map data and calling corresponding optimization parameters; in the mountain curve scenario, focusing on optimizing the assist characteristics and braking intervention timing during continuous steering, with learning data from different scenarios accumulated independently to avoid interference; optimization of fuzzy control rules and MPC weight coefficients through reinforcement learning algorithms to adapt to vehicle aging (such as tire wear) and driving habits; support for updating the basic parameters of the self-learning model via OTA.

[0104] Solution Analysis: In actual driving, vehicle characteristics (such as tire wear leading to a decrease in lateral stiffness), driving habits (such as long-term gentle driving or aggressive driving), and differences in scenarios (such as many curves in mountainous areas and congested urban roads) will cause the effect of fixed control parameters to decay, making it difficult for traditional systems to adaptively adjust.

[0105] This step's feedback and self-learning are based on the "closed-loop optimization" principle: the PID controller compares the error between feedback parameters (such as actual yaw rate) and control targets (such as ideal yaw rate), and corrects EPS assist and ESP braking pressure in real time (increasing braking pressure if the error is positive), forming a "measurement-comparison-correction" closed loop to reduce static errors. Scenario-based self-learning identifies scenarios using GPS and map data (such as "mountain curves" marked on the map) and calls upon historical optimization parameters for that scenario (such as earlier braking intervention on mountain curves). Its core is to leverage "scenario similarity" (similar control requirements for the same type of scenario) to improve parameter adaptation speed. Specialized optimization for mountain curves (such as smooth transition of assist characteristics during continuous steering) addresses the high-frequency needs of specific scenarios, avoiding insufficient adaptation of general parameters. Independent accumulation of learning data (separate storage of data from different scenarios) prevents urban road data from interfering with mountain curve parameter optimization, ensuring the purity of parameters for each scenario.

[0106] Reinforcement learning algorithms optimize control rules through a "trial and error-reward" mechanism: if a set of fuzzy control rules makes the vehicle more stable, a positive reward is given and the rule is strengthened; conversely, it is weakened, gradually adapting to changes in tire wear (e.g., requiring greater braking pressure after wear) and driving habits (e.g., requiring later intervention for aggressive driving). OTA updates allow manufacturers to push new basic parameters (e.g., optimized models for new road conditions), enabling continuous upgrades to system functions.

[0107] Technical Benefits: Real-time feedback and self-learning mechanisms enable the system to "continuously evolve," overcoming the limitations of traditional "fixed from the factory." PID closed-loop regulation reduces control errors caused by parameter drift (such as EPS assist deviation due to temperature changes) by correcting control quantities in real time, allowing the vehicle to stably track the control target under various operating conditions. Scenario-based self-learning significantly improves the system's adaptability to special scenarios. In mountain curves, optimized continuous steering assist characteristics make driving smoother, and precise adjustment of braking intervention timing reduces the risk of sideslip. Differentiated parameters for urban roads and highways respectively enhance flexibility in congested traffic and stability at high speeds.

[0108] Reinforcement learning algorithms enable the system to adapt to vehicle aging and changes in driving habits: after tire wear, the automatic adjustment of braking pressure and assist torque avoids the hassle of manual recalibration; after long periods of gentle driving, the system automatically intervenes in advance, aligning with the driver's preference for stability. OTA updates allow the system to continuously receive optimized parameters for new scenarios and road conditions, extending the technology's lifespan. Overall, this step upgrades the system from "fixed control" to "intelligent evolution," significantly improving control accuracy and user experience over long-term use.

[0109] Step 9: Switch to Fail-Safe Mode

[0110] Technical content: Real-time monitoring of sensor, EPS, and ESP actuator status; activation of fail-safe mode when critical component failure is detected; graded response based on fault severity: minor faults (such as non-critical sensor failure) activate partial redundancy scheme; moderate faults (such as partial EPS function limitation) activate degraded control mode to retain core functions; severe faults (such as complete ESP failure) switch to mechanical backup mode, activate hazard warning lights, send fault information and location via vehicle network, limit maximum vehicle speed (≤40km / h), and guide the driver to the nearest repair station using map data.

[0111] Solution Analysis: Faults in vehicle control systems (such as sensor failure or actuator failure) may lead to control strategy failure or even safety accidents. Traditional system fault handling is mostly a single "emergency stop" mode, which lacks a balance between flexibility and safety.

[0112] This multi-level fault response process is based on the principle of "risk grading": For minor faults (such as tire pressure sensor failure, a non-core parameter), local redundancy is activated (such as estimating tire pressure through wheel speed difference) to ensure that core control functions are not affected and to avoid system degradation due to non-critical faults; for moderate faults (such as partial failure of the EPS power assist motor), degraded control retains core functions (such as maintaining only basic power assist and disabling advanced correction functions) to reduce the impact of the fault while ensuring basic driving capabilities; for severe faults (such as complete failure of ESP, making braking intervention impossible), a mechanical backup mode (such as switching the steering system to pure mechanical transmission) ensures that the vehicle can still be steered, speed is limited (≤40km / h) to reduce the risk of accidents, hazard warning lights alert other vehicles, and the combination of vehicle networking and map data quickly guides the driver to the nearest repair shop, forming a "safety-rescue" closed loop.

[0113] The fault monitoring mechanism identifies faults by comparing redundant sensor data (such as the deviation between wheel speed and GPS speed) and actuator feedback signals (such as the deviation between the actual torque and the commanded torque of the EPS motor) in real time, ensuring that risks are detected and dealt with early.

[0114] Technical Benefits: The multi-level fail-safe mode significantly enhances the system's fault tolerance and safety, achieving "no stopping during a fault, guaranteed safety." Local redundancy for minor faults avoids unnecessary functional degradation, ensuring driving continuity; degradation control for moderate faults retains core functions while limiting unnecessary functions, reducing the risk of fault escalation and allowing the driver to continue safely to the repair shop; mechanical backup and speed limiting measures for severe faults ensure basic vehicle controllability even in extreme situations, avoiding the risk of loss of control; hazard warning lights and vehicle-to-everything (V2X) notifications increase the alertness of surrounding vehicles, reducing the risk of secondary accidents; and map guidance quickly solves the problem of "not knowing where to repair after a fault," shortening fault handling time.

[0115] Overall, this step establishes a complete safety mechanism from "fault detection" to "tiered response" and then to "rescue guidance," significantly improving the safety and reliability of the system under abnormal conditions and providing drivers with comprehensive fault protection.

[0116] Working principle: This solution mainly achieves adaptive control of the steering stability of light passenger vehicle chassis through a closed-loop control of the entire process of "perception-estimation-decision-execution-optimization-fault tolerance".

[0117] First, a multi-dimensional sensor cluster (Step 1) is used to collect parameters such as steering operation, vehicle status, and road environment in real time. This data, combined with a dynamic estimation model of suspension stiffness, corrects the roll stiffness, providing comprehensive and accurate raw data for subsequent analysis. Second, the collected data is preprocessed (Step 2) through filtering, anomaly detection, redundancy fusion, and health assessment to ensure data reliability and lay the foundation for state estimation. Next, a dual-model parallel strategy (Step 3) is employed, combining the outputs of an extended Kalman filter fusion model (fast calculation) and a multibody dynamics model (refined characteristics) to estimate the vehicle's actual speed, ideal yaw rate, and instability tendency index in real time. Based on the estimation results, control objectives are dynamically set (Step 4), prioritizing ease of use and stability according to vehicle speed, and adjusting the yaw rate threshold based on driver style. To achieve personalized goals, the system dynamically switches between various control strategies based on the matching results of goals and states (step 5), progressively adjusting the coordinated action of EPS and ESP from regular assistance to emergency intervention, and achieving smooth switching through a buffer mechanism. For the current mode, the coordinated control quantity is calculated (step 6), and the EPS assistance and ESP braking pressure are optimized using fuzzy control and MPC algorithms, introducing energy optimization and coupling compensation to avoid conflicts. Considering actuator delay and aging (step 7), the system ensures that the control action is synchronized with the vehicle's needs through advance compensation and state prediction. The control effect is fed back in real time (step 8), and the parameters are continuously optimized through PID closed-loop adjustment and scenario-based self-learning to adapt to vehicle aging and driving habits. Finally, through multi-level fail-safe modes (step 9), the system responds in stages according to the severity of the fault to ensure driving safety under fault conditions.

[0118] The entire process achieves full-scenario coverage from parameter acquisition to fault handling through close collaboration of each step. The core is to use dynamic perception, collaborative control, self-learning and fault tolerance mechanisms to adapt to changes in vehicle status, environment and driver needs in real time.

[0119] Technical effects of implementing this solution: Through multi-dimensional innovative design, this solution comprehensively solves the limitations of existing technologies in terms of working condition adaptability, control coordination, and response speed, and significantly improves the steering stability and driving safety of light passenger vehicles.

[0120] In terms of adaptability to different operating conditions, multi-dimensional parameter acquisition and dynamic correction of suspension stiffness enable the system to accurately capture the influence of complex factors such as load changes, road surface undulations, and tire aging; dual-model parallel estimation and scenario-based self-learning further ensure the accuracy of state estimation and the adaptability of control parameters under different operating conditions, completely solving the problem of insufficient adaptability of traditional systems with "fixed parameters" to complex scenarios.

[0121] In terms of control coordination, the dynamic switching of multimodal strategies and the calculation of coordinated control quantities break the independent working mode of EPS and ESP. Through progressive intervention and coupling compensation, the action coordination of the two is achieved, avoiding the cancellation of control effects and optimizing the global dynamic characteristics of the vehicle. The energy optimization objective function also reduces energy consumption while ensuring stability, taking into account both safety and economy.

[0122] In terms of response speed and accuracy, the actuator dynamic response compensation solves the problems of delay and aging, while the advanced prediction and PID closed-loop regulation ensure the timeliness and accuracy of the control action; the transition buffer mechanism avoids the impact of mode switching and improves driving comfort.

[0123] In terms of personalization and evolution capabilities, driver style recognition and manual mode enable personalized control that is "one face for a thousand people"; reinforcement learning and OTA updates enable the system to continuously adapt to vehicle aging and new scenarios, and have long-term evolution capabilities.

[0124] In terms of safety and reliability, the multi-level fail-safe mode ensures basic driving safety under different fault levels through a graded response strategy, while vehicle networking and map guidance improve fault handling efficiency; the anomaly detection and redundancy design throughout the process further reduce the risk of system failure.

[0125] In summary, this solution, through the deep integration of dynamic perception, collaborative control, intelligent optimization, and safety tolerance, significantly improves the steering stability of light commercial vehicles under complex road conditions, varying loads, and diverse driving styles, providing comprehensive protection for the driving safety of commercial light commercial vehicles.

[0126] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive control method for steering stability of a light passenger vehicle chassis, characterized in that: The control method is as follows: Step 1: Real-time acquisition of multi-dimensional operating parameters A cluster of sensors installed at various locations on the vehicle collects key parameters affecting steering stability in real time. These parameters include: using a steering wheel angle sensor to collect the steering wheel angle and rotation speed; using wheel speed sensors to obtain the real-time rotational speed of each wheel; using an acceleration sensor to measure the vehicle's longitudinal and lateral acceleration; using a sideslip angle sensor to collect the instantaneous sideslip angle of the vehicle body; using a yaw rate sensor to record the vehicle's rotational speed around its vertical axis; using a load sensor to detect the total mass and center of gravity distribution of the vehicle body; using a road surface recognition sensor to identify the road surface type and estimate the road adhesion coefficient; using a tire pressure monitoring system (TPMS) to obtain the real-time tire pressure and wear parameters; and using force sensors on the suspension springs and shock absorbers to collect the vertical load on the suspension, and displacement sensors to collect the suspension compression. All collected parameters are sampled at a frequency of 100Hz to ensure real-time data accuracy. Step 2: Preprocessing of operating parameters and diagnosis of anomalies The collected multi-dimensional parameters are preprocessed, including data filtering, outlier removal, and time synchronization. A Kalman filter algorithm is used to filter noise from dynamic parameters such as sideslip angle and yaw rate, improving measurement accuracy. Anomalies caused by sensor faults are identified and removed using the 3σ criterion combined with a deep learning anomaly detection model; the deep learning model can detect gradual sensor faults 0.5-1 second in advance. Data from different sensors is synchronized and aligned based on timestamps. Redundant sensor data is weighted and fused, with the fusion weights dynamically adjusted according to the historical errors of each sensor. A sensor health assessment model is constructed to diagnose faults by comparing the deviations of redundant sensor data; if an anomaly is detected, the system automatically switches to backup sensor data or activates the parameter estimation model. Step 3: Real-time estimation of vehicle dynamics Based on the preprocessed operating parameters, a dual-model parallel estimation strategy (a simplified two-degree-of-freedom model and a refined multibody dynamics model) is adopted, combined with the Extended Kalman Filter (EKF) algorithm to construct a vehicle dynamics state estimation model. The two-degree-of-freedom model is used for fast calculation (cycle ≤ 10ms), while the multibody dynamics model includes detailed characteristics of the suspension and tires (cycle 50ms), and the output is fused through Kalman filtering. Specifically, this includes: estimating the actual vehicle speed based on wheel speed, longitudinal acceleration, and road adhesion coefficient; estimating the ideal yaw rate using the two-degree-of-freedom vehicle model by combining steering wheel angle, vehicle speed, and center of gravity position; calculating the instability tendency index based on the deviation between the actual and ideal yaw rates and the slip angle data; and correcting the tire lateral stiffness based on tire condition and road adhesion coefficient. Step 4: Dynamically setting stability control targets The control objectives are dynamically set based on the vehicle's current state: at low speeds (≤30km / h) and large turns, ease of use is prioritized, with a small steering resistance coefficient; at medium speeds (30-80km / h), a balance between ease of use and stability is achieved, with a medium steering gain coefficient; and at high speeds (≥80km / h), stability is the core focus, with a large steering damping coefficient. A driver style recognition module is introduced, which analyzes historical data such as steering operation and pedal intensity to classify drivers into aggressive, moderate, and cautious types, and adjusts the yaw rate threshold for different types. The system also supports manual selection of control modes by the driver for personalized settings. Step 5: Dynamic switching of multimodal control strategies Based on the dynamic state estimation results and control objectives, a multimodal control strategy library (normal stable mode, mildly corrected mode, and emergency intervention mode) is constructed and dynamically switched. In the normal mode, only EPS assist is adjusted. In the mildly corrected mode, EPS assist characteristics are adjusted, and ESP applies slight braking (≤5 bar) to a single wheel. In the emergency intervention mode, EPS provides reverse assist, and ESP applies coordinated braking (5-15 bar) to multiple wheels while reducing engine output torque. A transition condition buffer mechanism is added, and a smooth switching is achieved through a control variable ramp transition algorithm (brake pressure change rate ≤2 bar / ms, EPS assist adopts an exponential transition curve). A mode switching hysteresis threshold is established to avoid frequent jitter. Step 6: Calculation of coordinated control quantities between steering and braking systems For the current control mode, the coordinated control quantities of EPS and ESP are calculated; the EPS system calculates the target assist torque through a fuzzy control algorithm (fuzzy rules optimized by a genetic algorithm) to achieve nonlinear dynamic adjustment; the ESP system calculates the target braking pressure of each wheel based on the yaw rate deviation and sideslip angle using a model predictive control (MPC) algorithm; an energy optimization objective function is introduced to dynamically adjust the MPC weight coefficients, prioritizing braking wheels and assist curves with lower energy consumption; a coupling compensation model of EPS and ESP control quantities is established to correct the assist torque in real time to avoid the control effects canceling each other out. Step 7: Actuator Dynamic Response Compensation Considering the dynamic response delays of the EPS motor (approximately 50ms delay) and the ESP brake actuator (approximately 100ms delay), a dynamic response model for the actuators is constructed to provide advance compensation for the control input. By collecting signals of motor current, temperature, and brake line pressure fluctuations, an actuator aging model is established to estimate changes in response characteristics in real time and dynamically adjust the compensation time (0.1-0.2 seconds). The vehicle state 0.1-0.2 seconds later is predicted, and the control input is corrected to ensure that the actuator action is synchronized with the vehicle's dynamic requirements. Step 8: Real-time feedback on control effects and parameter self-learning Real-time acquisition of feedback parameters such as yaw rate and sideslip angle, comparison with control targets to calculate errors, and closed-loop adjustment of EPS assist torque and ESP braking pressure through PID controller; construction of a scenario-based self-learning model, dividing operating conditions into typical scenarios such as urban roads and highways, identifying scenarios through GPS and map data and calling corresponding optimization parameters; optimization of fuzzy control rules and MPC weight coefficients through reinforcement learning algorithms to adapt to vehicle aging and driving habits; support for OTA updates of basic parameters of the self-learning model. Step 9: Switch to Fail-Safe Mode The system monitors the status of sensors, EPS, and ESP actuators in real time. When a fault is detected in a critical component, a fail-safe mode is activated. The system responds according to the severity of the fault: for minor faults, a partial redundancy scheme is activated; for moderate faults, a degraded control mode is activated to retain core functions; for severe faults, the system switches to mechanical backup mode, activates hazard warning lights, sends fault information and location via the vehicle network, limits the maximum vehicle speed (≤40km / h), and guides the vehicle to a safe stop.

2. The adaptive control method for steering stability of a light passenger vehicle chassis according to claim 1, characterized in that: In step 1, a dynamic estimation model for suspension stiffness is established by combining suspension vertical load and compression data, and the vehicle roll stiffness parameters are corrected in real time to improve the accuracy of dynamic state estimation during sharp turns, road undulations, and load switching.

3. The adaptive control method for steering stability of a light passenger vehicle chassis according to claim 1, characterized in that: In step 2, the parameter fusion algorithm reduces the weight of GPS when the signal is weak and increases the weight of GPS when the wheel speed sensor is abnormal due to slippage; the deep learning anomaly detection model improves the ability to identify gradual faults such as slow sensor drift by training a large number of normal and abnormal operating condition parameter features.

4. The adaptive control method for steering stability of a light passenger vehicle chassis according to claim 1, characterized in that: In step 3, the dual-model parallel estimation strategy mainly uses the results of the two-degree-of-freedom model under simple working conditions such as flat roads, and increases the weight of the multibody dynamics model under complex working conditions such as bumpy roads and continuous curves to improve the robustness of state estimation.

5. The adaptive control method for steering stability of a light passenger vehicle chassis according to claim 1, characterized in that: In step 4, the driver style recognition module increases the allowable threshold for yaw rate for aggressive drivers to reduce intervention, and lowers the threshold for cautious drivers to intervene earlier, balancing the needs for driving control and stability.

6. The adaptive control method for steering stability of a light passenger vehicle chassis according to claim 1, characterized in that: In step 5, the transition condition buffer mechanism avoids vehicle body attitude fluctuations during mode switching by limiting the maximum rate of change of braking pressure and the EPS assist transition curve, and the lag threshold prevents frequent mode switching jitter under critical conditions.

7. The adaptive control method for steering stability of a light passenger vehicle chassis according to claim 1, characterized in that: In step 6, the energy optimization objective function prioritizes braking the inner rear wheel to reduce kinetic energy loss under the mildly corrected mode. When calculating the EPS assist torque, the motor efficiency characteristics are considered. Under the braking energy recovery condition, the ESP braking distribution strategy is adjusted to balance recovery efficiency and stability.

8. The adaptive control method for steering stability of a light passenger vehicle chassis according to claim 1, characterized in that: In step 7, when the brake actuator response delay increases from 100ms to 150ms, the actuator health status monitoring automatically extends the compensation time from 0.1 seconds to 0.15 seconds. When the performance degradation exceeds the threshold, a maintenance reminder is triggered.

9. The adaptive control method for steering stability of a light passenger vehicle chassis according to claim 1, characterized in that: In step 8, the scenario-based learning mechanism focuses on optimizing the power assist characteristics and braking intervention timing during continuous steering in mountainous curve scenarios. Learning data from different scenarios are accumulated independently to avoid interference.

10. The adaptive control method for steering stability of a light passenger vehicle chassis according to claim 1, characterized in that: In step 9, the multi-level fault response strategy, in addition to limiting vehicle speed, pushes fault information in real time through the vehicle network backend for serious faults and guides the driver to the nearest repair station in combination with map data.

Images