DE10028878A1 - Identification and evaluation of leaks in suction pipe of IC engines - Google Patents

Abstract

The identification and evaluation method is carried out in 3 steps. Th steps involved are: Measuring of air mass flow behavior and the associated behaviors of further measurement, adjustment and regulating variables of the electronic engine control, during a time span in the different rpm and engine loads. Computing theoretical air mass flow behavior during the time span, from the adjusting, measuring and-or regulating variables of the electronic engine control. Forming a differential curve between the theoretical air mass flow behavior and the measured air mass flow behavior, and using the differential curve for the computing of an effective leak cross section in the suction pipe.

Description

The invention relates to a method for detecting and evaluating leaks in the intake manifold of internal combustion engines.

If there is a leak in the intake manifold in an internal combustion engine Throttle valve and the cylinders is located, so-called "false air" is sucked in. This means that the amount of air entering the cylinders can no longer be precisely regulated , which means that an exact setting of the lambda value is no longer possible. A precise setting of the lambda value is for a flawless and in particular but also low-pollution engine operation is of central importance.

In conventional or serial control units, there are leaks in the intake manifold only as a general error in the area of mixture formation / lambda control detected. The diagnostic procedures applied are based on a. on the adaptation factors of the lambda control loop. A differentiation between the multitude of possible The causes of errors have not yet occurred. With previous diagnostic procedures cannot be determined whether the registered error is due to a suction pipe leak or to another Cause is due. An estimate of the size of a possibly in the intake manifold Leakage is not possible. However, such knowledge is relevant when it is to be decided whether and, if so, which repair measures are seizing.

Based on this prior art, the object of the invention is a method for To create leak detection in the intake manifold of an internal combustion engine, by means of whose leaks in the intake manifold are reliably detected and their size is estimated can.

This object is achieved by a method with the features of claim 1.  

The method according to the invention is based on the physical effect that the Leakage current due to a constant leak cross section at low intake manifold pressures is particularly large, while the regular way in such operating points sucked air mass is relatively small. At high intake manifold pressures, however, is the leakage mass flow flowing through the leak cross section is particularly low, while the regular intake air volume takes very high values.

This effect is used to the effect that over a certain period of time in which different speeds occur, the air mass flow behind Throttle valve measured in the intake manifold and with a calculated Air mass flow curve is compared. The resulting difference course is then used to estimate the effective leak cross section.

A restriction to a finite number of discrete values within the courses of the Measured or calculated quantities according to claim 2 contributes to the arithmetic Manageability of the problem.

The arrangement of the discrete values in a grid according to claim 3 and the subtraction of values averaged over all or more grid fields contributes to the robustness of the Diagnostics for measurement data with errors.

Further advantages result from the following subclaims. In particular through the use of neural approximators according to claim 11 in connection with Claim 8, the numerical computational effort can be reduced so far that the Estimation of the leak cross section according to the invention also on an engine ECU low computing power can be performed.

The invention will now be described in more detail using an example with reference to the figures described.

Show it:

Fig. 1 is a schematic flow diagram of the signal preprocessing and

Fig. 2 is a schematic flow diagram of the optimization of the focus.

The diagnostic procedure is in signal preprocessing and neural pattern recognition structured.

The signal preprocessing SVV is shown schematically in FIG . The engine speed n _VM (t), the injection time t _EV (t), the lambda voltage U _λ (t) and the measurement signal L (t) for measuring the air mass flow in the intake manifold are used as input variables RD.

The signal from the engine speed sensor, the lambda probe voltage, the current injection times and the measurement signal from the air mass flow sensor are evaluated. As an alternative to the signal from an air mass sensor, the signal from an intake manifold pressure sensor can also be used. The air mass flow _{L, DK} can be calculated both from the signal from an air mass sensor and from the signal from an intake manifold pressure probe. Since the four necessary signals are available in all common electronic motor controls, no additional sensors are required to implement the diagnostic function.

In a first method step S1, static values are determined from the dynamic courses of the input variables. This is accomplished as follows:
Both during the dynamic driving cycles used for exhaust gas tests and during normal road driving, there are phases in which speed and load remain approximately constant over several work cycles. Such phases are suitable for determining static values. To determine such phases, barriers must be defined for the speed and the load value, in which the variables belonging to the same static support point, which is a pair of values from speed and air mass flow, may move. In the first method step S1, the mean values of the variables speed, measured air mass flow, average injection time and the air ratio measured on the lambda sensor are formed for such “static” support points in the control device.

When creating a diagnostic tool, a speed / load range that is relevant for the pattern recognition that takes place in a later method step can first be defined. The size "load" z. B. characterized by the measured air mass flow. The selected speed / load range should be selected according to the following criteria:

• - The selected area should include both the legally specified emissions test driving cycles, as well as in "normal" everyday operation as well as possible with static ones Support points must be covered.
• - The slope of the leakage mass flow change should be in the selected range be as large as possible.

In a second method step S2, the theoretical total air mass flow _{L, th} into the cylinders is estimated from the static values of the speed n _VM , the injection time t _EV and the air ratio λ determined during a driving cycle, which lie in the relevant load / speed range. For example, a Static Neural Network (SNN) of the following form can be used for this:

_{L, th} = SNN (n _VM , t _EV ) .λ (1)
with λ = f (U _{λ , VK} ),

where U _{λ , VK is} the lambda voltage of a lambda probe, which is located in the raw exhaust gas flow, that is, in front of a possibly present catalyst.

The difference between the estimated theoretical air mass flow in which the cylinders _{L, th} and the measured air mass flow _{L, DK} corresponds to the _leak mass flow _{L, leak} :

_{L, leak} = _{L, th} - _{L, DK} (2)

This difference is calculated in the third method step S3.

For the sample evaluation in the selected area, in the fourth method step S4 both the speed axis and the air mass flow axis are divided into a number of sections to be selected, which span a grid in the speed / air mass flow level. The centers of the grid fields are through the i × j matrix with ij = m

given, the n _{1.. . i} selected speeds and the _{L, 1. . . j are} selected air mass flows.

In method step S5, each _leak mass flow _{L, leak} estimated according to equation (2) is assigned to the grid in M, to the center of which the load value formed from the speed and air mass flow has the smallest distance. If there is at least one estimated _leak mass flow value _{L, leak} for each of the m grid fields in (3), then Δ _{L, 1} is used to determine the air mass flow difference values _{. . . m} in the m grid mean points from the m _leak mass flow values _{L, leak} requires an interpolation function. Empirical studies have shown that the functional relationship

_{L, leak_1. . . m} = f (n _{VM_1... m} , t _{EV_1... m} ) (4)

the special form of the plane equation

_{L, leak_1. . . m} = at _{VM_1. . . m} + bt _{EV_1. . . m} (5)

is more suitable than the general level equation

_{L, leak_1. . . m} = at _{VM_1. . . m} + bt _{EV_1. . . m} + c. (6)

The parameters a and b of the compensation plane can be determined using the "least squares" method. Using the level equation (5) obtained, the estimated air mass flow difference values Δ _{L, 1. . . m} are calculated in the grid center points in method step S6 (cf. matrix M from equation (3)):

Test bench measurements and practical tests have shown that air mass flow values and Injection times in many cases still involve offsets. A reason for one such an offset can e.g. B. be a temperature drift of the air mass meter, either is caused by a relatively high or a relatively low intake air temperature. A number has similar effects as an offset directly on the air mass meter other factors, such as dirt on the injection valves, different Fuel qualities, etc. If the fuel quality is poorer, there is a Increase in injection time. This increase should contribute to the same air mass values different speeds have approximately the same percentage value. Unfortunately here the non-linear characteristic of the injection valves prevents an exactly the same percentage increase in injection time.

The problem of offsets is therefore solved in the following way: The air mass flow values _{L, DK} associated with the air mass flow difference values Δ _L on the air mass meter are constant line by line in the matrix Δ _L from equation (8) (see equation (3)). An equal offset of the air mass measurement for the same air mass flow values _{L, DK} therefore has the same effects on all values of a line. Such offsets can be eliminated by subtracting the mean of all elements in a row from the individual row elements:

This takes place in step S7.

The matrix elements Δ _{N, 1} thus obtained _{. . . i, 1. . . j} should be used to estimate the effective leak diameter using a neuroapproximator. However, it is not recommended to use all matrix elements of Δ _N as inputs of the neuroapproximator, but to combine several matrix elements into one input each. If all matrix elements were used as individual inputs, the problems known as "curse of dimensions" would arise. In step S8, the dimensions of the pattern are therefore reduced.

For the scanning of the pattern obtained with a freely selectable number of n _E neuro-approximator inputs, e.g. B. a corresponding number of focal points in the examined speed / air mass flow range can be set arbitrarily. All elements of the Δ _N matrix are weighted in each center of gravity according to the distance to this center of gravity with a Gauss curve according to equation (9). The following applies to the red entrance:

Air mass flow centers of the n _E inputs:
_{S, 1} , _{S, 2} , _{S, 3,.} , , _{S, nE}
Speed focus of the n _E inputs:
n _{S, 1} , n _{S, 2} , n _{S, 3,.} , , n _{S, nE}
Standardization of the air mass flow:

Standardization of the speed:

For each center of gravity, a weighting matrix W _{r is obtained} with which the elements of Δ _{N are} weighted. Since these weighting matrices are kept constant after they have been defined in this way, the matrix values can be stored in the control unit and do not have to be calculated online.

The following calculation applies to the value of the rth input:

il | r = trace (W _r ^T .Δ _N ) (10)

Signal preprocessing is now complete. The values obtained here (cf. Equation 10) are now classified, that is, they become the size of one possible leaks estimated. Such a classification is always included different methods and procedures possible. That also includes  Pattern recognition method because, as described above, the size of the Leakage mass flow has typical courses depending on the speed.

One way to classify is to use neuro classifiers. Such neuro-classifiers, which, for. B. on a simple static multi-layer Perceptron based, can be parameterized automatically and are for the present Task particularly suitable. The structure of the multi-layer perceptron - especially that Parameters "number of inputs" and "number of weightings" - must be used can be optimized with the help of empirical studies. As the output variable of the Multi-layer perceptrons can use the cross-sectional area of the leak. There the estimated leak cross-section is faulty, that of the neuro-approximator is required Estimated leakage cross-section usually after signal processing. On The control unit should only diagnose a certain diagnosed leak cross section as an error are displayed if the diagnosed leak cross-section is sufficient Safety margin to the estimation error. Furthermore, there is also a multiple It makes sense to repeat the diagnosis with new measured values. The different Estimates can be statistically evaluated.

It is now shown how the location of the focal points can be optimized. As shown in FIG. 2, the raw data are first fed to the signal preprocessing (see also FIG. 1). Both training and validation data are generated using signal preprocessing. The neuroclassifier described above is trained with the training data, that is to say the neuroclassifier learns the patterns to be recognized by it later. The neuroclassifier is then tested with the validation data. The results of this test are subjected to a genetic algorithm. If the results are unsatisfactory, the focus of the signal preprocessing is shifted and the cycle is repeated. The loop is terminated if the results of the neuroclassifier test are of a sufficiently high quality.

Claims (11)

1. Procedure for the detection and evaluation of leaks in the intake manifold of internal combustion engines with the following steps:

• Measuring at least one air mass flow curve and the associated curve of further measurement / actuating and control variables of the electronic engine control system during a period in which different speeds and engine loads occur,
• - Estimating at least one theoretical air mass flow curve during the period from the manipulated, measured and / or controlled variables of the electronic engine control,
• Formation of a difference profile between the theoretical air mass flow profile and the measured air mass flow profile,
• - Using the difference course to estimate an effective leak cross-section in the intake manifold.

2. The method according to claim 1, characterized in that from the courses a plurality of the measured and calculated and / or estimated quantities generated discrete values and used to estimate the effective leak cross section is applied. 3. The method according to claims 1 to 2, characterized in that the Values sorted in a grid depending on at least one load size become. 4. The method according to claims 1 to 3, characterized in that from the discrete difference values interpolation lines or interpolation planes are formed the properties of which are used to estimate the effective leak cross section be used.   5. The method according to any one of claims 2 to 4, characterized in that difference values sorted by or from one or more load variables formed interpolation values are examined for the presence of certain patterns become. 6. The method according to any one of claims 3 to 5, characterized in that a special speed / load range is selected so that the amount of derivative the difference values according to selected load sizes within this range takes the greatest possible values. 7. The method according to any one of claims 1 to 6, characterized in that eventu All offsets and / or sensor errors by subtracting the mean of the Differences between measured and estimated air mass flows for a similar one Load but different speeds can be eliminated. 8. The method according to claims 3 to 7, characterized in that the by the total number of interpolated or measured function values of the Patterns formed with only a few focal points is evaluated, the individual function values of the halftone dots in each case weighted according to their distance from the respective center of gravity. 9. The method according to claim 8, characterized in that focal points with Be determined with the help of genetic algorithms or other search algorithms. 10. The method according to any one of the preceding claims, characterized in that fuzzy techniques are used to estimate the effective leak cross section become. 11. The method according to any one of the preceding claims, characterized in that at least one neural approximator to estimate the effective Leak cross section is used.