Method for determining a vehicle reference speed and vehicle controller having such a method

Abstract

A method for determining a vehicle reference speed and a vehicle controller having such a method, in which directly or indirectly determined or estimated vehicle status signals, including weighting factors associated with each of the vehicle status signals, are merged in a merging module, wherein the merging module includes at least two stochastic estimators, which exchange signals with one another that correspond to physical vehicle parameters, wherein the association of the estimators is selected in accordance with a physics model for the vehicle behavior.

Claims

1. A vehicle reference speed determination method, comprising the steps of: receiving, by a vehicle controller, sensor signals from vehicle sensors; determining, by a first stochastic estimator of the vehicle controller, angular speeds of wheels on a first axle of the vehicle based at least in part on the sensor signals; determining, by a second stochastic estimator of the vehicle controller, angular speeds of wheels on a second axle of the vehicle based at least in part on the sensor signals; determining, by a third stochastic estimator of the vehicle controller, a speed of the vehicle by fusing driving state signals in a fusion model while incorporating respective weighting factors, the driving state signals based at least in part on the angular speeds of the wheels on the first axle and the second axle wherein the stochastic estimators are selected in accordance with a physical model based on a behavior of the vehicle.

2. The method as claimed in claim 1, wherein the fusion model is structured in at least two hierarchically structured model calculation levels, wherein each model calculation level models a sub region of a vehicle and each level comprises at least one stochastic estimator.

3. The method as claimed in claim 2, wherein besides stochastic estimators the fusion model additionally comprises a further stochastic estimator, that works according to the least squares method, and/or a physical computer model.

4. The method as claimed in claim 1, wherein besides stochastic estimators the fusion model additionally comprises a further stochastic estimator, that works according to the least squares method, and/or a physical computer model.

5. The method as claimed in claim 1, wherein the driving state signals are derived from the ESC-driving state sensors, selected from the group consisting of: i) wheel speeds (.sub.i) and the ii) vehicle longitudinal acceleration (a.sub.x,sensor), which comprise the vehicle longitudinal speed (v.sub.ref), the vehicle longitudinal acceleration (a.sub.ref) and the road gradient angle (.sub.ref).

6. The method as claimed in claim 1, wherein the model calculation levels comprise the levels vehicle model, tire model and drive train model, wherein the vehicle model is a longitudinal dynamic model of the vehicle.

7. The method as claimed in claim 1, wherein a wheel-specific slip curve estimation is used for determining the parameters of the tire characteristics.

8. The method as claimed in claim 1, wherein a stochastic estimator is used for each axle that determines the slip adjusted wheel speeds from the measured wheel speeds and the estimated vehicle speed.

9. The method as claimed in claim 1, wherein a determination of the vehicle speed and vehicle longitudinal acceleration is carried out, during which the following are used a stochastic estimator with linear dynamics for the fusion of a measured vehicle acceleration, compensated for long-term offset and gradient, a vehicle acceleration calculated from the model, and four wheel speeds adjusted for slip.

10. The method as claimed in claim 1, wherein determination of a model vehicle acceleration (a.sub.x,model) is carried out based on the circumferential forces from the tire characteristics and the vehicle mass based on a vehicle longitudinal model.

11. The method as claimed in claim 1, wherein a stochastic determination or estimation of the road gradient is carried out from the measured vehicle acceleration and the estimated vehicle acceleration.

12. A vehicle controller, implementing the method according to claim 1.

Description

BRIEF DESCRIPTION OF THE DRAWING

(1) Other preferred embodiments emerge from the dependent claims and the following description of an exemplary embodiment using a FIGURE.

(2) In the FIGURE

(3) FIG. 1 shows a representation of a vehicle reference speed determination method with a division into a plurality of interconnected stochastic estimators.

DETAILED DESCRIPTION OF THE INVENTION

(4) In FIG. 1 an estimator 10 for determining the vehicle reference speed is shown with the main calculation blocks, such as can be used in an electronic vehicle controller, such as for example in an ESC/ESP brake controller or a vehicle central computer. The electronic controller receives signals from vehicle sensors that are connected thereto or contained within the controller, such as wheel revolution rate sensors, a yaw rate sensor or a longitudinal acceleration sensor and processes said signals in a microcontroller. Because of the internal digital signal processing, discrete Kalman filters are therefore particularly advantageous for the fusion of states.

(5) The estimator 10 is a hierarchical filter formation with three levels. The state space model can be expressed for an all-wheel drive vehicle or for a two-wheel drive vehicle by a general state equation producing a mathematical relationship between the wheel speeds per wheel .sub.i (i=1 to 4), the vehicle speed v.sub.x and the acceleration a.sub.x in the longitudinal direction (x) of the vehicle.

(6) The estimator 10 processes the following digitally converted sensor signals: .sub.i=wheel revolution rate signals 9 (referred to as RDF in FIG. 1), a.sub.x,sensor=acceleration sensor signal 8, T.sub.eng=engine torque signal 11 and P.sub.THZ=driver's initial pressure 12 (THZ, brake torque box in FIG. 1).

(7) The observation of the above-mentioned computing-intensive state equation is divided in the estimator 10 into three independent calculation levels, each of which can be considered as an independent vehicle model: tire model 200 including , curve estimator 2 for determining the slip characteristic, drive train model 300 including wheel dynamics model 3 and longitudinal dynamics model 100 of the vehicle.

(8) The method according to the example can be understood as a subdivided, in particular at least partially hierarchically structured, structure of stochastic estimators. Thus an observer at a higher level can combine and assess the information at the lower levels. The resulting hierarchical structure has the advantage that task separation is carried out. The lower estimator of such a structure produces pre-filtering, whereas the upper level is responsible for the fusing. There can be an adaptive self-organization of the top-down and bottom-up influences. In addition, the estimators in the lower levels profit from the top-down influence of the hierarchical structure. This means that the lower levels can use the predictions of the upper levels, which can produce more accurate results, for example because of the higher information content of the plurality of inputs or of the more accurate model. Furthermore, the priorities determined in the upper levels can be handed down to support the lower observer with a more specific view of things. For the upper levels, by contrast, the bottom-up influence is important because the upper levels only have to manage the predictions of the lower levels that they receive as a measurement.

(9) A stochastic estimator is for example a known Kalman filter, in particular a discrete Kalman filter. With said filter it is possible to fuse together both measurement signals determined by sensors (e.g. measurement variables of physical driving states, such as for example the longitudinal acceleration measured with a longitudinal acceleration sensor), and also so-called pseudo-measurement signals (e.g. physical driving state variables that are calculated or estimated from or with other signals) with simultaneous weighting depending on the fuzziness of the estimated signals (confidence intervals or quality). In order to always be able to include the signal quality during the processing of the signals, the transfer of the signals is preferably carried out by transferring or handing over the current signal value to the next computing module together with the current quality of the signals currently being transferred.

(10) Because a discrete Kalman filter is a filter that is particularly suitable for linear equation systems, it is advantageous to perform a linearization in the region of the working point, for example by the particularly preferred use of an extended Kalman filter (EKF) or of an unscented Kalman filter (UKF). For an implementation in a vehicle controller, extended Kalman filters are more particularly to be preferred because the same have advantages related to the necessary capacity of the physical computers (resources) used.

(11) The modeling of the tire road contact results in a nonlinear overall system, which can only be treated numerically with a relatively high computing time requirement. It can therefore be advantageous not to carry out the exclusive use of simple Kalman filters that are primarily conceived for linear systems. It has been shown that EKF filters are particularly suitable for the purpose described here, because it is possible using said filters to linearize the system about the current working point.

(12) Another possibility of dealing with the nonlinearity of the system is the use of a UKF filter.

(13) The stochastic estimators used in the example comprise at least the components 1) to 7) described below. Said numbering should not be understood as a limitation of the example of the invention. Rather, it is possible to remove individual components or to add further components depending on the current requirements.

(14) Because of the difference of the wheel dynamics and vehicle dynamics, for example the estimation of the reference speed and the vehicle acceleration is separated from the estimation of the angular speeds. The aim is the estimation of the vehicle reference speed, therefore the estimator for the same is placed in the upper level. In order to further simplify the model of the angular speeds, there is a division into front and rear axles. The choice of said configuration has the advantage, in contrast to the division into the left and right wheels, that the wheels on both axles are strongly coupled to each other.

(15) An estimator formed in this way is relatively accurate and comprises no particular numerical problems as a result of excessively different dynamics. In contrast to an estimator for the complete model, the division and the self-organizing structure also give the advantage of an easier ability to parameterize. The division into front and rear axles also enables the observer to more easily configure for other types of vehicle. Thus all in all the design of the estimator 10 facilitates an implementation in a vehicle computer.

(16) At the lower level 300 there are two EKF filters 3, which estimate the slip-affected angular speeds .sub.FL and .sub.FR at the front axle and .sub.RL and .sub.RR at the rear axle. The estimators use as inputs the driving torque of the engine T.sub.eng 11 as well as the brake torques T.sub.brk 12. The brake torques can be calculated from the brake pressures of the vehicle brake system.

(17) When using modeling of a vehicle with two drive units, the estimator of the front axle uses the engine torque of the front engine and the estimator of the rear axle uses the engine torque of the rear engine. Likewise, the brake torques are divided as a result of the separation of the connection between the axles.

(18) At the upper level 100 there is a Kalman filter 7 that is used as a fusion filter. It estimates the speed v.sub.ref and the acceleration a.sub.ref of the vehicle. There is thus a total of three estimators in use in the illustrated example, one for the dynamics of the front axle, one for the dynamics of the rear axle and one for the overall dynamics of the vehicle. In the lower part of level 100 there is a block 5 for producing pseudo measurements for the acceleration in the form of the four slip-compensated wheel speeds that are made available to the fusion filter. Furthermore, the same uses the measurement signal of the acceleration sensor a of Block 8. In order to be able to exploit the strengths of a hierarchical structure, the standard deviations determined by the filters of the lower level are forwarded to the fusion filters. Moreover, the estimators for the front and rear axles exchange their estimates with each other and use the estimation of the speed and acceleration of the fusion filters as an input. When modeling a vehicle with two drive units, the use of the estimates of the lower level among themselves is omitted.

(19) 1) Gradient Angle Estimator Component

(20) From the determined vehicle reference acceleration v.sub.ref as well as the measured acceleration a.sub.ref from 7 and the measured acceleration a.sub.x,sensor from 10, a road gradient angle .sub.ref is estimated by means of a stochastic filter.

(21) 2) Slip Characteristic Estimator, ,-Curve Estimator Component

(22) Using the offset-corrected longitudinal acceleration of the vehicle a.sub.x,sensor,corr, the coefficient of friction used .sub.used can be calculated and the coefficient of friction can be divided among the four wheels depending on the type of drive (two-wheel drive, all-wheel-center differential). Based on a parameterized tire characteristic and using the coefficients of friction used .sub.used,i and the calculated slips .sub.i of the wheels from 4, the parameters c.sub.0, c.sub.1 and c.sub.2 of the reference characteristic are determined by a least squares method and the maximum coefficients of friction .sub.max, i are also determined therefrom. The model coefficients of friction .sub.model, i and the associated peripheral wheel forces
F.sub.x,i=F.sub.n,i.Math..sub.model,i
can be derived from the characteristic obtained depending on the slips.

(23) 3) Revolution Rate Estimator Component

(24) By analysis of the engine torque signal T.sub.eng, the revolution rate signals .sub.i, the estimated vehicle acceleration a.sub.ref from 7, of the estimated gradient angle .sub.ref from 1, the determined peripheral tire forces F.sub.x, i and the tire properties (slip curve) from 2, the noise-reduced wheel speeds {tilde over ()}.sub.i are determined by means of a stochastic estimating method (e.g. Extended Kalman Filter) taking into account the drive train model (incl. the tire dynamics model).

(25) 4) Slip Calculation Component

(26) The wheel slips .sub.i are calculated using the estimated vehicle speed v.sub.ref from 7 and the estimated wheel revolution rates {tilde over ()}.sub.i from 3.

(27) 5) Model-Acceleration Component

(28) A model vehicle acceleration a.sub.x,model is calculated from the sum of the estimated tire forces F.sub.x, i from 2 by division by the mass.

(29) 6) Slip Adjusted Wheel Speeds Component

(30) The estimated slips .sub.i from 4 are subtracted from the measured wheel speeds .sub.i and the wheel speeds {tilde over (v)}.sub.wh, i corresponding to the vehicle speed and adjusted for slip are thus calculated.

(31) 7) Fusion Filter Component

(32) In a stochastic fusion filter (e.g. Kalman filter), the vehicle speed V.sub.ref and the vehicle acceleration a.sub.ref are estimated from the following information by means of a linear dynamic model: {tilde over (v)}.sub.wh, i of the component 6), the measured vehicle acceleration a.sub.x,sensor, after subtracting the effect of the gradient angle .sub.ref, and the model vehicle acceleration a.sub.x,model from the component 5).

(33) The entire framework undergoes stochastic signal processing. This means that in addition to the estimated signals themselves the components also transfer the fuzziness of the estimated signals (confidence intervals), which are taken into account in the further signal processing. As can be seen, the number of measurements is greater than the number of states and the structure of the Kalman filter enables measurement signals to be fused together with pseudo measurement signals. The Kalman filter filters the most trustworthy information from the redundant speed signals and acceleration signals using the signal noise.