Vehicle mass estimation procedure

Abstract

Vehicle mass estimation method for estimating the mass (m G ) of a vehicle, in which several models (26) are created for the vehicle to estimate the mass, with the models (26) each at least one parameter (e.g. A ) is estimated, the measurable parameter on the vehicle (e.g. A ) is measured and an evaluation (16) of the models (26) by means of a comparison of the estimated and the measured parameter (z A ) takes place.

Description

[0001] The invention relates to a vehicle mass estimation method by which it is possible to estimate the mass of a vehicle, in particular a passenger car. Furthermore, the vehicle mass estimation method relevant here also makes it possible to estimate the mass of a combination, i.e., a vehicle with a trailer, as well as a trailer alone.

[0002] Such an estimation of the vehicle mass and in particular the trailer mass serves as the basis for the dynamic stabilization of the vehicle and, if applicable, its trailer.

[0003] Various methods are known for estimating the vehicle mass from the vertical dynamics of the respective vehicle, which in particular work with low-pass filters of the forces of the associated wheel suspension and with nonlinear state observers (EKF).

[0004] For example, EP 1863659 B1 discloses a method for determining the mass of a vehicle based on a vertical movement between a vehicle body and the vehicle wheels, wherein the mass estimation is carried out by means of a non-linear state observer based on the vertical dynamics of the vehicle.

[0005] EP 1430276 B1 discloses a method for determining the mass of a motor vehicle under various driving situations and evaluating the respective vehicle acceleration. In addition to the driving force of a drive unit, the respective resistance forces resulting from rotational forces, air resistance, rolling resistance, and slope force, as well as braking forces on the associated friction brakes, are taken into account. However, such mass estimates based on vehicle acceleration, i.e., on the basis of the vehicle's longitudinal dynamics, always include the trailer mass. This results in a comparatively large number of necessary model assumptions that cannot be determined with a high degree of certainty.

[0006] DE 102006045305 B3 discloses a method for determining the mass of a motor vehicle by measuring the ride height of the wheel suspension and calibrating it using a vehicle mass estimate from longitudinal dynamics. However, when using a level control system in the vehicle, a mass estimate based on the evaluation of ride height measurements is not possible, since the expected change in ride height is precisely what is compensated for by the control system.

[0007] DE 19744066 B4 describes a method for detecting trailer operation in a motor vehicle. A pressure signal that records the pressure applied to the trailer is evaluated. For trailer detection, the trailer must be mechanically connected to the towing vehicle via a trailer coupling and additional connections. Furthermore, an additional sensor is required for pressure measurement. A statement regarding the trailer's mass cannot be made.

[0008] According to US 2009 / 0 306 861 A1, it is possible to detect a trailer by comparing the expected dynamics of the vehicle with measured sensor values. However, if the trailer's center of gravity is close to a wheel axle, the trailer's influence on the vehicle's dynamics only exists during significant yaw rate changes. In general, the trailer's influence on the vehicle's dynamics is small in this configuration.

[0009] According to the invention, a vehicle mass estimation method for estimating the mass of a vehicle is provided, in which multiple models are created for estimating the mass of the vehicle, at least one measurable parameter on the vehicle is estimated using each of the models, the measurable parameter is measured on the vehicle, and the models are evaluated by comparing the estimated and measured parameters. In particular, when parameterizing the models according to the invention, only the parameter of the vehicle's mass is set as different or different in each case.

[0010] According to the invention, the multiple models are preferably created or executed cyclically, and for a new estimation of the vehicle's mass, the model that previously received a good rating compared to the other models is selected. In particular, multiple models (all with different masses) are initialized. These are then executed cyclically (preferably according to the Kalman filter). All models are evaluated in each step. This then provides a rating or probability for each model at any time. A cyclical procedure can be used to determine which model has received a good rating multiple times, and then to identify this model as the model most likely to correctly predict the vehicle's mass. Preferably, after a certain time, a "mass region" of models is rated as more likely than others.Then you can, so to speak, reinitialize in the previously found region with a higher density of models. To re-estimate the vehicle's mass, α models with the lowest probability are preferably deleted from the previous models and α new models are created. This allows a discretization to be refined in the probable range. Furthermore, for a re-estimation of the vehicle's mass, the models are advantageously optimized using the previously measured parameters. The multiple models for estimating the mass and their evaluation then run in parallel or simultaneously. Furthermore, to evaluate the models, it is advantageous to create a weighted average over at least two, preferably over all models. The mass of the respective model and its probability or quality are taken into account for the weighted average.

[0011] In other words, the invention uses a multi-model approach for mass estimation. For mass estimation, measured and estimated parameters are compared. This evaluates the quality of the models, which are parameterized under the assumption of different masses. In particular, the mass is selected according to the model with the highest quality.

[0012] According to the invention, the plurality of models are preferably created with a linear equation of state each.

[0013] During the evaluation, a previously defined probability for the model is advantageously used as an evaluation criterion.

[0014] A parameter of the vertical dynamics of the vehicle, in particular a ride height of a wheel suspension, a spring stiffness of a wheel suspension, a damping value of a wheel suspension, and / or a wheel acceleration of a vehicle wheel, is particularly preferably used as a measurable parameter. Alternatively or additionally, a parameter of the longitudinal dynamics or the lateral dynamics or the steering angle of the vehicle, in particular an overall acceleration, is used as a measurable parameter.

[0015] Furthermore, the models provided according to the invention preferably estimate a partial mass of the vehicle, preferably using four multi-model approaches for each quarter of the vehicle mass. The total mass of the vehicle is then determined as the sum of the partial masses.

[0016] In order to further improve the vehicle mass estimation method according to the invention, it is further preferably provided that braking and / or acceleration processes as well as cornering and steering processes of the vehicle are taken into account in the models according to the invention.

[0017] The invention is based on the finding that methods for determining chassis parameters, especially vehicle mass, are well known from vertical dynamics. However, these methods have the particular disadvantage that nonlinear state observers are relatively complex and require comparatively large amounts of computing time. Furthermore, unstable or diverging state estimates can result due to insufficient parameterization quality.

[0018] With the solution according to the invention, however, it is possible to reduce the computing time per model to approximately 1 / 10 compared to the use of non-linear state observers. The solution according to the invention is further based on the fact that many variables required for state estimation are precalculated once using associated equations. Thus, these variables do not have to be recalculated in each time step or each estimation run, as is otherwise usual. Furthermore, more robust modeling and parameterization are created. The functionality according to the invention also results in a level control system installed in the vehicle. In particular, according to the invention, no reference ride height, i.e. the height of the vehicle body with the spring of the associated wheel suspension relaxed, is required.

[0019] In an advantageous embodiment of the inventive solution, the weighted mean of all model masses is determined, weighted by the respective model quality. Well-known methods evaluate the model errors based on the variance of a normal distribution—this is not done here. According to the invention, the same variance is used for the evaluation for all models. The variance corresponds to a design parameter: the larger the assumed variance, the slower the mass estimate converges. Well-known methods also assume a normal distribution for the model errors (Kalman filter)—this limitation is resolved with the proposed method.

[0020] By combining the mass estimation according to the invention with a well-known mass estimation from vehicle longitudinal dynamics, the distribution of masses (vehicle mass, trailer mass) in a vehicle-trailer combination can be determined. In such a known mass estimation method, as explained above with regard to EP 1430276 B1, the mass of a combination is determined based on the total inertial mass determined during a braking or acceleration process.

[0021] An exemplary embodiment of the inventive solution is explained in more detail below with reference to the attached schematic drawings. It shows: Fig. 1 the functional diagram of a quarter vehicle used according to the invention, Fig. 2 a sequence of the method according to the invention as a whole, Fig. 3 a sequence of the method according to the invention at one of its time steps, Fig. 4 a first graph for the evaluation of the models according to the invention and Fig. 5 a second graph for the inventive evaluation of the models.

[0022] Fig. Figure 1 illustrates the procedure according to the invention, according to which a vehicle is divided into four parts, so-called quarter vehicles 10. Each quarter vehicle 10 includes a wheel suspension with an associated vehicle wheel, which has a height of R The associated vehicle body has (simplified according to the invention) a partial mass m A and an altitude z A Between the vehicle body and the vehicle wheel is the spring and damping system of the wheel suspension with a spring constant F c and a damping value F d .

[0023] For the dynamics of the thus reduced quarter vehicle, several linear state observers of the form x=(zA−zR;z˙A−z˙R;zR;z˙R;z¨R) with the measurement vector z=(zA−ZR;z¨R) used.

[0024] For parameterization, the following parameters are used, in particular the spring stiffness F c and the damping value or damping parameter F d of the vehicle suspension. In a vehicle with air springs, which can be installed on one or two axles, the internal pressure can be measured using an air pressure sensor, for example, and the spring stiffness F can be calculated from a lookup table. c be determined.

[0025] In Fig. Figure 2 illustrates the essential steps of the procedure. First, the vehicle is divided into four quarter vehicles 10, as explained. For each of the quarter vehicles 10, a simulation or estimation 12 is performed using a multi-model approach with multiple models. The results 14 of the estimation 12 are subjected to an evaluation 16 by comparison with measured vehicle dynamics data. Based on a probability 18, the model judged to be the best is determined, and this is used to determine a total mass 22 of the vehicle in an overall estimate 20. Furthermore, the models are corrected or optimized 24 based on the measured vehicle dynamics data.

[0026] As in Fig. 3, in this way, at a time step of the method according to the invention, several models 26 with the running number i, i+1, ... for the partial mass mass m Aset up. For each model 26, a predicted state 30 is determined on the basis of a previous state 28 by means of the estimation 12. This is then converted into a corrected state 32 by means of the aforementioned optimization 24 and into a quality value 34 by means of the evaluation 16. From this, a model probability 36 is determined, based on which the best model 26 is selected for the mass estimation. For the optimization 24 and the evaluation 16, as explained, a measurement 38 of measurable parameters of the respective model 26 is also taken into account in each time step. The measurement therefore occurs once in each time step, from sensors on the real vehicle. This measurement is then compared with correspondingly estimated states of the respective models.

[0027] The correction or optimization 24 of the models 26 is similar to a Kalman filter: x^priori,k=Fx^posteriori,k−1

[0028] Here F is the system matrix and x̂ posteriori,k-1 the state estimate from the last time step. x^posteriori,k=x^priori,k+Lk(zk−Hx^priori,k)

[0029] z k is the measurement in time step k and H is the observation matrix. L k corresponds to the stationary value of the Kalman gain of a quarter vehicle, which has an approximately expected quarter vehicle mass.

[0030] The steady-state Kalman gain is used as the stationary value.

[0031] The stationary values ​​can be determined by multiple evaluation of Ppriori,k=FPposteriori,k−1F'+Q Sk=HPpriori,kH'+R Lk=Ppriori,kHSk−1 Pposteriori,k=(I−LkH)Ppriori,k be calculated.

[0032] The Kalman filter is a state observer that, when used normally, calculates a complete state estimate (including all equations) in time-synchronous fashion. In this case, only the equations x̂ priori,k = Fx posteriori,k-1 and x̂ posteriori,k = x priori,k + L k (e.g. k - Hx priori,k ) is used synchronously. Equations (1) and (4), however, are calculated only once and separately. When calculated once, the calculation is performed cyclically until the Kalman gain has "settled."

[0033] The Kalman gain L is set constant in this approach and can therefore be precalculated separately, depending on the selected process noise matrix Q and the measurement noise matrix R, and depending on the model that specifies F. In concrete terms, this means that a model of a mass of 400 kg will have a slightly different F than a model of a mass of 500 kg. In practice, one should use a mass model for which suitable measurements have been made. To do this, equations (1) to (4), the so-called Kalman equations, are calculated in a loop until they settle and the Kalman gain reaches a steady-state value. This is then saved and can be used at any time.

[0034] A solution using the so-called Riccati equation P=F(P−PH'(HPH'+R)−1HP)F'+Q However, it is not recommended as an alternative in this case, since the variances of z R and ż Rdiverge.

[0035] The procedure mentioned can be carried out before use in the vehicle, since the Kalman filter allows the gains to be adjusted independently of the measurements. k are.

[0036] Where P posteriori,k-1 is the estimated covariance matrix of the last state vector, Q is the discretized covariance matrix of the system noise, R is the covariance matrix of the measurement noise, and I is the identity matrix.

[0037] The Kalman gain L is therefore the same for all models at every time step. With the one-time parameter selection of Q and R, no attention is paid to ensuring the filter is consistent. This would be possible for a single mass, but with multiple masses, the parameters Q and, if necessary, R would have to be changed accordingly. In practice, however, not all necessary measurement series are always available. This leads to the covariances calculated by the Kalman filter not being used to evaluate the models.

[0038] As already mentioned Fig. 3, a multi-model approach is carried out, in which a total of N models 26 with different masses m are used for the movement of each wheel. i be initialized. This shows Fig. 3 a model 26 for a mass estimate i and another model 26 for a mass estimate i+1.

[0039] By K consecutive measurements z of the vertical dynamics, in this case the altitude z A and preferably also a wheel acceleration, the evaluation 16 of the individual models 26 with the different body masses or partial masses m A For this purpose, for each model i with mass m i preferably a specially defined quality measure 34 for an error E i determined to Ei|K=∑k=1K|hk−h^i,k|δ with the measurement h k = z k (1) and estimate ĥ i,k = x i,k (1) the altitude z Afrom model i, weighted by the exponent δ. For example, if δ = 2 is set, this corresponds to the optimization criterion of a least-squares optimization, and the score 16 becomes similar to a likelihood score or probability score (assuming a normal distribution).

[0040] For each model i after K measurement steps, the following preferred model probability 36 can be defined: Pi|K=e−Ei|K∑je−Ej|K

[0041] The vehicle mass of the quarter vehicle 10 is then determined to be the mass m̂ K , which for the model with the smallest quality measure E (i|K) , respectively the greatest probability P (i|K ), is used: iK=arg maxi∈NPK(i) m^K=m(iK)

[0042] To estimate the quality of the mass estimation and possible termination criteria when a sufficiently reliable estimate is reached, the shape of the curve is used, as shown in Fig. 4. The vertical diagram axis shows the model probability 36 over the corresponding vehicle mass of the quarter vehicle 10 (in kg) on ​​the horizontal diagram axis.

[0043] Fig. 5 illustrates how, after half of the calculation time, the model probabilities 36 lead to a “broader” curve than at the end of the calculation (see Fig. 4) As time increases, the probabilities focus on a specific area.

[0044] After a certain cycle time, improbable models 26 are discontinued to save computing power. In the range of high model probabilities 36, new models 26 are initialized to improve the resolution of the mass estimate.

[0045] For example, from the N models 26, α models (α < N - 2) with the lowest probabilities 36 are deleted and at the same time α new models 26 are initialized in the following way: Most likely model lies at the edge (i = 1 or i = N) : Δm neu > Δm alt (Width of discretization of masses is increased) α new models set the simulation on the side of the model with P max away. Most likely model is not at the boundary (1 < i < N): Δm neu < Δm alt (Width of discretization of masses is reduced) α new models around the model with P max divide the most probable intervals.

[0046] Finally, it should be noted that to calculate the Kalman gain, in addition to the quarter vehicle mass, the variance entries for the system noise Q and the measurement noise R can be specified. An optimization algorithm ("differential evolution") can be used to determine the process noise parameters, the measurement noise parameters, and the exponent δ. This algorithm determines the optimal parameters based on training data, or rather, test runs of vehicles with known mass, by evaluating how accurately the mass (and other dynamic variables) were estimated.

[0047] The mass m G of the entire vehicle is ultimately the sum of the four individual results m A from the four-car models: mVehicle=∑i=14m^i

[0048] For estimating the mass of a trailer, mass estimates from longitudinal dynamics are well known. In the known methods, the total mass of the combination is estimated if a trailer is present. In the method presented here, mass estimation from vertical dynamics, the pure vehicle mass m G (plus the relatively small static trailer load on the trailer coupling of the towing vehicle).

[0049] With m^trailer=m^trailer-m^vehicle The mass of the trailer can thus be determined.

[0050] In an area where it is safely assumed that a trailer is present, the vehicle mass - and thus also the trailer mass - can still be corrected by the expected static coupling forces that were added to the vehicle in the mass estimation by the vertical dynamics: m^vehicle, corr=m^vehicle−m^corr m^Anganger, corr=m^Anha¨nger+m^corr m^corr=FClutch, statg with the acceleration of gravity g=9.81ms2.

[0051] Systematic errors in the estimation of the vehicle mass, which occur due to trailer properties (mass, moment of inertia,...), can be taken into account in a second correction: m^corr, 2=f(m^trailer, corr, …) List of reference symbols 10 quarter vehicle 12 Estimate 14 results 16 reviews 18 Probability 20 Total estimate 22 Total mass 24 Optimize 26 Model 28 Condition 30 condition 32 condition 34 Quality value 36 Model probability 38 Measurement

Claims

[1] Vehicle mass estimation method for estimating the mass (m G ) of a vehicle, in which several models (26) are created for the vehicle to estimate the mass, with the models (26) each at least one parameter (e.g. A ) is estimated, the measurable parameter on the vehicle (e.g. A ) is measured and an evaluation (16) of the models (26) by means of a comparison of the estimated and the measured parameter (z A ) takes place. [2] A vehicle mass estimation method according to claim 1, wherein the plurality of models (26) are created cyclically and are used for re-estimating the mass (m G ) of the vehicle, the model (26) is selected which has previously received a good rating (16) in comparison with the other models (26). [3] Vehicle mass estimation method according to claim 2, wherein for a re-estimation of the mass (m G) of the vehicle from the previous models (26) α models (26) with the lowest probability (18) are deleted and α new models (26) are created. [4] Vehicle mass estimation method according to claim 2 or 3, wherein for a re-estimation of the mass (m G ) of the vehicle, the models (26) are optimized using the previously measured parameters. [5] Vehicle mass estimation method according to one of claims 1 to 4, wherein the plurality of models (26) are each created with a linear equation of state. [6] Vehicle mass estimation method according to one of claims 1 to 5, wherein a previously defined probability (18) for the model is used as an evaluation criterion in the evaluation (16). [7] Vehicle mass estimation method according to one of claims 1 to 6, in which a parameter (z A ) of the vertical dynamics of the vehicle, in particular a height (e.g.A ) of a wheel suspension, a spring stiffness (F c ) of a wheel suspension, a damping value (F d ) of a wheel suspension and / or a wheel acceleration of a vehicle wheel. [8] Vehicle mass estimation method according to one of claims 1 to 7, in which a parameter of the longitudinal dynamics or the lateral dynamics of the vehicle, in particular a total acceleration, is used as the measurable parameter. [9] Vehicle mass estimation method according to one of claims 1 to 8, in which the models (26) are used to estimate a partial mass (m A ) of the vehicle is estimated, preferably using four multi-model approaches for each quarter vehicle mass. [10] Vehicle mass estimation method according to one of claims 1 to 9, in which braking and / or acceleration processes and / or cornering or steering processes of the vehicle are taken into account in the models (26).

Images