Method for assessing vehicle rollover risk, method for controlling vehicle rollover and system thereof

Abstract

The application discloses a vehicle rollover risk evaluation method, a vehicle rollover control method and a system thereof, and relates to the active safety technical field of a vehicle. The evaluation method comprises the following steps: acquiring a collection value of a first parameter of a vehicle at a first time step; inputting the collection value of the first parameter into a prediction model to obtain a prediction value of a second parameter at a second time step; constructing a risk measurement index based on a differential energy Lyapunov index according to the prediction value of the second parameter; and evaluating the rollover risk of the vehicle according to the value and variation trend of the risk measurement index. Through the above technical scheme, the vehicle rollover problem is abstracted as a stability problem of a dynamic system, a provable theoretical basis is provided for risk evaluation and control, the generalization ability and robustness are stronger, and the precision of control and the adaptability to complex working conditions are significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of vehicle active safety technology, in particular to a vehicle rollover risk assessment method, a vehicle rollover control method and a system thereof. BACKGROUND

[0002] Vehicle rollover is a serious type of road traffic accident, and poses a particularly serious threat to vehicles with high centers of gravity, such as sport utility vehicles, trucks and buses. The dynamic process of rollover accidents is extremely complex, nonlinear and changeable, which poses high requirements on any system aimed at preventing such accidents.

[0003] From the perspective of dynamics, rollover accidents can be divided into untripped rollover and tripped rollover. The key limitation of the prior art is that it cannot effectively deal with these two complex rollover scenarios at the same time. In order to quantitatively assess the rollover risk, the prior art has developed various evaluation indicators, but all have inherent defects that are difficult to overcome.

[0004] Therefore, there is an urgent need for a new prediction paradigm that must be able to overcome the defects of the prior art and provide accurate, reliable and physically interpretable decision-making basis for active control systems. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a vehicle rollover risk assessment method, a vehicle rollover control method and a system thereof, which abstracts the vehicle rollover problem as a stability problem of a dynamic system, providing a provable theoretical basis for risk assessment and control, having stronger generalization ability and robustness, and significantly improving the accuracy of control and the adaptability to complex working conditions.

[0006] To achieve the above purpose, the first aspect of the present application provides a vehicle rollover risk assessment method, the assessment method comprising: obtaining a collection value of a first parameter of the vehicle at a first time step; inputting the collection value of the first parameter into a prediction model to obtain a predicted value of a second parameter at a second time step, wherein the second time step is after the first time step, and the loss function of the prediction model during training includes a data-driven loss, a kinematic consistency loss and a dynamic consistency loss; constructing a risk measurement index based on a differential energy Lyapunov index according to the predicted value of the second parameter, wherein the risk measurement index includes an asymmetry factor and an energy magnitude factor; and assessing the rollover risk of the vehicle according to the value and the trend of change of the risk measurement index.

[0007] This application also provides a method for controlling vehicle rollover, the method comprising: determining the value and trend of a risk metric of the vehicle according to the vehicle rollover risk assessment method described above; determining the energy change rate of the vehicle according to the trend of the energy magnitude factor in the risk metric; and performing graded intervention control on the vehicle according to the risk metric and / or the energy change rate.

[0008] This application, in another aspect, provides a vehicle rollover risk assessment system, the assessment system comprising: a data acquisition unit for acquiring the acquired values ​​of a first parameter of the vehicle at a first time step; a processing unit for inputting the acquired values ​​of the first parameter into a prediction model to obtain the predicted values ​​of a second parameter at a second time step, wherein the second time step is after the first time step, and the loss function of the prediction model during training includes data-driven loss, kinematic consistency loss, and dynamic consistency loss; an output unit for constructing a risk metric based on the differential energy Lyapunov exponent according to the predicted values ​​of the second parameter, wherein the risk metric includes an asymmetric factor and an energy level factor; and an assessment unit for assessing the rollover risk of the vehicle based on the value and trend of the risk metric.

[0009] Through the above technical solution, this invention creates a novel prediction paradigm that overcomes the shortcomings of existing technologies, providing accurate, reliable, and physically interpretable decision-making basis for active control systems. Based on Lyapunov stability theory, this invention constructs a function that characterizes the system's energy state, assessing rollover risk from a more fundamental physical level. This significantly improves control accuracy and adaptability to complex operating conditions, and possesses extremely high versatility and adaptability, realizing a paradigm shift from "passive response" to "active prevention." Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0010] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings:

[0011] Figure 1 A flowchart illustrating the method for assessing vehicle rollover risk according to this application is shown;

[0012] Figure 2 The overall architecture and workflow diagram according to this application are shown;

[0013] Figure 3A schematic diagram of the PINN composite loss function and training process according to this application is shown;

[0014] Figure 4 A schematic diagram of the core prediction method according to this application is shown;

[0015] Figure 5 A flowchart illustrating the vehicle rollover control method according to this application is shown;

[0016] Figure 6 A multi-level control logic diagram based on Lyapunov theory according to this application is shown;

[0017] Figure 7 A diagram illustrating the condition-adaptive differentiated control strategy according to this application is shown.

[0018] Figure 8 A schematic diagram of the structure of a vehicle rollover risk assessment system according to this application is shown;

[0019] Figure 9 A schematic diagram of the structure of a vehicle rollover control system according to this application is shown;

[0020] Figures 10a-10c Simulation calculation diagrams for non-tripping rollover conditions according to embodiments of this application are shown respectively;

[0021] Figures 11a-11c The following are schematic diagrams illustrating simulation calculations under a tripping and rollover condition according to embodiments of this application;

[0022] Figures 12a-12f The simulation verification comparison diagrams of the embodiments and comparative examples according to this application under the dual-wheel obstacle crossing condition are shown respectively;

[0023] Figures 13a-13f The following schematic diagrams show the simulation verification comparison between the embodiments of this application and the comparative examples under the J-turn extreme turning-extreme non-overturning condition;

[0024] Figures 14a-14f The following schematic diagrams show the simulation verification comparison between the embodiments of this application and the comparative examples under the J-turn extreme turn-just rollover condition;

[0025] Figures 15a-15f The simulation verification comparison diagrams of the embodiments and comparative examples according to this application under the single-sided obstacle crossing condition are shown respectively.

[0026] Figures 16a-16f The simulation verification comparison diagrams of the embodiments and comparative examples according to this application under the Fishhook condition are shown respectively. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0028] It should be noted that if the embodiments of this application involve directional indications (such as up, down, left, right, front, back, etc.), these directional indications are only used to explain the relative positional relationships and movement of the components in a specific posture (as shown in the attached figures). If the specific posture changes, the directional indications will also change accordingly. Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" can explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0029] The applicant first provides a brief description of the existing technology. As mentioned in the background section, from a dynamic perspective, rollover accidents can be divided into non-tripping rollovers and tripping rollovers. A key limitation of the existing technology is its inability to effectively address both of these complex rollover scenarios simultaneously. To quantify rollover risk assessment, various evaluation indicators have been developed, but all have inherent and insurmountable flaws, as described below.

[0030] 1. Lateral Load Transfer Rate (LTR): This metric quantifies rollover risk by monitoring the difference in vertical load between the left and right wheels of a vehicle. However, LTR has two major drawbacks: First, its calculation relies on the real-time vertical load of the tires, a physical quantity that is difficult to measure directly and accurately using conventional onboard sensors, leading to high cost and insufficient reliability in practical applications. Second, LTR suffers from a fatal signal saturation problem. Once one wheel leaves the ground, the LTR value is "clamped" at 1, and remains unchanged regardless of how the rollover process evolves, completely losing the ability to quantitatively assess the subsequent severity. This means that control systems relying on LTR are effectively "blind" at the moment when accurate information is most needed.

[0031] 2. Energy-based Rollover Index (ERI): This method assesses risk by comparing the vehicle's current total energy with a preset critical energy threshold. The core flaw of ERI stems from its formula construction: its key gravitational potential energy term mg(hh) init The ERI (Energy Retention Index) is only related to the absolute height h of the vehicle's center of gravity, therefore it cannot physically distinguish whether the rise in the center of gravity is caused by a dangerous single-wheel lift (such as a single wheel hitting a curb) or a safe double-wheel lift (such as going over a speed bump). In the latter case, the vehicle's total energy may increase dramatically in an instant, causing the ERI to generate serious false alarms and greatly reducing the system's reliability.

[0032] 3. Time-to-Rollover (TTR): This metric, based on a simplified vehicle dynamics model, predicts the time required for a vehicle to reach a critical rollover angle under the current driving conditions. However, the accuracy of TTR predictions heavily relies on the simplified assumptions of its underlying model. For example, it typically assumes that the vehicle's lateral acceleration or steering input remains constant during the prediction period, which is almost impossible in rapidly changing real-world driving scenarios. Therefore, TTR has poor predictive ability for nonlinear, time-varying input conditions (such as continuous steering corrections by the driver) and is extremely sensitive to small errors in the initial state (such as roll angle and roll rate), resulting in unstable prediction results and low reliability.

[0033] In summary, existing technologies urgently require a novel predictive paradigm that can overcome all the aforementioned shortcomings and provide accurate, reliable, and physically interpretable decision-making basis for active control systems. To address this, this invention is based on Lyapunov stability theory, constructing a function that characterizes the system's energy state to assess rollover risk from a more fundamental physical level. Specifically, this application first provides a method 100 for assessing vehicle rollover risk, such as... Figure 1 As shown in the flowchart, the evaluation method 100 may include steps S110-S140.

[0034] Step S110: Obtain the collected values ​​of the first parameter of the vehicle at the first time step.

[0035] The first parameter can include: lateral and longitudinal acceleration of the vehicle body; angular velocities of the vehicle body across its three axes; steering wheel angle; vertical acceleration of the left and right sprung masses; and vertical acceleration of the left and right unsprung masses. It is understood that the accelerations of the left and right sprung masses and unsprung masses can be further subdivided into accelerations at the front and rear suspension points. This distinction will not be made here or in subsequent formulas, but can be refined according to the specific application scenario during calculation.

[0036] Specifically, the first parameter can be used as the input vector X(t) to characterize the following sensor measurements or their calculated values ​​at time t, as expressed by the following formula:

[0037]

[0038] Among them, a x a y For the lateral and longitudinal accelerations of the vehicle body, g x g y g z Delta is the angular velocity of the vehicle's three axes. sw This refers to the steering wheel angle. The core input consists of four vertical accelerations: az... sml , az smr The vertical acceleration of the sprung masses on the left and right sides (e.g., can be obtained from the onboard IMU's a) z and roll acceleration It can be calculated based on geometric relationships (or measured directly by installing an accelerometer at the corresponding location), az uml , az umr The vertical accelerations of the left and right unsprung masses are measured directly by independent sensors. The first time step can be represented as sequence. length This is defined as the length of the input historical time series. For example, take 50 time steps (assuming a sampling frequency of 100Hz, then it corresponds to the data of the past 0.5 seconds).

[0039] Step S120: Input the collected value of the first parameter into the prediction model to obtain the predicted value of the second parameter at the second time step.

[0040] The second parameter may include: the vertical acceleration of the left and right sprung masses; the vertical acceleration of the left and right unsprung masses; the roll angle; the roll angular velocity; and the sideslip velocity. Similarly, the accelerations of the left and right sprung masses and unsprung masses can be further subdivided into accelerations at the front and rear suspensions, which will not be distinguished here or in subsequent formulas. Specifically, the second parameter can be used as an output vector Y(t) to characterize the prediction of the core higher-order dynamic state at a series of future time steps. Its composition should be consistent with the objective defined in the prediction steps of this application, as expressed by the following formula:

[0041]

[0042] Among them, az smlp , az smrp , az umlp , az umrp It is the predicted vertical acceleration of the unsprung mass on the left and right sides, φ. PIt is the predicted future roll angle. It is the predicted future roll rate, v yp This is the predicted future sideslip velocity. Higher-order dynamic quantities, such as vertical acceleration, which are directly predicted, will be used as inputs in the "Calculation and Reconstruction" step of this application to derive lower-order quantities such as displacement and angle required for the final calculation and evaluation indicators.

[0043] Specifically, the second time step, which occurs after the first time step, can be represented as a prediction. length This is defined as the length of the predicted future time series, for example, 50 time steps (corresponding to a prediction of 0.5 seconds in the future).

[0044] For example, the prediction model can be a Physics-Informed Neural Network (PINN) model, which is enforced to follow the fundamental laws of physics during training. It can be understood that the PINN prediction module, the core of this invention, is designed to achieve accurate, long-term, and physically consistent predictions of vehicle dynamics. Specifically, the network structure of the PINN prediction module used in this invention can employ an encoder-decoder architecture based on Long Short-Term Memory (LSTM) and a sequence-to-sequence (SeqtoSeq) architecture. This architecture is particularly suitable for processing time-series data, capable of learning complex dynamic patterns in historical sequences, and generating a prediction of future time series. Additionally, the network structure can also include an output layer located at the output of the decoder. Specifically, it includes the following structural functions.

[0045] 1) Encoder: Consists of two stacked LSTM network layers, each containing 128 hidden units. The encoder's role is to read the historical time series of the input and compress it into a fixed-length context vector, which contains all the key dynamic information of the historical sequence.

[0046] 2) Decoder: Also composed of two stacked LSTM layers, each containing 128 hidden units. The decoder receives the context vector generated by the encoder as its initial state, and then progressively and autoregressively generates the predicted value for each time step of the future time series.

[0047] 3) Output layer: At the output of the decoder, a fully connected (Dense) neural network layer is connected. This layer is responsible for mapping the hidden state of the decoder LSTM to the dimension of the physical quantity that needs to be predicted.

[0048] The network inputs and outputs of the PINN model can include the following:

[0049] 1) Network Input: The input can be a three-dimensional tensor with the following shape:

[0050] (batch size ,sequence length ,num features ),

[0051] batch size Batch size refers to the number of samples processed at a time, and can be adjusted based on the computer's computing power; sequence length The length of the input historical time series is defined; for example, 50 time steps (assuming a sampling frequency of 100Hz, this corresponds to the data from the past 0.5 seconds); num features The number of physical quantities input at each time step is defined. As mentioned above, the input vector X(t) at time t consists of the following sensor measurements or their calculated values:

[0052] The sensor unit includes a six-axis inertial measurement unit (IMU) mounted at the center of the vehicle body (sprout mass), and vertical acceleration sensors mounted on the left and right wheel assemblies (unsprout mass) (sensors measuring front and rear suspension acceleration have been omitted). IMU data is used to acquire the vehicle's three-axis angular velocity, longitudinal and lateral acceleration in real time, and combined with vehicle geometry parameters (such as wheelbase) to further calculate the vertical acceleration of the sprout mass on the left and right sides. The system is also supplemented by a steering wheel angle sensor.

[0053] 2) Network Output: The output is also a three-dimensional tensor with the following shape:

[0054] (batch size ,prediction length ,num predictions ),

[0055] batch size For batch size; prediction length The length of the predicted future time series is defined; for example, 50 time steps (corresponding to a prediction of 0.5 seconds in the future); num predictions The number of physical quantities to be predicted at each time step is defined. As mentioned above, the output vector Y(t) is a prediction of the core higher-order dynamic states at a series of future time steps, and its composition is consistent with the objective defined in the "prediction" step, namely:

[0056]

[0057] The overall workflow of the PINN prediction module of this invention is as follows: Figure 2 As shown, a central processing unit (ECU, Electronic Control Unit) can be used: employing a high-performance onboard domain controller (such as NVIDIA DRIVEORI or Qualcomm Snapdragon Ride platform), responsible for embedding and running the core PINN prediction model of this invention, processing sensor data streams in real time, and outputting risk prediction curves and hierarchical control commands. Specifically, the system first collects real-time dynamic data of the vehicle through onboard sensor units and performs necessary preprocessing (such as filtering and synchronization). Then, the processed data is constructed into a fixed-length historical time series, which serves as the input to the PINN model. After processing this series, the PINN model outputs a prediction of physical quantities for future time series. This prediction result is used, on the one hand, to calculate the future DELI risk curve through the "reconstruction" step; on the other hand, during the training phase, this prediction result also needs to be verified by the physical constraint layer to calculate the physical loss, thereby guiding the optimization of model parameters.

[0058] The key point of this invention is that the composite loss function during the training of the predictive model can include data-driven loss, kinematic consistency loss, and dynamics consistency loss. Specifically, the core reason this invention employs a physically-informed neural network is its unique composite loss function, which, during training, not only relies on data but is also constrained by the fundamental laws of vehicle dynamics. Figure 3 As shown, its total loss function L total It consists of three parts:

[0059] L total =L data +λ kinematic ·L kinematic +λ dynamics ·L dynamics ,

[0060] Among them, L data For data-driven loss; L kinematic Loss of kinematic consistency; L dynamics For dynamic consistency loss; λ kinematic and λ dynamicsThese are the weighting coefficients of different loss terms, used as hyperparameters to balance the weights of different loss terms. It is evident that this loss function not only includes traditional data prediction loss terms but also additional constraint loss terms based on first principles of physics. It is important to note that in the implementation of this invention, the numerical differentiation and integration operations involved will be processed using numerical stability methods known to those skilled in the art (such as high-order precision algorithms, adaptive step-size strategies, filtering and smoothing techniques, etc.) to reduce or avoid errors that may be introduced by discretization processing, ensuring the accuracy and reliability of the calculation results. The loss function can be determined using the Mean Squared Error (MSE) function, specifically including the following:

[0061] 1) Data-driven loss L data This is the supervised learning loss term for traditional neural networks. It guides training by comparing the network's predicted output with the "true value" obtained from simulation or real-world testing. The mean squared error can be used to calculate the data-driven loss L. data The following formula represents it:

[0062]

[0063] Among them, Y predicted It is the predicted value output by the prediction model, Y. ture It corresponds to the actual data.

[0064] 2) Loss of kinematic consistency L kinematic This loss term mandates that the different state variables predicted by the network must satisfy a calculus relationship, ensuring that the prediction results are kinematically consistent. Specifically, this constraint requires that the network predict the future roll angle time series (φ). predicted The result obtained by numerical differentiation must be compared with the time series of future roll angular velocity directly predicted by the network. To maintain consistency, the loss function penalizes the difference between the two, thereby forcing the network's multiple outputs to be kinematically consistent. Its form can be expressed as follows:

[0065]

[0066] in, For the time series of roll angular velocity output by the prediction model, d / dt(φ) predicted ) represents the derivative of the tilt angle time series output by the prediction model.

[0067] 3) Dynamic consistency loss L dynamicsThis loss term is the essence of PINN, requiring the network's predictions to follow Newton's second law (or Lagrange's equation). Specifically, in the rollover problem, i.e., "force / torque balance," it can be understood as the conservation of angular momentum: based on the rigid body rotation law (M=I...). α The loss requires that the “state of motion” (such as angular acceleration) predicted by the network must be explainable by the “system forces” (such as total roll moment) predicted by it. This loss can be constructed by the following steps.

[0068] Kinematic state derivation: When calculating the physical constraint loss, the vertical acceleration time series of sprung mass (vehicle body) and unsprung mass (wheel) predicted in parallel by the neural network are numerically integrated in real time to derive the corresponding vertical velocity and displacement time series.

[0069] Suspension state calculation: Based on the vertical velocities and displacements of the sprung and unsprung masses derived in step 1, calculate the real-time compression (Z) of the left and right suspensions. susp ) and compression speed In other application scenarios, it can be further subdivided into the real-time compression amount and compression speed of the left front, left rear, right front, and right rear suspensions.

[0070] Force / Torque Estimation: Estimation can be performed separately for the front and rear axles of the vehicle. Based on the suspension states (compression displacement and velocity) of each wheel obtained from network prediction and integration, the supporting forces generated by the suspension springs and dampers on both sides, as well as the anti-roll bar restoring torque caused by the suspension displacement difference, are estimated using the suspension physical model. Combined with the lateral inertial force estimated from the predicted lateral acceleration values, the total roll torque acting on the front and rear of the vehicle is calculated.

[0071] M estimated .

[0072] Suspension force estimation: based on simplified suspension physics models (such as linear spring-damping models).

[0073] Using the suspension compression and compression rate calculated above, estimate the suspension forces (F) on the left and right sides. sl,estimated F sr,estimated ).

[0074] Anti-roll bar moment estimation: The restoring moment generated by the anti-roll bar is related to the roll deformation of the axle, and its estimated value is M. arb,estimated It will be calculated in the overall moment equation.

[0075] Total roll moment calculation: Substitute the estimated suspension forces from above into the roll moment balance equation to calculate the total estimated roll moment ∑M of the axle. x,estimatedThe equation is as follows:

[0076]

[0077] Where: s is the wheel spacing of the axle. F sl,estimated and F sr,estimated These are the estimated forces (springs + dampers) on the left and right suspensions; m s a y h s g and g represent the sprung mass, lateral acceleration, height of the sprung center of mass, and gravitational acceleration, respectively; K arb It is the equivalent roll stiffness of the suspension anti-roll bar of the axle; φ in the equation arb The approximate calculation is performed by using the ratio of the difference in compression displacement between the left and right suspensions to the wheel spacing ls.

[0078]

[0079] Dynamic equation constraints: According to the rigid body rotation law M = I α The motion (α) generated by network prediction predicted It must be able to be determined by the force (M) corresponding to its predicted state. estimated This can be explained by the following formula: The loss term is the difference between the two, expressed as follows:

[0080] L dynamics =MSE(M estimated I x ·α predicted ),

[0081] Among them, M estimated To predict the torque output by the model, I x Let α be the vehicle's roll moment of inertia. predicted This is the roll acceleration output by the prediction model.

[0082] This constraint forces the network to learn the vehicle's intrinsic, unseen dynamic parameters and relationships (such as equivalent roll stiffness and damping), enabling it to make more reasonable and robust predictions when faced with unfamiliar operating conditions. This process constructs a physically self-consistent closed-loop constraint, forcing the network to learn the vehicle's inherent dynamic laws.

[0083] Additionally, it should be noted that the fixed parameters of the vehicle involved in the above loss function, such as mass, moment of inertia, stiffness coefficient, etc., can be directly found in the simulation software. In practical applications, they can be obtained by consulting the vehicle's factory parameters, conducting calibration tests, or directly learning from the PINN network.

[0084] Step S130: Based on the predicted value of the second parameter, construct the differential energy Lyapunov exponent.

[0085] DELI (Differential Energy Lyapunov Index) is a risk metric.

[0086] The risk measurement index DELI is expressed by the following formula:

[0087]

[0088] The first item in DELI This is an asymmetry factor used to quantify the degree of unevenness in energy distribution. (SGE) side For unilateral surrogate energy of the vehicle, it can include the left-side SGE. left and the right side SGE right , representing the generalized surrogate energy on the left and right sides of the vehicle, respectively, are determined by the real-time compression of the suspension on that side and the equivalent vertical velocity of the sprung mass on the left and right sides. The compression is estimated by performing two time integrals on the relative vertical acceleration between the vehicle body (sprung mass) and the wheels (unsprung mass). This relative vertical acceleration can be calculated from the data difference between the vehicle body IMU and the unsprung mass vertical acceleration sensor, while the equivalent vertical velocity can be obtained by integrating the sprung mass vertical acceleration. Specifically, it is expressed as follows:

[0089]

[0090] Among them, the first item The second term represents the elastic potential energy of the suspension. k represents the equivalent kinetic energy. susp d is the suspension stiffness coefficient of the vehicle. side The dynamic displacement (compression or extension) of the vehicle's left or right suspension, m eff The equivalent mass of the vehicle on the left or right side can be approximated as half the vehicle body mass, v side The equivalent vertical velocity of the vehicle's sprung mass on the left or right side.

[0091] The second item in DELI V is an energy magnitude factor used to quantify the total rollover-inducing energy of the system. DLEI The generalized total energy causing a vehicle rollover is rigorously constructed as a Lyapunov function in this invention. Its physical meaning lies in quantifying the "energy norm" of the vehicle's dynamic system deviating from its stable equilibrium point (i.e., the ideal upright driving state). According to Lyapunov's second method, as long as this function value is bounded and its time derivative d(V) DLEI If ) / dt is negative definite, the system is asymptotically stable. Therefore, V DLEI The value of V and its changing trend fundamentally reflect the stability margin of the system.DLEI From the tilt energy (V roll ) and weighted sideslip energy (V lateral Composed of ), the specific expression is:

[0092]

[0093] Among them, V roll For the vehicle's roll energy, V lateral φ represents the vehicle's weighted sideslip energy; φ represents the vehicle's roll angle. v is the vehicle's roll rate. y The lateral speed of the vehicle can be predicted by the PINN model of this invention based on the input signals from the sensor units; x For the vehicle's roll moment of inertia, K φ denoted as the vehicle's equivalent roll stiffness, w as the vehicle's sideslip energy weighting coefficient, and m as the vehicle's total mass, all of which are pre-calibrated vehicle physical parameters.

[0094] E critical The critical energy threshold represents the minimum potential energy required for a vehicle to overturn, expressed by the following formula:

[0095]

[0096] h cg T is the height of the vehicle's center of gravity. w Let g be the wheelbase of the vehicle, and g be the acceleration due to gravity.

[0097] Step S140: Assess the rollover risk of the vehicle based on the value and trend of the risk measurement index.

[0098] Through the above steps, the evaluation method of this invention employs an innovative three-step prediction strategy of "deconstruction-prediction-reconstruction," see [link to relevant documentation]. Figure 4 This includes the following content.

[0099] 1) Deconstruction: The complex DELI index is mathematically broken down into its most basic, directly measurable physical components, including: the vertical displacement (d) of the left and right suspensions. sl d sr ) and velocity (v) sl v sr ), the roll angle (φ) and roll rate of the vehicle body and the lateral speed of the vehicle body (v) y ).

[0100] 2) Prediction: Using an advanced time-series neural network (such as PINN), taking real-time sensor data of the vehicle as input, it predicts in parallel the evolution trend of a set of core high-order dynamic states over a future period (e.g., 500 milliseconds). The core states to be directly predicted are: the vertical acceleration of the left and right sprung masses (az). sml , az smr ), vertical acceleration of the unsprung masses on the left and right sides (az) uml , az umr The roll angle (φ) and roll rate of the vehicle body. Vehicle lateral speed (v) y ).

[0101] 3) Calculate & Reconstruct: This step transforms the higher-order quantities predicted by the neural network into lower-order quantities required to compute DELI, including the following.

[0102] Suspension state calculation: First, based on the predicted vertical acceleration time series, the relative vertical acceleration (a) on the left and right sides is calculated. rel =az sm -az um Subsequently, two successive numerical integrations are performed on this relative acceleration sequence (i.e., the first time integration of acceleration yields velocity, and the second time integration yields displacement), thereby deriving the suspension compression velocity (v) over future time series. sl v sr ) and suspension compression displacement (d sl d sr ).

[0103] DELI curve reconstruction: Reconstructing all future physical component values ​​(v) calculated through prediction and integration. sl v sr ,d sl d sr ,φ, v y Substituting these values ​​into the original mathematical formula for DELI, a physically self-consistent DELI risk prediction curve over a future time series is calculated.

[0104] In addition, embodiments of this application provide a vehicle rollover control method 200, such as... Figure 5As shown in the flowchart, the control method 200 may include steps S210-S230: Step S210, determining the value and trend of the vehicle's risk measurement index according to the vehicle rollover risk assessment method described above; Step S220, determining the vehicle's energy change rate according to the trend of the energy magnitude factor in the risk measurement index; Step S230, performing graded intervention control on the vehicle according to the risk measurement index and / or energy change rate.

[0105] Through the above steps, this invention provides a forward-looking multi-level vehicle rollover control method based on Lyapunov stability theory. This method is based on both the predicted value of the final risk index DELI and the energy change rate d(V). DLEI The predicted trend of ) / dt is used for graded intervention. Its core objective is to apply optimal control input u (such as differential braking force) to actively stabilize the Lyapunov function VDLEI and ensure its time derivative d(V) / dt. DLEI ) / dt tends to be negative. This can be achieved, for example, through actuator units, including the vehicle's Electronic Stability Control (ESC) / Anti-lock Braking System (ABS) module for precise braking of individual wheels; and the vehicle's active / semi-active suspension controller (if equipped) for adjusting suspension stiffness and damping.

[0106] Among them, such as Figure 6 As shown, tiered intervention can include a first-level threshold (early warning), a second-level threshold (proactive stabilization intervention), and a third-level threshold (emergency sedation).

[0107] Specifically, step S230 may include any one or a combination of the following steps.

[0108] Step S231: Initiate an early warning if the value of the risk measurement indicator exceeds the first safety threshold.

[0109] This step is to execute the first-level threshold (early warning), which can be understood as: when the system predicts that the future DELI value will exceed a lower first safety threshold, an early warning is initiated.

[0110] Step S232: If the risk metric exceeds the second safety threshold and the energy change rate is positive, control the vehicle to perform active stability intervention. The second safety threshold is greater than the first safety threshold.

[0111] This step, in order to execute the second-level threshold (active stabilization intervention), can be understood as follows: when the system's predicted final risk indicator DELI exceeds the intervention threshold (second safety threshold), and simultaneously the system's energy change rate d(V) is predicted... DLEIWhen d(V) / dt is positive, it indicates that the vehicle is not only in a risky state, but is also actively sliding towards a more dangerous situation. At this time, the goal of the control system is to apply a stabilizing control torque to counteract the positive energy injection, making d(V) / dt remain stable. DLEI ) / dt is suppressed to zero or negative values.

[0112] Step S233: If the risk metric exceeds the third safety threshold, or if the energy change rate is greater than the fourth safety threshold after active stabilization intervention, control the vehicle to perform emergency stabilization intervention. The third safety threshold is greater than the second safety threshold.

[0113] This step, in order to perform the three-level threshold (emergency sedation), can be understood as follows: when the system predicts that the DELI value will exceed the emergency threshold (third safety threshold), or the predicted d(V) value will exceed the emergency threshold (third safety threshold), the system will perform the emergency sedation. DLEI The value of ) / dt remains significantly positive after applying secondary intervention (greater than the fourth safety threshold; the specific value is related to the vehicle's parameters, such as weight and other vehicle conditions; for example, it could be 0.5 × 10). 5 ~2×10 5 When the vehicle is in a rapidly diverging, unstable state (between V and V), the control objective is to apply maximum control authority, thereby generating the strongest negative d(V). DLEI ) / dt, which means dissipating the rollover energy of the system at the fastest speed to pull the vehicle back from the edge of overturning.

[0114] In addition, after step S230, the control method 200 may further include: step S240, in the case of an increase in the risk measurement index, controlling the vehicle to perform corresponding actions based on the changing trend of the energy level factor and / or asymmetric factor in the risk measurement index.

[0115] like Figure 7 As shown, this step, based on graded intervention, can also analyze the energy injection leading to rollover online and find (d(V)). DLEI The source of ) / dt>0) is to find the most efficient way to make d(V) DLEI The optimal control strategy is to make ) / dt turn negative.

[0116] Specifically, step S240 may include performing step S241 and / or step S242.

[0117] Step S241: If the increase in risk measurement index is dominated by asymmetric factors, determine that the vehicle is in a tripping rollover scenario, and control the vehicle to perform differential braking and / or active suspension adjustment.

[0118] This can be understood as follows: when the system determines that the increase in the DELI value is mainly dominated by a sharp increase in the "asymmetry factor," it typically corresponds to a typical tripping-and-rollover scenario, such as a single wheel colliding at high speed with a curb or obstacle. Under this condition, the system identifies a positive rate of energy change d(V). DLEI The force ) / dt primarily originates from a brief but significant external asymmetric input (i.e., impact torque). Therefore, the control strategy focuses on instantly generating a restoring torque in the opposite direction to this impact torque to directly counteract the vehicle's rollover tendency. Specific control execution details are as follows.

[0119] 1) Differential Braking (Core Actuator): The control system immediately instructs the Electronic Stability Control (ESC) module to execute precise differential braking. For example, when the right front wheel experiences an impact causing the vehicle to roll to the right, the system applies rapid and forceful braking to the left (the side opposite to the lifted wheel) front and rear wheels. This braking action not only generates a stabilizing yaw moment that causes the vehicle to yaw to the left, but also instantaneously increases the ground pressure on the left wheel through the dynamic transfer of vertical load, thereby creating a restoring moment that directly resists the vehicle's tilt to the right.

[0120] 2) Active Suspension (if equipped): If the vehicle is equipped with active or semi-active suspension, the control system will adjust accordingly. In the example above, the controller will instantaneously adjust the stiffness and damping of the left suspension to the maximum to provide maximum physical support, while possibly reducing the stiffness of the right suspension to absorb some impact energy.

[0121] In summary, these two measures can be combined to reduce the d(V) caused by external physical impact. DLEI Suppressing or eliminating the positive value of ) / dt is the most direct and effective engineering method to deal with the above working conditions.

[0122] Step S242: When the increase in risk measurement index is dominated by energy level factor, determine that the vehicle is in a non-tripping rollover scenario, and control the vehicle to perform dynamic weighted global energy reduction and attitude stability management.

[0123] This can be understood as the system determining that the increase in the DELI value is mainly due to the "energy level factor" (V). DLEI When the value remains high and the "asymmetry factor" gradually increases, it usually corresponds to non-tripping rollover scenarios, such as vehicles performing extreme obstacle avoidance maneuvers like J-turns or fishhooks at high speeds.

[0124] Under these operating conditions, the system identifies the positive d(V) DLEIThe energy dt primarily originates from the enormous centrifugal force generated by the vehicle's high-speed maneuvering. The control strategy employs a coordinated control method that prioritizes reducing the total system energy while also considering attitude management.

[0125] 1) Global Energy Reduction (Basic Braking): The primary action of the control system is to apply a basic braking force to all four wheels via the braking module. Its core objective is to systematically reduce the vehicle's total kinetic energy by decreasing the vehicle's longitudinal speed. Due to V DLEI Both roll and sideslip energy are highly correlated with vehicle speed; therefore, reducing vehicle speed is the fundamental way to reduce V-forces. DLEI Energy base is a prerequisite for mitigating the risk of rollover.

[0126] 2) Attitude Stability Management (Differential Braking Overlay): While applying basic braking, the system overlays an additional differential braking force on the outer wheel during cornering, based on the rollover tendency. This serves two main purposes: first, to generate a stabilizing yaw moment to suppress oversteer and help the vehicle maintain the correct trajectory; second, to directly counteract the roll moment generated by centrifugal force, suppressing the continuous increase of the roll angle, thereby controlling V-axis stability. DLEI V in roll (Tilt energy) component is managed.

[0127] 3) Dynamic Weight Allocation: The intensity of basic braking and differential braking is not in a fixed ratio. The control system dynamically and adaptively adjusts the weight allocation between the two based on the PINN model's prediction of future risks. For example, in the initial stage of maneuvering, when excessive vehicle speed is the primary concern, control may prioritize basic braking; when vehicle speed has decreased but roll rate becomes the main threat, the weight of differential braking will be increased accordingly to achieve better control of d(V). DLEI ) / dt provides the most effective and stable calming effect.

[0128] In summary, this invention provides a predictive vehicle rollover prevention method based on physical information neural networks and differential energy indices, and particularly relates to an intelligent control system and method capable of predicting vehicle rollover risk in real time and actively intervening to prevent accidents. This invention technically integrates three interdisciplinary fields: vehicle dynamics, artificial intelligence predictive models, and real-time embedded control systems.

[0129] Compared with the prior art, the present invention has the following significant advantages.

[0130] 1) Predictability and foresight: The greatest advantage of this system lies in its foresight. It does not react only when danger occurs, but intervenes in advance based on predictions of the future, gaining valuable reaction time to prevent accidents and realizing a paradigm shift from "passive response" to "proactive prevention".

[0131] 2) High accuracy and low false alarm rate: This invention achieves a physical-level "logical AND gate" by multiplying the "asymmetric factor" and the "energy level factor", which fundamentally solves the inherent defects of traditional indicators.

[0132] 3) Overcoming the limitations of a single factor: If only an asymmetric factor is available, the system may generate false alarms under asymmetric but low-energy, non-dangerous conditions such as low-speed scraping of curbs; if only an "energy level factor" (V) is available... DLEI The system may generate false alarms under certain high-energy but controllable stable conditions (such as controlled high-speed slalom or drift). In J-turn maneuvers, the vehicle is in a high-energy, nonlinear but quasi-stable state for an extended period, and indicators based solely on energy levels may fail to provide timely warnings because the system does not immediately become unstable. This invention only issues an alarm when both the "energy level factor" confirms high-risk potential and the "asymmetric factor" confirms a rollover tendency, thus ensuring sensitivity to real dangers and robustness to non-dangerous conditions.

[0133] 4) Interpretability and Robustness (White-box AI): The decision-making process of this system is based on explicit physical quantities and laws, rather than an uninterpretable "black box". This first-principles-based constraint gives it stronger generalization ability and robustness when faced with extreme conditions not seen in the training data.

[0134] 5) Control precision and adaptability to working conditions: By analyzing the different components of the DELI index, the system can effectively distinguish between different rollover causes such as tripping and non-tripping, and execute the optimal differentiated control strategy that matches them, which significantly improves the control precision and adaptability to complex working conditions.

[0135] 6) Hierarchical Closed-Loop and Adaptability: The system achieves a complete closed loop from risk perception and decision-making to control execution. The hierarchical intervention strategy ensures that the system provides maximum safety while minimizing interference with normal driving. Furthermore, this PINN framework can be used for online system identification and reverse inference of vehicle parameters such as suspension stiffness, thereby automatically adapting to parameter changes caused by different vehicle models or load variations, exhibiting extremely high versatility and adaptability.

[0136] 7) Theoretical Rigor and Provable Stability: This invention does not simply combine sensor signals, but rather abstracts the vehicle rollover problem into a stability problem of a dynamic system. Its core idea includes the following parameters: energy function (V... DLEI ), as the Lyapunov function of the system, has a value V. DLEI >0 represents the "energy distance" of the system from its stable equilibrium point; the rate of energy change (d(V)) DLEId(V) / dt), as the derivative of the Lyapunov function, directly reflects the real-time trend of stability. DLEI ) / dt>0 means there is external power injection, and the system is becoming unstable;

[0137] d(V DLEI A value of 0 for dt / dt < 0 indicates that the system energy is dissipating and is approaching stability. The Ultimate Risk Index (DELI), by introducing a "symmetry factor," becomes a physically complete decision indicator. It ensures that only when the rollover energy (V) is within the acceptable range can the system achieve a stable state. DLEI The system is considered to have a real risk only when both conditions of being huge and having a highly asymmetrical energy distribution are met simultaneously.

[0138] Therefore, this assessment framework based on Lyapunov theory provides a provable theoretical basis for risk assessment and control.

[0139] On the other hand, this application also provides a vehicle rollover risk assessment system 300, such as... Figure 8 As shown in the structural diagram, the evaluation system 300 may include: a data acquisition unit 310, used to acquire the acquired values ​​of the first parameter of the vehicle at the first time step; a processing unit 320, used to input the acquired values ​​of the first parameter into the prediction model to obtain the predicted values ​​of the second parameter at the second time step, wherein the second time step is after the first time step, and the loss function of the prediction model during training includes data-driven loss, kinematic consistency loss, and dynamic consistency loss; an output unit 330, used to construct a risk metric based on the differential energy Lyapunov exponent according to the predicted values ​​of the second parameter, wherein the risk metric includes an asymmetric factor and an energy level factor; and an evaluation unit 340, used to evaluate the rollover risk of the vehicle according to the value and trend of the risk metric.

[0140] On the other hand, this application also provides a vehicle rollover control system 400, such as... Figure 9 As shown in the structural diagram, the control system 400 may include: a vehicle rollover risk assessment system 300, used to determine the value and trend of the vehicle's risk measurement index; a determination device 410, used to determine the vehicle's energy change rate based on the trend of the energy magnitude factor in the risk measurement index; and a control device 420, used to perform graded intervention control on the vehicle based on the risk measurement index and / or the energy change rate.

[0141] To further illustrate the essential features of the present invention, several embodiments are provided below.

[0142] Example 1: Non-tripping rollover (high-speed emergency obstacle avoidance).

[0143] 1. Scenario: An SUV equipped with the system of this invention is performing an emergency obstacle avoidance maneuver (such as a "fishhook test" or J-turn) on a highway, with the driver making rapid left-right consecutive turns. Figure 10a This is a simulation diagram of the DELI index under this operating condition; Figure 10b This is a risk trend chart under this operating condition; Figure 10c This is the risk phase plane diagram under this operating condition.

[0144] 2. Risk Identification: The system detects through sensor data that due to the vehicle's high initial longitudinal speed, a significant lateral acceleration will be generated during emergency steering, leading to a substantial sideslip speed (v). y ) and vehicle roll angle / angular velocity These physical quantities make the "energy level factor" (V) in the DELI index... DLEI The body roll remains at a high level. Simultaneously, the body roll causes the "asymmetry factor" to gradually increase. For example... Figure 10a As shown, the PINN model predicts that within the next few hundred milliseconds, the combined effect of the two factors will cause the total DELI value to exceed the second or even third level threshold.

[0145] 3. Differentiated Control: Based on the above-described scheme of this application, the system identifies this as a typical "energy-driven" rollover risk. Therefore, its control strategy emphasizes both "deceleration" and "righting." For example, this can be executed through the ECU command actuator:

[0146] 1) Reduce speed: Apply moderate braking to the vehicle to quickly reduce the vehicle's kinetic energy and weaken the "energy level factor" at its source.

[0147] 2) Righting: Simultaneously, additional braking force (differential braking) is applied to the wheels on the outer side of the steering wheel to generate a stabilizing torque that resists rollover and suppresses the continued growth of the "asymmetry factor." The entire control process is gradual and coordinated, designed to smoothly mitigate risks.

[0148] Specifically, in such Figure 10b The graph for the "J-turn rollover condition" shows that starting from t=4.5 seconds, the vehicle's high-speed maneuvering results in enormous roll and sideslip energy, causing the absolute value of the DELI index to continuously increase, eventually far exceeding -10.0. This clearly reflects that the "energy level factor" is at a high level, indicating a significant risk of rollover for the vehicle. Specifically, this includes the following:

[0149] 1) The “asymmetry factor” gradually increases: It is precisely because the vehicle roll causes differences in the left and right suspension states and tire loads that the “asymmetry factor” deviates from zero and multiplies with the high energy factor, ultimately leading to a significant decrease in the DELI value.

[0150] 2) Prediction will exceed the threshold: At t≈3 seconds, the DELI index will cross the first-level threshold; at t≈4.5 seconds, due to the continuous decrease in the DELI value and the energy change rate d(V) DLEI ) / dt is positive (see risk phase plane) Figure 10c This will trigger a secondary intervention.

[0151] The system then identified this as a typical "energy-driven" rollover risk, and the control strategy was to balance "deceleration" and "righting." Specifically, this included the following:

[0152] 1) Speed ​​Reduction (Reducing Energy Quantity Factor): The control system applies braking to the entire vehicle to reduce its kinetic energy. This directly reduces (V... DLEI The value of ) such as Figure 10c The "risk phase plane" diagram has the effect of preventing the trajectory from continuing to diverge in a direction away from the origin (left or right).

[0153] 2) Righting (Suppressing Asymmetry Factor): Simultaneously, differential braking is applied to the outer steering wheel to generate a restoring torque that resists rollover. This operation aims to directly counteract the growth of the "asymmetry factor," such as... Figure 10b On the "risk phase plane" diagram on the right, its effect is to apply a factor that causes the rate of energy change d(V) to... DLEI The force that turns ) / dt into a negative value pulls the trajectory back to the stable equilibrium point (origin).

[0154] Example 2: Tripping-type rollover (single wheel impact with curb).

[0155] 1. Scenario: While the vehicle is traveling at a moderate speed, its right front wheel suddenly hits the curb. Figure 11a This is a simulation diagram of the DELI index under this operating condition; Figure 11b This is a risk trend chart under this operating condition; Figure 11c This is the risk phase plane diagram under this operating condition.

[0156] 2. Risk Identification: Sensors (especially suspension displacement sensors or IMUs) will detect a severe, high-frequency impact coming from the right side of the vehicle. For example... Figure 11a As shown, the PINN model immediately predicts that the "asymmetry factor" in the DELI index will experience a pulse-like, large spike, while the "energy level factor" may not change much. The system identifies this as a typical "asymmetric shock-driven" rollover risk.

[0157] 3. Differentiated Control: Based on the above-described scheme of this application, the danger at this moment is instantaneous and fatal. The system determines that the fastest and strongest response must be taken. Therefore, the ECU immediately triggers a level three emergency intervention, instructing the actuator to perform the following:

[0158] 1) Powerful Righting: Ignoring gentle deceleration, all control resources are used to resist rollover. The ESC is instructed to apply maximum braking force to the front and rear wheels on the left side (opposite to the impact side) to generate the strongest counter-stabilizing torque.

[0159] 2) Active Suspension (if applicable): Simultaneously instruct the active suspension controller to instantly adjust the stiffness and damping of the left suspension to the maximum, physically supporting the vehicle body and resisting the tendency to roll to the right. The goal of this control strategy is not ride comfort, but to counteract sudden, asymmetrical rollover moments with the fastest speed and greatest force.

[0160] Specifically, in Figure 11b As can be seen, the "asymmetry factor" experiences a pulse-like spike. At the moment of impact at t≈4.7 seconds, the DELI index (red curve) shows a precipitous, almost vertical drop. This is precisely because the external shock causes the "asymmetry factor" to surge instantaneously, resulting in a sudden change in the DELI value.

[0161] Furthermore, it is necessary to identify "asymmetric shock-driven" risks: the key to system risk identification lies in simultaneously monitoring the rate of energy change d(V). DLEI ) / dt. In such Figure 11b In the "single-sided obstacle crossing condition", it can be seen that d(V DLEI At the moment of impact, a huge, sharp positive pulse appeared, with a peak value exceeding 8 × 10⁻⁶. 5 This signal clearly tells the system that this is a dangerous event caused by a sudden, rapid energy injection, driven by an external impact, rather than normal driver operation. The danger is instantaneous and potentially fatal, and the system determines that the fastest and strongest response is necessary, directly triggering Level 3 emergency intervention. This includes the following:

[0162] 1) Powerful Righting: The control objective is to generate the strongest counter-stabilizing torque. Specifically, this involves instructing the ESC to apply maximum braking to the left-hand wheel (the side opposite the impact side). The goal of this action is to generate the strongest negative d(V) torque. DLEI ) / dt, to offset the huge positive energy injection brought about by the impact at the fastest speed.

[0163] 2) Active suspension (if applicable): Simultaneously, the stiffness and damping of the left suspension are adjusted to their maximum to provide physical support. These two measures work together to counteract the massive energy pulse seen in the diagram.

[0164] In addition, to verify the superiority of the differential energy Lyapunov exponent (DELI) proposed in this invention over the prior art (LTR, ERI), this paper also provides simulation verification comparative examples of the embodiments of this invention and comparative examples. Specifically, comparative tests were conducted on various typical working conditions in the TruckSim-Simulink co-simulation environment.

[0165] Comparative Example 1: Two-wheel obstacle crossing condition (false alarm test).

[0166] Scenario: A vehicle travels at normal speed over symmetrically placed speed bumps (both wheels cross the bumps simultaneously, and the bumps increase in size but do not cause a rollover). This is a typical safe operating condition and should not trigger a rollover warning. Simulation results are as follows: Figures 12a-12f As shown. Among them. Figure 12a It is a comparison chart of the three indicators LTR, ERI, and DELI under this operating condition. Figures 12b-12d This is a simulation calculation diagram of the three indices LTR, ERI, and DELI under this working condition. Figure 12e This is a risk trend diagram under this operating condition. Figure 12f This is the risk phase plane diagram under this operating condition.

[0167] 1) ERI (purple curve): Indicates a significant risk of false alarms. As shown in the graph, when the vehicle passes over a speed bump (approximately 5-11 seconds), the overall center of gravity of the vehicle rises due to the lifting of both wheels, causing a sharp fluctuation in the total energy of the system. This results in large positive and negative peaks in the ERI indicator. Such energy changes unrelated to rollover can easily trigger false alarms, reducing system reliability.

[0168] 2) LTR (blue curve): It basically stays near zero, but at about 9 seconds, due to the vehicle bumps, there is a tendency for a sudden change in the instantaneous load on one side of the wheel, and the LTR shows an instantaneous negative peak, which also poses a risk of false alarm.

[0169] 3) DELI (red curve, this invention): Exhibits excellent robustness. Although the total system energy (as shown in the dV / dt plot) fluctuates dramatically, the "asymmetry factor" in the DELI formula remains close to zero because the energy injection is essentially symmetrical. This results in only minor fluctuations in the final DELI index value throughout the process, always well below any meaningful warning threshold, successfully suppressing false alarms caused by safe vertical turbulence.

[0170] Comparative Example 2: J-turn extreme turning condition (sensitivity and discrimination test).

[0171] Scenario: By fine-tuning the vehicle's center of gravity height, two extremely close J-turn conditions were set up: "limited non-rollover" and "just rollover" (center of gravity increases by only 1mm). Simulation results are as follows. Figures 13a-13c as well as Figures 14a-14f As shown. Among them. Figure 13a Figures 14a and 14a are comparison charts showing the LTR, ERI, and DELI indicators together under the two operating conditions mentioned above. Figures 13b-13d 14b-14d are simulation calculation diagrams of the three indices LTR, ERI and DELI under the above two working conditions. Figure 13e 14e is a risk trend diagram under the two operating conditions mentioned above. Figure 13f 14f is the risk phase plane diagram under the above two operating conditions.

[0172] 1) LTR (blue curve): In both cases, the LTR curve drops rapidly to a saturation value of -1 after about 3 seconds, making it completely impossible to distinguish between the two drastically different outcomes of "near-overturning" and "just-overturning". Once the wheel leaves the ground, the LTR becomes completely ineffective, losing its ability to quantitatively assess the subsequent risk evolution.

[0173] 2) ERI (purple curve): Although the ERI value continues to decrease during the rollover process under the "just rollover" condition, its curve is very similar in the early stage (5-7 seconds) of the two conditions, making it difficult to make effective distinctions, resulting in untimely warnings and poor differentiation.

[0174] 3) DELI (red curve, this invention): Exhibits excellent sensitivity and discrimination. Specifically, in the "limited non-rollover" condition, the DELI value reaches a negative peak after approximately 7 seconds, then significantly decreases as the vehicle stabilizes. Its energy factor risk phase plane plot (dV / dt vs DELI) shows a closed-loop trajectory, indicating that although the system experienced a risky state, it eventually returned to stability. However, in the "just rollover" condition, the DELI value does not decrease after reaching a similar peak, but instead increases continuously and divergently (negatively), clearly indicating an irreversible rollover trend. Its energy factor risk phase plane plot shows a spiral divergence trajectory, with the energy change rate dV / dt remaining consistently positive, indicating that the system energy is being continuously injected and tending towards instability.

[0175] Conclusion: By comparing the DELI curves under the two operating conditions, this invention can identify the critical point of system instability earlier and more accurately, providing a clear basis for decision-making.

[0176] Comparative Example 3: Single-sided obstacle crossing condition (tripping and rollover test).

[0177] Scenario: A vehicle's single wheel strikes an obstacle 0.5 meters high, causing a typical trip-over rollover. Simulation results are as follows.Figures 15a-15f As shown. Among them. Figure 15a It is a comparison chart of the three indicators LTR, ERI, and DELI under this operating condition. Figures 15b-15d This is a simulation calculation diagram of the three indices LTR, ERI, and DELI under this working condition. Figure 15e This is a risk trend diagram under this operating condition. Figure 15f This is the risk phase plane diagram under this operating condition.

[0178] 1) LTR (blue curve): It saturates rapidly to -1 after the moment of impact (about 4.6 seconds). Although it can detect the wheel leaving the ground, it completely loses quantitative information on the severity of the subsequent rollover process.

[0179] 2) ERI (purple curve): It can reflect the overall trend of rollover, but its initial slope is gentler than DELI, and there is a certain lag.

[0180] 3) DELI (red curve, this invention): The response is the most rapid and severe. As can be seen from the figure, at the moment of impact, the "asymmetry factor" of DELI increases sharply due to the huge unilateral impact, causing the DELI value to drop precipitously immediately, providing the most timely warning. Its value continues to change throughout the rollover process, accurately quantifying the severity of the rollover. At the same time, its risk trend graph (dV / dt) shows a huge positive pulse at the moment of impact, which is a typical characteristic of "asymmetric impact-driven" rollover, and can be effectively identified by the control strategy of this invention.

[0181] Comparative Example 4: Fishhook Condition (LTR Saturation Failure Test).

[0182] Scenario: The vehicle performs a "fishhook" maneuver, causing the inner rear wheel to remain off the ground. Simulation results are as follows: Figures 16a-16f As shown. Among them. Figure 16a It is a comparison chart of the three indicators LTR, ERI, and DELI under this operating condition. Figures 16b-16d This is a simulation calculation diagram of the three indices LTR, ERI, and DELI under this working condition. Figure 16e This is a risk trend diagram under this operating condition. Figure 16f This is the risk phase plane diagram under this operating condition.

[0183] 1) LTR (blue curve): After about 3 seconds, as the inner wheel leaves the ground, the LTR index immediately saturates to around -1 and remains at this saturation value for up to 3 seconds, completely failing to reflect the vehicle's attitude changes and risk evolution during the process.

[0184] 2) ERI (purple curve): There is almost no obvious response under this condition, and the potential rollover risk is completely not identified.

[0185] 3) DELI (red curve, this invention): Excellent performance. After the wheels leave the ground and LTR saturates, the DELI value does not become ineffective, but rather maintains a high risk level after reaching a risk peak (approximately 4 seconds). This accurately reflects that the vehicle is in a high-risk, quasi-stable critical state, providing a reliable basis for subsequent continuous intervention (such as reducing total energy by slowing down).

[0186] Conclusion: Through the above multi-condition simulation comparison, it is proved that the DELI index proposed in this invention is significantly superior to the prior art in terms of suppressing false alarms, improving sensitivity under extreme conditions, effectively distinguishing critical states, and avoiding signal saturation.

[0187] On the other hand, this application also provides a vehicle that may include the vehicle rollover risk assessment system and / or the vehicle rollover control system described above.

[0188] The beneficial effects of the vehicle rollover risk assessment system and vehicle provided by this invention can be referred to the above description of a vehicle rollover risk assessment method, and will not be repeated here.

[0189] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0190] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for assessing vehicle rollover risk, characterized in that, The evaluation method includes: Obtain the collected value of the first parameter of the vehicle at the first time step; The collected value of the first parameter is input into the prediction model to obtain the predicted value of the second parameter at the second time step, wherein the second time step is after the first time step, and the loss function of the prediction model during training includes data-driven loss, kinematic consistency loss and dynamic consistency loss. Based on the predicted value of the second parameter, a risk metric based on the differential energy Lyapunov index is constructed, wherein the risk metric includes an asymmetric factor and an energy level factor; and The rollover risk of the vehicle is assessed based on the values ​​and trends of the aforementioned risk metrics. The second parameter includes: the vertical acceleration of the left sprung mass, the vertical acceleration of the right sprung mass, the vertical acceleration of the left unsprung mass, the vertical acceleration of the right unsprung mass, the roll angle, the roll rate, and the lateral velocity of the vehicle. The step of constructing a risk measurement index based on the differential energy Lyapunov index according to the predicted value of the second parameter includes: Calculate the relative vertical acceleration between the left and right sides based on the vertical acceleration of the left sprung mass, the right sprung mass, the left unsprung mass, and the right unsprung mass. By performing two successive numerical integrations on the time series of the relative vertical acceleration, the suspension compression velocity and suspension compression displacement in the future time series can be obtained. The risk metric is calculated by substituting the suspension compression speed, suspension compression displacement, roll angle, roll angular velocity, and vehicle lateral velocity into the formula for the risk metric.

2. The evaluation method according to claim 1, characterized in that, The first parameter includes: lateral and longitudinal acceleration of the vehicle body; angular velocity of the vehicle body on three axes; steering wheel angle; vertical acceleration of the left sprung mass and the right sprung mass; vertical acceleration of the left unsprung mass and the right unsprung mass.

3. The evaluation method according to claim 1, characterized in that, The prediction model is a physical information neural network model, comprising: an encoder and a decoder based on a long short-term memory network; and an output layer located at the output of the decoder. The shape of the three-dimensional tensor of the network input of the physical information neural network model is: ( The shape of the three-dimensional tensor output by the physical information neural network model is: ( ), in, For batch size, To input the length of the historical time series, The number of physical quantities input at each time step. To predict the length of future time series, The number of physical quantities that need to be predicted at each time step.

4. The evaluation method according to claim 1, characterized in that, The loss function includes: in, Data-driven loss; Loss of kinematic consistency; This represents the loss of dynamic consistency. and These are the weighting coefficients for different loss terms.

5. The evaluation method according to claim 4, characterized in that, The loss function is determined using the mean squared error (MSE) function. Among them, the data-driven loss The following formula represents it: in, It is the predicted value output by the prediction model. It corresponds to real data. The loss of kinematic consistency The following formula represents it: in, The time series of roll angular velocity output by the prediction model. The time series of the roll angle output by the prediction model. The derivative of the time series of the roll angle output by the prediction model is given. The dynamic consistency loss The following formula represents it: in, The torque output by the prediction model. Let be the roll moment of inertia of the vehicle. The time series of the derivative of the roll angular velocity output by the prediction model is the time series of the roll angular acceleration.

6. The evaluation method according to claim 1, characterized in that, The risk measurement index is The following formula represents it: in, For the aforementioned asymmetry factor, For the unilateral proxy energy of the vehicle, including the left side and the right side The following formula represents it: in, Let be the suspension stiffness coefficient of the vehicle. This refers to the dynamic displacement of the vehicle on the left or right side suspension. Let be the equivalent mass of the vehicle on the left or right side. The equivalent vertical velocity of the sprung mass of the vehicle on the left or right side. in, The energy level factor is... The generalized total energy that caused the vehicle to roll over, The critical energy threshold includes the following parameters: in, The roll energy of the vehicle. The weighted sideslip energy of the vehicle; Let be the vehicle's body roll angle. Let be the roll rate of the vehicle. Let be the lateral velocity of the vehicle; The roll inertia of the vehicle, The equivalent roll stiffness of the vehicle, This is the sideslip energy weighting coefficient for the vehicle. The total mass of the vehicle is denoted as . The height of the vehicle's center of gravity. Let g be the wheelbase of the vehicle, and g be the acceleration due to gravity.

7. A method for controlling vehicle rollover, characterized in that, The control method includes: The method for assessing vehicle rollover risk according to any one of claims 1-6 determines the value and trend of the risk measurement index of the vehicle; Based on the changing trend of the energy magnitude factor in the risk measurement index, the energy change rate of the vehicle is determined; and The vehicle is subject to graded intervention control based on the risk metric and / or the energy change rate.

8. The control method according to claim 7, characterized in that, The graded intervention control of the vehicle based on the risk metric and / or the energy change rate includes one or more of the following: An early warning is triggered if the value of the risk metric exceeds a first safety threshold. If the risk metric exceeds a second safety threshold and the energy change rate is positive, the vehicle is controlled to perform active stability intervention, wherein the second safety threshold is greater than the first safety threshold; If the risk metric exceeds the third safety threshold, or if the energy change rate is greater than the fourth safety threshold after the active stabilization intervention is performed, the vehicle is controlled to perform an emergency sedation intervention, wherein the third safety threshold is greater than the second safety threshold.

9. The control method according to claim 7, characterized in that, The control method further includes: when the risk metric index increases, controlling the vehicle to perform the following actions based on the changing trends of the energy level factor and / or asymmetry factor in the risk metric index: If the increase in the risk metric is dominated by the asymmetric factor, it is determined that the vehicle is in a tripping-induced rollover scenario, and the vehicle is controlled to perform differential braking and / or active suspension adjustment; or When the increase in the risk metric is dominated by the energy level factor, it is determined that the vehicle is in a non-tripping rollover scenario, and the vehicle is controlled to perform dynamically weighted global energy reduction and attitude stability management.

10. A vehicle rollover risk assessment system, characterized in that, The evaluation system includes: The acquisition unit is used to acquire the acquisition value of the first parameter of the vehicle at the first time step; The processing unit is used to input the collected value of the first parameter into the prediction model to obtain the predicted value of the second parameter at the second time step, wherein the second time step is after the first time step, and the loss function of the prediction model during training includes data-driven loss, kinematic consistency loss and dynamic consistency loss. The output unit is used to construct a risk metric based on the differential energy Lyapunov index according to the predicted value of the second parameter, wherein the risk metric includes an asymmetric factor and an energy level factor; and The assessment unit is used to assess the rollover risk of the vehicle based on the values ​​and trends of the risk measurement indicators. The second parameter includes: the vertical acceleration of the left sprung mass, the vertical acceleration of the right sprung mass, the vertical acceleration of the left unsprung mass, the vertical acceleration of the right unsprung mass, the roll angle, the roll rate, and the lateral velocity of the vehicle. The output unit is configured to perform the following operations when constructing a risk metric based on the differential energy Lyapunov index: Calculate the relative vertical acceleration between the left and right sides based on the vertical acceleration of the left sprung mass, the right sprung mass, the left unsprung mass, and the right unsprung mass. By performing two successive numerical integrations on the time series of the relative vertical acceleration, the suspension compression velocity and suspension compression displacement in the future time series can be obtained. The risk metric is calculated by substituting the suspension compression speed, suspension compression displacement, roll angle, roll angular velocity, and vehicle lateral velocity into the formula for the risk metric.

Images