KR0162671B1 - Vehicle slip angle measuring method and a device therefor - Google Patents

Abstract

The present invention provides a vehicle center slip angle measuring device that can be applied to the accurate measurement and the vehicle by the use of the neural network 32 that has been learned, and can also reduce the burden of the neural network. In the configuration, the body center slip angle measuring device includes a handle angle sensor 16, a vehicle speed sensor 18, a front and rear acceleration sensor 20, a horizontal acceleration sensor 22, and a yaw angle sensor 24. An approximate equation derived from a linear two degree of freedom vehicle model, and an approximation value A of the vehicle center slip angle based on the approximation equation from only the detection values of the sensors 16, 18, and 24. The approximate calculation block 26 to calculate, the preprocessing block 28 and memory 30 which preprocess the detected value of a sensor group, and generate input information, and input the said input information and the approximation value, Correction value corresponding to error between slip angle and the above approximation An adder for outputting a value obtained by adding the correction value and the approximation value output from the neural network 32 that has completed the learning and the neural network 32 that has completed the learning, to the approximate calculation block 26. It is characterized by including 36.

Description

Body center slip angle measuring device and body slip angle measuring method

1 is a view for explaining the body center slip angle.

2 is a schematic diagram showing a four-wheel steering vehicle incorporating a body center slip angle measuring device.

FIG. 3 is a block diagram of a portion of the controller of FIG. 2 that estimates the vehicle center slip angle. FIG.

4 is a graph showing a linear two degree of freedom vehicle model.

5 is a graph showing a steering input in a step shape.

6 is a schematic diagram showing the configuration of a neural network.

7 is a block diagram showing the learning system of the neural network of FIG.

FIG. 8 is a flow chart showing the calculation routine of the body center slip angle executed by the block diagram of FIG.

FIG. 9 is a graph showing the relationship between the calculated body center slip angle, actual body center slip angle, and an approximation for front wheel steering. FIG.

* Explanation of symbols for the main parts of the drawings

14 controller 16 handle angle sensor

18: vehicle speed sensor 20: front and rear acceleration sensor

22: horizontal acceleration sensor 24: yaw angle speed sensor

26: Approximate calculation block (computation means) 28: Preprocessing block (preprocessing means)

30 memory (preprocessing means) 32: neural network

36: addition unit (additional means)

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a body center slip angle measuring device suitable for a vehicle having a four-wheel steering system.

Since a vehicle that has a four-wheel steering system can also steer the rear wheel according to the steering of the front wheel, optimal control of the rear wheel steering can improve turning performance and steering stability performance compared to a normal vehicle.

However, in order to optimally perform the rear wheel steering control, it is necessary to accurately grasp various states indicating the behavior of the vehicle at the time of turning the vehicle, and the body center slip angle, which is one of the state quantities, performs the rear wheel steering, It becomes especially important. Here, the body-center slip angle is defined as an angle formed by the traveling direction of the vehicle and the direction of the vehicle body when turning the vehicle, as shown in FIG.

If the body center slip angle can be detected, the rear wheel steering can be controlled such that the body center slip angle becomes zero when the vehicle is turning, and as a result, the phase difference between the yaw angle velocity of the body and the transverse acceleration with respect to the steering Can theoretically be zero.

For the detection of the vehicle body center slip angle, for example, an optical non-contact ground speedometer can be used.

However, since this earth speedometer is optical, it is easy to be affected by noise depending on the road surface condition, and considering its size and price, it is incorporated in a motion control system of commercial vehicles to stabilize high accuracy detection. There is a problem that the work is actually difficult.

The present invention has been made on the basis of the above-described circumstances, and an object thereof is to provide a vehicle-centered slip angle measuring apparatus which can reduce the burden of a trained neural network and enable high-precision detection. .

The body-center slip angle measuring apparatus of the present invention includes first detection means for detecting a steering state of a front wheel of a vehicle, second detection means for detecting various movement states of the vehicle, and first and second detection means. Calculating means for calculating an approximation of the vehicle body slip angle from the front wheel steering information and the motion information detected by the front wheel steering information and the motion information detected by the first and second detecting means, and the calculated means calculated by the calculating means. When the approximation value is input and the deviation between the actual body center slip angle and the approximation value is learned as teacher data, and the approximation value is input in addition to the front wheel steering information and the exercise information, the correction value corresponding to the deviation value is inputted. A neural network that has finished learning to output a value, and an approximation value output from the computing means and a correction value output from the neural network which has completed learning. And an adding means for outputting the vehicle body center slip angle.

According to the above-described body center slip angle measuring device, when steering information and motion information of the vehicle are obtained by the first and second detection means, the calculation means calculates an approximation value of the body center slip angle from the steering information and the motion information. do.

On the other hand, the steering information and the motion information obtained by the first and second detection means are given as inputs to the completed neural network together with the approximation obtained by the calculation means. After completing this learning, the neural network is trained as teacher data on the deviation between the actual vehicle body slip angle and the approximation value. Therefore, the correction value corresponding to the deviation is output from the neural network. The approximation value is added by the adding means, and this addition value is output as the vehicle body center slip angle.

DETAILED DESCRIPTION Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Referring to FIG. 2, a four-wheel steering vehicle (4WS vehicle) equipped with a vehicle-centered slip angle measuring apparatus according to an embodiment of the present invention is schematically shown. The 4WS vehicle includes a power cylinder 2, a rack and a pinion mechanism, a front wheel steering system 6 for steering the front wheels FW _L and FW _R in response to an operation of the handle 4, and a rear wheel actuator composed of a hydraulic cylinder ( 8), the rear wheel steering system 10 for steering the rear wheels RW _L and RW _R is provided.

Regarding the rear wheel steering system 10, in more detail, the rear wheel actuator 8 is provided with an electromagnetic valve 12, which receives the control signal from the controller 14 and opens it. The valve direction and the opening degree are controlled, and as a result, the rear wheels RW _L and RW _R are steered through the rear wheel actuator 8.

The controller 14 is connected to various sensors used in the vehicle body center slip angle measuring device, and these sensors are the handle angle sensor 16, the vehicle speed sensor 18, the front and rear acceleration sensors 20, and the horizontal acceleration sensor. And a yaw angle sensor 24.

The steering wheel angle sensor 16 detects the operation angle of the steering wheel 4, that is, the steering wheel angle θ _H , and the vehicle speed sensor 18 detects the speed of the vehicle and the vehicle speed V. FIG. Here, the vehicle speed sensor 18 may be a speed meter of a vehicle or, in the case where a wheel speed sensor is attached to each wheel, the average front wheel speed obtained by averaging the outputs of the wheel speed sensors of the front wheels FW _L and FW _R. The vehicle speed V may be the smaller of the average rear wheel speeds obtained by averaging the outputs of the wheel speed sensors of the rear wheels RW _L and RW _R.

The front and rear acceleration sensor 20, the horizontal acceleration sensor 22, and the yaw angle sensor 24 detect the front and rear acceleration G _Z of the vehicle body, the horizontal acceleration G _Y of the vehicle body, and the yaw angle velocity ψ of the vehicle body, respectively. Therefore, the controller 14 inputs the steering wheel angle θ _H , the vehicle speed V, the front and rear acceleration G _{Z, the} horizontal acceleration G _Y, and the yaw angular velocity ψ from each sensor.

The controller 14 includes a function of measuring the vehicle body center slip angle from the detection signals from the various sensors described above, and controlling steering of the rear wheels RW _L and RW _R based on the measurement results. The controller 14, as shown in Figure 3, approximates the calculation block 26, the preprocessing block 28, the memory 30, the neural network 32 and the feedback control block 34. Equipped.

In the approximate calculation block 26, detection signals from the steering wheel angle sensor 16, the vehicle speed sensor 18, and the yaw angle speed sensor 24, that is, the steering wheel angle θ _H , the vehicle speed V and the yaw angle speed ψ are respectively input. Based on these, the approximation calculation block 26 calculates an approximation value A of the vehicle body center slip angle beta.

In the approximate equation calculation block 26, an approximation equation is given for the calculation of the approximation value A, which is derived from the general linear two degree of freedom vehicle model of FIG. That is, the equation of motion of this vehicle model is described by the following two equations.

m: body mass, Kf, Kr: equivalent nose owner power of the front and rear wheels,

I: yaw moment of inertia, Lf: center distance between front axles,

Lr: rear axle center distance, δf: front wheel angle,

Now

In addition, as shown in FIG. 5, the linear response of the response when the step-shaped steering input is applied to the equation. The solution of the slip angle β is obtained, respectively, and from these two solutions, an approximation equation shown in the following equation can be obtained.

β _o : (β normal value obtained from vehicle speed V and front wheel steering angle δf) / yaw angle velocity ψ

K _o : Correction coefficient in transient state according to paragraph 1.

Therefore, when the vehicle speed V, the front wheel steering angle δf, and the yaw angle velocity ψ are given to the approximation equation 4, the approximate value A of the vehicle center slip angle β can be calculated. In addition, the front wheel steering angle delta f can be calculated based on an approximation equation from the steering wheel angle θ _H in the approximate calculation block 26.

Here, the approximate expression of .4 expression holds only under the condition that the specific steering pattern expressed by the expression (3) is given. For this reason, the approximation A does not consider the change of the steering pattern or the nonlinear characteristics of the tire, so that the approximation A obtained from the approximation equation produces a deviation with respect to the actual body center slip angle β, but this deviation is as follows. This is eliminated by the output of the neural network 32 described below.

That is, when the information from the various sensors is input via the memory 30 and the preprocessing block 28, the neural network 32 has a correction value C corresponding to the deviation based on these input information. It is supposed to be output.

The input information to the neural network 32 will be described in detail. After the steering wheel angle θ _H from the handle and the sensor 16 is converted into the front wheel steering angle δf by arithmetic processing in the preprocessing block 28, the memory ( 30) is stored as a time series. For the previous example, the memory 30 has, for the moment the current steering angle δf _O, t ₁ time (0.1 seconds) prior to the current steering angle δf _1, t ₂ time (0.2 seconds) prior to the current steering angle δf _2, t3 time (0.3 seconds) The time series data of the current steering angle δf ₄ before the current steering angle δf ₃ and t ₄ hours (0.4 seconds) is stored as a parameter of the vehicle, and then this time series data is input to the neural network 32.

On the other hand, the vehicle speed V from the vehicle speed sensor 18, the front and rear acceleration sensor 20, the horizontal acceleration sensor 22 and the yaw angle sensor 24, the front and rear acceleration G _Z horizontal acceleration G _Y and the yaw angle velocity ψ are preprocessed blocks ( 28), and this preprocessing block 28, as another vehicle speed parameter, is a transverse acceleration-vehicle speed x yaw angular velocity (= G _Y- V x ψ), vehicle speed x yaw angular velocity (= V x ψ), yaw After calculating and calculating the angular velocity / vehicle velocity (= ψ / V), yaw angle acceleration (= ψ _A ), and front / rear acceleration × yaw angular velocity (= G _Z × ψ), each calculation result is input to the neural network 32. do.

Here, the input information from the preprocessing blur 28 to the neural network 32 is determined in consideration of the following equations obtained by modifying the above ① and ② expressions, respectively.

From the above equations (5) and (6), it can be seen that the body center slip angle beta is described as a linear sum of the state quantities indicated by the underline.

In the approximation formula, the coefficient of the state amount is an integer, but in the actual difference, it can be regarded as a function of tire non-linearity, load starting, and the like. If the state quantity is taken as input information to the neural network 32, the neural network 32 may approximate an error vector between the coefficient vector of the approximation equation and the coefficient vector of the actual difference. The structure of the neural network 32 includes a coordinate system transformation from the vehicle speed V in which the sign is reversed in the front-back direction to the body center slip angle β in which the sign is reversed in the left-right direction, and the input of the neural network 32 is inputted. This can simplify the function from the output to the output.

In the above state amount, the front wheel steering angle δf is input to the vehicle, and thus is given to the neural network 32 as a time series pattern as described above.

Regarding the state amount of V × (dβ / dt) in the equation (5), assuming that the body center slip angle β is sufficiently small, the lateral acceleration G _Y is written as V × (dβ / dt + ψ),

Therefore, the state amount of V × (dβ / dt) is

Can be displayed as

In addition to the state amount of G _Z x ψ with respect to the forward and backward acceleration G _Z , which is not considered in the linear two degree of freedom model, an approximation A is given as input information to the neural network 32.

The neural network 32 has a hierarchical structure as shown in FIG. 6, the input layer having an input unit IU for each input information, an intermediate layer having some intermediate unit MUs, and a correction value. It consists of an output layer with one output unit OU that outputs C.

The strength of the coupling between units in the neural network 32 is learned in advance by the method of back propagation, and this learning is performed using the learning system shown in FIG.

In addition to the functional blocks and sensor group of the controller 14 shown in FIG. 3, the learning system includes a vector tachometer 38 and a vector lateral tachometer 40. These tachometers 38, 40 ) Is connected to the vehicle body center slip angle calculation block 42. The body center slip angle calculation block 42 calculates the actual body center slip angle β _A based on a known formula from the outputs of the vector front and rear speedometer 38 and the vector horizontal accelerometer, that is, the front and rear acceleration and the horizontal acceleration. Calculate.

The actual vehicle center slip angle β _A , which is the output of the vehicle center slip angle calculation block 42, and the approximate value A from the approximation addition calculation block 26 are input to the subtraction section 44, and the subtraction section 44 is the actual subtraction section 44. The deviation D between the vehicle body center slip angle beta _A and the approximate value A is output as teacher data.

Then, the deviation D from the subtraction section 44 and the correction value C from the neural network 32 are input to the subtraction section 46, and the subtraction section 46 is an error between these deviations E and the correction value C. The error E is returned to the neural network 32.

The learning rule of backpropagation using Sigmoid function is, as is known, the strength (coupling amount) W of the coupling between units in the neural network 32 such that the error E is minimized. To learn.

More specifically, in the neural network 32, first, an output value X (X ₁ , X _m , X _o ) of each unit IU, MU, OU of each layer is calculated. These output values X accumulate the product of the input to each unit and the coupling amount in each layer, and are calculated | required as a value of the sigmoid function which made the accumulated value the parameter.

The error E ₀ of the output unit OU is based on the deviation D, the output value X _O (= correction value C) of the output unit OU, and the derivative value F _o of the sigmoid function indicating the input / output relationship of the output unit OU. This is calculated from the following equation.

In addition, the error E _M of each intermediate unit MU in the intermediate layer includes the differential value F _M of the sigmoid function indicating the input / output relationship of each intermediate unit MU, the error E _O of the output unit OU, and the respective intermediate unit MU and the output. Based on the coupling amount W _OM with the unit OU, it is calculated from the following equation.

For the error E ₁ of each input unit IU of the input layer, the derivative value F _l of the sigmoid function indicating the input relation of each input unit IU, the error E _M of each intermediate unit MU, and each intermediate unit MU Based on the coupling amount W _MI between the input unit IU and each input unit IU, it is calculated from the following equation.

Next, the learning constant n, the error E (E ₁ , E _M , E _O ) of each unit (IU, MU, OU) of each layer and the output value X (of each unit (IU, MU, OU) of each layer. Based on X ₁ , X _M , X _O ), the correction amount ΔW (ΔW _ID ,) of the combined amount W (W _ID , W _MI , W _OM ) with the input data of each unit (IU, MU, OU) of each layer. ΔW _MI , ΔW _OM ) are calculated by the following equation. In addition, the coupling amount W _ID indicates the coupling amount with the input data of the input unit IU.

ΔW = n × E × X

The calculated correction amount ΔW is added to the coupling amount W (W _ID , W _MI , W _OM ) with the historical data of each unit, so that an updated coupling amount W (= W + ΔW) can be obtained.

The above-described learning is repeatedly performed based on a large number of teacher data in consideration of various driving conditions, road surface conditions, and front wheel steering patterns.

In addition, learning is carried out by a driving test of a vehicle equipped with the system of FIG. 7 or by simulating and running a nonlinear six degree of freedom vehicle model on a large calculator to collect training data, and using the collected training data. You can also do it.

Therefore, the correction value C outputted from the neural network 32 that has been learned becomes a value corresponding to the deviation D of the approximation A with respect to the vehicle body center slip angle β, and thus, as shown in FIG. When the correction value C output from the neural network 32 and the approximate value A output from the approximation calculation block 26 are added by the adder 36, the addition value corresponds to the actual vehicle body slip angle β _A. The center slip angle β is obtained.

The estimated body center slip angle beta is input to the feed back control block 34 from the adder 36, and the feedback control block 34 is between the target value (zero) and the body center slip angle beta. On the basis of the deviation of, the control signal is output toward the solenoid valve 12 of the rear wheel actuator 8. As a result, the rear wheel actuator 8 causes the rear wheel RW _L so that the vehicle body center slip angle beta is zero. , Control the steering angle of RW _R.

The measurement procedure of the vehicle body center slip angle mentioned above can be displayed by the measurement routine of FIG.

Measurement routine

First, stored in if at step SI, the controller 14 handles each θ _H that is the front wheel steering angle δf is input to, 0 is set to the value of the counter i (step S2), the input front wheel steering angle δf is the memory 30 That is, it is saved (step S3). The front wheel steering angle δf is calculated by calculating from the steering wheel angle θ _H in the preprocessing block 28 as described above.

In the next step S4, after waiting for 0.1 second, the value i of the counter is increased by 1 (step S5), and it is determined whether the value i of the counter is less than 4 (step S6).

If the determination result of step S6 is YES, the step after step S1 is repeated, and if it is determined that it is up (No), it will progress to step S7. Therefore, when the determination result of step S6 becomes No, time series data regarding front wheel steering angle delta f, ie, delta f0, delta f1, delta f2, delta f3, delta f4, is saved in the memory 30.

In step S7, the vehicle speed V from the sensors 18, 20, 22, and 24, the front and rear speed _{GZ, the} acceleration speed G _Y and the yaw angle speed ψ are respectively input to the controller 14, From these inputs, after approximation A is calculated based on an approximation equation (step S8), the above-described G _Y -V × ψ, V × ψ, ψ / V, ψ _A, G _{Z ×} ψ is calculated by preprocessing. (Step S9), these values and the time series data are provided as input information to the trained neural network 32, and the neural network 32 estimates and corrects the correction value C (step S10). ), And the approximate value A and the correction value C are added to estimate the vehicle body center slip angle beta (step S11).

The estimated vehicle body center slip angle beta is output to the feedback control block 34 (step S12), and the following step S13 determines the end of the processing of this routine. That is, in this step, for example, it is determined whether or not the steering wheel 4 of the vehicle is being operated, and if the operation is in progress, the routine is repeatedly executed, and when the steering wheel 4 returns to the neutral position, the routine is processed. Quit.

Referring to FIG. 9, the estimated body center slip angle β and the actual body center slip angle β _A in addition to the changes in yaw angle ψ and lateral acceleration G _Y when the front wheel steering angle δf changes as shown in the figure. Although the relationship with is shown, these body center slip angles β and the actual body center slip angles β _A almost coincide with each other. As a result, according to the measuring device of the present invention, it is possible to accurately estimate the body center slip angles β. Able to know. In Fig. 9, the dashed dashed line indicates the change in output of the approximated value A.

As described above, according to the body center slip angle measuring apparatus of the present invention, steering information and motion information of the vehicle obtained by the first and second detection means are given to the calculation means, thereby obtaining an approximation of the body center slip angle. The approximation value is input to the neural network that has been learned along with the steering information and the exercise information. Since the neural network that has completed the learning is trained as the teacher data on the deviation between the actual body center slip angle and the approximation value, the correction value output from the Martin neural net for learning corresponds to the above deviation. The value obtained by adding the correction value to the approximation becomes the correct body center slip angle. By using the neural network that has been learned in this way, it is possible to measure the body center slip angle without the need of the existing optical earth speedometer, so that the vehicle can be mounted on top of the vehicle. Since the correction value of the approximate value is used as the output instead of the body center slip angle itself, the dynamic range of the output is reduced, and the burden of the neural network is also reduced.

Claims (6)

First detecting means 16 for detecting the steering state of the front wheels of the vehicle and outputting front wheel steering information; second detecting means 18, 20, 22, 24 for detecting the movement state of the vehicle and outputting the movement information; ; Calculating means (26) for calculating an approximation of the center slip angle of the vehicle body based on front wheel steering information output from the first detection means and motion information including at least the vehicle speed and yaw angle velocity output from the second detection means; Deviation in the approximate value of the vehicle slip angle based on the front wheel steering information output from the first detecting means, the motion information including at least the vehicle speed and yaw angle velocity output from the second detecting means, and an approximation calculated by the calculating means. A trained neural network 32 for estimating a correction value corresponding to the learning; And an addition means 36 for outputting the value calculated by correcting the approximation value calculated by the calculating means using the correction value estimated by the neural network that has completed the learning as the body center slip angle. Center slip angle measuring device. The said calculation means considers a vehicle as a linear hydraulic model, is derived from the motion system of this hydraulic model, and it calculates the said approximation only from a vehicle speed, a front wheel steering angle, and the yaw angle speed of a vehicle. A body center slip angle measuring device, comprising: an approximation equation that enables calculation. The preprocessing according to claim 1, wherein said computing means processes the front wheel steering information and the motion information before giving the front wheel steering information and the motion information detected by the first and second detection means to the neural network. An apparatus for measuring body center slip angle, comprising: a means. The vehicle detecting apparatus according to claim 2 or 3, wherein the first detecting means includes a steering wheel angle sensor for detecting a steering wheel angle of the vehicle, and the second detecting means includes a vehicle speed sensor for detecting a vehicle speed and a yaw of the vehicle. A yaw angular velocity sensor for detecting an angular velocity, a front and rear acceleration sensor for detecting front and rear acceleration of a vehicle, and a horizontal acceleration sensor for detecting lateral acceleration of the vehicle, the apparatus for measuring a vehicle center slip angle. 5. The apparatus according to claim 4, wherein the preprocessing means comprises: storage means for storing the front wheel steering angle of the past as time series data at the present time calculated based on the output of the handle angle sensor, and detecting each sensor of the second detection means; Value, the calculation means for calculating the transverse acceleration-vehicle speed x yaw angle speed, vehicle speed x yaw angle speed, yaw rate / vehicle speed, yaw angle acceleration, front-back acceleration, yaw rate, and calculated by the time series data of the storage means and the calculation means Body center slip angle measuring device, characterized in that to give each data to the neural net. (a) detecting the steering state of the front wheels of the vehicle and outputting front wheel steering information; (b) detecting an exercise state of the vehicle and outputting exercise information; (c) calculating an approximation of the body center slip angle based on the front wheel steering information output in step (a) and the motion information including at least the vehicle speed and yaw angle speed output in step (b); (d) In order to estimate the correction value corresponding to the deviation of this approximation, the trained neural network receives the front wheel steering information output in step (a) and at least the vehicle speed, yaw angle speed and horizontal acceleration output in step (b). Imparting exercise information and an approximation calculated in step (c); (e) outputting the value calculated by correcting the approximation value calculated in step (c) using the correction value estimated in step (d) as the vehicle body slip angle; Body center slip angle measuring device, characterized in that made.