Low adhesion road vehicle stability control method and device

Abstract

A low adhesion road vehicle stability control method and device, the method comprising: determining a transient impact event according to the three-phase current of the wheel hub motor, and extracting a current transient impact feature from the corresponding synthesized current waveform of the transient impact event; extracting a power spectrum feature according to the vertical acceleration of the vehicle suspension; inputting the current transient impact feature, the power spectrum feature and the wheel speed fluctuation frequency into a preset road identification model to obtain a road adhesion coefficient; solving a control instruction according to the road adhesion coefficient and a preset constraint condition, and performing stability control on the vehicle. The method can quickly and accurately obtain the road adhesion coefficient at the initial stage of the low adhesion road working condition, and can effectively avoid the control conflict between different control systems, thereby improving the stability of the vehicle on the low adhesion road.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, specifically to a method and device for controlling vehicle stability on low-adhesion road surfaces. Background Technology

[0002] Currently, the stability control systems of new energy vehicles (especially distributed drive electric vehicles) mainly rely on indirect road adhesion coefficient estimation methods based on IMU (Inertial Measurement Unit) and wheel speed sensor signals. Based on different technical architectures, they can be divided into three categories: external centralized architecture, which uses an independent ESP (Electronic Stability Program) controller, and collects wheel speed and yaw rate signals through a CAN (Controller Area Network) bus (250kbps). When the slip ratio is detected to exceed a fixed threshold (e.g., 20%), the ESP requests torque reduction from the engine or motor management system, or requests pressure build-up from the hydraulic unit. The response time of this architecture is about 50-100ms. The simple monitoring architecture of the drive body, in which the MCU (Microcontroller Unit) has a simple slip ratio comparison logic built in. When the rate of change of motor speed exceeds a set value (e.g., 500rpm / s), the output torque is directly reduced. This architecture has a fast response (about 10ms), but cannot coordinate braking and steering. The integrated control architecture based on MPC (Model Predictive Control) adopts a hierarchical MPC architecture, which coordinates yaw stability and drive anti-slip through a weighted summation objective function, and outputs additional yaw torque, additional front wheel steering angle and wheel drive torque.

[0003] In existing technologies, some solutions employ phase plane analysis, dynamic weight allocation, and quadratic programming, but these rely on visual judgment, which carries the risk of judgment failure under special working conditions. This results in the inability to quickly and accurately identify the road surface condition in the early stages of low-adhesion road conditions, leading to delayed control intervention. Other solutions employ weighted summation MPC, which focuses on drive anti-skid and fails to cover the full coordinated control of the drive, braking, and steering systems. This leads to control conflicts between systems, resulting in decreased vehicle stability and making it difficult to meet the requirements for all-weather, highly robust stability control. Summary of the Invention

[0004] This application provides a method and apparatus for controlling vehicle stability on low-adhesion road surfaces, which can solve the technical problems of existing technologies being unable to quickly and accurately identify road conditions in the early stages of low-adhesion road surface conditions, and the control conflicts between different control systems leading to a decrease in vehicle stability.

[0005] To achieve the above objectives, in a first aspect, this application provides a method for vehicle stability control on low-adhesion road surfaces, the method comprising: Transient impact events are determined based on the three-phase current of the hub motor, and transient impact characteristics are extracted from the synthetic current waveform corresponding to the transient impact events.

[0006] Power spectrum features are extracted based on the vertical acceleration of the vehicle suspension.

[0007] The transient impact characteristics of the current, the power spectrum characteristics, and the wheel speed fluctuation frequency are input into a preset road surface identification model to obtain the road surface adhesion coefficient.

[0008] Based on the road surface adhesion coefficient and preset constraints, control commands are calculated, and vehicle stability control is performed.

[0009] Furthermore, in one embodiment, determining the transient impact event based on the three-phase current of the hub motor includes: The three-phase currents are subjected to Clarke transformation to obtain the α-axis current component and the β-axis current component, and the composite current is calculated.

[0010] If the rate of change of the synthetic current exceeds a preset rate of change threshold within multiple consecutive preset first time periods, it is determined to be a transient impact event.

[0011] Furthermore, in one embodiment, the transient current impact characteristics include: the peak slope, duration, and rise time required from a first preset ratio to a second preset ratio of the synthesized current waveform; the first preset ratio is less than the second preset ratio.

[0012] Furthermore, in one embodiment, the peak slope is the maximum rate of change of the synthesized current waveform within the interval from the rising start point to the peak value.

[0013] The baseline current value before the transient impact event is determined by multiplying the baseline current value by a preset multiple by the standard deviation of the composite current before the transient impact event. The rising start point is the position where the amplitude of the composite current first exceeds this sum.

[0014] The duration is the time from the start of the transient impact event to the time when the deviation between the synthetic current amplitude and the current baseline value first becomes less than a preset error.

[0015] Furthermore, in one embodiment, the step of extracting power spectral features based on the vertical acceleration of the vehicle suspension includes: Within a preset second time period, the vertical acceleration of each suspension of the vehicle is subjected to a fast Fourier transform to obtain the corresponding complex spectrum.

[0016] The power spectral density of the corresponding suspension is calculated based on the complex spectrum, and the average power spectral density is obtained by averaging the power spectral densities of all suspensions.

[0017] Power spectral features are extracted from the average power spectral density, including the energy ratio of the low-frequency band to the mid-frequency band and the energy centroid frequency.

[0018] Furthermore, in one embodiment, the road surface recognition model is a machine learning model, including a classifier and a regressor.

[0019] The classifier takes the transient impact characteristics of the current, the power spectrum characteristics, and the wheel speed fluctuation frequency as inputs and outputs the road surface type.

[0020] The regressor takes the transient impact characteristics of the current, the power spectrum characteristics, the wheel speed fluctuation frequency, and the road surface type as inputs, and outputs the road surface adhesion coefficient.

[0021] Furthermore, in one embodiment, the wheel speed fluctuation frequency is obtained by performing a fast Fourier transform on the wheel speed and extracting the corresponding power spectrum centroid frequency.

[0022] Furthermore, in one embodiment, the control command is solved based on the road surface adhesion coefficient and preset constraint conditions, including: The desired yaw rate is calculated using a linear two-degree-of-freedom reference model based on the steering wheel angle and vehicle speed.

[0023] The limiting factor is calculated based on the road surface adhesion coefficient, and the desired yaw rate is limited to obtain the target yaw rate.

[0024] Based on the driving anti-slip constraint, yaw stability constraint, and active steering constraint in sequence, the model predictive control method is adopted to solve the control command according to the target yaw rate. The control command includes driving torque, braking pressure, and compensation angle.

[0025] Furthermore, in one embodiment, the driving anti-skid constraint condition is: the slip ratio does not exceed the optimal slip ratio corresponding to the road surface adhesion coefficient plus a preset margin; the optimal slip ratio is obtained by querying a preset relationship table between the road surface adhesion coefficient and the optimal slip ratio based on the road surface adhesion coefficient.

[0026] The yaw stability constraint is to minimize the yaw stability objective function.

[0027] The active steering constraint is to minimize the active steering objective function.

[0028] Secondly, this application provides a vehicle stability control device for low-adhesion road surfaces, the device comprising: The first feature extraction module is used to determine transient impact events based on the three-phase current of the hub motor, and to extract current transient impact features from the synthetic current waveform corresponding to the transient impact events.

[0029] The second feature extraction module is used to extract power spectrum features based on the vertical acceleration of the vehicle suspension.

[0030] The adhesion coefficient acquisition module is used to input the transient impact characteristics of the current, the power spectrum characteristics, and the wheel speed fluctuation frequency into a preset road surface recognition model to obtain the road surface adhesion coefficient.

[0031] The control module is used to solve for control commands based on the road surface adhesion coefficient and preset constraints, and to perform stability control on the vehicle.

[0032] The beneficial effects of the technical solutions provided in this application include: This application determines transient impact events based on the three-phase current of the hub motor and extracts transient impact features from the corresponding synthetic current waveform. Simultaneously, it extracts power spectrum features based on the vertical acceleration of the vehicle suspension. These features, combined with the wheel speed fluctuation frequency, are input into a pre-defined road surface identification model to obtain the road surface adhesion coefficient. This allows for rapid and accurate acquisition of the road surface adhesion coefficient in the initial stages of low-adhesion road conditions. Based on this, control commands are solved according to the road surface adhesion coefficient and pre-defined constraints, and vehicle stability control is implemented. This effectively avoids control conflicts between different control systems, thereby improving vehicle stability on low-adhesion roads. Attached Figure Description

[0033] Figure 1 This is a flowchart of a vehicle stability control method for low-adhesion road surfaces according to an embodiment of this application.

[0034] Figure 2 This is a detailed flowchart of step S2 in an embodiment of this application.

[0035] Figure 3 This is a detailed flowchart of step S4 in an embodiment of this application.

[0036] Figure 4 This is a block diagram of a vehicle stability control device for low-adhesion road surfaces according to an embodiment of this application. Detailed Implementation

[0037] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0038] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0039] In a first aspect, embodiments of this application provide a method for controlling vehicle stability on low-adhesion road surfaces.

[0040] In one embodiment, see Figure 1 As shown, the above-mentioned vehicle stability control method for low-adhesion road surfaces includes: S1. Determine the transient impact event based on the three-phase current of the hub motor, and extract the transient impact characteristics of the current from the synthetic current waveform corresponding to the transient impact event.

[0041] S2. Extract power spectrum features based on the vertical acceleration of the vehicle suspension.

[0042] S3. Input the transient impact characteristics of the current, the power spectrum characteristics, and the wheel speed fluctuation frequency into the preset road surface identification model to obtain the road surface adhesion coefficient.

[0043] S4. Based on the road surface adhesion coefficient and preset constraints, solve for the control commands and perform stability control on the vehicle.

[0044] In this embodiment, prior to determining the transient impact event, preprocessing of the acquired three-phase current and wheel speed is included, comprising wavelet denoising and RLS (Recursive Least Squares) adaptive filtering. Specifically, wavelet denoising performs a Db4 (Daubechies 4) wavelet transform on the three-phase current and wheel speed, decomposing them into 5 layers, and employs soft thresholding (threshold coefficient 0.7) to effectively filter out high-frequency texture noise from the road surface (frequency > 200 Hz, Hertz). Further, RLS adaptive filtering is applied, using a recursive least squares filter with a forgetting factor λ = 0.98 to adaptively filter the wheel speed and three-phase current, thereby suppressing noise interference and improving the signal-to-noise ratio.

[0045] In this embodiment, by determining transient impact events based on the three-phase current of the hub motor and extracting transient impact features from the corresponding synthetic current waveform, the high sensitivity and rapid response of the motor current to load changes can be utilized to capture transient electrical features that traditional wheel speed sensors cannot detect in the very early stages of changes in the microscopic dynamic contact between the wheel and the road surface. This provides valuable time for subsequent road condition identification. Subsequently, power spectrum features are extracted based on the vertical acceleration of the vehicle suspension. From the frequency domain distribution of the road surface vertical excitation, another dimension of feature information, which is different from the physical source of the current feature, is obtained. This feature is also sensitive to the road surface material and adhesion state, and can effectively compensate for the identification blind spots that may exist in single current features under specific operating conditions.

[0046] The extracted transient current impact characteristics, power spectrum characteristics, and wheel speed fluctuation frequency are then input into a pre-defined road surface recognition model. This model, by integrating feature parameters from multiple physical sources and different time scales, can stably and reliably output a quantitative estimate reflecting the current road surface adhesion capability before the vehicle exhibits significant slippage. This achieves accurate pre-identification based on vehicle signals, independent of external environmental information. Finally, control commands are solved based on the obtained road surface adhesion coefficient and pre-defined constraints to perform vehicle stability control. Since the constraints themselves embody a clear priority order among different control functions, the solved commands such as drive torque, braking pressure, and compensation steering angle naturally meet this priority requirement. This effectively avoids command oscillations or mutual cancellations caused by target conflicts between systems such as drive anti-slip, yaw stabilization, and active steering.

[0047] By executing and coordinating the above steps sequentially, a coherent, efficient, and conflict-free control chain can be formed throughout the entire process of driving on low-adhesion surfaces, from early perception to later execution, significantly improving the vehicle's driving safety and handling stability on low-adhesion surfaces such as ice, snow, and slippery surfaces.

[0048] Furthermore, in one embodiment, in step S1 above, determining the transient impact event based on the three-phase current of the hub motor involves the following steps: The Clarke transform is performed on the three-phase currents to obtain the α-axis current component and the β-axis current component, and the composite current is calculated.

[0049] If the rate of change of the synthetic current exceeds a preset rate of change threshold within multiple consecutive preset first time periods, it is determined to be a transient impact event.

[0050] The first time period mentioned above is set based on the transient response characteristics and sampling frequency of the hub motor current; in this embodiment, it can be 10ms. The change rate threshold mentioned above is set based on the statistical characteristics of the synthetic current change rate under no transient impact; in this embodiment, it can be 2000A / s.

[0051] In this embodiment, by performing Clarke transformation on the three-phase current of the hub motor, the originally coupled three-phase instantaneous values ​​are converted into two-phase orthogonal α-axis and β-axis current components, and then a scalar signal characterizing the overall current amplitude is synthesized. This synthesis process effectively eliminates the redundant information caused by the independent fluctuations of the three-phase currents and highlights the overall current change trend caused by load mutation.

[0052] Based on this, a short-term window matching the electromagnetic time constant of the motor is set as the first time period. The rate of change of the synthetic current is calculated by continuously sliding this window. If the rate of change in multiple consecutive such windows exceeds the rate of change threshold set based on the statistical characteristics of the non-impact condition, it is determined to be a transient impact event. This strategy of continuous multi-window joint determination utilizes the physical advantage of the high sampling frequency of the motor current, which can capture microsecond-level load fluctuations, ensuring that road surface changes can be detected even when the wheel slip rate is extremely low. At the same time, the dual screening mechanism of multi-window and threshold effectively avoids false triggering caused by noise or random disturbances at a single sampling point, thus achieving a good balance between recognition sensitivity and anti-interference capability. The detection method of this embodiment can reliably and stably locate transient impact events at the initial moment when the road surface adhesion state changes, providing accurate time anchors and data segments for subsequent feature extraction and road surface adhesion coefficient identification.

[0053] Furthermore, in one embodiment, in step S1 above, the transient current impact characteristics include: the peak slope, duration, and rise time required from the peak value to a first preset ratio to a second preset ratio of the synthesized current waveform; wherein, the first preset ratio and the second preset ratio are set according to the sampling period and the expected duration of the transient impact event, and the first preset ratio is less than the second preset ratio. In this embodiment, the first preset ratio can be 10%, and the second preset ratio can be 90%.

[0054] The aforementioned peak slope represents the maximum rate of change of the synthesized current waveform within the range from the starting point of the rise to the peak value.

[0055] The baseline current value before the transient impact event is determined by multiplying it by a preset multiple by the standard deviation of the synthetic current before the transient impact event. The rising starting point is the position where the amplitude of the synthetic current first exceeds this sum. The preset multiple is determined based on the statistical characteristics of the synthetic current noise signal; in this embodiment, it can be set to 3.

[0056] The aforementioned duration is the time from the start of the transient impact event to the time during which the deviation between the aforementioned synthetic current amplitude and the aforementioned current baseline value first becomes less than a preset error. The preset error is set based on the steady-state noise amplitude of the synthetic current signal before the transient impact event; in this embodiment, it can be taken as 5% of the current baseline value.

[0057] In this embodiment, after determining the transient impact event, the peak slope, rise time, and duration are further extracted from the synthetic current waveform corresponding to the event as transient impact characteristics. The physical principle is as follows: when a vehicle travels on a low-adhesion road surface (such as an icy or snowy surface), even if the wheel slip ratio has not yet increased significantly (e.g., less than 3%), micro-slippage has already occurred in the microscopic contact between the tire and the road surface. This micro-slippage directly causes microsecond-level fluctuations in the load torque of the wheel hub motor. Traditional wheel speed sensors, due to limitations in mechanical inertia and sampling rate (typically 100Hz), cannot capture this transient change.

[0058] The electromagnetic time constant of the motor current (<1ms) is extremely short, and its sampling frequency (greater than 20kHz) is much higher than that of traditional wheel speed sensors. The motor current signal can detect the initial signs of sudden road surface changes a considerable time window before wheel speed changes, making it the earliest available source of vehicle-mounted information. Utilizing this physical advantage, this embodiment first acquires the three-phase instantaneous current of the hub motor at a high frequency (20kHz), and then converts it into two orthogonal α-axis and β-axis current components using the Clarke transform. The composite current is then calculated, and the formula for calculating the composite current is: (1), in, Indicates the resultant current. Represents the α-axis current component. This represents the β-axis current component.

[0059] The above synthesis process effectively eliminates redundant information caused by the independent fluctuations of the three phases, highlighting the overall current change trend caused by load mutations.

[0060] Based on this, refined feature extraction is performed on the synthetic current waveform segment corresponding to the transient impact event: the peak slope is defined as the maximum rate of change of the synthetic current waveform within the interval from the rising point to the peak value. The rising point is set as the position where the synthetic current amplitude first exceeds the baseline value of the current before the impact plus three times the baseline standard deviation. This statistical threshold design can reliably eliminate steady-state noise interference and ensure accurate capture of the rising point. The rise time is defined as the time required for the synthetic current amplitude to rise from a first preset proportion of the peak value to a second preset proportion, for example, from 10% to 90%. This parameter reflects the steepness of the current impact front. The duration is defined as the length of time from the moment of impact to when the synthetic current amplitude falls back to a value less than a preset error (e.g., 5% of the baseline value). This parameter characterizes the overall width of the impact waveform.

[0061] The aforementioned features characterize the intensity, velocity, and duration of the impact from the time-domain waveform, and are independent and complementary to each other, collectively forming an electrical fingerprint reflecting changes in road surface adhesion. The transient impact features extracted in this embodiment can quantify the low adhesion characteristics of the road surface into stable and distinguishable feature vectors in the early stages when the wheel slip rate is extremely low, providing information-rich and physically meaningful input for subsequent road surface recognition models.

[0062] Furthermore, in one embodiment, see [reference needed]. Figure 2 As shown, in step S2 above, the power spectrum features are extracted based on the vertical acceleration of the vehicle suspension. The specific steps are as follows: S201. Within a preset second time period, perform a Fast Fourier Transform (FFT) on the vertical acceleration of each suspension of the vehicle to obtain the corresponding complex spectrum. The second time period is set based on the frequency domain characteristics of the road surface vertical excitation and the frequency resolution requirements of the FFT; in this embodiment, it can be 100ms.

[0063] S202. Calculate the power spectral density of the corresponding suspension based on the above complex spectrum, and take the average power spectral density of all suspensions to obtain the average power spectral density.

[0064] S203. Extract power spectral features from the above average power spectral density, wherein the power spectral features include the energy ratio of the low-frequency band to the mid-frequency band, and the energy centroid frequency.

[0065] In this embodiment, the vertical acceleration of the four suspension height sensors of the vehicle is collected at 1kHz. Within each 100ms time window, a Fast Fourier Transform is performed on the vertical acceleration signals of the four suspensions to calculate the four complex spectra in the corresponding 0-50Hz frequency band. Furthermore, the power spectral density of all suspensions is calculated and averaged to obtain the average power spectral density. Then, the energy ratio of the low-frequency (0-10Hz) band and the mid-frequency band (10-30Hz), as well as the energy centroid frequency, are extracted as power spectral features. The principle of power spectral feature extraction in this embodiment is as follows: the vertical excitation spectrum of low-adhesion road surfaces (such as gravel) is significantly different from that of high-adhesion road surfaces (asphalt). The vertical excitation spectrum of low-adhesion road surfaces is sensitive to road type and complementary to the motor current characteristics.

[0066] Furthermore, in one embodiment, the road surface recognition model in step S3 above is a machine learning model, including a classifier and a regressor.

[0067] The classifier takes the transient current impact characteristics, power spectrum characteristics, and wheel speed fluctuation frequency as inputs and outputs the road surface type.

[0068] The aforementioned regressor takes the aforementioned transient current impact characteristics, power spectrum characteristics, wheel speed fluctuation frequency, and road surface type as inputs, and outputs the road surface adhesion coefficient.

[0069] In this embodiment, the road surface recognition model can adopt the LightGBM (Light Gradient Boosting Machine) recognition model. The input features of the LightGBM recognition model are a six-dimensional vector, namely: the peak slope, rise time and duration extracted from the synthetic current waveform, the energy centroid frequency and low-to-mid frequency energy ratio extracted from the suspension vertical acceleration power spectral density, and the wheel speed fluctuation frequency extracted from the wheel speed signal. The model consists of a classifier and a regressor. The classifier takes the aforementioned six-dimensional vector as input and outputs the road surface type and its corresponding confidence score. The road surface types cover a variety of common low-adhesion and high-adhesion road surfaces, such as ice and snow, wet asphalt, dry asphalt, and gravel. The confidence score is used to quantify the reliability of the road surface type identification result. When the confidence score is higher than the preset confidence score threshold, the road surface type and adhesion coefficient output by the model directly participate in the subsequent stability control decision. When the confidence score is lower than the preset confidence score threshold, it is determined that the current feature information has noise interference or is in the coverage blind zone of the model training. At this time, the adhesion coefficient estimates from historical moments are automatically fused for a smooth transition, and the controller is triggered to switch to a more conservative control mode (e.g., reducing the upper limit of the expected yaw rate) to avoid sudden changes in control commands due to a single identification error. The regressor takes the aforementioned six-dimensional vector and the road surface type output by the classifier as input and outputs continuous road adhesion coefficient estimates.

[0070] The aforementioned model was trained using large-scale sample data collected under various road conditions (20 different road conditions, with 5000 frames of data collected for each condition). In practical applications, the model's inference computation latency is extremely short (<2ms), enabling accurate output of the road adhesion coefficient at an early stage before the wheel slip ratio increases significantly. Compared to traditional recognition schemes that rely on the slip ratio to increase before triggering, the pre-recognition timing in this embodiment is significantly earlier (completed at a slip ratio <3%, approximately 200ms earlier than traditional schemes), providing ample response time for subsequent stability control and thus achieving earlier and smoother control intervention.

[0071] Furthermore, in one embodiment, the wheel speed fluctuation frequency in step S3 above is obtained by performing a fast Fourier transform on the wheel speed and extracting the corresponding power spectrum centroid frequency.

[0072] Furthermore, in one embodiment, see [reference needed]. Figure 3 As shown, in step S4 above, the control command is solved based on the road surface adhesion coefficient and preset constraint conditions, and the vehicle stability control is performed. The specific steps are as follows: S401. Calculate the desired yaw rate using a linear two-degree-of-freedom reference model based on the steering wheel angle and vehicle speed.

[0073] S402. Based on the aforementioned road surface adhesion coefficient, calculate the limiting factor to limit the desired yaw rate, thus obtaining the target yaw rate. The calculation formula is as follows: (2), in, Indicates the target's yaw rate. This represents the limiting factor, with the range limited to [0.3, 1.0]. This represents the desired yaw rate. Represents gravitational acceleration. Indicates the steering wheel angle. Indicates longitudinal vehicle speed.

[0074] S403. Based on the driving anti-slip constraint, yaw stability constraint, and active steering constraint in sequence, the model predictive control method is adopted to solve the control command according to the target yaw rate. The subsequent constraint is solved within the solution set of the previous constraint. The control command includes driving torque, braking pressure, and compensation angle.

[0075] S404. Based on the above control commands, perform stability control on the vehicle.

[0076] In this embodiment, the desired yaw rate is first calculated using a linear two-degree-of-freedom reference model based on the steering wheel angle and vehicle speed. This calculation reflects the driver's steering intention on a good road surface. Based on this, a limiting factor calculated using the road adhesion coefficient is introduced to adaptively limit the desired yaw rate. When the road adhesion coefficient is low, the limiting factor decreases accordingly, thus actively limiting the target yaw rate to a safe range that matches the current road adhesion capability. This feedforward limiting strategy effectively avoids the controller generating excessively high yaw rate requests on low-adhesion surfaces, preventing the vehicle from becoming unstable due to exceeding the adhesion limit. For example, with a road adhesion coefficient of 0.15 (ice / snow road condition) and a limiting factor of 0.31, the target yaw rate is only 31% of that on a dry road surface, effectively avoiding oversteering requests.

[0077] Subsequently, a model predictive control method is employed to solve for control commands sequentially according to the constraints of traction control, yaw stability, and active steering. Each subsequent constraint must be optimized within the solution set of the preceding constraint. This lexicographical priority solution architecture ensures that, under any operating condition, traction control constraints take absolute priority over yaw stability and active steering objectives. When these three objectives conflict, the priority is to ensure that the wheel slip ratio does not exceed the optimal range corresponding to the adhesion limit. Under this premise, yaw stability is improved as much as possible, and finally, the energy consumption of active steering is considered.

[0078] Through the collaborative design of the above-mentioned layered amplitude limiting and priority solution, this embodiment enables the final output driving torque, braking pressure and compensation steering angle commands to not only adapt to the current road surface adhesion limit, but also follow the safety-first control logic, thereby effectively avoiding the command oscillation or control failure caused by target conflict in the traditional solution, and significantly improving the overall stability and reliability of the vehicle on low-adhesion road surfaces.

[0079] Furthermore, in one embodiment, the aforementioned driving anti-skid constraint condition is: the slip ratio does not exceed the optimal slip ratio corresponding to the aforementioned road surface adhesion coefficient plus a preset margin. Solving under this constraint yields the first control command solution set, the solution formula of which is: (3), in, This represents the first set of control commands. Represents the control input vector. Indicates the first k At the sampling time, the first i The first wheel in the j The predicted slip ratio for each prediction step. This represents the optimal slip ratio corresponding to the road surface adhesion coefficient. =0.1-0.2 indicates a dry road surface. =0.08 indicates a wet road surface. =0.05 represents an icy surface. If the solution set of the first control command is not empty, proceed to solve the solution set of the second control command. If the solution set of the first control command is empty, output the compromise solution that minimizes the slip ratio and trigger a warning.

[0080] The aforementioned preset margin is set based on the combined uncertainty of slip ratio measurement and control error and road surface adhesion coefficient estimation error. In this embodiment, it can be set to 0.02. The aforementioned optimal slip ratio is obtained by querying a preset relationship table between road surface adhesion coefficient and optimal slip ratio based on the aforementioned road surface adhesion coefficient. This relationship table is obtained through a large number of real vehicle calibration tests on different road surfaces.

[0081] The above yaw stability constraint is to minimize the yaw stability objective function. Under this constraint, the second control command solution set is obtained by solving within the first control command solution set. The solution formula is as follows: (4), in, This represents the solution set of two control commands. Indicates the length of the prediction time domain. The weight matrix represents the yaw rate.

[0082] The above active steering constraint is to minimize the active steering objective function. Under this constraint, the final control command solution is obtained by solving within the solution set of the second control command mentioned above. The solution formula is as follows: (5), in, This represents the final control command solution, including the final drive torque, braking pressure, and compensation steering angle. Indicates the first k Active steering compensation angle of the step, This represents the weighting coefficient for the active steering compensation angle.

[0083] In this embodiment, Model Predictive Control (MPC) is used to solve for the control commands. The parameters of this method are set as follows: prediction time domain. =15 (corresponding to 0.3 seconds), controlling the time domain =5 (corresponding to 0.1 seconds), sampling period T s =20ms. The weight matrix for yaw rate. =10, centroid side slip angle weight =5, yaw moment weight R mz =1, steering energy consumption weight R d=2. The constraints consist of four layers, namely, actuator physical constraints, linear braking pressure. P Not exceeding the maximum value related to the road surface adhesion coefficient μ P_max(μ) The motor torque does not exceed the maximum value. T_max Stability constraint: slip ratio λ Not exceeding the coefficient of adhesion to the road surface μ The corresponding optimal slip ratio Add 0.02; Coordination priority constraint: Acceleration Slip Regulation (ASR) is higher than Electronic Stability Program (ESP), which is higher than Active Rear Steering (ARS). Comfort constraint (yaw rate of change) dγ / dt The longitudinal acceleration change rate does not exceed the preset threshold. d ax / dt (Not exceeding 5 m / s³).

[0084] As shown in Table 1 below, Table 1 compares the constraints of the embodiments of this application with the advantages of the weighted summation method.

[0085] Table 1

[0086] As can be seen from Table 1, in terms of target conflict handling, the weighted summation method makes trade-offs through weight adjustment, while this embodiment prioritizes the achievement of high-priority targets. In terms of extreme operating conditions, the weighted summation method may sacrifice drive anti-slip for yaw stability, while this embodiment prioritizes drive anti-slip in all situations. In terms of weight tuning difficulty, the weighted summation method requires a lot of calibration work, while this embodiment does not require weight tuning and only needs to define the constraint thresholds for each priority. In terms of interpretability, the weighted summation method is a black box model, while this embodiment has a clear hierarchical decision-making logic.

[0087] After obtaining the final control commands, the drive torque command is sent to each wheel hub motor controller with a period of 10ms and an accuracy of ±0.5 N·m, the braking pressure command is sent to the electro-hydraulic brake controller with a period of 10ms and an accuracy of ±0.1 MPa, and the compensation angle command is sent to the active steering unit with a period of 20ms and an accuracy of ±0.1°. The target torque, target wheel cylinder pressure, and target compensation angle sent simultaneously meet the constraints of drive anti-slip, yaw stability, and active steering.

[0088] Meanwhile, the 12.3-inch instrument panel displays in real-time the current road surface type icon, road surface adhesion coefficient value, yaw rate tracking error progress bar, and bar charts of torque and braking force distribution for each wheel, and can record the historical data playback curve for the previous 30 seconds. Furthermore, the above control method also features a closed-loop self-learning mechanism: when the vehicle leaves the factory, model parameters trained on 20 typical road surfaces are loaded; 100 sets of feature and control effect data are collected every 100km to incrementally learn the LightGBM model; when the vehicle returns to the factory, desensitized data is uploaded to the cloud via OTA, and the baseline model is retrained and updated quarterly; simultaneously, the remaining thickness of the vehicle's brake pads (initial 12mm, wear limit 2mm) is estimated using the EHB (Electronic Hydraulic Brake) stroke-pressure relationship; when the remaining thickness is less than 5mm, the maximum braking torque limit is actively reduced; the formula for adjusting the maximum braking torque limit after brake pad wear is: (6), in, This indicates the upper limit of the adjusted maximum braking torque. This indicates the default maximum braking torque limit. This indicates the remaining thickness of the brake pad.

[0089] Tire tread depth is estimated by long-term comparison of wheel speed with GPS (Global Positioning System) vehicle speed. When the depth is less than 3mm, the optimal slip ratio target is lowered, and the user is prompted to replace the tires. The formula for adjusting the optimal slip ratio when the tire tread depth is insufficient is as follows: (7), in, This represents the adjusted optimal slip ratio target. This indicates the default optimal slip ratio.

[0090] Secondly, embodiments of this application also provide a vehicle stability control device for low-adhesion road surfaces.

[0091] In one embodiment, see Figure 4 As shown, the aforementioned vehicle stability control device for low-adhesion road surfaces includes a first feature extraction module, a second feature extraction module, an adhesion coefficient acquisition module, and a control module. Specifically: The first feature extraction module is used to determine transient impact events based on the three-phase current of the hub motor, and to extract transient impact features from the synthetic current waveform corresponding to the transient impact events.

[0092] The second feature extraction module is used to extract power spectrum features based on the vertical acceleration of the vehicle suspension. The adhesion coefficient acquisition module is used to input the transient impact characteristics of current, power spectrum characteristics, and wheel speed fluctuation frequency into a preset road surface recognition model to obtain the road surface adhesion coefficient.

[0093] The control module is used to solve for control commands based on the road surface adhesion coefficient and preset constraints, and to perform stability control on the vehicle.

[0094] In this application, by fusing the transient impact characteristics of the in-wheel motor current with the power spectrum characteristics of the suspension vertical acceleration, and combining them with the wheel speed fluctuation frequency, a lightweight gradient lift model can quickly and accurately pre-identify the road adhesion coefficient at an extremely early stage before the wheel slip ratio has significantly increased. This identification process is entirely based on onboard sensors, without relying on visual or cloud information, and has all-weather applicability. The identification time is extremely short and the accuracy is extremely high, providing ample response lead time for subsequent stability control.

[0095] At the control level, this application proposes a lexicographically optimized model predictive control architecture, which forms strict hierarchical hard constraints according to the constraint order of driving slip, yaw stability, and active steering. This architecture can prioritize the achievement of driving slip control objectives under extreme conditions, effectively avoiding the command conflict problem that may occur in the traditional weighted summation method, where driving slip control is sacrificed for yaw stability. This significantly reduces the peak value of the vehicle's yaw rate, greatly shortens the convergence time, and reduces the number of command conflicts to zero.

[0096] Furthermore, this application uses a visual instrument to display road surface type, adhesion coefficient, yaw rate tracking error, and torque and braking force distribution status of each wheel in real time, and records historical data playback curves, significantly shortening fault diagnosis time. The closed-loop self-learning mechanism includes factory model loading, incremental learning during driving, cloud model updates after returning to the store, and adaptive adjustment of actuator status based on brake pad wear and tire tread depth detection. This dynamically optimizes control boundaries, ensuring that the vehicle maintains excellent stability control performance throughout its entire lifecycle, with performance degradation kept at an extremely low level.

[0097] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0098] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0099] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0100] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0102] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for controlling vehicle stability on low-adhesion road surfaces, characterized in that, The method includes: Transient impact events are determined based on the three-phase current of the hub motor, and transient impact characteristics are extracted from the synthetic current waveform corresponding to the transient impact events. Power spectrum features are extracted based on the vertical acceleration of the vehicle suspension; The transient impact characteristics of the current, the power spectrum characteristics, and the wheel speed fluctuation frequency are input into a preset road surface identification model to obtain the road surface adhesion coefficient. Based on the road surface adhesion coefficient and preset constraints, control commands are calculated, and vehicle stability control is performed.

2. The vehicle stability control method for low-adhesion road surfaces as described in claim 1, characterized in that, The method of determining transient impact events based on the three-phase current of the hub motor includes: The three-phase currents are subjected to Clarke transformation to obtain the α-axis current components and β-axis current components, and the composite current is calculated. If the rate of change of the synthetic current exceeds a preset rate of change threshold within multiple consecutive preset first time periods, it is determined to be a transient impact event.

3. The vehicle stability control method for low-adhesion road surfaces as described in claim 1, characterized in that, The transient impact characteristics of the current include: the peak slope, duration, and rise time required from a first preset ratio to a second preset ratio of the peak value of the synthesized current waveform; the first preset ratio is less than the second preset ratio.

4. The vehicle stability control method for low-adhesion road surfaces as described in claim 3, characterized in that, The peak slope is the maximum rate of change of the composite current waveform within the interval from the rising start point to the peak value. The current baseline value before the transient impact event is determined by multiplying the current baseline value by a preset multiple by the standard deviation of the composite current before the transient impact event. The rising start point is the position where the amplitude of the composite current first exceeds the sum of the values. The duration is the time from the start of the transient impact event to the time when the deviation between the synthetic current amplitude and the current baseline value first becomes less than a preset error.

5. The vehicle stability control method for low-adhesion road surfaces as described in claim 1, characterized in that, The extraction of power spectrum features based on the vertical acceleration of the vehicle suspension includes: Within the preset second time period, the vertical acceleration of each suspension of the vehicle is subjected to a fast Fourier transform to obtain the corresponding complex spectrum. The power spectral density of the corresponding suspension is calculated based on the complex spectrum, and the average power spectral density is obtained by averaging the power spectral densities of all suspensions. Power spectral features are extracted from the average power spectral density, including the energy ratio of the low-frequency band to the mid-frequency band and the energy centroid frequency.

6. The vehicle stability control method for low-adhesion road surfaces as described in claim 1, characterized in that, The road surface recognition model is a machine learning model, which includes a classifier and a regressor; The classifier takes the transient current impact characteristics, power spectrum characteristics, and wheel speed fluctuation frequency as inputs and outputs the road surface type. The regressor takes the transient impact characteristics of the current, the power spectrum characteristics, the wheel speed fluctuation frequency, and the road surface type as inputs, and outputs the road surface adhesion coefficient.

7. The vehicle stability control method for low-adhesion road surfaces as described in claim 1, characterized in that, The wheel speed fluctuation frequency is obtained by performing a fast Fourier transform on the wheel speed and extracting the corresponding power spectrum centroid frequency.

8. The vehicle stability control method for low-adhesion road surfaces as described in claim 1, characterized in that, Based on the road surface adhesion coefficient and preset constraints, the control commands are solved, including: Based on the steering wheel angle and vehicle speed, the desired yaw rate is calculated using a linear two-degree-of-freedom reference model. Based on the road surface adhesion coefficient, a limiting factor is calculated to limit the desired yaw rate, thereby obtaining the target yaw rate. Based on the driving anti-slip constraint, yaw stability constraint, and active steering constraint in sequence, the model predictive control method is adopted to solve the control command according to the target yaw rate. The control command includes driving torque, braking pressure, and compensation angle.

9. The vehicle stability control method for low-adhesion road surfaces as described in claim 8, characterized in that, The driving anti-skid constraint condition is: the slip ratio does not exceed the optimal slip ratio corresponding to the road surface adhesion coefficient plus a preset margin; the optimal slip ratio is obtained by querying a preset relationship table between the road surface adhesion coefficient and the optimal slip ratio based on the road surface adhesion coefficient. The yaw stability constraint is to minimize the yaw stability objective function. The active steering constraint is to minimize the active steering objective function.

10. A vehicle stability control device for low-adhesion road surfaces, characterized in that, The device includes: The first feature extraction module is used to determine transient impact events based on the three-phase current of the hub motor, and to extract current transient impact features from the synthetic current waveform corresponding to the transient impact events. The second feature extraction module is used to extract power spectrum features based on the vertical acceleration of the vehicle suspension. The adhesion coefficient acquisition module is used to input the transient impact characteristics of the current, the power spectrum characteristics, and the wheel speed fluctuation frequency into a preset road surface recognition model to obtain the road surface adhesion coefficient. The control module is used to solve for control commands based on the road surface adhesion coefficient and preset constraints, and to perform stability control on the vehicle.

Images