DE102011104270A1 - Method for controlling driving dynamics of e.g. passenger car, involves evaluating stability of current dynamics state of vehicle, and performing stabilization measurement when evaluated stability is not sufficient - Google Patents

Abstract

The method involves dividing stationary parameters correlating with a steady state of the vehicle and intermittent parameters correlating with intermittent processes of the vehicle depending on variance of associated actual values. Nominal and actual values for current driving dynamics state of the vehicle are determined based on the stationary and intermittent parameters. Stability of the current vehicle dynamics state is determined by comparing the nominal and actual values. The stability is evaluated, and a stabilization measurement is performed when the stability is not sufficient.

Description

The present invention relates to a method for driving dynamics control of a road vehicle, in particular a utility vehicle, preferably a team.

In the recent past, vehicle control systems have been developed that recognize vehicle-dynamic unstable states at an early stage and stabilize the vehicle through targeted braking intervention on individual wheels. The basic principle is the following in 1 illustrated: Movement quantities of the vehicle are measured on the one hand, on the other hand calculated from the also measured steering angle course using suitable algorithms - as setpoints for the control. The calculated and measured quantities are now compared. A significant deviation is detected as instability and the control system is activated. Unilateral braking stabilizes the vehicle, as in 2 shown. It is exploited the property that the cornering ability of a braked wheel decreases, ie when oversteer is mainly the outside front wheel, understeer braked the inside rear wheel.

In the practical implementation of setpoint generation different ways are contested. From the article " Alberti, V .; Babble, E .; Drumm; A .; Driving stability control by active braking of individual wheels, Automatisierungstechnik 44 (1996) 5 ", Abbreviated hereafter to" Alberti ", and from the article" van Zanten, A .; Erhardt, R .; Pfaff, G .: FDR - The driving dynamics control of Bosch, Automobiltechnische Zeitschrift 96 (1994) 11 ", Hereinafter abbreviated to" van Zanten ", it is known to obtain the desired value of the yaw angular velocity from a stationary model. In Alberti, the difference between the setpoint and the actual value is filtered in triplicate in order to eliminate influences from vehicle parameter fluctuations, dynamics and measurement errors.

From the article " Müller, A .; Achenbach, W .; Schindler, E .; Wohland, Th .; Mohn, F.-W .: The new Electronic Stability Program Electronic Stability Program of Mercedes-Benz, Automobiltechnische Zeitschrift, 96 (1194) 11 , hereinafter abbreviated to "Müller", it is known to additionally take into account the lateral acceleration. In the article " Donges. e .; Naab, K .: Control Systems for Vehicle Guidance and Stabilization in Automotive Technology, Automatisierungstechnik 44 (1996) 5 ", Hereinafter abbreviated to" Donges ", refers to the need for an observer. This unit ensures, among other things, that the calculated and measured quantities do not "diverge" even with smaller deviations in the model parameters or due to disturbing influences.

Together to the o. G. Methods is that the calculation of the setpoints is done using fixed set constants, which are obtained from driver searches. For passenger cars, this approach is justified because the physical sizes of vehicles within a series vary little: the wheelbase is the same, the payload is compared to the tare weight relatively small and thus the center of gravity can be moved only within a narrow range. The relative uncertainty is caused by the tire properties: depending on the design, wear condition and internal pressure, the skew stiffness can vary widely.

For heavy commercial vehicles, the situation is different. The dimensions within a series can vary greatly: The wheelbase variants in current series are between 3.6 and 6.3 m. The differences between empty and loaded are also serious: a solo truck with a gross vehicle weight of 18 t weighs less than 7 tonnes empty, with a particularly large variation in the carrying capacity of the rear axle. Another important aspect is the treatment of vehicle trains: it is obvious that the behavior of a solo vehicle must differ from the behavior of a semitrailer already because of the increase in degrees of freedom. This fact is also reflected in the article " Hecker, F .; Jundt, O .; Leimbach, K.-D .; Faye, I .; Schramm, H .: Vehicle dynamics control system for semitrailer tractors, 6th specialist conference Tires - Chassis - Roadway Hannover 1007, VDI reports 1350 ", Hereinafter abbreviated to" Hecker "described.

Now a possibility is sought to determine for the practical application sufficiently accurate setpoint values of the vehicle movement. The objective here is to use a single configuration of software and hardware for all vehicle variants and thus to avoid programming at the end of the tape in production or a possible variety of parts. The already existing for the vehicle dynamics control sensor should be sufficient, namely in particular steering wheel angle and yaw rate and lateral acceleration (in semi-trailer or trailer operation on towing vehicle).

The present invention thus addresses the problem of providing an improved or at least another embodiment for the vehicle dynamics control of road vehicles especially suitable for commercial vehicles, which are offered in different variants. In particular, a simplified parameterization of the control units should be realized.

This problem is solved in the present invention, in particular by the subject of the independent claim. Advantageous embodiments are the subject of the dependent claims.

The invention is based, for example, on the general idea of dynamically and automatically carrying out the parameterization during the driving operation of the vehicle, in which the driving conditions are constantly detected by means of the existing sensor system and the reactions of the vehicle are monitored and evaluated for changes in state. This can be used to find the parameters contained in an existing computer model of the vehicle.

For example. Actual values for parameters that correlate with the driving dynamics of the vehicle can be recorded. The detected parameters can then be subdivided into stationary parameters which correlate with a stationary state of the vehicle and into non-stationary parameters which correlate with transient processes of the vehicle as a function of the variance of the associated actual values. A larger variance speaks for unsteady parameters, while a smaller variance speaks for stationary parameters. For example. For the variance, it is possible to predetermine a limit range below which the detected parameters are interpreted as stationary parameters above which the detected parameters are transient parameters.

Based on the stationary parameters, setpoint values for the current vehicle dynamics state of the vehicle can now be determined. Actual values for the current vehicle dynamics state of the vehicle can be determined on the basis of the transient parameters. Furthermore, a stability of the current vehicle dynamics state of the vehicle can be determined by a desired-actual comparison of the determined values for the current vehicle dynamics state. The determined stability can now be evaluated. If the determined stability is assessed as insufficient, a suitable stabilization measure can then be carried out automatically.

According to an advantageous embodiment, a high-pass filter can be used in the evaluation of the transient parameters. For example. Via the high-pass filter, stationary parameters can be separated from transient parameters.

According to another embodiment, it can be considered in the evaluation of the determined stability, whether the vehicle is still in a linear dynamic range. As long as the current driving dynamics state is within such a linear range, it can be assumed that the current driving dynamics state is sufficiently stable. However, as soon as the driving dynamics state leaves the linear range, the vehicle becomes unstable.

According to another expedient embodiment, the determination of the current stability with respect to the determined and evaluated actual values of the parameters can be performed recursively. As a result, the amount of data to be evaluated or stored can be significantly reduced, as a result of which the method can be implemented more simply within a corresponding control device. In principle, this recursive method can be split into old and new values.

In another variant of the method according to the invention, a function of "forgetting" can be realized. In particular, in determining the current stability, it is possible to weight the determined and evaluated actual values of the parameters with regard to their acquisition time. The actual values have expediently a decreasing weighting with increasing age. This reduces the influence of the actual values with their age. Through this functionality, the vehicle dynamics control of the vehicle is particularly adaptive and can adapt to time-varying boundary conditions.

Appropriately, a yaw rate of the vehicle can be determined as the driving dynamics state, ie the angular velocity of a rotation of the vehicle about a vertical vertical axis. The stability of the vehicle can be determined, for example, on the basis of a float angle or assessed on the basis of the float angle. The float angle is the angle between the direction of movement of the vehicle in the vehicle's center of gravity and the vehicle's longitudinal axis.

As a suitable stabilization measure, for example, a braking intervention on at least one vehicle wheel can be performed. This braking intervention is specifically asymmetrical, so that there is a stabilizing balancing moment, which acts in particular around the vertical vertical axis of the vehicle and should lead to a reduction of the slip angle.

In the following, the present invention will be explained in more detail by means of examples and in particular in conjunction with drawings.

In detail, the drawings show

1 the signal flow of the vehicle dynamics control,

2 the mode of operation of the vehicle dynamics control,

3 a block diagram for an ESP system,

4 a simplified flowchart for data sorting,

5 a simplified diagram for the division of the data into sections,

6 a simplified representation of a single-track model,

7 a diagram for illustrating a virtual steering game,

8th a simplified diagram illustrating a sensor arrangement for lateral acceleration measurement,

9 a simplified block diagram illustrating a hierarchy of identification blocks,

10 a simplified plan view of a test track "Dauerlaufkurs Süd",

11 Diagrams illustrating measurement data

12 Diagrams to illustrate a sorting of the measured data

13 Diagrams illustrating parameters for stationary behavior,

14 Diagrams illustrating yaw dynamics parameters,

15 Diagrams to illustrate parameters of lateral acceleration dynamics,

16 Diagrams to illustrate a simulation,

17 a simplified block diagram to illustrate a Simulink model of the system,

18 a diagram to illustrate the negative half of a complex number plane,

19 Diagrams showing different functionalities,

20 a block diagram to Luenberger.

The following is based on the 3 to 20 explains a preferred procedure for parameter recognition for an electronic stability program, ESP, under real conditions. The following description is structured to maintain clarity with headings.

Summary

The basic functionality of the electronic vehicle dynamics stabilization system is the stabilizing, wheel-selective brake intervention to prevent excessive oversteering or understeering of the vehicle. In order to detect an unstable driving condition, is in accordance with 3 compared the measured and a calculated yaw rate. The calculation of the quantities takes place with the aid of a simple calculation model on the basis of the current steering wheel angle position. For larger deviations, a targeted braking intervention is initiated.

Furthermore, the slip angle of the vehicle is determined with the help of an observer, with implausibly high values is also intervened. A suitable observer is z. As the Luenberger observers, which will be explained in more detail below.

In order to obtain mathematical values, vehicle-specific parameters must be available in addition to the calculation algorithm. If the vehicle characteristics are within narrow limits, the parameters may be obtained from experimental data and stored in the controllers. If the vehicle characteristics vary greatly, it is advisable to adjust the parameters automatically.

The following explains how parameter recognition is possible under operating conditions with sufficient accuracy, speed and stability. In the choice of methods and algorithms the practical feasibility is always considered. A consideration with regard to computer performance and memory requirements should be made in a further step.

In order to test the method, measurement data from vehicle dynamics measurements with a tractor, CAN records from the continuous running of various test vehicles and simulated data of a 6 × 2 vehicle were used. The results show that the parameters of the vehicles are detected so accurately after a relatively short time that a simulation running in parallel can reproduce the vehicle dynamic state variables even with small deviations. However, a more precise time specification makes little sense at this point because the process is not time-dependent but event-dependent.

The identification method

The method chosen here to identify parameters of a dynamic system is briefly presented: It is first necessary to describe a simple model of the system to be identified with a linear or linearized (differential) equation or a system of equations. The equation contains constant parameters and variables, which in the present case are quantities of the vehicle movement. The parameters of this equation are searched.

Under experimental conditions or in real operation, the variables are measured in rapid succession at discrete times. Since each measurement satisfies the equation, in principle a number of measurements corresponding to the number of unknown parameters would suffice. In reality, the measurements are superimposed by perturbations, inaccuracies and unmodeled effects. Therefore, the measurements are continuously made in large numbers. This results in an over-determined system of equations, which is also evaluated continuously. For this, the least squares method is used. This method yields the best approximation of the parameters as a compensation calculation and determines the variance of the entire equation system as well as the parameters. The identification equation for n parameters has the following form:

The vectorial notation in the second line should indicate that the scalar on the left side and the row vector contain the variable measures, while the column vector contains the constant parameters. By a multiplicity of k of the measurements the measuring vector b and the measuring matrix A arise:

The solution according to the method of least squares is: x = (A ^T * A) ^-1 * A ^T * b

Weighting column by column:

The measured variables can assume completely different orders of magnitude due to the physical units used. The result is, on the one hand, that numerical inaccuracies occur in the equation solution and, on the other hand, that the individual parameters are calculated with very different accuracy. One remedy is the possibility of weighting the individual variables according to their variance. This simply means that the measured quantities are converted into a different unit of measurement, which matches the size better. The results may need to be recalculated to the original unit.

In a non-recursive application, the variance of the measured quantities can be calculated and used exactly. In the recursive application, the weighting factors for the measured quantities must be determined before the calculation. They can be determined from preliminary examinations in practice.

Weighting line by line:

In order to largely exclude outliers, the individual measurements can be weighted. The recursive version checks how well the new measurement fits the previously calculated parameters. Starting from the equation u _{0, j} = x ₁ · u _{1, j} + x ₂ · u _{2, j} + ... + x _n · u _{n, j} the weighting of the jth measurement is formulated:

Now both sides, or all measurands are multiplied by the weighting factor: _{j j} u u _{, j} = x ₁ · g _j · u _{1, j} + x ₂ · g _j · u _{2, j} + ... + x _n · g _j · u _{n, j}

error control

To follow the quality of identification, the following evaluations can be made. The vector of errors is calculated as e = A · x - b

The mean square error is

The coefficient of determination R ² , assuming that the mean value of the measurement vector b is zero:

Finally, the variance of the parameters: var (x) = var (e) * diag (A ^T * A) ^-1

However, before the inversion of the matrix A ^T · A, which is essential for the calculation, is carried out, it should be checked whether this operation is feasible and the result has the necessary relevance. As an indicator the condition of the matrix is used, which is defined mathematically as the product of the norms of the original and the inverted matrix and can also be expressed as a quotient of the largest and smallest eigenvalues of this matrix:

The condition of the matrix is ideally one, otherwise larger; the limit of usability still needs to be studied in practical examples. The determination of the size is offered in the program package used. Furthermore, if the software itself is created, the function can be realized by means of a diagonal / triangle decomposition of the matrix. This can be realized, for example, as follows:

Equation calculation with a symmetric matrix

A symmetric, positive definite matrix can be decomposed as follows: U · D · U ^T = A

U is an upper triangular matrix with 1s on the main diagonal, D is a diagonal matrix. The multiplication (using the example of a 4 × 4 matrix);

wobei die Indizes von hinten nach vorn durchlaufen werden. Die Diagonalmatrix besteht aus den Eigenwerten der Matrix A, so dass die gesuchte Konditionszahl

ist. Die Lösung des Gleichungssystems geschieht folgendermaßen: A·x = U·D·U^T·x = y The calculation rule, in general:

with the indices going from back to front. The diagonal matrix consists of the eigenvalues of the matrix A, so that the sought condition number

is. The solution of the equation system is as follows: A · x = U · D · U · ^T x = y

There are two auxiliary vectors v and w defined when inserting backwards: U · (D · U ^T · x) = U · v = y D * (U ^T * x) = D * w = v U ^T · x = w Calculation rule:

Recursive version

In order to be able to evaluate the measurements continuously, but not have to cancel the individual measurement results, the "condensed" data can be refreshed. It suffices to store the matrix A ^T * A and the vector A ^T * b. The new matrix, or new vector, results when the new measurement is added as follows. (A ^T * A) _j = (A ^T * A) _j-1 + u T / j * u _j (A ^T * b) _j = (A ^T * b) _j-1 + u T / j * u _{0, j} With u _j = (u _{1, j} u _{2, j} ... u _{n, j} )

The mean square error can also be calculated recursively:

For this, only the scalar b ^T · b has to be stored and updated: (b ^T * b) _j = (b ^T * b) _j-1 + u 2/0, j

As the measuring time increases, the amounts of the matrix elements become larger and larger so that the maximum amount that can be represented in the memory is exceeded. To prevent this, the amounts of both the stored quantities and the new receipts can be multiplied by an ever decreasing number <1. Another possibility is to reduce only the old values. Thus, the newer measurements are weighted more, while the old ones are successively "forgotten". This second approach is particularly useful when increasing the time the criterion of selection is tightened, so only "good" measurements are stored.

Derivatives, filtering

For the evaluation of dynamic processes, it is necessary to use derivatives, even higher, measured time courses. In the interests of further usability, they must be free from spurious ripple, and synchronous with the original size as well as with each other. These requirements are fulfilled with the aid of a condition filter. A detailed description will be given below.

Low pass filter and 4th order differentiator with Butterworth characteristics

In the present analyzes derivations of measured data are required. However, these are superimposed with noise, so that a direct derivation leads to useless results. With higher derivatives, the problem is increasing.

In order to still obtain useful time courses of the derivatives, the state filter and differentiator is used. The basic idea is the fact that a low-pass filtered signal from the o. G. Interference is freed. The higher the filter order, the more practically useful derivatives can be formed.

A state model is set up, whereby the first state variable is the output variable - thus the filtered signal -, the further state variables are the derivatives in increasing order.

The formal state space representation is: x. = Ax + bu

Detailed in the regulatory normal form:

The transfer function of the matrix equation with the output variable y = x ₁ , as can be seen directly from the last line:

The matrix elements α ₄₁ to α ₄₄ must be designed so that the transfer function has the desired characteristic. The 4th order Butterworth filter has poles that lie on a semicircle to the zero point on the complex number plane in the negative half, symmetric to the real axis. The angles to the real axis are equally distributed, ie 22.5 and 67.5 degrees, the radius of the semicircle - ie the distance to the zero point - is the corner frequency of the filter ω ₀ . In the 18 The representation shown was normalized to the corner frequency.

The design is done in several steps. First, the denominator of the transfer function by means of the presettable eigenvalues of the matrix λ ₁ ... λ ₄ written in the following form: (s - λ ₁ ) (s - λ ₂ ) (s - λ ₃ ) (s - λ ₄ ) = s ⁴ - ((λ ₁ + λ ₂ ) + (λ ₃ + λ ₄ )) s ³ + (( λ ₁ λ ₂ ) + (λ ₁ + λ ₂ ) (λ ₃ + λ ₄ ) + (λ ₃ λ ₄ )) s ² - ((λ ₁ λ ₂ ) (λ ₃ + λ ₄ ) + (λ ₁ + λ ₂ ) (λ ₃ λ ₄ )) s + (λ ₁ λ ₂ ) (λ ₃ λ ₄ )

From the coefficient comparison arise directly α ₄₁ = - (λ ₁ λ ₂ ) (λ ₃ λ ₄ ) α ₄₂ = ((λ ₁ λ ₂ ) (λ ₃ + λ ₄ ) + (λ ₁ + λ ₂ ) (λ ₃ λ ₄ )) α ₄₃ = - ((λ ₁ λ ₂ ) + (λ ₁ + λ ₂ ) (λ ₃ + λ ₄ ) + (λ ₃ λ ₄ )) α ₄₄ = (λ ₁ + λ ₂ ) + (λ ₃ + λ ₄ ) b ₄ = -α41

Now the conjugate complex eigenvalue pairs are determined accordingly: Re (λ ₁ ) = Re (λ ₂ ) = ω ₀ · cos 22.5 ⁰ Im (λ ₁ ) = -Im (λ ₂ ) = ω ₀ · sin 22.5 ⁰ Re (λ ₃ ) = Re (λ ₄ ) = ω ₀ · cos 67.5 ⁰ Im (λ ₃ ) = -Im (λ ₄ ) = ω ₀ · sin67.5 ⁰

Used in the equations: α ₄₁ = - (λ ₁ λ ₂ ) (λ ₃ λ ₄ ) = -1 · ω 4/0 α ₄₂ = ((λ ₁ λ ₂ ) (λ ₃ + λ ₄ ) + (λ ₁ + λ ₂ ) (λ ₃ λ ₄ )) = -2,613 · ω 3/0 α ₄₃ - ((λ ₁ λ ₂ ) + (λ ₁ + λ ₂ ) (λ ₃ + λ ₄ ) + (λ ₃ λ ₄ )) = -3,414 · ω 2/0 α ₄₄ = (λ ₁ + λ ₂ ) + (λ ₃ + λ ₄ ) = -2,613 · ω ₀ b ₄ = -α ₄₁ = ω 4/0

The transfer function of the filter:

The frequency response of the filter describes the amplitude and phase relationship between the input and output signals when the input is harmonic. The quantitative determination is done by replacing the Laplace operator with jω in the transfer function.

The remaining steps are calculated numerically. The representation in diagram form takes place in 19 ,

The lower left diagram shows the Bode diagram, the upper left diagram shows the reinforcement in double logarithmic representation, and the right diagram shows the phase in simple logarithmic representation.

mit F = P^–1·Q und g = P^–1·b vereinfacht sich die Berechnung zu

The calculation of the filtered function and the derivatives is done by means of the bilinear transformation:

with F = P ^-1 · Q and g = P ^-1 · b, the calculation becomes easier

The matrices are:

The matrix P is inverted numerically before use.

Special features in the present application:

In real driving conditions also occur that are not useful for the parameter identification. First, it is the stoppage and the very slow ride, and the ride (almost) straight, but rarely also loss of stability, so intervention case for the scheme. The possible cases are described in the flow chart according to 4 shown.

As indicated above, different criteria must be applied to the separation of operating conditions. Speed and path radius can be determined by means of individual values. However, to be able to evaluate variance and correlation, an interval must be considered. This consideration then leads to the following procedure: According to 5 the measured values of a 2 s long section are buffered. The data will be as in the flowchart of 4 displayed, checked and then added to the stored data of the corresponding group, or discarded. Thereafter, the first stored data of the buffer are deleted and filled with new measurement data until the section of 2 s is complete. Now the process described is repeated.

The linear single-track model, transfer functions

• • Dynamisches Gleichgewicht in Querrichtung und um die Hochachse (aus Fahrzeugmasse m, Trägheitsmoment um die Hochachse J_z Querbeschleunigung α_y, Reifenseitenkräfte vorn und hinten FV und FH sowie Abstände der Vorder- und Hinterachse zum Schwerpunkt IV und IH): ma_y = F_y + F_H J_ZΨ .. = F_Vl_V – F_Hl_H
• • Kinematische Beziehungen zwischen Schwimmwinkel β, Gierwinkelgeschwindigkeit Ψ ., Lenkwinkel
  
  und Schräglaufwinkeln der Vorder- und Hinterachse α_V, α_H:
  
• • Reifen-Seitenkraft aus Schräglaufsteifigkeit und Schräglaufwinkel an Vorder- und Hinterachse: F_V = c_V·α_V F_H = c_H·α_H
• • Die Querbeschleunigung lässt sich aus Fahrgeschwindigkeit, Gierwinkelgeschwindigkeit sowie aus der Ableitung des Schwimmwinkels berechnen: α_y = ν(ψ . + β .) (x)

A simple, and for the present case sufficiently accurate model of vehicle lateral dynamics is the in 6 reproduced linear single-track model. The system of equations derives from the following individual equations:

• Dynamic equilibrium in the transverse direction and about the vertical axis (from vehicle mass m, moment of inertia about the vertical axis J _z lateral acceleration α _y , front and rear tire lateral forces FV and FH and distances of the front and rear axle to the center of gravity IV and IH): ma _y = F _y + F _H J _Z Ψ .. = F _V I _V - F _H I _H
• • Kinematic relationships between slip angle β, yaw rate Ψ., Steering angle
  
  and slip angles of the front and rear axles α _V , α _H :
  
• • Tire side force from skew stiffness and slip angle at front and rear axle: F _V = c _V · α _V F _H = c _H · α _H
• • The lateral acceleration can be calculated from the driving speed, yaw rate and the derivative of the slip angle: α _y = ν (ψ. + β.) (x)

The equations are ordered according to the system of differential equations. The coefficients contain the driving speed in different ways:

By eliminating the slip angle, or using equation (x), the transfer function can be expressed from the steering wheel angle to the yaw angular velocity or lateral acceleration. The dependence of the parameters on the velocity resolved, this is specified explicitly.

What remains is the stationary yaw gain G _stat , the velocity ν, as well as five constant coefficients, which in combination with the complex variables of the Laplace transform s, which indicates the degree of the derivative, describe the dynamic behavior.

The stationary yaw gain of the steering wheel angle results from the speed, the steering ratio, the wheelbase and the Eigenlenkgradienten. Since the physical quantities in the identification can not be separated, the parameters q ₁ and q _{2 are} used.

From this it is possible to express the identification equation for the parameters of the stationary yaw gain.

It can be seen that the steering angle during steady-state driving consists of a path-curvature proportional and a lateral acceleration-proportional component

The steady yaw gain is therefore a quantity that is non-linear over the travel speed. It follows that at least two speeds are required for identifiability, constant speed measurement procedures are not suitable for determining these parameters.

Once the stationary gain is known with sufficient accuracy, it will be used to calculate the dynamic quantities. The equation of identification results after conversion from equation (Y)

The physical contents of the individual parameters are:

While the meter constant is simply the float angle gradient of the vehicle, the tag constants are complexed together.

To identify the missing counter constants of the lateral acceleration, the already determined ones are used:

The physical contents can be reduced to simple forms:

Practical suitability

In order to test the practical applicability of the theoretically derived formulas, data from vehicle dynamics measurements were used. At the proving grounds in Pagenburg, driving-dynamically relevant measurement data were taken on an Axor tractor unit empty, with a loaded semi-trailer and with a load frame as a solo vehicle in various test procedures and in practice-like overland travel. The vehicle was extensively equipped with sensors, so that even sizes are available that are not available in a production vehicle.

In addition, some driving dynamics simulations were carried out to test the suitability of the system in conjunction with other variants (eg 6 × 2) as well as at low friction coefficient.

Stationary parameters, zero offset

For the test of the stationary parameters the driving maneuver "stationary circle run" appeared at first most suitable. Here, the speed is slowly increased to a constant radius of 80 m, a dynamic steering excitation does not occur. However, it was found that the determined parameters clearly differ from each other in the left and the right-hand circle. The evaluation of the relationship between steering wheel angle and Radlenkwinkel, which was initially suspected directly, shows according to 7 in the maneuver "sinusoidal steering excitation", that in the steering a significant "game" is present. The explanation is that the power steering around the center position has only the low stiffness of the torsion spring of the servo control valve, which appears in this representation as a game. Furthermore, one can not assume that steering angle and yaw angular velocity sensor are aligned with each other, a zero deviation is expected.

The consequence of the equation of identification is that the zero deviation has to be recorded separately page by page, ie to the left and to the right. The identification equation thus expands

Stationary parameters, data storage

In connection with the storage of the measured data, that is to say the matrix A ^T · A and the vector A ^T · b, the following special feature still has to be taken into account. The quotient equation, once the product of the yaw angle and the driving speed, appears once in the equation of identification. It is obvious that measurements must be made at different speeds, otherwise the matrix will remain singular. Geometrically expressed: to construct the yaw gain parabola above the velocity, only one point is available except the zero point.

Now, if the already described possibility of exponential forgetting is used, and the vehicle runs for a long time at a nearly constant speed, z. On the highway, the new measurements displace older ones made at lower speeds. The result is that the matrix first loses stamina, then rank, and becomes unusable. To avoid this undesirable effect, the measurements must be divided into speed ranges (eg, three) stored in separate matrices. The total matrix to be evaluated is added together.

Instationary parameters, high-pass filtering

As shown in the equation of identification for parameters of the dynamics, one succeeds to the parameters, in which the left side calculates the difference of the steering angle under stationary and general (thus also transient) conditions. A zero deviation between these variables leads to the error, but it is present and can not be determined around the zero crossing. An improvement in accuracy is achieved by passing the quantities through a simple high-pass filter. The high-pass filter has the property of always leaving the station portion to zero and thus to hide it from the equations.

Lateral acceleration measurement

A difficulty in the measurement - and the subsequent evaluation - the lateral acceleration represents the fact that due to roll movement, the orientation of the sensor deviates from the horizontal. Not only the angle component reduced size, but also the component of the gravitational acceleration is detected. In the measurements used, this effect was corrected with the help of the additionally installed measuring technology, the so-called horizontal lateral acceleration was stored.

Under production conditions, the use of a modified sensor could help. The idea is, according to 8th with two horizontally different, best symmetrically measured at 45 degrees employed sensors. From the measurement signals α ₁ and α ₂ , the direction can be determined in addition to the absolute magnitude of the acceleration. From this, apart from the horizontal lateral acceleration, the sensor rotation, ie the absolute roll angle of the vehicle can be calculated.

The amount of resulting acceleration:

The angle between the resulting acceleration and the symmetry axis of the sensors:

The lateral acceleration:

The angle between the vertical and the sensor axis:

In reality, the gravitational acceleration assumed here is superimposed by the vertical excitation of the roadway, which must be filtered out.

Evaluations, results

To check the procedure, the procedure as it would result on the continuous course is simulated as follows:
First, the quantities such as time, vehicle speed, lateral acceleration and yaw angular velocity are read in from the measurements, filtered and divided into segments, as already described. Subsequently, the segments are processed sequentially. First, they are tested for correlation and stationarity to determine if they are relevant and suitable for determining parameters of steady-state or transient events. The measured values sorted in this way are then stored in the respective initially empty memory matrix and continuously evaluated.

The process starts with the data for stationary operations. As soon as these are available, the calculation of the parameters for dynamic processes with respect to the yaw angular velocity is started. To Then the parameters for dynamic processes with respect to lateral acceleration can be determined. Finally, the calculation of the current calculated values for yaw angular velocity and lateral acceleration starts. As time progresses, the database is filled to saturation if necessary, the parameters converge to their optimum. 9 shows the associated data flow.

The test track "Dauerlaufkurs Süd" will run in 10 shown. The results - first from the measurement of the unloaded tractor - show the next diagrams according to 11 ,

In the diagrams of 11 the output data, such as driving speed, steering wheel angle and yaw rate are shown.

The start takes place at the gyro at low speed. Already in the acceleration phase and shortly thereafter, some curves (radii of curvature between 50 and 500 m) are passed through. It can be seen that both stationary and dynamic states occur, so that measurement data for the identification equations are collected.

In the diagrams of 12 the breakdown of the data is shown. In the upper part, under "dynamics", the variance of the measured data is plotted in the individual blocks. The blocks below the red line are used to determine the steady state, above the green line of the dynamic parameters. The lines therefore do not run horizontally because the limit was chosen as a function of the speed.

The lower image shows the correlation within the data blocks. If this remains below the value of 0.9, the block is discarded as unusable. Likewise, measurements are taken if the orbit radius is over 500 m.

The diagrams of 13 show the parameters for describing the stationary yaw gain. The curvature parameter converges fastest to its final value, followed by the lateral acceleration parameter. The zero offsets to the left and right are "difficult to calm down" because they represent absolute quantities that are influenced by external influences, such as external factors. B. road-bank are strongly disturbed.

In the diagrams of 14 the parameters of the yaw dynamics are displayed. As already discussed, the meter constant (that is, the proportionality factor of the steering speed) is difficult to determine because, under normal conditions, this quantity only weakly affects the overall equation and its influence decreases with increasing speed.

The two denominator constants, which describe the momentum of the vehicle, already approximated the true value when it appeared.

The remaining parameters, namely the two denominator coefficients for describing the lateral acceleration dynamics are shown in the diagrams of the 15 shown. It can be seen that these values take the longest to reach their final value.

In the diagrams of 16 Finally, measured and calculated values are compared. The measurement is shown with a dashed line, also recognizable by the greater noise than the calculation (solid). The simulation starts approx. 35 s after the beginning of the measurement until the values of the calculation assume zero.

The measured and simulated values of the yaw rate coincide very well. In the more noisy lateral acceleration, deviations are to be discovered in some places; the underlying parameters are used in the practical application in the observer for the float angle calculation.

To check the function was an in 17 presented Simulink model of the system created.

The measurement data was merged into a matrix in a ".mat" file, which represents the data input in the simulation. The additional data block - products and quotients - are generated in the block "Add data". Then in the block "filtall_sub" the curves are low-pass filtered and the derivatives are formed. For the evaluations of the dynamic processes they are still high-pass filtered. The block "param_sub" forms the core of the system. The measurement data are sorted, collected and evaluated. The now available parameters, together with the measurement data, are sent to the forwarded further blocks. In the "simu_sub", the setpoint calculation of the yaw angular velocity takes place, in the block "beob_sub" the estimation of the non-measurable slip angle. In this version, the setpoint of the lateral acceleration and the yaw angular velocity reconstructed in the observer, which are not needed in the "real" application, are determined as control.

simulation

The transfer function in general:

The approaches:

The approaches used and rearranged:

The Matlab / Octave function:

The Luenberger observer

In many cases, the entire state vector x is needed, but not all state variables can be measured at a reasonable cost. In the following, the totality of all measured quantities is denoted by y. Then the measurement equation applies: y = C * x + D * u

Problem: how can one approximate the entire state vector from the measuring vector y and the control vector u?

Problem solving according to DG Luenberger (1964-66) is in 20 illustrated.

L is to be chosen such that for t → ∞ x ^ (t) → x (t) strives. x ^ is then called the estimated value of x. Equation of the Luenberger observer:

Freely selectable is L.

Interpretation of the observer:

One writes the eigenvalues of the observer, d. H. of the matrix (A - L · C) to the left of the eigenvalues of the observed system so that the observer operations decay faster than the system operations. Choose L such that Q = (A - L · C) has these eigenvalues (pole specification).

der Messvektor besteht aus der Querbeschleunigung und der Gierwinkelgeschwindigkeit:

Des Weiteren besteht der Zusammenhang α_y = ν·(ψ . + β .), die folgendermaßen umgestellt wird:

In the present case, the state variables of the slip angles and the yaw angular velocity are:

the measurement vector consists of the lateral acceleration and the yaw angular velocity:

Furthermore, there is the connection α _y = ν · (ψ. + β.), which is changed as follows:

On the other hand, according to the state equation system: β. = α ₁₁ · β + α ₁₂ · ψ. + b ₁ · δ

So the lateral acceleration can be expressed from the state and input quantities:

This results in the matrices C and D too

The 2 × 2 matrix Q required here can be defined with given natural frequency ω and damping D as follows:

Since the calculation takes place with a frequency of 25 Hz with the Euler one-step method, the eigenvalue should be as far below this frequency as possible. The matrix L is then added L = (A - Q) · C ^-1

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited non-patent literature

• Alberti, V .; Babble, E .; Drumm; A .; Driving stability control through active braking of individual wheels, Automatisierungstechnik 44 (1996) 5 [0003]
• van Zanten, A .; Erhardt, R .; Pfaff, G .: FDR - The driving dynamics control of Bosch, Automobiltechnische Zeitschrift 96 (1994) 11 [0003]
• Müller, A .; Achenbach, W .; Schindler, E .; Wohland, Th .; Mohn, F.-W .: The new driving safety system Electronic Stability Program of Mercedes-Benz, Automobiltechnische Zeitschrift, 96 (1194) 11 [0004]
• Donges. e .; Naab, K .: Control Systems for Vehicle Guidance and Stabilization in Automotive Engineering, Automatisierungstechnik 44 (1996) 5 [0004]
• Hecker, F .; Jundt, O .; Leimbach, K.-D .; Faye, I .; Schramm, H .: Vehicle dynamics control for semitrailer tractors, 6th technical conference Tires - Chassis - Road Hannover 1007, VDI reports 1350 [0006]

Claims (8)

Method for driving dynamics control of a road vehicle, In which actual values for parameters which correlate with the driving dynamics of the vehicle are recorded, In which the detected parameters are subdivided according to the variance of the associated actual values, stationary parameters which correlate with a stationary state of the vehicle, and transient parameters which correlate with transient processes of the vehicle, In which reference values for the current vehicle dynamics state of the vehicle are determined on the basis of the stationary parameters, In which actual values for the current vehicle dynamics state of the vehicle are determined on the basis of the transient parameters, In which a stability of the current vehicle dynamics state of the vehicle is determined by a nominal-actual comparison of the vehicle dynamics state, - in which the determined stability is assessed, - in which a stabilization measure is carried out if the stability determined is assessed as insufficient. A method according to claim 1, characterized in that in the evaluation of the transient parameters, a high-pass filter is used. A method according to claim 1 or 2, characterized in that is taken into account in the assessment of the determined stability, whether the vehicle is still in a linear dynamic range. Method according to one of claims 1 to 3, characterized in that the determination of the current stability with respect to the determined and evaluated actual values of the parameters takes place recursively. Method according to one of claims 1 to 4, characterized in that in determining the current stability, the determined and evaluated actual values of the parameters are weighted with respect to their detection time, so that the actual values have a decreasing weighting with increasing age. Method according to one of claims 1 to 5, characterized in that a yaw rate of the vehicle is determined as the driving dynamics state. Method according to one of claims 1 to 6, characterized in that a float angle is determined as the stability of the vehicle. Method according to one of claims 1 to 7, characterized in that as a stabilizing measure a braking intervention is performed on at least one vehicle wheel.

Images