DE102018111668A1 - Measurement-based method for detecting a measuring point stability of an internal combustion engine - Google Patents

Abstract

The invention relates to a method for operating a test apparatus (1) for checking an internal combustion engine (2) in real time, wherein the test apparatus (1) has a memory (4) and at least one sensor (5) for detecting at least one state variable of the internal combustion engine (2). and using measured sensor values (32.1, 32.2, 32.3) a stability value of the sensor values (32.1, 32.2, 32.3) is determined, which is a measure of a correlation between the sensor values (32.1, 32.2, 32.3).

Description

The invention relates to a method for operating a test apparatus for checking an internal combustion engine, wherein the test apparatus has a sensor for detecting a state variable of the internal combustion engine.

From the EP 2 078 945 A2 is such a method known. In this method, a quality feature is determined, which is a measure of a noise or a scattering of sensor values detected by the sensor. In the EP 2 078 945 A2 It is proposed to use statistical parameters such as the standard error, the mean square deviation, the coefficient of variation and / or the median deviation to determine values of quality features. Here, with the help of user-defined limits, a check of individual quality characteristics is adapted to an internal combustion engine to be tested. The disadvantage here is that the limits must be reentered for each test object and these can also vary depending on the knowledge of a user. This may result in the process being performed unnecessarily long, depending on the knowledge of the user.

The object of the present invention is therefore to provide a method for operating a testing device for an internal combustion engine, with which an unnecessarily long operation of the testing device for testing the internal combustion engine can be prevented.

This object is achieved by a method according to claim 1 and a test device according to claim 13. Further advantageous embodiments will become apparent from the dependent claims.

To solve this problem, a method for operating a test apparatus for checking an internal combustion engine in real time is proposed. The test apparatus has a memory and at least one sensor for detecting at least one state variable of the internal combustion engine. The method comprises the following steps. In a first step, the internal combustion engine is operated at an operating point. In a second step, a plurality of sensor values of the sensor are measured in a first time period. In a third step, a first and / or second stability value of the sensor values are determined using a calculation rule. The calculation rule has as variable variables exclusively the measured sensor values, the first and / or second stability value being a measure of a correlation between the sensor values. In a fourth step, it is checked whether a first and / or second stability criterion is fulfilled. The checking takes place with the aid of the first and / or second stability value and independently of a value of a parameter that can be set as a function of the internal combustion engine for influencing the first and / or second stability value and / or the first and / or second stability criterion. In a fifth step, a measurement is triggered if the first and / or second stability criterion is met, wherein at least one of the sensor values and / or a further sensor value measured with the sensor for detecting the operating point as a stationary operating point of the internal combustion engine in the Memory is stored.

Due to the fact that the calculation rule exclusively has the sensor values as variable variables, it is possible to dispense with a setting process for adapting the testing device to the internal combustion engine. A parameter which can be set as a function of the internal combustion engine for influencing one of the two stability values and / or one of the two stability criteria can be any parameter within the meaning of the invention which serves to adapt the calculation rule or the stability criteria to the internal combustion engine to be tested.

In particular, the measurement also includes detecting further sensor values of a further sensor, which measures a further state variable of the internal combustion engine. The measurement may also include sensing environmental conditions such as humidity, ambient pressure, and / or room temperature. In particular, a transient phase, which begins with a start of operation of the internal combustion engine in the operating point, can be terminated if the first and / or second stability criterion is met. During the measurement, the sensor values, the further sensor value and / or further sensor values measured with the sensor can be used to determine a stationary measured value. For example, the stationary measured value can be calculated from an average value of the further sensor values. Preferably, the stationary operating point can be described by the stationary measured value and further stationary measured values, which are obtained by means of further sensors according to the proposed method.

The correlation between the sensor values means a relationship between the sensor values that can be described using a function without a random variable. The function may be, for example, a straight line or a periodic function which describes a measurement series formed from the sensor values. In any case, the correlation does not mean a stochastic distribution of the sensor values in the first period. A measure of the random distribution of the sensor values may be calculated with a standard deviation of the sensor values from an average of the sensor values in the first period and is referred to in the art as a possible quality characteristic of a series of measurements. With such a quality feature, however, no correlation between the sensor values of the measurement series can be detected.

The advantage of the method according to the invention is that with the determination of the first and / or second stability value, additional information about the measurement series can be obtained in addition to a quality feature, for example how high and / or significant the correlation between the sensor values is. This additional information can be used to make a decision about continuing to operate the internal combustion engine at the operating point and / or triggering the measurement of the internal combustion engine.

In an advantageous embodiment, the operation of the internal combustion engine is terminated at the operating point when the first and / or second stability criterion is met. In this case, a delay time may elapse between the termination of the operation and a point in time at which the first and / or second stability criterion is fulfilled. During the delay time, for example, the measurement can be performed.

Advantageously, the operation of the internal combustion engine at the operating point can also be terminated if, on the basis of an evaluation of a value of a quality feature, for example by calculating a variance of the sensor values, this operation alone would not be terminated. This is especially useful if, due to the first stability value, a correlation of the sensor values is detected, which can be described by a constant function, and at the same time a variance of the sensor values lying above a specification is detected. If only the variance were evaluated as in the prior art, the operation might not be interrupted in this case. Often, however, a reduction in the variance can not be achieved in such a case, because an overall system consisting of the tester and the internal combustion engine is in a steady state. The proposed method can therefore be prevented from operating the test apparatus unnecessarily long at the same operating point of the internal combustion engine.

With the proposed method, the measurement can be triggered if a variance of the sensor values lying above a specification is detected and the first and / or second stability criterion are met. This is particularly advantageous if a change in the quality of the series of measurements over time is very small, but the sensor values have no drift. Thus, it can be prevented that the test apparatus is operated unnecessarily long, if no improvement in the quality of the measurement series is to be expected.

The proposed method further has the advantage that user-defined inputs are not necessary and therefore a duration of the method is independent of a user's knowledge. This is achieved by determining the first or the second stability value independently of a value of a parameter that can be set as a function of the internal combustion engine for influencing the first and / or second stability value or the first and / or second stability criterion.

In an advantageous embodiment, it is provided that the sensor values are used to approximate a time-dependent first function for describing a time profile of the sensor values in the first time period and to determine a corresponding function value with the first function for one of the sensor values. The first stability value may be a slope of the first function in this embodiment. A refinement of this refinement can provide that the first and / or second stability value is determined as a function of a first sum which has differences between in each case one of the sensor values and the function value corresponding to the sensor value.

The first function can be determined by a regression method. Furthermore, the first function may be described by a model having individual subfunctions. The model is preferably linear or quadratic dependent on time, such as a linear polynomial model. This has the advantage that the approximation of the first function with the model can be performed in real time.

Advantageously, it can be checked whether the sensor values during the first time period increasingly deviate from a course of a constant function in the first time period, ie the sensor values have a drift. The higher the slope of the first function, the higher the drift of the sensor values can be.

The advantage of approximating the first function with the sensor values is that using parameters describing the first function, such as slope, the correlation between the sensor values can be quantified.

$$M{S}_{j,\text{data}}:=\frac{1}{n-1}{\displaystyle \sum _{i=1}^n{\left(y_{j+i}-{\overline{y}}_j\right)}^2,}$$

wobei eine erste mittlere quadratische Abweichung MS_j,model der Sensorwerte y_j+i bezogen auf die jeweiligen korrespondierenden Funktionswerte ŷ_j+i in Abhängigkeit der ersten Summe aus den jeweiligen Quadraten der Differenzen zwischen einem der Sensorwerte y_j+i und einem zu dem jeweiligen Sensorwert korrespondierenden Funktionswert ŷ_j+i gebildet wird. Vor dem ersten Zeitraum, in oder nach welchem das Betreiben der Verbrennungskraftmaschine bevorzugt beendet wird, können weitere Zeiträume liegen, in denen weitere Sensorwerte gemessen wurden.In an advantageous development, the first stability value is determined as a function of a second sum, which has differences between in each case one of the sensor values and an average value of the sensor values. The second sum represents a measure of a scattering of the sensor values, ie a criterion for determining a quality of the measurement series. In one variant, the first sum can be compared with the second sum. In this case, the first stability value can be formed from a quotient of the first and the second sum. Advantageously, the first stability value F _{1 j} , for a period j, in particular for the first period, J determined in accordance with the following calculation rule: $$\begin{array}{ccc}{F}_{1_j}:=\frac{M{S}_{\text{j}\text{, model}}}{M{S}_{\text{j}\text{, data}}}.& \text{With}& M{S}_{j.\text{model}}:=\frac{1}{n-d{f}_{\text{model}}}{\displaystyle \underset{i=1}{\overset{n}{\Sigma }}{\left(y_{j+i}-{\widehat{y}}_{j+i}\right)}^2\text{and}}\end{array}$$

$$M{S}_{j.\text{data}}:=\frac{1}{n-1}{\displaystyle \underset{i=1}{\overset{n}{\Sigma }}{\left(y_{j+i}-{\overline{y}}_j\right)}^2.}$$

wherein a first mean square deviation MS _{j, model of} the sensor values y _{j + i} relative to the respective corresponding function values ŷ _{j + i} as a function of the first sum of the respective squares of the differences between one of the sensor values y _{j + i} and one to the respective Sensor value corresponding function value ŷ _{j + i is} formed. Before the first period, in or after which the operation of the internal combustion engine is preferably ended, there may be further periods during which further sensor values were measured.

The first index j denotes a number of a period, wherein for an initial period of time the first index assumes the value zero, the value following thereafter the value of one and, for the first period, preferably the maximum value of j. The second index i denotes a number of a detected sensor value in the period j, wherein the second index for the first sensor value in a period j takes the value one. This means that the same measurement of one of the sensor values can be used for a calculation of first stability values in different overlapping time periods.

From the first sum, the first mean square deviation MS _{j, model is} calculated by dividing the first sum by a difference between a number n of the measured sensor values in a period j and a number of degrees of freedom df _{model of} the model. If the first function is constant, linear or quadratic, the number of degrees of freedom df _{model is} correspondingly equal to one, two or three.

The first stability value F _{1 j} , for a period j, in particular for the first J period, is preferably based on a quotient of the first mean square deviation MS _{j, model} and a second mean square deviation MS _{j, data of} the sensor values y _{j + i of} the period j with respect to Average y _{j of} the sensor values.

The second mean square deviation MS _{j, data} is a function of the second sum of the respective squares of the differences between each one of the sensor values y _{j + i} and the average value of the sensor values y _j calculated. From the second sum, by dividing the second sum by a difference between the number n of the sensor values in a period j and one, the second mean square deviation MS _{j, data is} formed.

By comparing the first with the second sum, in particular with the aid of the first stability value, a significance of the detected drift of the sensor values as a function of the scatter of the sensor values can be determined. The lower the scatter, the more significant a drift of the sensor values can be detected with the first stability value. With the first stability value F _{1 j} can be checked if Residues between sensor values and the corresponding function values in the first period are significantly different to residuals between the sensor values and the mean value of the sensor values.

Advantageously, a statistical test method, such as an F-test method, is used to check whether the first stability criterion has been met. If the F-test method is used, it can be assumed as a null hypothesis that the sensor values assume such values over the first time period as has a function of a sum of a constant value and a Gaussian distributed over the first time period. Assuming this null hypothesis, a probability distribution can be determined, which for arbitrary values of the first stability value F _{1 j} each indicates a probability. The J probability distribution can be a Fisher distribution, ie F distribution. The F distribution depends on the number of degrees of freedom df _{model of} the model and a number of degrees of freedom n-1 in the determination of the second mean square deviation MS _{j, data} .

If a first value according to one of the variants described above is determined with the proposed method for the first stability value, then using the probability distribution a first probability can be determined for the first stability value to assume the first value if the null hypothesis is to apply. If the first probability lies below a first barrier stored in the memory, for example in the value of 0.01, then it is assumed that the null hypothesis does not hold. In this case the first stability criterion is considered not fulfilled. In the other case, the first stability criterion is fulfilled.

A further embodiment provides that the first stability value is determined according to one of the variants described above and a comparison of the first stability value with a constant first comparison value, for example 0.95, is carried out. On the basis of a result of the comparison, it is checked whether the first stability criterion is fulfilled. For example, the first stability criterion may be considered satisfied if the first stability value is greater than or equal to 0.95.

In an advantageous variant, depending on a third sum, which has differences between in each case two of the sensor values which follow one another directly, the second stability value is determined. A development of this variant provides that the second stability value is determined as a function of the first and the third sum.

wobei die reduzierte quadratische Varianz MS_{j, diff} in Abhängigkeit der dritten Summe aus den jeweiligen Quadraten der Differenzen zwischen einem ersten Sensorwert y_j+i und einem unmittelbar auf den ersten Sensorwert zeitlich folgenden zweiten Sensorwert y_j+i+1 gebildet wird. Aus der dritten Summe wird durch ein Dividieren der dritten Summe durch die doppelte Differenz zwischen der Anzahl n der Sensorwerte in einem Zeitraum j und eins die reduzierte quadratische Varianz MS_{j, diff} errechnet.Advantageously, the second stability value F ₂ for a period j, in particular for the first period, is calculated from a quotient with the first mean square deviation MS _{j, model} as a dividend and a reduced quadratic variance MS _{j, diff} from the differences of the immediately successive sensor values y _{j + i} and y _{j + i + 1 are} calculated as a divisor according to the following calculation rule: $$\begin{array}{ccc}{F}_{2_j}:=\frac{M{S}_{\text{j}\text{, model}}}{M{S}_{\text{j}\text{diff}}}.& \text{With}& M{S}_{j.\text{diff}}:=\frac{1}{2\left(n-1\right)}{\displaystyle \underset{i=1}{\overset{n-1}{\Sigma }}{\left(y_{j+i+1}-y_{j+i}\right)}^2.}\end{array}$$

wherein the reduced quadratic variance MS _{j, diff is formed} as a function of the third sum from the respective squares of the differences between a first sensor value y _{j + i} and a second sensor value y _{j + i + 1} which follows immediately after the first sensor value. From the third sum, the reduced quadratic variance MS _{j, diff is} calculated by dividing the third sum by the double difference between the number n of the sensor values in a time period j and one.

In the same way as for the first stability value, it can be checked by means of the F-test method whether the second stability criterion corresponding to the second stability value is fulfilled. In this case, the same null hypothesis is preferably used and a second probability for the calculated value of the second stability value is determined on the assumption of the null hypothesis. If the second probability lies below a second barrier stored in the memory, for example in the value of 0.01, it is assumed that the null hypothesis does not hold. In this case the second stability criterion is considered not fulfilled. In the other case, the second stability criterion is met.

Alternatively it can be provided that a comparison of the second stability value is carried out with a constant second comparison value. On the basis of a result of the comparison, it is checked whether the second stability criterion is fulfilled. For example, the second stability criterion may be considered satisfied if the second stability value is greater than or equal to 250. The second comparison value is preferably significantly greater than one. This has the advantage that the second stability criterion is satisfied even if only comparatively small fluctuations in the sensor values occur.

In a further embodiment, the second stability value F _{2 j} for a period j, in particular for the first period, from a quotient with the first mean square deviation MS _{j, model} as divisor and the reduced quadratic variance MS _{j, diff} calculated as a dividend according to the following calculation rule: $$F_{2_j}:=\frac{M{S}_{\text{j}\text{diff}}}{M{S}_{\text{j}\text{, model}}}.$$

In this embodiment, the second stability criterion is preferably met when the second stability value is less than or equal to 0.004.

The first and second comparison values are each independent of a value of a parameter that can be set as a function of a test object, such as the internal combustion engine, for influencing the first or second stability value or the first or second stability criterion. Preferably, the first and / or the second comparison value are values independent of the test object and of the test apparatus, i. universally accepted sizes.

Because the second stability value is formed as a function of the third sum, additional information about the measurement series relating to a correlation of adjacent, in particular directly adjacent, sensor values of the measurement series can be obtained with the aid of the second stability value. Another advantage that results from using the reduced squared variance is the following. The reduced quadratic variance can be considered to be a correct estimate of a quadratic variance for a squared pure noise signal by a constant value if the sensor values are not correlated. However, if the sensor values are correlated, the reduced quadratic variance deviates from the quadratic variance, causing a change in the second stability value.

While with the first stability value, in particular with the aid of the first function, information about an approximated course of the sensor values during the first time period can be obtained, it is possible that the second stability value can be used to detect preferably all existing correlations between directly adjacent sensor values. For this reason, it is possible with the second stability value to detect a vibration-like course of the sensor values in the first time period. When calculating the first stability value, the oscillation may possibly not be detected in the case of a vibration-like progression of the sensor values by a constant value. Such a course can, for example, be caused by a swinging control loop, in which the sensor values are actual values. Advantageously, a warning signal is output if the second stability criterion is not met.

Alternatively, the second stability value can be calculated from a quotient with a constant second value as a dividend and the reduced squared variance MS _{j, diff} as a divisor. As a result, the first sum is not included in the calculation of the second stability value. This has the advantage that in the event that there is a very slight scattering of the sensor values and the first sum approaches a value of zero, the second stability value likewise does not approach zero. Thus, a correlation of the sensor values can be detected even with a very small scattering. Furthermore, a calculation of the second stability value according to this alternative can possibly be carried out faster than a calculation of the first stability value, since an approximation of the first function is not necessary and thus computing time can be saved. In particular, if sensor values from a plurality of sensors are detected on a plurality of channels of the test apparatus, this alternative may allow the proposed method to have real-time capability.

The second stability criterion according to this alternative is preferably satisfied when the second stability value for a time permanently stored in the memory is greater than a third barrier. The third bound preferably assumes a value in the range of one thousand to ten thousand, preferably the value two thousand. The fixed time preferably assumes a value in the range of ten to fifty seconds, preferably thirty seconds.

It is also possible that the dividend is the second value or the first mean square deviation MS _{j, model} , depending on which of the two values is greater. A check of the second stability criterion can be made according to which of the two values is greater, according to the different types described above.

Advantageously, with the second stability value, vibrations of the sensor values can be detected from a lower limit frequency, which depends on a duration of the first time period. In a variant, a duration of successive time periods j can be changed in order to change the lower limit frequency and to increase a frequency range of the detectable oscillations. The frequency range is limited by an upper limit frequency, which depends on a sampling rate at which the sensor values are detected.

des zweiten Zeitraums beziehungsweise für die quadratische Varianz ${\sigma }_{\mathrm{1,}data}^2$

des ersten Zeitraums verwendet werden.In a further refinement, a third stability value is determined from a variance of the sensor values in the first period and a variance of the sensor values in a second period preceding the first. In this embodiment, the third stability value is a measure of a difference between the two variances. The third stability value is preferably calculated from a root of a quotient having the second mean square deviation MS _{2, data of} the second time period as a dividend and the second mean square deviation MS _{1, data of} the first time period as a divisor. The second mean square deviation MS _{2, data of} the second time period and the second mean square deviation MS _{1, data of} the first time period can each be used as an estimate for the quadratic variance ${\sigma }_{}^{}$${}_{\mathrm{2,}data}^2$

of the second period or for the quadratic variance ${\sigma }_{}^{}$${}_{\mathrm{1,}data}^2$

of the first period.

wobei die untere und die obere Grenze bevorzugt durch einen festen Wert r_MSStable beschrieben werden. Der feste Wert r_MSStable ist vorzugsweise etwa gleich 1,1.If the third stability value lies between a lower and an upper limit, a third stability criterion is satisfied. The lower and upper limits are stored in the memory and preferably not changeable, in particular independent of the internal combustion engine to be tested. The third stability criterion can be checked according to the following formula: $$\text{min}\left(r_{MSStabl{e}^{.}}1/r_{MSStable}\right)<\underset{=\frac{{\mathrm{<figure-callout\; id="2"\; label="\sigma "\; filenames="DE102018111668A1\_0036.png"\; state="\{\{state\}\}">\sigma </figure-callout>}}_{\mathrm{2,}data}}{{\mathrm{<figure-callout\; id="1"\; label="\sigma "\; filenames="DE102018111668A1\_0037.png"\; state="\{\{state\}\}">\sigma </figure-callout>}}_{\mathrm{1,}data}}}{\underset{\}}{\sqrt{\frac{\mathrm{<figure-callout\; id="2"\; label="M\; S"\; filenames="DE102018111668A1\_0036.png"\; state="\{\{state\}\}">M\; </figure-callout>}{S}_{}{}_{\mathrm{2,}data}}{\mathrm{<figure-callout\; id="1"\; label="M\; S"\; filenames="DE102018111668A1\_0037.png"\; state="\{\{state\}\}">M\; </figure-callout>}{S}_{}{}_{\mathrm{1,}data}}}}}<\text{Max}\left(r_{MSStabl{e}^{.}}1/r_{MSStable}\right).$$

the lower and upper limits are preferably described by a fixed value r _MSStable . The fixed value r _MSStable is preferably about 1.1.

Using the third stability value, the measurement series can be assessed as stable if a temporal change in the variances of sensor values of chronologically consecutive periods is below a threshold value. The greater a time offset of the time periods, for example thirty seconds, the better a change in the variances can be detected. Preferably, the time offset is changed with multiple calculation of the third stability value.

Due to the fact that the third stability value can only be changed by the sensor values of the first and second time periods, when using this stability value it is also possible to dispense with an adaptation of the testing device when changing the internal combustion engine.

In a further advantageous embodiment, it is provided that as a function of a difference between the mean value y _{j, data of} the sensor values and a function value ŷ _model corresponding to one of the sensor values, in particular to the last-detected sensor value, is checked as to whether a fourth stability criterion has been met. The fourth stability criterion is preferably satisfied when an amount of the difference, preferably an entire duration of the first period, is less than a fourth bound, which can be expressed by the following equation: $$\left|{\widehat{y}}_{\text{model}}-{\overline{y}}_{j.\text{data}}\right|<r_{\text{tol}}\cdot \Delta y_{tOl}$$

The fourth barrier is advantageously formed from a product of a maximum resolution Δy _{tol of} the sensor, for example seventeen grams per second, and a factor r _to1 , the resolution and the factor being stored in the memory and being independent of the internal combustion engine. The function value ŷ _model corresponding to the sensor value can be determined with the first function. Particularly advantageously, the corresponding function value ŷ _{model is} calculated with a third function. The third function is preferably approximated using sensor values which are acquired in a fourth period beginning before the first period. Thus, information of the sensor values of the first period are compared with information of the sensor values, which are detected before the beginning of the first period and in the form of the approximated third function.

It is possible that the first stability criterion is fulfilled and the fourth stability criterion is not met. This can occur if a model for describing the first function can be formed in the determination of the first stability criterion, but this is not significant, i. E. the first stability value deviates so little from the value one that in particular the null hypothesis can not be rejected with satisfactory certainty. In such a case, however, the fourth stability criterion can still be considered as not satisfied because, due to a current sensor value, the model can be changed automatically such that the difference between the mean value of the sensor values and the function value corresponding to the current sensor value exceeds the fourth limit. The use of the fourth stability criterion can thus further increase accuracy of information about the stability of the sensor value.

In a further development of the method is dependent on a mean error Δ y _{j, data of} an average value of the sensor values and a difference between an average value y _{j, data of} the sensor values and a set value of the sensor values or an average value y _{j, ref} of changing setpoint values of the sensor values checked whether a fifth criterion is met. The mean error is formed from the root of a quotient with the second mean square deviation MS _{j, data} as a dividend and the number n of the sensor values in a period j, in particular in the first period. The fifth criterion is preferably satisfied when an amount of the difference between the average of the sensor values and the mean of the target value is smaller than a product of the mean error and a constant value, for example three, which is described by the following formula: $$\begin{array}{ccc}\Vert {\overline{y}}_{\text{j}\text{, data}}-{\overline{y}}_{\text{j}\text{, ref}}\Vert \le r_{\text{ref}}\cdot \Delta {\overline{y}}_{\text{j}\text{, data}}& \text{, With}& \Delta {\overline{y}}_{\text{j}\text{, data}}=\sqrt{\frac{M{S}_{j.\text{data}}}{n}}=\frac{{\sigma }_{\text{j}\text{, data}}}{\sqrt{n}},\end{array}$$

The constant value r _ref is independent of a value of a parameter which can be set as a function of the internal combustion engine. The value r _ref is preferably equal to ten. A use of the fifth criterion has the advantage that, with a low mean error, it can be detected whether the mean value of the sensor values deviates from the desired value or from a fluctuating desired value. On the other hand, if only one quality criterion were used, such as the standard deviation of the target values, then such a deviation could not be detected. Therefore, the fifth criterion is preferably checked in addition to the first, second and / or third stability criterion.

A fluctuating setpoint, for example, can be caused by a complex control system with multiple controllers. For example, a NO _x -Emmissionswert vary as a target value due to a variation of a position of an actuator of an exhaust gas recirculation valve. The fifth criterion can preferably be used to check how strongly a deviation of the sensor values from the nominal value of random signal components, such as, for example, a fluctuating measured value or a further fluctuating setpoint value, depends.

Advantageously, it is checked whether the first and / or second and the third and / or fourth stability criterion for a stored in the memory period is met. The period of time is preferably three to four seconds. If the first and / or second and the third and / or fourth stability criterion for the period of time is met, the measurement is triggered. Due to the fact that at least two stability criteria have to be fulfilled simultaneously during the time span, the stability of the sensor values can be checked more accurately in this embodiment.

Particularly advantageously, the measurement is triggered if at least the first and / or second stability criterion and a quality criterion which describes the quality of the measurement series are fulfilled. The quality criterion may be a standard deviation of the sensor values from the mean of the sensor values in the first period. This makes it possible to check whether, on the one hand, the course of the sensor values in the first period is stable and, on the other hand, the sensor values have a low noise, and the measurements are triggered when both criteria are met.

A particular embodiment provides that the first function for the first period and a second function for a third period ending before the first period are each approximated using a regression method. The first and the second, preferably linear, function have at least one respective first and a second parameters. The two functions are determined by a respective first and second parameter value of the first and second parameters. Continue to describe The functions each have a time profile of the sensor values in the first or third time period and are used to calculate the first and / or second stability value for the first and third time periods according to the variants described above. The first period and the third period are divided into several, preferably three hundred, time intervals.

The first and the third time periods overlap, in particular such that the first time period begins and ends a time interval later than the third time period.

In the regression method, a fourth sum is formed for a calculation of the first and second parameter values of the first and second parameters of the second function and stored in the memory. The summands of the fourth sum have at least several, preferably all, of the sensor values detected in the third time period. For a calculation of the first and second parameter values of the first and second parameters of the first function, a fifth sum is formed using the stored fourth sum and a sensor value acquired after the third time period.

$$,\text{ mit}{\left({\text{X}}^T\text{X}\right)}^{-1}={\left[\begin{array}{cc}n& {\displaystyle \sum _{i=1}^{n}i}\\ {\displaystyle \sum _{i=1}^{n}i}& {\displaystyle \sum _{i=1}^n{i}^2}\end{array}\right]}^{-1}$$

The regression method can be a method for minimizing error squares, in which the respective first parameter value b ₁ and second parameter value b _{2 of} the first and second parameters of the first and second functions are calculated by converting a normal equation. For this purpose, a first matrix M ₁ is preferably multiplied by a first vector ν ₁ , as shown in the following formula: $$\left(\begin{array}{c}{b}_1\\ b_2\end{array}\right)=\underset{{\text{<figure-callout id="1" label="M" filenames="DE102018111668A1\_0037.png" state="{{state}}">M</figure-callout>}}_1}{\underset{\}}{{\left({\text{X}}^T\text{X}\right)}^{-1}}}\cdot \underset{{\mathrm{<figure-callout\; id="1"\; label="\nu "\; filenames="DE102018111668A1\_0037.png"\; state="\{\{state\}\}">\nu </figure-callout>}}_1}{\underset{\}}{{\text{X}}^T\text{y}}}.{\text{X}}^T\text{y =}{\left[{\displaystyle \underset{i=1}{\overset{n}{\Sigma }}{y}_i{\displaystyle \underset{i=1}{\overset{n}{\Sigma }}i\cdot y_i}}\right]}^T.{\mathrm{\nu }}_{1_3}=\left[\begin{array}{c}{S}_{j.n}^0\\ S_{j.n}^1\end{array}\right]={\left[{\displaystyle \underset{i=1}{\overset{n}{\Sigma }}{y}_i{\displaystyle \underset{i=1}{\overset{n}{\Sigma }}i\cdot y_i}}\right]}^T$$

$$.\text{With}{\left({\text{X}}^T\text{X}\right)}^{-1}={\left[\begin{array}{cc}n& {\displaystyle \underset{i=1}{\overset{n}{\Sigma }}i}\\ {\displaystyle \underset{i=1}{\overset{n}{\Sigma }}i}& {\displaystyle \underset{i=1}{\overset{n}{\Sigma }}{i}^2}\end{array}\right]}^{-1}$$

und eine zweite Komponente des ersten Vektors ν_{1 3} für den dritten Zeitraum aus einer sechsten Summe $S_{j,n}^1$

gebildet. Eine erste Komponente des ersten Vektors ν_{1 l} für den ersten Zeitraum wird aus der fünften Summe aus dem nach dem dritten Zeitraum erfassten Sensorwert y_j+n+1 und einer Differenz zwischen der vierten Summe $S_{j,n}^0$

und dem ersten erfassten Sensorwert y_j+1 im dritten Zeitraum ermittelt. Eine zweite Komponente des ersten Vektors ν_{1 l} für den ersten Zeitraum kann aus einer siebten Summe aus einem Produkt des nach dem dritten Zeitraum erfassten Sensorwertes y_j+n+1 mit der Anzahl n der Zeitintervalle des ersten Zeitraums und aus einer Differenz zwischen der sechsten Summe $S_{j,n}^1$

und der vierten Summe $S_{j,n}^0$

berechnet werden. $${\mathrm{\nu }}_{1_1}=\left[\begin{array}{c}{S}_{j,n}^0-y_{j+1}+y_{j+n+1}\\ S_{j,n}^1-S_{j,n}^0+n\cdot y_{j+n+1}\end{array}\right]$$

Preferably, a number n of the time intervals of the first and third time periods and the duration of the two time periods are the same, for example thirty seconds. This allows the first matrix to be used to calculate the two parameter values for both the first and third time periods. The first vector is dependent on the sensor values and different for both periods. A first component of the first vector ν _{1 3} for the third period will be the fourth sum $S_{j.n}^0$

and a second component of the first vector ν _{1 3} for the third period from a sixth sum $S_{j.\mathrm{<figure-callout\; id="1"\; label="n"\; filenames="DE102018111668A1\_0037.png"\; state="\{\{state\}\}">n</figure-callout>}}^1$

educated. A first component of the first vector ν _{1 l} for the first period, the fifth sum of the sensor value y _{j + n + 1} detected after the third period and a difference between the fourth sum $S_{j.n}^0$

and the first detected sensor value y _{j + 1 determined} in the third period. A second component of the first vector ν _{1 l} for the first time period, a seventh sum of a product of the sensor value y _{j + n + 1} detected after the third time interval with the number n of the time intervals of the first time period and a difference between the sixth sum $S_{j.\mathrm{<figure-callout\; id="1"\; label="n"\; filenames="DE102018111668A1\_0037.png"\; state="\{\{state\}\}">n</figure-callout>}}^1$

and the fourth sum $S_{j.n}^0$

be calculated. $${\mathrm{<figure-callout\; id="1"\; label="\nu "\; filenames="DE102018111668A1\_0037.png"\; state="\{\{state\}\}">\nu </figure-callout>}}_{1_1}=\left[\begin{array}{c}{S}_{j.n}^0-y_{j+1}+y_{j+n+1}\\ S_{j.n}^1-S_{j.n}^0+n\cdot y_{j+n+1}\end{array}\right]$$

The fact that in the calculation of the two components of the first vector ν _{1 l} for the first period, the components of the first vector ν _{1 3} can be used for the third period, ie the fourth and sixth sum, can be saved computation time and made that the method is real-time capable. The method preferably provides for sensor values to be detected in further periods lying before the first and third time periods and these to be approximated using further functions. To the parameter values for each period according to the variant described above, the sensor values first acquired in the respective period are stored in the memory, preferably in a ring memory of the memory.

beispielsweise durch eine Division der vierten Summe $S_{j,n}^0$

durch die Anzahl n der Zeitintervalle des dritten Zeitraums, abgeschätzt werden. Dadurch kann auf ein Abspeichern der in dem jeweiligen Zeitraum zuerst erfassten Sensorwerte verzichtet werden.To save storage space, the first detected sensor value y _{j + 1} in the third period can be calculated using the fourth sum $S_{j.n}^0.$

for example, by dividing the fourth sum $S_{j.n}^0$

by the number n of the time intervals of the third period. This makes it possible to dispense with storing the sensor values first acquired in the respective period.

In a further advantageous embodiment of the method, it is provided that the first and / or second stability value for the first time period is determined with the aid of the fifth sum, and a first and / or second stability value for the third time period is determined with the aid of the fourth sum. With this refinement, an arithmetic time can be shortened when updating the first and / or second stability value. Updating here means that the first and / or second stability value for the first period is calculated temporally after the first and / or second stability value for the third period. The computing time can be shortened in particular because the fifth sum is formed with the aid of the fourth sum and the fourth sum has been stored in the memory before the calculation of the fifth sum.

As already described, the first stability value for a period j, in particular for the third period, can be calculated from the quotient with the first mean square deviation MS _{j, model} as a dividend and a second mean square deviation MS _{j, data} as a divisor. The third period is assigned the index j in the following.

aus den Quadraten der Sensorwerte y_j+i und einem Skalar T_j für den dritten Zeitraum ermittelt, wie in folgenden Formeln dargestellt ist: $$P_{j,n}^2:={\displaystyle \sum _{i=1}^n{y}_{j+1}^2,M{S}_{model,j}=\frac{1}{n-1}\left(P_{j,n}^2-T_j\right)}$$

The first mean square deviation MS _{j, model} for the third period is preferably using an eighth sum $P_{j.\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="DE102018111668A1\_0036.png"\; state="\{\{state\}\}">n</figure-callout>}}^2$

from the squares of the sensor values y _{j + i} and a scalar T _j for the third period, as shown in the following formulas: $$P_{j.n}^2:={\displaystyle \underset{i=1}{\overset{n}{\Sigma }}{y}_{j+1}^2.M{S}_{mOdel.j}=\frac{1}{n-1}\left(P_{j.n}^2-T_j\right)}$$

und der sechsten Summe $S_{j,n}^1$

berechnet werden. Für den Fall, dass die zweite Funktion eine lineare Funktion ist, ergibt sich vorteilhaft folgende Berechnungsvorschrift für den Skalar T_j für den dritten Zeitraum: $$T_j={\left[\begin{array}{c}{S}_{j,n}^0\\ S_{j,n}^1\end{array}\right]}^T\cdot {\left[\begin{array}{cc}n& {\displaystyle \sum _{i=1}^{n}i}\\ {\displaystyle \sum _{i=1}^{n}i}& {\displaystyle \sum _{i=1}^n{i}^2}\end{array}\right]}^{-1}\cdot \left[\begin{array}{c}{S}_{j,n}^0\\ S_{j,n}^1\end{array}\right]$$

The scalar T _j for the third period may vary depending on the fourth sum $S_{j.n}^0$

and the sixth sum $S_{j.\mathrm{<figure-callout\; id="1"\; label="n"\; filenames="DE102018111668A1\_0037.png"\; state="\{\{state\}\}">n</figure-callout>}}^1$

be calculated. In the case where the second function is a linear function, the following calculation rule for the scalar T _j for the third period advantageously results: $$T_j={\left[\begin{array}{c}{S}_{j.n}^0\\ S_{j.n}^1\end{array}\right]}^T\cdot {\left[\begin{array}{cc}n& {\displaystyle \underset{i=1}{\overset{n}{\Sigma }}i}\\ {\displaystyle \underset{i=1}{\overset{n}{\Sigma }}i}& {\displaystyle \underset{i=1}{\overset{n}{\Sigma }}{i}^2}\end{array}\right]}^{-1}\cdot \left[\begin{array}{c}{S}_{j.n}^0\\ S_{j.n}^1\end{array}\right]$$

$$T_{j+1}={\left[\begin{array}{c}{S}_{j+\mathrm{1,}n}^0\\ S_{j+\mathrm{1,}n}^1\end{array}\right]}^T\cdot {\left[\begin{array}{cc}n& {\displaystyle \sum _{i=1}^{n}i}\\ {\displaystyle \sum _{i=1}^{n}i}& {\displaystyle \sum _{i=1}^n{i}^2}\end{array}\right]}^{-1}\cdot \left[\begin{array}{c}{S}_{j+\mathrm{1,}n}^0\\ S_{j+\mathrm{1,}n}^1\end{array}\right],$$

An advantageous embodiment may provide that the first mean square deviation MS _{j + 1, model} is calculated for the first period according to the following formulas, in which the index is assigned to the first period j + 1: $$\begin{array}{ccc}{P}_{j+\mathrm{1,}\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="DE102018111668A1\_0036.png"\; state="\{\{state\}\}">n</figure-callout>}}^2=P_{j.n}^2-y_{j+1}^2+y_{j+n+1}^2.& & M{S}_{\text{model}\text{, j + 1}}=\frac{1}{n-1}\left(P_{j+\mathrm{1,}n}^2-T_{j+1}\right).\end{array}$$

$$T_{j+1}={\left[\begin{array}{c}{S}_{j+\mathrm{1,}n}^0\\ S_{j+\mathrm{1,}n}^1\end{array}\right]}^T\cdot {\left[\begin{array}{cc}n& {\displaystyle \underset{i=1}{\overset{n}{\Sigma }}i}\\ {\displaystyle \underset{i=1}{\overset{n}{\Sigma }}i}& {\displaystyle \underset{i=1}{\overset{n}{\Sigma }}{i}^2}\end{array}\right]}^{-1}\cdot \left[\begin{array}{c}{S}_{j+\mathrm{1,}n}^0\\ S_{j+\mathrm{1,}n}^1\end{array}\right].$$

und der siebten Summe $S_{j+\mathrm{1,}n}^1$

berechnet mit: $$S_{j+\mathrm{1,}n}^0=S_{j,n}^0-y_{j+1}+y_{j+n+1}$$

$$S_{j+\mathrm{1,}n}^1=S_{j,n}^1-S_{j,n}^0+n\cdot y_{j+n+1}$$

In this variant, a scalar T _{j + 1} for the first period is calculated using the fifth sum $S_{j+\mathrm{1,}n}^0$

and the seventh sum $S_{j+\mathrm{1,}\mathrm{<figure-callout\; id="1"\; label="n"\; filenames="DE102018111668A1\_0037.png"\; state="\{\{state\}\}">n</figure-callout>}}^1$

calculated with: $$S_{j+\mathrm{1,}n}^0=S_{j.n}^0-y_{j+1}+y_{j+n+1}$$

$$S_{j+\mathrm{1,}\mathrm{<figure-callout\; id="1"\; label="n"\; filenames="DE102018111668A1\_0037.png"\; state="\{\{state\}\}">n</figure-callout>}}^1=S_{j.n}^1-S_{j.n}^0+n\cdot y_{j+n+1}$$

mithilfe der im Speicher gespeicherten achten Summe $P_{j,n}^2$

ermittelt werden. Somit können bei der Berechnung des ersten Stabilitätswertes für den dritten Zeitraum die im Speicher gespeicherte vierte, sechste und achte Summe verwendet werden. Dadurch ist es nicht notwendig, die fünfte, siebte und neunte Summe explizit durch eine Summierung aller im ersten Zeitraum gemessenen einzelner Sensorwerte y_j+1+i zu berechnen, nachdem die vierte, sechste und achte Summe bereits berechnet wurde. Dies kann in Abhängigkeit von vorgegebenen Hardwarekomponenten der Prüfvorrichtung, insbesondere eines Steuergerätes der Prüfvorrichtung, eine Echtzeitfähigkeit des vorgeschlagenen Verfahrens ermöglichen.By calculating the fifth and the seventh sum from the fourth and the sixth sum stored in the memory, calculation time can be saved when calculating the scalar T _{j + 1} for the first time period. Also, a ninth total $P_{j+\mathrm{1,}\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="DE102018111668A1\_0036.png"\; state="\{\{state\}\}">n</figure-callout>}}^2$

using the eighth sum stored in memory $P_{j.\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="DE102018111668A1\_0036.png"\; state="\{\{state\}\}">n</figure-callout>}}^2$

be determined. Thus, in calculating the first stability value for the third period, the fourth, sixth, and eighth sum stored in memory may be used. Thus, it is not necessary to calculate the fifth, seventh and ninth sums explicitly by summing all the individual sensor values y _{j + 1 + i} measured in the first period after the fourth, sixth and eighth sums have already been calculated. This can enable a real-time capability of the proposed method as a function of predetermined hardware components of the test apparatus, in particular of a control unit of the test apparatus.

• 1 eine Prüfvorrichtung zur Überprüfung einer Verbrennungskraftmaschine;
• 2 einzelne Schritte eines Verfahrens zum Betreiben des Prüfsystems nach 1;
• 3 ein Diagramm mit Sensorwerten und modellierten Sensorwerten zur Berechnung eines ersten Stabilitätswertes;
• 4 ein weiteres Diagramm mit Sensorwerten und modellierten Sensorwerten zur Berechnung eines zweiten Stabilitätswertes;
• 5 ein weiteres Diagramm mit Sensorwerten und modellierten Sensorwerten zur Berechnung eines zweiten Stabilitätswertes.

Further advantages, features and details of the invention will become apparent from the following description of at least one preferred embodiment and with reference to the figures. These show in:

• 1 a testing device for checking an internal combustion engine;
• 2 individual steps of a method for operating the test system according to 1 ;
• 3 a diagram with sensor values and modeled sensor values for calculating a first stability value;
• 4 another diagram with sensor values and modeled sensor values for calculating a second stability value;
• 5 another diagram with sensor values and modeled sensor values for calculating a second stability value.

1 shows a tester 1 for checking an internal combustion engine 2 in real time, with the tester 1 a control unit 3 with a memory 4 and at least one sensor 5 for detecting at least one state variable of the internal combustion engine 2 having. The tester 1 In an advantageous embodiment, it is also suitable for checking a machine system 6 that the internal combustion engine 2 and an electric motor 7 having. The control unit 3 includes a non-transitory computer-readable storage medium having information stored thereon when executed by a processor of the controller 3 effect an implementation of the following method.

In 2 the individual steps of the proposed method are shown. In a first step 11 becomes the internal combustion engine 2 operated at an operating point. In a second step 12 be multiple sensor values of the sensor 5 in a first period 33 measured. In a third step 13 become a first stability value F _{1 j} and / or second stability value F _{2 j} the sensor values are determined with a calculation rule. The calculation rule has as variable variables only the measured sensor values, wherein the first and / or second stability value is a measure of a correlation between the sensor values. In a fourth step 14 it is checked whether a first and / or second stability criterion is met. The testing takes place with the aid of the first and / or second stability value and independently of a value of one as a function of the internal combustion engine 2 adjustable parameter for influencing the first and / or second stability value and / or the first and / or second stability criterion. In a fifth step 15 a measurement is triggered when the first and / or second stability criterion is met, wherein in the measurement of at least one of the sensor values and / or a further sensor value measured by the sensor for detecting the operating point as a stationary operating point of the internal combustion engine is stored in the memory.

$$M{S}_{j,\text{data}}:=\frac{1}{n-1}{\displaystyle \sum _{i=1}^n{\left(y_{j+1}-{\overline{y}}_j\right)}^2}$$

3 shows a diagram with a timeline 31 and a measurement series of at least the sensor values 32.1 . 32.2 . 32.3 that in the first period 33 with the sensor 5 be recorded. The sensor values 32.1 . 32.2 . 32.3 , hereinafter also referred to as sensor values y _{j + i} , can qualitatively on an ordinate 34 be read. The following describes how the first stability criterion can be checked. The first stability value F ₁ corresponding to the first stability criterion _j for a period of time j, especially for the first period 33 , is derived from a quotient of a first mean square deviation MS _{j, model of} the sensor values y _{j + i} with respect to respective function values ŷ _{j + i} and a second mean square deviation MS _{j, data of} the sensor values y _{j + i of} the period j an average 35 the sensor values y _{j + i} , hereinafter also referred to as mean value y _j , determined according to the following formula: $$\begin{array}{ccc}{F}_{1_j}:=\frac{M{S}_{\text{j}\text{, model}}}{M{S}_{\text{j}\text{, data}}}.& \text{With}& M{S}_{j.\text{model}}:=\frac{1}{n-d{f}_{\text{model}}}{\displaystyle \underset{i=1}{\overset{n}{\Sigma }}{\left(y_{j+i}-{\widehat{y}}_{j+i}\right)}^2}\end{array}\text{and}$$

$$M{S}_{j.\text{data}}:=\frac{1}{n-1}{\displaystyle \underset{i=1}{\overset{n}{\Sigma }}{\left(y_{j+1}-{\overline{y}}_j\right)}^2}$$

The second mean square deviation MS _{j, data} is a function of the second sum of the respective squares of the differences between each one of the sensor values y _{j + i} and the mean 35 calculated. In 3 is shown as the sensor values 32.1 . 32.2 . 32.3 around the mean 35 the sensor values fluctuate. Based on the sensor values 32.1 . 32.2 . 32.3 becomes a first function 36 to describe a time history of the sensor values in the period 33 approximated and to each one of the sensor values 32.1 . 32.2 . 32.3 the corresponding function value ŷ _{j + i} , or 36.1, 36.2, 36.3, using the first function 36 calculated. The first function 36 can be formed by a model with 2 degrees of freedom, ie df _model is equal to two in this case.

Out 3 It can be seen that the differences between the respective sensor value 32.1 . 32.2 . 32.3 and the mean 35 the sensor values of the amount higher than the corresponding differences between the respective sensor value 32.1 . 32.2 . 32.3 and the corresponding function value 36.1 . 36.2 . 36.3 are. Assuming a null hypothesis in which the sensor values 32.1 . 32.2 . 32.3 have no drift, a probability can be calculated for the calculated first stability value F _{1 j} to obtain. If this probability is smaller than a confidence probability, then the null hypothesis is rejected and the first stability criterion is not fulfilled, as it is in 3 example shown is the case. As a result, the first stability value deviates from one in this case.

4 and 5 show an application of the second stability criterion. 4 shows another diagram with a timeline 41 and a measurement series of at least the sensor values 42.1 . 42.2 . 42.3 that in the first period 33 with the sensor 5 be recorded. The sensor values 42.1 . 42.2 . 42.3 , hereinafter also referred to as sensor values y _{j + i} , can qualitatively on an ordinate 44 be read. The second stability value F ₂ corresponding to the second stability criterion _j is calculated from a quotient with a first mean square deviation MS _{j, model} as a dividend and a reduced quadratic variance MS _{j, diff} from differences of immediately successive sensor values y _{j + i} and y _{j + i + 1} , for example a difference from the sensor value 42.1 and the sensor value 42.2 , calculated as a divisor according to the following calculation rule: $$\begin{array}{ccc}{F}_{2_j}:=\frac{M{S}_{\text{j}\text{, model}}}{M{S}_{\text{j}\text{diff}}}.& \text{With}& M{S}_{\text{j}.\text{diff}}:=\frac{1}{2\left(n-1\right)}{\displaystyle \underset{i=1}{\overset{n-1}{\Sigma }}{\left(y_{j+i+1}-y_{j+i}\right)}^2}\end{array}$$

In the in 4 The example shown is the sensor values 42.1 . 42.2 . 42.3 and 42.4 correlated. It follows that the second stability value deviates from a value that the second stability value F _{2 j} would assume if the sensor values were uncorrelated. As for the first stability criterion, an F-test procedure can be used to determine that the second stability criterion is not met in this case.

5 shows an example of a series of measurements of sensor values 52.1 . 52.2 . 52.3 . 52.4 that in the first period 33 with the sensor 5 be detected and have no correlation and for which the second stability criterion is met. In the 5 shown sensor values 52.1 . 52.2 . 52.3 . 52.4 and the respective corresponding function values 56.1 . 56.2 . 56.3 are used in the same way for the calculation of the second Stability value used as the corresponding sensor values 42.1 . 42.2 . 42.3 . 42.4 and the respective corresponding function values 46.1 . 46.2 . 46.3 ,

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• EP 2078945 A2

Claims (13)

Method for operating a test device (1) for checking an internal combustion engine (2) in real time, wherein the test device (1) has a memory (4) and at least one sensor (5) for detecting at least one state variable of the internal combustion engine (2), with following steps: Operating the internal combustion engine (2) at an operating point, Measuring a plurality of sensor values (32.1, 32.2, 32.3) of the sensor (5) in a first period, Determining a first and / or second stability value of the sensor values (32.1, 32.2, 32.3) with a calculation specification which has exclusively the measured sensor values (32.1, 32.2, 32.3) as variable variables, the first and / or second stability value being a measure of is a correlation between the sensor values (32.1, 32.2, 32.3), Checking whether a first and / or second stability criterion is satisfied, wherein the checking using the first and / or second stability value and independently of a value of a parameter that can be set as a function of the internal combustion engine (2) influences the first and / or second stability value and / or the first and / or second stability criterion, Triggering a measurement, if the first and / or second stability criterion is met, wherein at least one of the sensor values (32.1, 32.2, 32.3) and / or a further sensor value measured with the sensor (5) for detecting the operating point as a stationary operating point of the internal combustion engine (2) in the memory (4) is stored. Method according to Claim 1 , characterized in that the operation of the internal combustion engine is terminated at the operating point when the first and / or second stability criterion is met. Method according to Claim 1 or 2 , characterized in that the sensor values (32.1, 32.2, 32.3) approximate a time-dependent first function for describing a chronological progression of the sensor values (32.1, 32.2, 32.3) in the first period to one of the sensor values (32.1, 32.2, 32.3 ) determines a corresponding function value with the first function and the first and / or second stability value is determined as a function of a first sum which has differences between in each case one of the sensor values (32.1, 32.2, 32.3) and the function value corresponding to the sensor value. Method according to one of the preceding claims, characterized in that the first stability value is determined as a function of a second sum which has differences between in each case one of the sensor values (32.1, 32.2, 32.3) and an average value of the sensor values (32.1, 32.2, 32.3) becomes. Method according to one of the preceding claims, characterized in that the second stability value is determined as a function of a third sum which has differences between in each case two of the sensor values (32.1, 32.2, 32.3, 32.4) which follow one another directly. Method according to one of the preceding claims, characterized in that a third stability value consists of a variance of the sensor values (32.1, 32.2, 32.3) in the first period and a variance of the sensor values (32.1, 32.2, 32.3) in a second period preceding the first is determined, wherein the third stability value is a measure of a difference between the two variances and a third stability criterion is checked as a function of the third stability value. Method according to one of the preceding Claims 4 to 6 , characterized in that, depending on a difference between the mean value of the sensor values (32.1, 32.2, 32.3) and a function value corresponding to one of the sensor values (32.1, 32.2, 32.3), it is checked whether a fourth stability criterion is satisfied. Method according to one of the preceding claims, characterized in that in dependence on a mean error of an average value of the sensor values (32.1, 32.2, 32.3) and a difference between an average value of the sensor values (32.1, 32.2, 32.3) and a set value of the sensor values (32.1, 32.2, 32.3) or an average value of variable setpoint values of the sensor values (32.1, 32.2, 32.3), it is checked whether a fifth criterion is fulfilled. Method according to one of the preceding claims, characterized in that it is checked whether the first and / or second and the third and / or fourth stability criterion for a time stored in the memory (4) is satisfied and the measurement is triggered when the first and / or second and the third and / or fourth stability criterion for the period is met. Method according to one of the preceding claims, characterized in that the measurement is triggered when a quality criterion, which describes a quality of the sensor values (32.1, 32.2, 32.3), is met. Method according to one of the preceding Claims 3 to 10 characterized in that the first function for the first period and a second function for a third period ending before the first period, wherein the first and third periods overlap, are each approximated using a regression method in which a calculation of a first period and a second parameter value of the second function, a fourth sum whose summands has at least a plurality of sensor values detected in the third time period is formed and stored in the memory and for calculating a first and a second parameter value of the first function a fifth sum using the stored fourth sum and a sensor value detected after the third time period. Method according to Claim 11 characterized in that the first and / or second stability value for the first period is determined using the fifth sum, and a first and / or second stability value for the third period is determined using the fourth sum. Test device (1) for checking an internal combustion engine (2) in real time, wherein the test device (1) has a control device (3) with a memory (4) and at least one sensor (5) for detecting at least one state variable of the internal combustion engine (2), wherein the control device (3) has a non-volatile computer-readable storage medium with information stored thereon, which, when executed by a processor of the control device (3), performs an implementation of the method Claim 1 cause.

Images