Method and a Device for Determining the Propulsion Torque

Abstract

In a test run, in order to easily provide a high-quality propulsion torque of a torque generator based on the partially low-quality measured variables available on the test bench, it is foreseen that an inner torque (M.sub.i) of the torque generator (D) is measured and based on the measured inner torque (M.sub.i), from an equation of motion, including the measured inner torque (M.sub.i), a dynamic torque (M.sub.dyn) and a shaft torque (M.sub.w) measured on the output shaft of the torque generator (D), a correction torque ({circumflex over (M)}.sub.cor) is estimated, and from the estimated correction torque ({circumflex over (M)}.sub.cor) and the measured inner torque (M.sub.i), the propulsion torque (M.sub.v) according to the relation M.sub.v={circumflex over (M)}.sub.cor+M.sub.i is computed.

Claims

1. A method for determining the propulsion torque (M.sub.v) of a torque generator (D), which is mounted on a test bench, wherein an inner torque (M.sub.i) of the torque generator (D) is measured and, based on the measured inner torque (M.sub.i), from an equation of motion, comprising the measured inner torque (M.sub.i), a dynamic torque (M.sub.dyn) and a shaft torque (M.sub.w) measured on an output shaft of the torque generator (D), a correction torque ({circumflex over (M)}.sub.cor) is estimated, and from the estimated correction torque ({circumflex over (M)}.sub.cor) and from the measured inner torque (M.sub.i), the propulsion torque (M.sub.v) is computed according to the relation M.sub.v={circumflex over (M)}.sub.cor+M.sub.i.

2. The method according to claim 1, wherein the equation of motion is averaged over a certain time period and the average value is used as an estimated correction torque ({circumflex over (M)}.sub.cor).

3. The method according to claim 2, wherein the correction torque ({circumflex over (M)}.sub.cor) is formed by a basic correction torque (M.sub.cor,0) and at least one term κ{dot over (φ)}, as a function of angular velocity ({dot over (φ)}) and the basic correction torque (M.sub.cor,0) and the parameter (κ) are determined by at least two averaging operations on the equation of motion.

4. The method according to claim 2, wherein a characteristic map of the correction torque ({circumflex over (M)}.sub.cor) is formed over angular velocity ({dot over (φ)}).

5. The method according to claim 1, wherein from the equation of motion a state observer is provided, which estimates the correction torque ({circumflex over (M)}.sub.cor).

6. The method according to claim 5, wherein the equation of motion is written with an estimated rotation angle ({circumflex over (φ)}), or its temporal derivatives ({circumflex over ({dot over (φ)})}) and ({circumflex over ({umlaut over (φ)})}), and with the estimated correction torque ({circumflex over (M)}.sub.cor) and a target function (Z) is set, wherein the target function (Z) comprises the estimated rotation angle ({circumflex over (φ)}), and a measured rotation angle (φ.sub.m), as well as the estimated correction torque ({circumflex over (M)}.sub.cor)) and wherein the estimated correction torque ({circumflex over (M)}.sub.cor) is determined by and that optimizing the target function (Z).

7. The method according to claim 6, wherein an estimated value of the rotation angle ({circumflex over (φ)}) is computed in an iterative method from the equation of motion with an estimated value for the correction torque ({circumflex over (M)}.sub.cor), and therefrom by optimizing the target function (Z), a new estimated value for correction torque ({circumflex over (M)}.sub.cor) is computed, wherein at the beginning an initial value of the correction torque ({circumflex over (M)}.sub.cor) is defined and the iteration is continued until a defined stop criterion is met.

8. The method according to claim 6, wherein the target function (Z) contains weighting factors (λ.sub.φ, λ.sub.{dot over (φ)}, λ.sub.M).

9. The method according to any of claims 6, wherein with the correction torques ({circumflex over (M)}.sub.cor) determined by the state observer, a mathematical model for the correction torque ({circumflex over (M)}.sub.cor) is trained.

10. The method according to claim 9, wherein the mathematical model is corrected on the basis of current estimates of the correction torque ({circumflex over (M)}.sub.cor).

11. A method according to claim 1 in a test run for a test object on a test bench, wherein the test object comprises the torque generator (D) as a real component and at least one simulated virtual component, wherein the virtual component of the test object complements the real component of the test object, and the simulation of the virtual component processes the computed propulsion torque (M.sub.v).

12. The method according to claim 11, wherein the torque generator (D) comprises an n-cylinder internal combustion engine and the propulsion torque (M.sub.v) of the n-cylinder internal combustion engine is computed from the inner torque (M.sub.i), which is measured on at least one cylinder (Zn) of the internal combustion engine.

13. A device for determining the propulsion torque (M.sub.v) of a torque generator (D), which is mounted on a test bench, wherein an indicating measurement system (MS) is arranged on the test bench, which measures an inner torque (M.sub.i) of the torque generator (D), and that a correction torque computing unit and a propulsion torque computing unit are provided, wherein the correction torque computing unit estimates a correction torque ({circumflex over (M)}.sub.cor) with the measured inner torque (M.sub.i) from an equation of motion, comprising the measured inner torque (M.sub.i), a dynamic torque (M.sub.dyn) and a shaft torque (M.sub.w) measured on an output shaft of the torque generator (D), and the propulsion torque computing unit computes, from the estimated correction torque ({circumflex over (M)}.sub.cor) and the measured inner torque (M.sub.i), the propulsion torque (M.sub.v) according to the relation M.sub.v={circumflex over (M)}.sub.cor+M.sub.i.

14. The device according to claim 13, wherein the torque generator (D) comprises an n-cylinder internal combustion engine, and on the test bench, on at least one cylinder (Zn), an indicating measuring system (MSn) is arranged.

15. The device according to claim 13, wherein the torque generator (D) comprises an n-cylinder internal combustion engine, and on the test bench, at least one cylinder (Zn) of the n-cylinder internal combustion engine is mounted.

16. The method according to claim 3, wherein a characteristic map of the correction torque ({circumflex over (M)}.sub.cor) is formed over angular velocity ({dot over (φ)}).

17. The method according to claim 7, wherein the target function (Z) contains weighting factors (λ.sub.φ, λ.sub.{dot over (φ)}, λ.sub.M).

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The present invention is explained in the following with reference to FIGS. 1 to 2, which show, as an example, schematically and in a non-limiting way, advantageous embodiments of the invention. In particular

[0031] FIG. 1 is a typical test object configuration on a test bench and

[0032] FIG. 2 is an assembly according to the invention for determining the propulsion torque.

DETAILED DESCRIPTION

[0033] In FIG. 1 an exemplary test bench configuration 1 on a test bench 2 is shown. On the test bench 2 as the real component of the test object a hybrid transmission with a real internal combustion engine 3 and a real electric motor 4 is mounted. “Real” in this case means that these real components are physically present hardware. The internal combustion engine 3 and the electric motor 4 are mechanically connected to each other by a connection shaft 6 with a clutch 7. On the output shaft 8 of the hybrid transmission acts the shaft torque M.sub.2. The output shaft 8 receives from the torque generator D a propulsion torque M.sub.v. A load machine 5 (dynamometer) is mechanically connected with the output shaft 8 via a dynamometer shaft 9 and a clutch 10. The load machine 5 generates, according to the setting of the test run to be executed, a load torque M.sub.D, which influences also the shaft torque M.sub.w.

[0034] In a test bench control unit 11 the test run to be executed is implemented. To this end, in the test bench control unit 11, a simulation model 12 (which may also be comprised of various individual interacting partial models) is implemented, which simulates virtual components of the test object. Virtual components may in this case be, for example, a gearbox, a differential gear, a clutch, the mass inertias of the virtual internal combustion engine, a battery, tires, the vehicle, the environment of the vehicle, the interaction of the vehicle with the environment, etc. The combination of real and virtual components adds up to the test object. Depending on the test run, it is obvious that any test object configurations (real and virtual components) and test bench configurations may be used. For example, in the case of a real four-wheel-drive transmission, four load machines 5, each for a respective side shaft of the transmission, may be provided. However, the invention does not deal with concrete configurations of the test object and of the test bench configuration.

[0035] The test bench control unit 11 also determines, according to the predetermined test run, the control variables Sn for the test bench components and for the test object, with which in particular also the real components of the test object configuration and the load machine 5 are controlled on the test bench 2, as shown in FIG. 1. The test bench control unit 11 may also capture different measurement variables from the test bench or from the real components of the test object, such as for example the speed n, of the internal combustion engine 3, of the electric motor 4 n.sub.E and of the load machine 5 n.sub.D, a rotation angle φ as well as the active shaft torque M.sub.w on the output shaft 8, if a suitable torque sensor is mounted, or alternatively also the load torque M.sub.D or a torque determined by a pendulum support.

[0036] In order to determine the propulsion torque M.sub.v of interest, it is not possible to assume a high quality detected shaft torque M.sub.w, since such a high-quality measurement value is normally absent, as noted earlier. Completely to the contrary, it has to be assumed that the shaft torque M.sub.w and/or the rotation angle φ is present as a low-quality measurement signal, i.e. with a low temporal or value resolution and/or noisy signal. Therefore, according to the invention, the inner torque Mi of the torque generator D (indicated in FIG. 1) is used. This is explained with reference to FIG. 2.

[0037] In FIG. 2, an n-cylinder internal combustion engine 3 as a torque generator D1 is arranged on the test bench 1. The internal combustion engine 3 in this example is a four-cylinder engine. On each cylinder Z1 . . . Zn an indicating measurement system MS1 . . . MSn is arranged. An indicating measurement system MS is known to detect thermodynamic variables of the combustion in the cylinders Z1 . . . Zn, as in particular the change of inner pressure over time or equivalently the change of inner pressure depending on crank angle, from which the inner torque M.sub.iT of internal combustion engine 3 is obtained as a sum torque. The detected indicating measurement variables I1 . . . In are transmitted to the test bench control unit 11. An indicating variable I1 . . . In may already represent the inner torque M.sub.iT of the internal combustion engine 3. Alternatively, the inner torque M.sub.iT may also be computed from the indicating variables I1 . . . In in the test bench control unit 11. In a further alternative, the inner torque M.sub.iT of the internal combustion engine 3 may also be provided by an engine control apparatus ECU to the test bench control unit 11, as shown in FIG. 2, if this can be performed with sufficient speed and precision.

[0038] In case of an electric motor 4 as a torque generator D2, the indicating variables I1 . . . In may for example comprise the electric motor current and the electric motor voltage, which are detected by the indicating measurement system MS and which may then be converted into an inner torque M.sub.E of the electric motor 4 (air gap torque).

[0039] If various interconnected torque generators D1, D2 are present, as shown for example in FIG. 1, the individual inner torques M.sub.iT, M.sub.iE of individual torque generators D1, D2 add with a correct sign in order to form the inner torque M.sub.i. The inner torque M.sub.i of torque generator D is therefore generally obtained as

[00001]$M_i=\underset{D_j}{.Math.}.Math..Math.M_{{\mathrm{iD}}_j}$

with j∈□.sub.I. In the test bench control unit 11, the sum of all indexed torques of individual torque generators D1, D2 is known in real time as inner torque Mi of torque generator D on test bench 2. Therefore, in the following a generalized torque generator D is mentioned, which may be comprised of a plurality of individual torque generators D1, D2.

[0040] On the mass inertia of the torque generator D further torques are also acting, which influence the propulsion torque M.sub.v provided by the torque generator D based on inner torque M.sub.i and which are summed to a correction torque M.sub.cor. These further torques typically cause a reduction of the propulsion torque M.sub.v of torque generator D. Typical is a friction torque M.sub.fric, which detects, for example, the friction effects in the internal combustion engine 3 or in the electric motor 4. The correction torque M.sub.cor may also be comprised of further torques, which act on the mass inertia of the torque generator D. For example, torques M.sub.auxm caused by a number m of secondary assemblies connected to the crank shaft or engine shaft may be considered. Such secondary assemblies may for example be a water pump, an oil pump, a conditioner compressor, a starter-engine/generator, etc. The correction torque M.sub.cor would then be obtained as follows:

[00002]$M_{\mathrm{cor}}=M_{\mathrm{fric}}+\underset{n}{.Math.}.Math..Math.M_{\mathrm{auxn}},$

wherein torques are obviously algebraically introduced (and therefore with their proper sign). In order to consider the correction torque Maw in determining the propulsion torque M.sub.v, in the test bench control unit 11 a correction torque computing unit 14 is provided, in which the correction torque M.sub.cor is computed.

[0041] For the propulsion torque M.sub.v of the generalized torque generator D, due to a simplified definition equation, it must hold that M.sub.v=M.sub.i+M.sub.cor, wherein the torques are algebraic variables and therefore have to be provided with their correct sign.

[0042] In the same way, the Euler equation of motion in the form M.sub.dyn=M.sub.v+M.sub.w (torque equilibrium) has to be considered. The dynamic torque M.sub.dyn is obtained in the simplest case notoriously from J{umlaut over ( )}, where J is the mass moment of inertia acting on the crank shaft of the internal combustion engine 3 or on the shaft of the electric motor 4 and {umlaut over (φ)} is the angular acceleration. The mass moment of inertia J may also be dependent on the rotation angle φ , as typical for a crank shaft, and therefore J(φ). In the same way, the dynamic torque M.sub.dyn may consider not only the generalized angular acceleration {umlaut over (φ)} but other additional dynamic torques, in particular a centrifugal torque in the form of

[00003]$\frac{1}{2}.Math.\frac{\mathrm{dJ}\left(\varphi \right)}{d.Math..Math.\varphi }.Math.{\stackrel{.}{\varphi }}^2,$

which is typical for an internal combustion engine 3, since the mass moment of inertia J varies along a rotation of the crank shaft. The dynamic torque M.sub.dyn may then follow as

[00004]$M_{\mathrm{dyn}}=J\left(\varphi \right).Math.\stackrel{\mathrm{.Math.}}{\varphi }+\frac{1}{2}.Math.\frac{\mathrm{dJ}\left(\varphi \right)}{d.Math..Math.\varphi }.Math.{\stackrel{.}{\varphi }}^2.$

In the same way, the dynamic torque M.sub.dyn could for example also consider the fact that a mass moment of inertia varies, when a clutch 7 between internal combustion engine 3 and electric motor 4 is opened or closed. The equation of motion follows then as

[00005]$J\left(\varphi \right).Math.\stackrel{\mathrm{.Math.}}{\varphi }+\left[\frac{1}{2}.Math.\frac{\mathrm{dJ}\left(\varphi \right)}{d.Math..Math.\varphi }.Math.{\stackrel{.}{\varphi }}^2\right]=\underset{\underset{M_v}{}}{M_i+M_{\mathrm{cor}}}+M_w.$

[0043] In square brackets the optional term of centrifugal torque is provided as mentioned above.

[0044] The rotation angle φ, the angular velocity {dot over (φ)}, or the angular acceleration {umlaut over (φ)} may be measured or may be derived from the detected speed n.sub.(v,E).

[0045] From this, the requested propulsion torque M.sub.v may be derived by measuring the shaft torque M.sub.w, directly from the equation of motion. The problem, however, is the normally very low quality of the measured value of shaft torque M.sub.w, which is frequently noisy and has a low temporal and value resolution. Moreover, also the angular acceleration {umlaut over (φ)} is extremely noisy, since it is not directly measured, but is obtained from the angular velocity {dot over (φ)} by time differentiation, or from the rotation angle φ through double time differentiation. The directly obtained propulsion torque M.sub.v would therefore also be almost useless, for example for use in a simulation, or it should be correspondingly processed (for example by filtering), which however causes a loss of information.

[0046] In order to circumvent this problem, according to the invention another path is followed, in that from the inner torque Mi, which is known, is of high quality and often has a high frequency resolution and is also almost free of dead time, and from the noisy shaft torque M.sub.w and the noisy acceleration signal 0, a high-quality estimate of the correction torque M.sub.cor is first determined. From the above definition of propulsion torque M.sub.v=Mi+M.sub.cor a high-quality (i.e. non-noisy and high-frequency) propulsion torque M.sub.v is then determined. In the test bench control unit 11 a propulsion torque computing unit 13 is provided to this end, which computes the relevant propulsion torque M.sub.v and provides it to other components of the test bench 1, in particular to the simulation by means of simulation model 12 of virtual components of the test object. The inner torque M.sub.i which is directly computed from measurements, provides an additional measurement variable, which allows the determination of both variables, i.e. the correction torque M.sub.cor and the propulsion torque M.sub.v.

[0047] Obviously, the correction torque computing unit 14 and the propulsion torque computing unit 13 may be provided as independent hardware, may be integrated in a single hardware, or may also be provided as software modules in the test bench control unit 11.

[0048] The determination of the correction torque M.sub.cor is based, according to the invention, on an estimate based on the high-quality inner torque M.sub.i and low-quality measurement values for shaft torque M.sub.w and/or rotation angle φ. The estimate may be provided in different ways, as shown in an example in the following by means of advantageous embodiments.

[0049] From the above equation of motion the correction torque M.sub.cor may be computed from the relation M.sub.cor=M.sub.dyn−M.sub.i−M.sub.w. In this case the circumstance that the correction torque M.sub.cor normally varies very slowly over time is exploited. The correction torque M.sub.cor may therefore be considered in certain periods as a quasi-static variable, so that M.sub.corM.sub.cor=const., with an average correction torque M.sub.cor as an estimate {circumflex over (M)}.sub.cor of the correction torque. For example, a friction torque M.sub.fric depends on parameters like temperature, air humidity, age, etc., however these parameters vary only very slowly in time. Secondary assemblies also normally cause a torque M.sub.auxm, which is time-independent. This allows to estimate the correction torque M.sub.cor in that it is averaged over a relatively long period of time (relative to the desired real-time computing). Due to this temporal averaging, irregularities in the shaft torque M.sub.w and dynamic torque M.sub.dyn (for instance due to a noisy measurement of angle of rotation φ) are also averaged out and the influence of such irregularities is reduced. If for example the averaging is done over a working cycle of a four-cylinder internal combustion engine,

[00006]${\int }_{\varphi =0}^{4.Math..Math.\pi }.Math.\left(J\left(\varphi \right).Math.\stackrel{\mathrm{.Math.}}{\varphi }+\left[0.5.Math.\frac{\delta .Math..Math.J\left(\varphi \right)}{\mathrm{\delta \varphi }}.Math.{\stackrel{.}{\varphi }}^2\right]\right).Math.d.Math..Math.\varphi -{\int }_{\varphi =0}^{4.Math..Math.\pi }.Math.\left(M_i+M_w\right).Math.d.Math..Math.\varphi =4.Math..Math.\pi .Math..Math.{\stackrel{\_}{M}}_{\mathrm{cor}}.Math.$

follows from the equation of motion, from which the correction torque {circumflex over (M)}.sub.cor may be estimated as an average correction torque M.sub.cor. In the square brackets the optional term of centrifugal torque is provided as mentioned above. It is hereby sufficient to estimate the average correction torque M.sub.cor once, for example, and then retain the same for the following cycle, or a plurality of following cycles. Alternatively, for the average correction torque M.sub.cor, a model for different angular velocities may be construed, which is continuously estimated and corrected. In the same way, the average correction torque M.sub.cor may also be computed continuously, in the form of a shifting average value. Instead of a working cycle the averaging may also be provided over any other period (time or angle).

[0050] The integral

[00007]${\int }_{\varphi =0}^{4.Math..Math.\pi }.Math.M_i.Math.d.Math..Math.\varphi$

may be set equal to the variable indicated as Indicated Mean

[0051] Effective Pressure (IMEP) in the indicating measurement technique, which is a variable which is normally directly provided by the indicating measurement or which is provided in an engine control ECU.

[0052] The model of the correction torque M.sub.cor normally does not depend on time, or depends on it only in a very slowly way. The correction torque M.sub.cor may, however, depend on the angular velocity {dot over (φ)}, in the form M.sub.cor({dot over (φ)}). Also in this case the correction torque M.sub.cor may be easily estimated from the equation of motion, if for example the estimated correction torque {circumflex over (M)}.sub.cor({dot over (φ)}) is written as the sum of a basic correction torque M.sub.cor,0 and a term κ{dot over (φ)} dependent from the angular velocity {dot over (φ)}, so that {circumflex over (M)}.sub.cor({dot over (φ)})=M.sub.cor,0+κ{dot over (φ)}. The term M.sub.cor,0 and parameter κ vary only very slowly over time. From the equation of motion it follows then again

[00008]${\int }_{\varphi =0}^{\theta }.Math.\left(J\left(\varphi \right).Math.\stackrel{\mathrm{.Math.}}{\varphi }+\left[0.5.Math.\frac{\delta .Math..Math.J\left(\varphi \right)}{\mathrm{\delta \varphi }}.Math.{\stackrel{.}{\varphi }}^2\right]\right).Math.d.Math..Math.\varphi -{\int }_{\varphi =0}^{\theta }.Math.\left(M_i+M_w\right).Math.d.Math..Math.\varphi ={\int }_{\varphi =0}^{\theta }.Math.\left(M_{\mathrm{cor},0}+\kappa .Math.\stackrel{.}{\varphi }\right).Math.d.Math..Math.\varphi ,.Math.$

from which both variables M.sub.cor,0 and parameter κ may be computed. To this end, either the integration limit θ or the angular velocity {dot over (φ)} may be varied, wherein at least two variations are necessary in order to compute both variables.

[0053] It is readily apparent that with above-mentioned averaging of the equations of motion it is also possible to estimate a characteristic map (model) for the correction torque {circumflex over (M)}.sub.cor depending on the angular velocity {dot over (φ)}, which may then be used for computing the propulsion torque M.sub.v.

[0054] In this way, obviously, other dependencies of the correction torque M.sub.cor may be considered, in that the estimated correction torque {circumflex over (M)}.sub.cor is completed with further or other terms. Instead of the above linear relation M.sub.cor,0+κ{dot over (φ)} it is for example possible to determine online, a more complex, in particular non-linear, model for the correction torque M.sub.cor as a function of the crank angle φ and/or the angular velocity {dot over (φ)} or even the angular acceleration {umlaut over (φ)}, which may also be continuously corrected online.

[0055] Obviously, from known variables M.sub.dyn, M.sub.i, M.sub.w, a mathematical model for estimating the correction torque {circumflex over (M)}.sub.cor as a function of crank angle φ and/or angular velocity {dot over (φ)} or speed n may be trained, for example in the form of a neuronal network. The parameters of a physical model of the estimated correction torque {circumflex over (M)}.sub.cor could also be determined as a function of measured variables, for example with known methods of parameter estimation.

[0056] An estimated value of the correction torque {circumflex over (M)}.sub.cor may also be estimated from the known inner torque M.sub.i, according to the invention, by means of a state observer. The general procedure is again based on above mentioned equation of motion of form

[00009]$J\left(\varphi \right).Math.\stackrel{\mathrm{.Math.}}{\varphi }+\left[\frac{1}{2}.Math.\frac{\delta .Math..Math.J\left(\varphi \right)}{\mathrm{\delta \varphi }}.Math.{\stackrel{.}{\varphi }}^2\right]=\underset{\underset{M_v}{}}{M_i+M_{\mathrm{cor}}}+M_w.$

[0057] If estimated values are indicated by “̂”, the equation of motion may be written in the following form.

[00010]$J\left(\hat{\varphi }\right).Math.\stackrel{\mathrm{.Math.}}{\hat{\varphi }}+\left[\frac{1}{2}.Math.\frac{\delta .Math..Math.J\left(\varphi \right)}{\mathrm{\delta \varphi }}.Math.{\stackrel{.}{\hat{\varphi }}}^2\right]=M_i\left(t\right)+{\hat{M}}_{\mathrm{cor}}\left(t\right)+M_w\left(t\right)$

[0058] For that, an arbitrary target function Z is defined as a function of an estimated rotation angle {circumflex over (φ)}, or its temporal derivatives {circumflex over ({dot over (φ)})} and {circumflex over ({umlaut over (φ)})}, and of the estimated correction torque {circumflex over (M)}.sub.cor,which is minimized, Z.fwdarw.min.

[0059] As a target function Z, an integral in the form

[00011]${\int }_{t=0}^{T_{\max }}.Math.\left\{{{\lambda }_{\varphi }\left({\varphi }_m\left(t\right)-\hat{\varphi }\left(t\right)\right)}^2+{\lambda }_M.Math.{\hat{M}}_{\mathrm{cor}}^2\left(t\right)\right\}.Math.\mathrm{dt}=\min .Math..Math.\mathrm{or}.Math..Math.\mathrm{an}.Math..Math.\mathrm{integral}.Math..Math.\mathrm{in}.Math..Math.\mathrm{the}.Math..Math.\mathrm{form}.Math.$${\int }_{t=0}^{T_{\max }}.Math.\left\{{{\lambda }_{\varphi }\left({\varphi }_m\left(t\right)-\hat{\varphi }\left(t\right)\right)}^2+{{\lambda }_{\stackrel{.}{\varphi }}\left({\stackrel{.}{\varphi }}_m\left(t\right)-\stackrel{.}{\hat{\varphi }}\left(t\right)\right)}^2+{\lambda }_M.Math.{\hat{M}}_{\mathrm{cor}}^2\left(t\right)\right\}.Math.\mathrm{dt}=\min$

is used, wherein “m” indicates measured variables and with weighting factors λ.sub.φ, λ.sub.{dot over (φ)}, λ.sub.M.

[0060] The weighing factors λ.sub.φ, λ.sub.{dot over (φ)}, λ.sub.M are manually defined or may be defined by known mathematical methods. The determination of weighing factors λ.sub.φ, λ.sub.{dot over (φ)}, X.sub.M by means of already known Kalman filtering is mentioned, for example, as described for instance in the document by S. Jakubek, et al., “Schätzung des inneren Drehmoments von Verbrennungsmotoren durch parameterbasierte Kalmanfilterung”, Automatisierungstechnik, 57 (2009) 8, pages 395-402. The Kalman filtering has the advantage, in this case, that with it the quality of the measured values is considered in determining the weighing factors λ.sub.φ, λ.sub.{dot over (φ)}, λ.sub.M, which is very advantageous in the application according to the invention, in which very noisy or not well-resolved measured values may be present.

[0061] It is expressively to be noted that the above-mentioned target function is only an example, and that also any other target function Z could be equally used. In particular in the target function Z, temporal derivatives of the correction torque {circumflex over (M)}.sub.cor could also be contained.

[0062] The searched estimate value for the correction torque {circumflex over (M)}.sub.cor is then determined by minimization (optimization) of target function Z. To this end a variety of methods are known, which cannot all be mentioned in this document. An example is an analytical solution of the optimization problem, which may be derived for instance when using linear target functions Z (Ricatti-equation). Also iterative methods may be used, as described in the following.

[0063] To this end, at the beginning an initial value for correction torque {circumflex over (M)}.sub.cor is provided. From the equation of motion in each iteration step the estimated rotation angle {circumflex over (φ)}, or its temporal derivatives {circumflex over ({dot over (φ)})} and {circumflex over ({umlaut over (φ)})} are calculated. This may happen algebraically. With the estimated rotation angle {circumflex over (φ)}, or its temporal derivatives {circumflex over ({dot over (φ)})} and {circumflex over ({umlaut over (φ)})}, from the optimization of the target function Z(t), a new estimated value for the correction torque {circumflex over (M)}.sub.cor is computed and the above-mentioned steps are repeated, until a predetermined stop criterion for optimization is fulfilled. The estimation of the correction torque {circumflex over (M)}.sub.cor may be performed online in a continuous way during a test run.

[0064] It is also possible that with the estimate of the correction torque {circumflex over (M)}.sub.cor a model for the correction torque {circumflex over (M)}.sub.cor is trained, for example in the form of a neuronal network. With such a model, then, depending on determined variables, such as for example an angular velocity {dot over (φ)}, the correction torque {circumflex over (M)}.sub.cor for a test run may be determined. The model may obviously be updated continually with update measurement values and the above-mentioned method.

[0065] In this context it is also known that in the optimization boundary conditions for the variables of the target function may be defined, which are considered in the optimization.

[0066] Then, with the estimated value of the correction torque {circumflex over (M)}.sub.cor, which is determined with the above-mentioned method and is therefore known, from the above equilibrium equation, the propulsion torque M.sub.v of the torque generator D may be determined in the form M.sub.v={circumflex over (M)}.sub.cor+M.sub.i.

[0067] This allows to provide the propulsion torque M.sub.v for a test run, but also for other applications, in particular for simulations in a simulation model 12. This calculation takes place for the test run in predetermined periods of time, for instance every millisecond or every 1 degree of rotation angle φ, i.e. in real time. Therefore, the propulsion torque M.sub.v is available in each desired time step, for example in order to be processed in a simulation model 12 for a virtual component of the test object.

[0068] Besides the propulsion torque M.sub.v, the measured shaft torque M.sub.w may also be rendered plausible. From the knowledge of the propulsion torque M.sub.v and of the dynamic torque M.sub.dyn, from the above equation of motion, an estimated/computed shaft torque {circumflex over (M)}.sub.w may be determined. In this way, the measurement of the shaft torque M.sub.w may be rendered plausible, for instance in order to identify a shaft rupture on the test bench 2. It is also possible to correct the measured shaft torque (noisy and/or imprecise) M.sub.w or it may be replaced with the computed estimated shaft torque {circumflex over (M)}.sub.w. For a simulation in a simulation model 12 or for other components of the test bench 2, it is therefore possible to provide a better-quality shaft torque M.sub.w.

[0069] Due to the knowledge of a correction torque M.sub.cor, in fact an estimated value for the correction torque {circumflex over (M)}.sub.cor, it is possible to study different influences on the propulsion torque M.sub.v on the test bench 2. In particular, it is possible to analyze the influence of different torques considered in the correction torque M.sub.cor.

[0070] As an example, a certain test run is assumed in which an internal combustion engine 3 is operated on the basis of settings of the test run and wherein the exhaust gas emissions are measured. It would now be possible to analyze how the gas emissions vary, when another conditioner compressor (which is simulated as the virtual component of the test object) or when a different lubrication oil (for example through correction factors in the determination of the correction torque M.sub.cor) is used. These analyses may be executed without the respective components (in this case the conditioning compressor or the lubrication oil) having to be present in real. It is sufficient that these components are virtually present, which represents a great simplification in the development of a test object. In particular this is also due to the fact that at the time of first test runs, often not all components which are driven by the torque generator D are actually available.

[0071] With the procedure according to the invention, another test scenario is also feasible. If in the simulation model 12 a virtual complete vehicle with a multi-cylinder internal combustion engine is simulated, at the interface between the computing unit 13 and the simulation model 12 the propulsion torque M.sub.v of the multi-cylinder internal combustion engine is expected. If, however, on the test bench 1 only a single cylinder internal combustion engine is mounted, a test run may nonetheless be performed. The missing cylinders are simulated in the computing unit 13. This takes place, in the simplest case, by multiplication of all measurement variables of the actually mounted cylinder with a corresponding factor and if necessary also with a corresponding phase shift and correction of the dynamic torques M.sub.dyn (in particular in the case of an internal combustion engine). This is particularly interesting in the development of large engines, for instance ship engines with a high number of cylinders, whereby first test runs are possible even before the large engine is built as a whole.

[0072] The simulation of missing cylinders may also be necessary if on the test bench 2 not all cylinders of the internal combustion engine 3 are provided with an indicating measurement system M. In this case the cylinders without the indicating measurement system M would be simulated. The simulation model 12 would therefore always receive the propulsion torque M.sub.v of the expected multi-cylinder engine, possibly with all secondary assemblies.

[0073] This has also the inestimable advantage that on the test bench 2, the interfaces, for example for components of the simulation model 12, may be unchanged, regardless of which components of the test object are real or virtual.

[0074] Equally, the method according to the invention may be expanded with further degrees of freedom of movement. In this case one would not assume an equation of motion in one degree of freedom of movement, here the rotation angle φ, but a system of equations according to the number of degree of freedom of movement. This is for instance interesting when the torque generator D is modeled with a non-rigid suspension, such as for instance in an internal combustion engine 4 in a vehicle. The active forces or torques cause, due to machine dynamics, also a corresponding movement of the torque generator D with respect to the vehicle. In this way, multidimensional equations of motion would be obtained, which in the sense of the invention are considered as the above-mentioned equation of motion. Nothing is therefore changed in above-mentioned procedure according to the invention.