DE10214833A1 - Method for determining an indicated actual engine torque for an internal combustion engine, using Fourier analysis to calculate appropriate Fourier coefficients from a signal for crankshaft speed - Google Patents

Abstract

An engine speed signal (22) is derived from an internal combustion engine (46) and subjected to Fourier analysis (48) to produce characteristics (50) in the form of first-order Fourier components that are passed along with other engine parameters, e.g. manifold passage pressure (52), to a neuronal network (54) that calculates indicated average pressure. Rotary speed of a crankshaft is scanned via a crankshaft angle to produce a signal for crankshaft speed. The indicated actual engine torque is calculated from the Fourier coefficients.

Description

The invention relates to a method for determining an indexed actual engine torque an internal combustion engine with a crankshaft, according to the preamble of Claim 1.

The thermodynamic processes in the internal combustion engine of a motor vehicle, due to the stricter exhaust gas limit values and those to be reduced Fleet consumption is a technical process that involves control engineering new requirements with regard to modeling are constantly being made. Various process variables are not already used directly in engine control units measured, but determined via a model. Of these indirectly determined Process variables, torque is of the greatest importance, since it is only available on the one hand high effort can be measured and on the other hand a central parameter represents the coordination of the requirements for the engine and the control interventions performs. A setpoint for the engine torque is derived from the internal and external Torque requirements are determined and compared with the current actual engine torque, which is based on a torque model in the form of stored tables and Characteristic curves is determined. Due to the diverse dependencies, the Parameterization or application of these tables difficult because of constant external Conditions with variation of only one input variable have to be established. So relevant parameters such as engine speed, Air mass flow and ignition angle kept constant and the influence of Air-fuel ratio λ determined by stepwise variation. Then will another size varies under otherwise constant conditions. That kind of Parameterization presupposes that only derivatives of the target variable according to the Input variables are available. The system has to vary all Behavior in the same way as for the additive overlay of different ones Stationary states with modification of only one size. However, there are additional partial derivatives or other cross-couplings before, so the overall system this approach no longer correctly captures what is happening in dynamic operation Leads to model errors.

Conventional torque models use various measuring and Actuating variables that influence the torque as parameters of the combustion to have. With an increasing number of adjustment options (continuous Camshaft adjustment, intake manifold changeover etc.) automatically takes the number the input variables as well as the parameterization effort to the actual torque to be able to determine exactly. Aging and cylinder-specific remain Differences not taken into account. An alternative are torque models that Analyze the result of the combustion instead of the various influences to be included in the combustion. The most reliable method is the Measurement of the cylinder pressure and the thermodynamic calculation of the indicated Work represents. However, this method has the disadvantage that the cylinder pressure sensors are costly and complex.

DE 199 42 144 A1 describes a method for adaptive vibration damping known by means of neural networks. Here the course of the vibrations is indicated by a intelligent method for nonlinear function approximation, such as a "Harmonically activated neural network", in the spectral range for variable Operating points identified and reproduced for compensation. It will Frequency components of the torque vibrations determined and via an electrical Machine supplied to the crankshaft in opposite directions, so that there is a compensation of Sets torque vibrations.

DE 199 62 799 A1 describes a system and a method for detecting Engine misfires using frequency analysis. in this connection an angular acceleration of the crankshaft is detected and a discrete one Fourier analysis (DFT). The frequency components derived from this are compared to a predetermined reference value to misfire the Detect internal combustion engine.

The present invention has for its object a method of to improve the above-mentioned type regarding the realism of the torque model and Simplify the application at the same time.

This object is achieved by a method of the above. Kind of with the in Characteristics characterized claim 1 solved. Advantageous embodiments of the Invention are specified in the dependent claims.

For this purpose, it is provided according to the invention that a course of rotation of the crankshaft over the Crank angle sampled and from this scan a crankshaft speed signal the crank angle is generated that from the crankshaft speed signal by means of a Fourier analysis corresponding Fourier coefficients are calculated and that from the Fourier coefficient, the indexed actual engine torque is calculated.

This has the advantage that the actual engine torque taking into account aging and Evaluation of the combustion result is determined, but none complex devices for measuring the cylinder pressure are required.

A cylinder-specific determination of the indicated actual engine torque is achieved in that the Fourier analysis for the crankshaft speed signal between two Ignitions is applied.

In a preferred embodiment, the Fourier coefficients and the at least one motor parameter is fed to a neural network as an input variable, wherein the neural network determines an indexed actual mean pressure from which a motor constant, the indexed actual motor torque is calculated. The weights of the neural networks are preferred by comparison with measured ones Cylinder pressure data from which the indexed engine torque is calculated, predetermined. The input values are expediently changed with each ignition cycle Internal combustion engine supplied to the neural network. Particularly advantageous in the Application of a neural network is that contrary to the parameterization of others model-building procedures no steady states in the training phase of the neural network must be set to obtain suitable data. Instead of rather, data sets from transient processes can also be used become. This results in the proximity of the model to the real state significant improvements. Accordingly, the simplified Parameterization.

To calculate the indicated actual engine torque, at least one is also used other engine parameters used, such as engine speed, Air mass flow and / or intake manifold pressure includes.

Further features, advantages and advantageous configurations of the invention result from the dependent claims, as well as from the following description of the Invention with reference to the accompanying drawings. These show in

Fig. 1 is a graphical representation of a high-resolution detection of the rotational speed of the crankshaft,

Fig. 2 is a graphical representation of the Fourier Amplitude 1. Order of the speed signal above the indicated medium pressure in bar at various loads and a speed of 3,500 rpm.

Fig. 3 is a graphical representation of the Fourier Amplitude 1. Order of the speed signal above the indicated mean pressure in bar at various loads and a speed of 4,500 rpm.

FIG. 4 shows a schematic block diagram of an exemplary embodiment of a neural network and used according to the invention

Fig. 5 is a schematic block diagram of an exemplary application of the inventive method.

Fig. 1 shows the course of the engine speed of an internal combustion engine. In this example, a high-resolution speed measurement is carried out at idle speed on a 4-cylinder engine with a 60-2 lock washer on the crankshaft. The sampling points are plotted on a horizontal axis 10 and the speed value is plotted on a vertical axis 12 . A load change occurs at 14 , reference number 16 denotes a working cycle of 720 ° KW, at 18 there is a tooth gap and 20 denotes the 60-2 tooth lock washer. As can be seen directly from FIG. 1, a periodic fluctuation of the speed signal or crankshaft signal or crankshaft speed signal 22 occurs when the measurement is correspondingly high-frequency scanning, the main cause of which can be seen in the compression and expansion of the cylinder in the combustion phase. To determine the torque using a model approach, suitable features are extracted from the high-resolution measured speed signal 22 , such as the speed difference within a crankshaft segment, one segment being 180 ° KW in the 4-cylinder, 4-stroke engine under consideration. This speed difference is not to be confused with the speed difference of successive segments, also known as smoothness or uneven running, which represents a measure of the comfort of the vehicle, but does not allow any conclusions to be drawn about the individual combustion in the individual cylinders. Corresponding input variables for a neural network are obtained from the speed signal 22 by means of a discrete Fourier analysis, a corresponding speed model being parameterized using corresponding correlation measurements with the indicated mean pressure.

The speed signal 22 could serve directly as an input for a model. With the high-resolution measurement of the speed signal 22 , however, large amounts of data arise and a corresponding computing effort is necessary. It is therefore particularly advantageous, instead of using the sensor data directly, ie the speed signal 22 , as inputs of an empirical model to form features from the measured data as possible input variables of a model. Such a suitable feature forms, for example, a differential speed. To form this differential speed, two crank angle positions are determined at the beginning and at the end of the expansion phase of the respective cylinder, for example 20 ° KW and 70 ° KW, and their associated speed values are subtracted from the speed signal 22 . This method requires less computing effort and also has a good correlation to the indicated medium pressure at low speeds, with cylinder-specific differences occurring. However, the correlation factor between the differential speed and the indicated mean pressure deteriorates with increasing speed, which is probably due to a superimposition of the indicated gas torque with the mass torque, the amplitude of which increases in proportion to the speed. Even at 2,000 rpm, the spread is so high that exact assignment is no longer possible.

According to the invention, a Fourier analysis of the speed curve is carried out, Fourier amplitudes being determined, for example, of the 1st order. Surprisingly, even at high speeds, there is a small scatter band in the correlation between the Fourier amplitude 1 . Order and the indicated medium pressure, ie a good correlation, so that an exact allocation to the indicated medium pressure is possible. This can be seen from FIGS. 2 and 3, in which the Fourier amplitude 1 . Order is plotted on the vertical axis 24 above the indicated mean pressure in bar on the horizontal axis 26 . The determined Fourier amplitudes are 1 in FIG. 2. Order shown at 3,500 rpm and various loads 28 , 30 , 32 , 34 . The determined Fourier amplitudes are 1 in FIG. 3. Order shown at 4,500 rpm and various loads 28 , 30 , 32 , 34 .

FIG. 4 shows an exemplary structure of a speed model with its input variables in the form of engine speed 36 , intake manifold pressure 38 and Fourier components 40 of the speed signal 22 . A value for the indicated mean pressure is output at the output 42 , from which the indicated actual engine torque is then calculated using a motor constant. The inputs 36 , 38 and 40 are fed to the neural network with each ignition cycle of the internal combustion engine, which corresponds to 180 ° KW for a 4-cylinder, 4-cycle engine. Subsequently, the inputs 36 , 38 and 40 are used to calculate the indicated mean pressure, which differs from the indicated actual engine torque only by an engine constant. The parameters or weights 44 of the neural network are predetermined offline using a suitable parameterization method. The parameter sets are different for each cylinder and can therefore take into account the cylinder-specific torque generation.

Fig. 5, the application illustrates schematically the inventive method in an internal combustion engine 46 from the derived speed signal 22 and the Fourier analysis 48 is supplied. The features 50 obtained from the Fourier analysis 48 , for example in the form of Fourier components 1 . Order, possibly together with other engine parameters, for example the intake manifold pressure 52 , are fed to the neural network 54 , which calculates the indicated mean pressure, which is converted with the engine constant to the indicated actual engine torque 56 . At 58 , the indexed actual engine torque 56 obtained from the neural network 54 is compared with an indexed target engine torque 60 , which is derived, for example, from an accelerator pedal position and possibly other variables. A torque controller 62 and an EGAS controller 64 act in a corresponding manner on an actuator 66 for changing the engine torque, such as a throttle valve. The position of the actuator 66 then causes an intake manifold pressure 52 downstream of an intake manifold 68 .

Test trials in which the measured indicated mean pressure with the initial values of the neural network for the indicated mean pressure have shown that the deviation (error) between measured and from the model determined indicated mean pressures moved in a band of +/- 5%.

Claims (8)

1. A method for determining an indexed actual engine torque of an internal combustion engine with a crankshaft, characterized in that a course of rotation of the crankshaft is sampled via the crank angle and a crankshaft speed signal is generated via the crank angle from this scan that corresponding Fourier coefficients are calculated from the crankshaft speed signal by means of a Fourier analysis and that the indexed actual engine torque is calculated from the Fourier coefficients. 2. The method according to claim 1, characterized in that the Fourier analysis on the crankshaft speed signal is applied between two ignitions. 3. The method according to claim 1 or 2, characterized in that the Fourier coefficients and the at least one motor parameter a neural Network are supplied as an input variable, the neural network one indexed actual mean pressure is determined, from which the indexed actual engine torque is calculated. 4. The method according to claim 3, characterized in that weights of neural network by comparison with measured cylinder pressure data, from which the indexed engine torque is calculated, predetermined. 5. The method according to claim 3 or 4, characterized in that the Input values for each ignition cycle of the internal combustion engine in the neural network be fed. 6. The method according to at least one of the preceding claims, characterized characterized that additionally for the calculation of the indicated actual engine torque at least one other engine parameter is used. 7. The method according to claim 6, characterized in that the at least one Engine parameters include engine speed, air mass flow and / or intake manifold pressure. 8. The method according to at least one of the preceding claims, characterized characterized in that the Fourier analysis is a discrete Fourier analysis.