Automobile aligning torque estimation method based on multi-sensor coupling

Abstract

The invention relates to the technical field of automobiles, in particular to an automobile aligning torque estimation method based on multi-sensor coupling, which comprises the following steps: S1, collecting data in different sensors, including steering wheel angle, wheel speed, longitudinal vehicle speed, IMU, radar and camera, S2, preprocessing the collected data, and obtaining a pre-processed data; s3, coupling data of multiple sensors, S4, establishing an aligning torque estimation model, and S5, establishing a parameter adaptive module. The aligning torque estimation method based on multi-sensor coupling does not depend on a single sensor, and different information is coupled to improve the estimation accuracy; sensor data are processed and coupled by using a complex algorithm, so that the sensor has faster response capability; the method has higher dynamic adaptability, can adjust the estimation result in real time, and improves the system stability.

Description

Technical Field

[0001] The present invention relates to the technical field of automobiles, and particularly to a method for estimating the return torque of an automobile based on multi-sensor coupling. Background Art

[0002] During the driving of a vehicle, the resistance torque felt by the driver when turning the steering wheel is called road feel. Road feel enables the driver to perceive the motion state of the vehicle, road surface state information, etc., and is a key link in constructing the "human-vehicle-road" closed-loop system. In a traditional mechanical steering system, the return torque is an important part of road feel simulation and is a key dynamic parameter generated when the tire contacts the ground during vehicle steering. The wheel and the ground generate a side slip to produce a return torque, and then the wheel return torque is transmitted to the steering wheel through the steering mechanism. The return torque directly affects the return characteristics of the steering wheel, handling stability, and driver road feel feedback. However, in a steer-by-wire system, the mechanical connection between the steering wheel and the steering wheel is cancelled, and the steering wheel and the steering wheel are connected by an electrical signal. In this way, the driver cannot directly receive the feedback information of the road surface state through the steering wheel, that is, the intuitive perception of "road feel" is lost. This seriously affects the driver's ability to judge the dynamic state of the vehicle and driving safety, and this has also become one of the core problems faced by the widespread application of the current steer-by-wire system.

[0003] In modern vehicle control, accurately estimating the return torque is crucial for maintaining the vehicle's straight-line driving ability and the driver's feeling of turning driving speed and driving performance, and is one of the key technologies affecting vehicle maneuverability and stability; accurate return torque can provide clear and real road feel, enabling the driver to better perceive the road surface conditions and vehicle dynamics, and enhancing the driving experience; it is of extremely important significance for improving vehicle maneuverability, enhancing vehicle stability, improving driving safety, and optimizing vehicle design. The steer-by-wire system can also adjust the return torque according to different road surface conditions to ensure that sufficient return ability can be provided on low-adhesion coefficient road surfaces, adapt to different road conditions, and improve the stability and safety of the vehicle.

[0004] Traditional methods for estimating the return torque often rely on single or a few sensors, or vehicle dynamics models for estimation, and have the following problems:

[0005] 1. Limitations of a single sensor: A single sensor cannot comprehensively reflect dynamic working conditions (such as changes in road surface friction coefficient, nonlinear characteristics of tires), and the information obtained is limited, resulting in a significant decrease in estimation accuracy under low-adhesion road surfaces;

[0006] 2. Insufficient model robustness: A mathematical model with fixed parameters is difficult to adapt to complex working conditions and changing driving modes (such as high-speed steering, low-adhesion road surfaces, load changes);

[0007] 3. Noise Sensitivity: Sensor noise and time delay cause the accumulation of estimation errors. Summary of the Invention

[0008] The purpose of the present invention is to solve the problems existing in the prior art, and a method for estimating the return torque of an automobile based on multi-sensor coupling is proposed.

[0009] To achieve the above purpose, the present invention adopts the following technical solutions:

[0010] A method for estimating the return torque of an automobile based on multi-sensor coupling includes the following steps:

[0011] S1. Collect data in different sensors, including steering wheel angle, wheel speed, longitudinal vehicle speed, IMU, radar, and camera;

[0012] S2. Preprocess the collected data, including limiting the steering wheel angle data, removing high-frequency noise in the wheel speed data, performing coordinate transformation on the IMU data, frequent state prediction, low-frequency updating of radar and camera data, and finally queuing and aligning the sensor data according to the time stamp;

[0013] S3. Couple the multi-sensor data, that is, extend the Kalman filter to couple the multi-sensor data. First, determine the system state vector and establish the system state equation and observation equation:

[0014] x k = f(x k-1 , u k ) + w k

[0015] Among them, the state vector x k includes position (x, y), vehicle speed v, yaw rate ω r , lateral acceleration a y ; the input δ is the front wheel angle; w k is the process noise;

[0016] z k = h(x k ) + v k

[0017] The observation vector z k is the multi-sensor raw data, z k = [a xm , a ym , ω rm , v xm , μ] T , v k is the observation noise, directly measure the front wheel angle and use it as the input of the dynamic model;

[0018] Then predict the state vector and error covariance matrix:

[0019] Using the state estimate of the previous moment and the state equation of the system, predict the state vector at the current moment; meanwhile, according to the dynamic model of the system and the statistical characteristics of the process noise, predict the error covariance matrix at the current moment:

[0020]

[0021] P ′ (k) = F(k - 1)P(k - 1)F T (k - 1) + Q(k - 1);

[0022] Jacobian matrix linearization: Perform a first-order Taylor expansion on the non-linear state equation and measurement equation to obtain the linearized equation;

[0023] Then, according to the observation equation and the measurement data at the current moment, calculate the Kalman gain and perform state update and covariance update:

[0024] Calculate the Kalman gain: K(k) = P ′ (k)H T (k)(H(k)P ′ (k)H T (k) + R(k)) -1

[0025] H k is the measurement matrix; R(k) is the measurement noise covariance matrix;

[0026] Update the state vector and error covariance matrix:

[0027] State update equation:

[0028] where h is the measurement equation;

[0029] Covariance update equation: P(k) = (I - K(k)H(k)P ′ (k));

[0030] Finally, perform an iterative loop, repeat the prediction and update steps, and continuously update the state estimate with new measurement data;

[0031] S4. Returning moment estimation model based on dynamic radial basis function neural network (RBFNN);

[0032] S5. Parameter adaptive module: Dynamically adjust the weight matrix W of RBFNN by recursive least squares (RLS).

[0033] Preferably, the RBFNN model structure:

[0034] Input layer: Coupled feature vector F v ;

[0035] Hidden layer: Number of nodes m = 10 (empirical value);

[0036] The activation function is the Gaussian kernel function. The number of nodes is initialized according to experience, and the output of each node is

[0037]

[0038] Output layer: Estimated value of the restoring moment

[0039] Initialization parameters:

[0040] Fixed center and width:

[0041] Center initialization: Cluster the input space of the samples using clustering, and take the cluster center as the initial center;

[0042] Width initialization: Set to the empirical value σ = 0.5;

[0043] Weights and RLS parameters:

[0044] Initial weight w(0) = 0;

[0045] Covariance matrix P j (0) = λ -1 I, where λ = 0.99 (forgetting factor).

[0046] Preferably, theoretical value calculation: Calculate based on the two-degree-of-freedom vehicle dynamics model

[0047] RBFNN forward propagation: Calculate the hidden layer output φ j (t), predict the restoring moment M z (t);

[0048] RLS weight update:

[0049] Calculate the error:

[0050] Calculate the gain vector

[0051] Update the weight matrix w j (t) = w j (t - 1) + K j (t)e j (t);

[0052] Update the covariance matrix

[0053]

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] 1. By coupling data from multiple sensors, the accuracy of the return torque estimation under complex working conditions is improved;

[0056] 2. The combined algorithm of the extended Kalman filter and the radial basis function neural network realizes high-precision modeling of nonlinear systems, and the response time is reduced;

[0057] 3. Dynamic parameter adjustment reduces the response delay in high-speed emergency lane-changing scenarios;

[0058] 4. The extended Kalman filter reduces noise interference and the estimation error is reduced. Description of the Drawings

[0059] Figure 1 It is a schematic structural diagram of the system architecture of a method for estimating the return torque of an automobile based on multi-sensor coupling proposed by the present invention;

[0060] Figure 2 It is a flow chart of EKF multi-sensor coupling;

[0061] Figure 3 It is a schematic diagram of the RBFNN model structure and parameter adaptation. Specific Embodiments

[0062] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0063] Refer to Figures 1 - 3 , a method for estimating the return torque of an automobile based on multi-sensor coupling, characterized by comprising the following steps:

[0064] S1. Collect data in different sensors. The data includes the steering wheel angle, wheel speed, longitudinal vehicle speed, IMU, radar, and camera. The above-mentioned steering wheel angle sensor is installed at the bottom of the steering column to collect the steering wheel angle and convert it into an electrical signal and transmit it to the data processing unit. The wheel speed sensors are respectively installed at the wheels of the vehicle to collect the wheel speed and longitudinal vehicle speed. The IMU is installed at the vehicle center of gravity position and includes a gyroscope and an accelerometer, which are respectively used to collect the angular velocity and acceleration. The millimeter-wave radar and the camera are respectively installed inside the front bumper of the vehicle and above the windshield, and the two are fused to calculate the road surface adhesion coefficient.

[0065] S2. Preprocess the collected data, including limiting the steering wheel angle data to prevent unstable subsequent calculations caused by abnormally large angle inputs, using a filtering algorithm to remove high-frequency noise from the wheel speed sensor data, performing standardization processing such as coordinate transformation on the IMU data, handling the data frequencies of different sensors. The high-frequency IMU data requires frequent state prediction, while the radar and camera data are updated at a lower frequency. Predict when the IMU data arrives and update when other sensor data arrives. Finally, queue and align the sensor data according to the timestamps.

[0066] S3. Couple the multi-sensor data, that is, use the extended Kalman filter to couple the multi-sensor data. First, determine the system state vector x = [x, y, v x , ω r , a y T , establish the state transition equation, adopt a simplified bicycle model, and establish a non-linear state transition equation

[0067] x k = f(x k-1 , u k ) + w k

[0068] where f(·) is the non-linear state transition function, which can be obtained by discretizing the vehicle dynamics equation.

[0069] where the state vector x k includes the position (x, y), vehicle speed v, yaw rate ω r , lateral acceleration a y ; the input δ is the front wheel angle; w k is the process noise;

[0070] z k = h(x k ) + v k

[0071] The observation vector z k is the multi-sensor raw data, z k = [a xm , a ym , ω rm , v xm , μ] T , v k is the observation noise, directly measure the front wheel angle and use it as the input of the dynamics model;

[0072] Then predict the state vector and the error covariance matrix:

[0073] ​Predict the state vector at the current time using the state estimate value at the previous time and the state equation of the system; meanwhile, predict the error covariance matrix at the current time according to the dynamic model of the system and the statistical characteristics of the process noise:

[0074]

[0075] P ′ (k) = F(k - 1)P(k - 1)F T (k - 1) + Q(k - 1);

[0076] Jacobian matrix linearization: Calculate the Jacobian matrix: Perform a first-order Taylor expansion on the non-linear state equation and measurement equation to obtain the linearized equations;

[0077] Then, according to the observation equation and the measurement data at the current time, calculate the Kalman gain and perform state update and covariance update:

[0078] Calculate the Kalman gain: K(k) = P ′ (k)H T (k)(H(k)P ′ (k)H T (k) + R(k)) -1

[0079] H k is the measurement matrix; R(k) is the measurement noise covariance matrix;

[0080] Update the state vector and error covariance matrix:

[0081] State update equation:

[0082] where h is the measurement equation;

[0083] Covariance update equation: P(k) = (I - K(k)H(k)P ′ (k));

[0084] Finally, perform an iterative loop, repeat the prediction and update steps, and continuously update the state estimate with new measurement data.

[0085] Perform the above prediction and update steps on the data of multiple sensors respectively to obtain the state estimate value and covariance matrix of each sensor, and construct a coupling matrix; then multiply the eigenvectors of each sensor after extended Kalman filter processing by the corresponding weights to obtain the weighted eigenvectors; use the weighted average method to fuse the weighted eigenvectors to obtain the final coupled eigenvector.

[0086] S4. Returning moment estimation model based on dynamic Radial Basis Function Neural Network (RBFNN). RBFNN model structure:

[0087] Input layer: Coupled feature vector F v ;

[0088] Hidden layer: Number of nodes m = 10 (empirical value);

[0089] Activation function is Gaussian kernel function. The number of nodes is initialized according to experience. The output of each node is

[0090]

[0091] Output layer: Estimated value of returning moment

[0092] Initialization parameters:

[0093] Fixed center and width:

[0094] Center initialization: Cluster the input space of samples by clustering, and take the cluster center as the initial center;

[0095] Width initialization: Set it to the empirical value σ = 0.5;

[0096] Weights and RLS parameters:

[0097] Initial weight w(0) = 0;

[0098] Covariance matrix P j (0) = λ -1 I, where λ = 0.99 (forgetting factor).

[0099] S5. Parameter adaptive module: Dynamically adjust the weight matrix W of RBFNN by Recursive Least Squares (RLS). Theoretical value calculation: Calculate based on the two-degree-of-freedom vehicle dynamics model

[0100] RBFNN forward propagation: Calculate the hidden layer output φ j (t), predict the returning moment M z (t);

[0101] RLS weight update:

[0102] Calculate error:

[0103] Calculate gain vector

[0104] Update weight matrix w j (t) = w j (t - 1) + K j (t)ej (t);

[0105] Update covariance matrix

[0106]

[0107] In summary: By coupling the data of multiple sensors, each sensor can play its role in its proficient field, avoiding waste of resources. Each sensor will generate certain errors during the measurement process. By using the extended Kalman filter algorithm to couple and process the multi-sensor data, these errors can be analyzed and corrected, reducing the accumulation of the overall system error caused by the errors of a single sensor and improving the reliability and accuracy of the data.

[0108] The extended Kalman filter algorithm can effectively fuse and filter multi-source data while considering system noise and measurement noise, improving the reliability and accuracy of the data. By weighting the data of different sensors, important information can be highlighted and noise interference can be suppressed, thus obtaining more accurate fused data. As a non-linear filtering method, the extended Kalman filter algorithm linearizes the non-linear function, transforms the non-linear problem into a linear problem, and then uses the idea of Kalman filtering for state estimation, which can more accurately describe the state change of the system and improve the accuracy of state estimation.

[0109] Compared with the traditional return torque estimation method, the return torque estimation method based on multi-sensor coupling does not rely on a single sensor, couples different information to improve the estimation accuracy; uses complex algorithms to process and couple sensor data, enabling it to have a faster response ability; has higher dynamic adaptability, can adjust the estimation result in real time, and improve the system stability.

[0110] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for estimating the vehicle self-aligning torque based on multi-sensor coupling, characterized in that: The steps include: S1. Collect data from different sensors, including steering wheel angle, wheel speed, longitudinal speed, IMU, radar and camera; S2, preprocessing the collected data, including limiting the steering wheel angle data, removing high-frequency noise in the wheel speed data, coordinate conversion of IMU data, frequent state prediction, low-frequency update of radar and camera data, and finally queuing and aligning the sensor data according to the timestamp; S3, coupling the multi-sensor data, that is, extending the Kalman filter to couple the multi-sensor data, first determine the system state vector, and establish the system state equation and observation equation: x k =f(x k-1 ,u k )+w k Among them, the state vector x k Including position (x, y), vehicle speed v, yaw rate ω r , lateral acceleration a y ; Input δ is the front wheel steering angle; w k is the process noise; z k =h(x k )+v k Observation vector z k is the raw data of multiple sensors, z k =[a xm ,a ym ,ω rm ,v xm ,μ] T , v k To observe the noise, the front wheel angle is directly measured as the input of the dynamic model; Then predict the state vector and error covariance matrix: The state vector at the current moment is predicted using the state estimate at the previous moment and the state equation of the system. At the same time, the error covariance matrix at the current moment is predicted based on the dynamic model of the system and the statistical characteristics of the process noise: P ′ (k)=F(k-1)P(k-1)F T (k-1)+Q(k-1); Jacobian Matrix Linearization: Compute the Jacobian matrix: Perform first-order Taylor expansion on the nonlinear state equation and measurement equation to obtain the linearized equation; Then, according to the observation equation and the measurement data at the current moment, the Kalman gain is calculated to perform state update and covariance update: Calculate the Kalman gain: K(k) = P ′ (k)H T (k)(H(k)P ′ (k)H T (k)+R(k)) -1 H k is the measurement matrix; R(k) is the measurement noise covariance matrix; Update the state vector and error covariance matrix: State update equation: Where h is the measurement equation; Covariance update equation: P(k) = (IK(k)H(k)P ′ (k)); Finally, the iterative loop repeats the prediction and update steps, continuously updating the state estimate with new measurement data; S4, a return moment estimation model based on a dynamic radial basis function neural network (RBFNN); S5, parameter adaptation module: dynamically adjust the weight matrix W of RBFNN through recursive least squares (RLS).

2. The method for estimating the vehicle aligning torque based on multi-sensor coupling according to claim 1, characterized in that: RBFNN model structure: Input layer: coupled feature vector F v ; Hidden layer: number of nodes m = 10 (empirical value); The activation function is the Gaussian kernel function, the number of nodes is initialized based on experience, and the output of each node is Output layer: Estimated value of the positive moment Initialization parameters: Fixed center and width: Center initialization: cluster the input space of the sample using clustering, and take the cluster center as the initial center; Width initialization: set to the empirical value σ = 0.5; Weights and RLS parameters: Initial weight w(0) = 0; Covariance matrix P j (0) = λ -1 I, where λ = 0.99 (forgetting factor).

3. The method for estimating the vehicle aligning torque based on multi-sensor coupling according to claim 2, characterized in that: Theoretical value calculation: Calculated based on a two-degree-of-freedom vehicle dynamics model RBFNN forward propagation: Calculate the hidden layer output φ j (t), predicted return torque M z (t); RLS weight update: Calculation error: Calculate the gain vector Update the weight matrix w j (t) = w j (t-1)+K j (t)e j (t); Update the covariance matrix

Images