DE102006048730A1 - Measuring device e.g. engine test stand, measuring accuracy determining device for e.g. passenger car, has module implementing measurement during operation of measuring device for online determination of plausibility of measuring data - Google Patents

Abstract

The device has a module implementing a dead time testing before start of operation of a measuring device. Another module implements measurement during start of operation of the measuring device for determining the reproducibility of the measuring data. A third module is provided for implementing a basic measurement during current operation of the measuring device for online determination of the plausibility of the measuring data. An independent claim is also included for a method for determining measuring accuracy of a measuring device e.g. engine test stand.

Description

The The invention relates to a method and a device for determination the measuring accuracy of a measuring device.

Preferably If the measuring device is a test bench device, in particular an engine test bench.

through the method and the device, the measurement accuracy of the Test facility especially in the engine test bench, recorded or monitored will be displayed as well as errors or deviations.

modern Internal combustion engines from the passenger car or commercial vehicle sector require for the fulfillment of the Emissions legislation and the ever-increasing demands for performance, Driveability and consumption an ever higher application effort.

• – Drosselklappe
• – Zündzeitpunkt oder Förderbeginn
• – Vor- und Nacheinspritzungen
• – Variable Turbolader oder Abgasventile
• – Klappen zur Beeinflussung der Zylinderinnenströmung
• – Ventilhub, Dauer und Winkellage
• – AGR-Rate
• – Sekundärluft
• – Einspritzdruck

The number of actuators that influence combustion, performance and emissions is steadily increasing. In the meantime, the following actuating or controlled variables must be optimized for almost every motor:

• - Throttle
• - Ignition point or start of delivery
• - Pre and post injections
• - Variable turbocharger or exhaust valves
• - Flaps for influencing the cylinder internal flow
• - Valve lift, duration and angular position
• - EGR rate
• - secondary air
• - Injection pressure

Around to determine the exact Optima and settings on the engine increases in addition to the number of actuators and the number of measuring points, too the number of measured variables is steady at.

Even though the quality the measuring devices has increased in recent years, is the failure of a sensor or the wrong calibration of the same is still the main cause for incorrect measurements and associated additional expenses.

One preferably on a supplementary computer present program make it possible the measurement data quality and the profitability from engine test stands too increase. This affects all Test bench applications such as application, endurance tests or other research test benches which internal combustion engines are equipped with a larger number of sensors.

Backgrounds are one with all engine test stands very high failure rate of one of the measuring instruments used and the ever more extensive Number of measurands.

Also the sequence of measurement points in modern surveying strategies, where frequent several manipulated variables at the same time be adjusted makes the traceability of the measured value changes almost impossible. The assessment how, for example, the nitrogen oxide emission at the same time increase of rail pressure, increase the EGR and early adjustment of the start of funding is behaving, is also for trained Staff hardly accomplish.

Often shows longer Time after the measurement, whether the data met the requirements, often with the result that the measurement must be repeated. Because the Motors often on pallet systems only for the duration of one day at the test bench Lingering, the repetition of the measurement is usually a very huge Time delay and organizational effort.

The Exam, whether the dates are right or wrong is under these conditions only with a complex Formula work, coupled with a wealth of experience of a test engineer possible. Exactly this functionality should e.g. by means of a software, preferably on a separate test computer, already online while be provided to the measurement.

It are therefore test routines necessary, on the one hand, are functional with a minimum of system knowledge, and on the other, calculate so fast that the online capability is guaranteed. The objective is only with the anyway in the test computer existing engine information. These are e.g. Information about drilling, Stroke, number of cylinders, charging method and engine type. Preferably should neither assuming a previous measurement of the engine, still a noticeable additional Measuring effort required be. The goal is a system that is used in large numbers in the test field can be.

The strongly growing number of measurement and calculation quantities at one engine test stand do it to the test stand driver or test bench engineer increasingly difficult to control all measurement data on the engine test bench. The can lead to implausible measuring points over a longer Period without being noticed. The Repeat measurements cause higher test bench times and test bench costs. additionally extend the development times. If the measurement errors are not noticed, can these lead to wrong decisions in the development process.

Of the The invention is therefore based on the technical problem, an improved Device and an improved method for determining the measurement accuracy a measuring device, in particular a motor test bench, to accomplish.

The solution the technical problem results according to the invention by things the claims 1 and 5. Further advantageous embodiments of the invention result from the dependent claims.

Of the The invention is based on knowledge, an online diagnosis to integrate at the engine test stands, which indicates a measurement error at any time and thus the test stand driver allows to fix the error.

It a procedure has been developed, which is already in progress before the beginning and while of the experiment controls the engine measurement data indicating the measurement error and thus the test bench driver or the test bench engineer allows to fix the error. Furthermore, it is possible to subsequently the Measurement and calculation variables at the workplace on their plausibility to investigate. The control methods used are based on mathematical Formulas and motor expert knowledge.

It is therefore a method according to the invention and an associated one Device for determining the measuring accuracy of a measuring device, in particular an engine test bench, proposed, comprising at least one arithmetic unit, the Acquisition of measurement data of the measuring device is suitable, with a Plurality of modules for reference measurement of the measured data provided is, by means of which the measured data are compared with reference values, wherein at least a first module before a standstill investigation Commissioning of the measuring system, a second module a pre-measurement while the commissioning of the measuring device to determine the reproducibility the measurement data and a third module a main measurement during the ongoing operation of the measuring device for the online determination of the plausibility perform the measurement data. The comparison of the measured data with reference values can be next to a simple comparison of, for example, fixed values or Expert knowledge (database) also include the use of formulas over which the readings are processed, these results then with Reference values and / or other measurement data can be compared. The Formulas can with physical, empirical or measurement data based (polynomial models) be constructed.

In an advantageous embodiment is a fourth module to carry out a re-measurement after completion of the operation of the measuring device intended for determining the offline plausibility of the measured data.

In Another advantageous embodiment is another Module for implementation a scatter band measurement during the commissioning of the measuring device for the evaluation of the measured data over a long time Time period provided.

In In another advantageous embodiment, the reference measurements are automatically by means of electronic data transmission to a before selectable Receiver can be transmitted.

The Invention will be described below with reference to a preferred embodiment explained. In the associated Drawings show

1 schematically the timing of the procedure,

2 schematically a comparison of the exhaust gas temperature with the intake temperature;

3 schematically the stepwise sequence in the module for performing a final measurement for the offline determination of the plausibility of the measured data,

4 schematically the three pillars of diagnosis in fault detection,

5 schematically the finding of outliers by median formation,

6 a diagram for the evaluation of the drift at coarse measuring range resolution,

7 a diagram comparing different temperatures,

8th a diagram for assessing drifting channels,

9 a diagram for the representation of digitization noise,

10 the scheme of a conventional engine measurement,

11 a schematic representation of methods for slope determination,

12 a schematic block diagram of a test setup,

13 a schematic replacement model for a measuring channel,

14 a first schematic representation of the quality of the modeling,

15 a second schematic representation of the quality of the modeling,

16 a schematic representation of the error function of a PDT2 member,

17 a schematic representation of error detection method,

18 a schematic representation of carbon paths on the internal combustion engine,

19 a schematic diagram of the exhaust gas analysis,

20 a schematic representation of various moisture correction approaches,

21 a tabular representation of areas of composition based on various analyzes,

22 a schematic representation of the model quality of a carbon balance,

23 a schematic diagram of an exhaust gas temperature model,

24 a schematic representation of the calculation of the exhaust gas temperature of diesel engines,

25 a schematic representation of the characteristics of a compressor,

26 a schematic representation of the maps of a turbine,

27 a schematic representation of the raw data of a compressor map,

28 a schematic representation of the definition of the isentropic compressor efficiency,

29 the speed line of a centrifugal compressor,

30 a schematic representation of the compressor torque over the flow coefficient,

31 a schematic representation of the interpolation and extrapolation of the compressor map,

32 a schematic representation of the structure of the neural networks used,

33 a schematic representation of the processes on a neuron,

34 a schematic representation of the transfer functions used,

35 a schematic representation of the results of a map comparison (6-cylinder commercial vehicle diesel),

36 a schematic representation of the features of the exhaust pressure measurement,

37 a schematic representation of a method for estimating the number of errors,

38 a schematic representation of thresholds for error-free models,

39 a schematic representation of threshold values for faulty models,

40 a further schematic representation of threshold values for faulty models,

41 a further schematic representation of threshold values for faulty models,

42 another, but no longer meaningful schematic of thresholds for two simultaneous errors,

43 a screen output of a detail level,

44 a schematic representation of detection and isolation rates of a gasoline engine,

45 a schematic representation of detection and isolation rates of a commercial vehicle diesel engine.

• 1. Stillstanduntersuchung (oder auch "Nullpunktuntersuchung"),
• 2a. Reproduzierbarkeit,
• 2b. Streubandmessung (optional)
• 3. Online-Plausibilität
• 4. Offline-Plausibilität (optional)

The developed method and the associated device has the following individual modules:

• 1. standstill investigation (or "zero-point investigation"),
• 2a. Reproducibility,
• 2 B. Scattering measurement (optional)
• 3. Online plausibility
• 4. Offline plausibility (optional)

The respective individual modules check the measurement and calculation data of the motor over time. The timing is shown schematically in 1 shown.

However, the subject of measurement plausibility can also be divided into the following individual topics:
Standstill analysis (zero point analysis) ⇒ Reproducibility ⇒ Offline plausibility ⇒ Online diagnosis

The respective individual modules are described below:

1. First Module for implementation a standstill investigation

The The basic idea of the standstill investigation is the engine on the test bench to start only when the test bench is faultless. To this to be checked at the cold engine temperatures and pressures with the ambient condition compared. Furthermore have to e.g. the torque and fuel consumption are zero. This procedure allows sensor failures, incorrect calibration and drift phenomena determine. The methods used are empirical and logical connections built up.

The Standstill measurement and evaluation, for example, runs every morning fully automatic at 5:30. The preferred field is the protocol subsequently the clerk by e-mail sent. The case officer then decides on the basis of the protocol if he can start the engine.

2. Second module for execution a pre-measurement to determine the reproducibility of the measured data

The reproducibility study is preferably the first study (see 1 Pre-measurement) while the engine is running. It should check the engine in warm operating condition. If no errors occur, you can start with the actual measurements.

• • Drehzahl 2000 1/min, Nutzmitteldruck 2 bar
• • Drehzahl 3000 1/min, Nutzmitteldruck 5 bar
• • Schlepppunkt
• • Nennleistungspunkt

The examination is carried out, for example, preferably in the morning after the standstill measurement. For this purpose, the engine is run warm in the respective reference points until a stationary engine operation is established. For example, the following reference points should be used:

• • Speed 2000 1 / min, medium pressure 2 bar
• • speed 3000 1 / min, working pressure 5 bar
• • Tow point
• • rated power point

Then be the reference measurements preferably further automatically in a tool evaluated and sent to the clerk by e-mail. In the Evaluation becomes the measurement and calculation quantities of the respective reference points e.g. compared with those of the previous days. If the measuring or calculation variables the previously defined tolerances, they are displayed as errors and enable the clerk, this before the actual important main measurements to fix. Thus, component faults, leaks and instrumentation errors detect.

It It is also conceivable that the operating points every morning in different Try measured and then be evaluated manually. Here are the creation of the graphics and the scaling of all axes made by hand. Every project preferably has its own templates, which must be kept up to date.

The The aim of the standardized reference experiments is to achieve a uniform To offer a template that is easy for a project and the respective operating point can be adjusted. The measurement of all reference points should then take place in an experiment and run automatically.

The Evaluation is preferably carried out immediately after the end of the experiment automatically and then it will be e.g. as an Excel evaluation or as a PDF file to the corresponding e-mail distributor sent.

2 B. Another module for execution a scatter band measurement

The Stray-band measurement can be optionally selected by the user. It should always be done at the beginning of a motor project. With the scatter band measurement it is possible, the measurement or calculation size over a longer To look at time period. This is a constant stationary operating point (for example, 2000 rpm and 2 bar) and the measured and calculated quantities become for example, measured every 15 minutes 15 times in succession. These will be over afterwards represented the time and the scattering measure specified. Both the measurement as well as the subsequent one Evaluation preferably takes place automatically and becomes the clerk available by e-mail posed.

3. Third module to carry out a Main measurement for determining the online plausibility of the measured data (hereinafter: "online plausibility")

The Module "Online plausibility" comes preferably then used when the standstill and reproducibility study performed successfully on the engine and no errors have occurred. The third module "online plausibility" can be used as a diagnostic tool integrated on the engine test stands at any time the measurement and calculation quantities during the To be able to control the main measurement. kick while If there is an error in the trial period, it will preferably appear on the Screen a warning message. At the same time, more preferably an indication of the probable fault location.

• • Grenzwertbetrachtungen: Vergleich der Mess- und Berechnungsgrößen mit den vorgegebenen Grenzwerten
• • Vergleiche/Ungleichungen: z.B. Vergleich der höheren Abgastemperatur mit der niedrigen Ansaugtemperatur (2)
• • Bilanzen: Physikalisch basierte Massen- und Energiebilanzen
• • Motormodelle: Physikalisch basierte Motormodelle, empirische Motormodelle, messdaten-basierte Polynommodelle

The methods used for error detection are based on mathematical equations and motor expert knowledge. The following methods are preferred:

• • Limit value considerations: Comparison of the measured and calculated quantities with the specified limit values
• • comparisons / inequalities: eg comparison of the higher exhaust gas temperature with the lower intake temperature ( 2 )
• • Balance sheets: Physically based mass and energy balances
• • Motor models: Physically based motor models, empirical motor models, measurement data-based polynomial models

Of the Analyzed errors usually do not yet give the exact cause of the error again. A logical probability-based method is for error isolation been developed to narrow the cause of the error.

4th module to carry out a Final measurement to determine the offline plausibility of the measured data (hereafter: "Offline plausibility")

With the module "Offline Plausibility" it is possible that the measurement data can also be checked after a measurement. For this purpose, the measured data are preferably read into a tool and the operating point for the operating point is checked ( 3 ). For example, errors that occur are displayed on the graphic interface. Standard images can be created to give an overview of the quality of the measurement data.

Furthermore Preferably there is an information menu by using all plausibility methods in a nutshell become. The methods used for fault detection and fault isolation are analogous to the methods in the module "online plausibility".

The error detection itself thus builds fundamentally on three pillars, as in 4 shown.

• 1. Gerätecheck
• 2. Statische Funktionsprüfung im Stillstand
• 3. Reproduzierbarkeitstest (Streubandmessung)
• 4. Beharrungserkennung zur Fallunterscheidung des transienten oder stationären Motorbetriebs
• 5. Bildung von Motorteilmodellen zur Prüfung der Plausibilität von Messgrößen bei laufendem Motor
• 6. Entwicklung eines Fehlerisolationsmechanismus für die Fehlerausgabe am Bildschirm

The necessary development steps for the method according to the invention and the associated device are therefore subdivided into the six main areas of work

• 1. Device check
• 2. Static function test at standstill
• 3. Reproducibility test (scatter band measurement)
• 4. steady state detection for case distinction of transient or stationary engine operation
• 5. Formation of engine submodels to check the plausibility of measured variables with the engine running
• 6. Development of a fault isolation mechanism for the error output on the screen

All Partial aspects are to develop against the background that both the type of engine as well as the metering channel equipment, i. the number and the type of metrics, can change daily. firm Decision structures, fault trees or encounter pattern recognition This quickly reaches its limits.

A Diagnosing motor measurement data may be more physical or logical Basis only if the individual channels have specific physical Sizes on the engine can be assigned. This requires a standardization of the name or the introduction of certain Measuring classes. Such a standard name list is the basis for the identification of all sizes that every inspection task must go ahead.

It also gives with the introduction such a list the possibility Knowledge in the form of empirical values, limit values and evaluation variables of a to assign each measurand. It has been shown that such a list is an essential part the examination diagnosis is.

A further boundary condition for the functioning of physically based diagnostic methods is the conversion all Sizes up International SI units: For example, by means of a separate Subprogram can a large number of units are automatically identified and converted become. This includes in addition to simpler things like the conversion of ° Celsius in Kelvin, the conversion of mg / stroke into mass flows, which as such already one Assignment of cylinder number and speed of the engine requires. Just the ECU sizes that often include a nearly redundant measurement data set, require this greater attention.

equipment check

Lots more complex analyzers such as fuel consumption meters and exhaust gas analyzers allow a status query and communication over network connections or the so-called AK-protocol. This is a textual protocol with which for short command codes, error codes or device values can be queried. At the same time, the self-diagnosis of measuring instruments has always been in recent years got better, so the gauges in some cases already a device internal Can show errors.

The accumulating information for two reasons Aspects are used. For one thing, the self-sufficient Status messages of the individual devices collected and output centrally on the diagnostic interface. In order to Although no new information is created, however, it does occur the error messages in all clear, so that no Measuring time is lost due to ignored error messages. So far it is quite possible that error messages are displayed on an exhaust tower, this Condition is perceived only with considerable delay.

To the others can by means of the meanwhile existing communication possibilities a much deeper diagnosis is made. You already have to existing data preferably only newly managed and evaluated become.

through a small database can have a history of pressures, temperatures and calibration values of the analyzers are constructed so that even minor bugs are discovered before the actual device failure can.

Advantageous It is at this point that the corresponding measuring devices are test stand-specific Measuring device like exhaust towers or fuel metering devices. They are subject therefore not the changes the other sensors that come with each new engine. The conditions for Permanent data monitoring is therefore very good and justifiable the unique structure and the installation of the corresponding files. Since the additional cost is only once, he is not the introduction this system in the way.

Static function test (standstill test)

The static functional test receives their justification by the problem that a more accurate signal analysis and checking calibration of various sizes while the engine is running more is possible.

So can e.g. at an oil temperature, when the engine is warm, 80 ° C indicating that 120 ° C would be the more correct value. It is So always a big one Range of uncertainty present.

Indeed it is natural desirable, a more precise error indication than 40 ° C to be able to represent.

There Most engines are either remounted on the test bench or overnight the possibility of cooling was given, the measurement of this condition offers itself in the morning the first fired measurement.

In that moment can much more precise Information about the setpoints of various sensors are made, namely independently from the influences of the engine. This affects all Temperatures, most pressures, Exhaust emissions, performance-determining parameters and the important ones Flow values of consumption and air mass measurement. These sizes will be preferably subjected to various tests to both the signal quality and to be able to determine the correctness of the values. The four tests carried out are intended will be briefly explained below.

1) limit value monitoring

The Most measuring data take place in the test stand standstill characteristic values, which are called environmental conditions can. This concerns e.g. the temperatures, all around the Ambient temperature have to move around. Out of this idea results the query of limit temperature violations, e.g. outside from 18 and 30 ° C lie. For air-conditioned test benches, such a window can be used Naturally tighter, making fault detection more sensitive becomes.

The same can be applied when monitoring pressures that must also indicate all ambient pressures. A difference to the temperatures, however, is that the measurement of the pressures can be absolute or relative to the environment, which makes automatic detection necessary. For this, the pressures in groups around 0 and in another group around 1000 mbar are sorted and evaluated separately. For physical plausibility with the engine running, important information can be obtained at this point as to how the individual pressure channels are to be converted to absolute pressure. A user input can thus be bypassed for this case.

The Exhaust gas analyzers need When the engine is at standstill, deliver values similar to those of fresh air same, that is: the trace gases must 0, the carbon dioxide analyzer 0.05 and the oxygen analyzer Show 20.8%.

Of Furthermore, all flowmeters such as fuel consumption or air mass, show the speed and also the torque zero.

Indeed are the measuring devices not so precise that all these values can be kept exactly. Out For this reason, preferably both a minimum value and Also, a maximum value specified as expert knowledge in the background is stored.

2) Comparison of the values among themselves

The Limit value query used only to analyze gross errors or sensor failures is preferably in a next step to the comparison various values added.

to compare let yourself be on the test bench usually only pressures and temperatures, since these sensors are used in larger numbers. After channels are detected and sorted out with limit violations, so only considered correct or almost correct channels.

By median value formation of these values, a representative environment can be calculated which any outliers that may occur during median value formation have no influence, as in 5 shown.

In order to There is now a reference size available, the a much more precise one Error analysis allows than the previous limit monitoring. By differentiating the individual channels with the "true" value, incorrect calibrations or contact resistances are detected.

Considered must be further preferably the digital resolution of AD card in conjunction with the measuring range, as it produces a minimal Difference between two temperature points or pressure points results. Is this difference over? the required deviation between the different pressures or Temperatures, the limit for the channel in question is preferred automatically adjusted so that a hasty error message is avoided becomes.

Example:

at a measurement configuration for the oil pressure if the sensor is e.g. For a measuring range of 0-10 bar designed. The Associated AD converter card features for example about a resolution of 10 bits. This corresponds to a step size of 9.7 mbar. In this case would a default of 5 mbar as the maximum permissible deviation from the reference value cause an inappropriate error message. The analysis of the digital resolution shows the problem, i. it automatically becomes a deviation around Increment (9.7 mbar) from the reference value. The error message is suppressed.

The digital resolution does not have to be communicated to the program, but preferably finds it within the standstill measurement an analysis of the smallest found increment instead of. Because the channels mostly have noise around one to two digital digits, This analysis can be very simple and requires no input on the part of the test stand staff.

3) measured value stability

Of the The term "measured value stability" describes the change, which the measured value passes through in a specific time window. is the test bench longer Time off, e.g. above Night, so an analyzer may ideally no changes show in his temporal behavior. This peculiarity can with a Curve approximation and subsequent curve discussion evaluated become.

The change of the measured value is extrapolated, for example, to 10 minutes and compared with corresponding limit values. The extrapolation is preferably carried out with a system identification, whereby a PT1 behavior is assumed. This is especially common with temperature sensors located near warming gauges. They then dodge a few degrees Celsius the environment. The PT1 behavior takes into account the decaying gradient, so that a more conservative estimation and error output takes place.

Here is the following to note:

A Drift can preferably be quantified only when the Measured value exceeds three digital increments. at only two increments, there is the possibility that an insignificant Drift of the analog signal in the middle between two stages a lot bigger drift in the recorded signal. The signal is in this Case rewritten to much more distant digital values. This connection, known as digitizing noise, can be found in unfavorable Position of the analog signal lead to mathematically excessive gradients. The digital resolution should therefore be taken into account, accordingly unnecessary To avoid error messages.

In 6 the difference of the gradient is shown when considering the analog and the resulting digital signal. It can clearly be seen that the gradient of the digital signal is almost twice as large. The actual conditions are therefore not reflected correctly.

4) Examination of the standard deviation

In addition to the drift, the noise of the measured values is a peculiarity that can not be completely prevented. However, the strength of the noise is a measure that can be used for a diagnosis. Investigations have shown that depending on the test bench and sensor type, a characteristic noise level is established. 7 illustrates this similarity with the example of five NiCrNi thermocouples.

Becomes the measuring chain, for example due to magnetic interference fields from high current applications (load units) disturbed, can the expression this channel characteristic differ significantly. For NiCrNi thermocouples is on VKM test rigs e.g. 0.012% of the measuring range end. As a measure of the noise level is For example, the standard deviation is used. At the calculation and standard deviation diagnosis are preferably two important ones To distinguish and consider aspects:

Drift:

The actual measurement noise can only be meaningfully calculated if the signal is drift-free. Otherwise, the value of the standard deviation dominated by the influence of the temporal change of the measured value.

Around but to be able to perform a clearer error analysis in this program preferably the pure noise values are investigated without drift, since this is more preferable already in a separate program part is done.

How to get in 7 For example, all NiCrNi elements have a similar noise characteristic. If one were to calculate the standard deviation without prior drift correction, the result would be a standard deviation, which is greater by a factor of 7 than that of the other sensors, at the top of the image for the course of the measurement data. This would provide a wrong image of the sensor. For that reason, the course of the measurement is preferably first occupied with a curve approximation and the difference between curve and measured data is recalculated. The result is a drift-free measurement data course with its original noise characteristic. 8th shows the result of this procedure.

Based For example, the present signal may be more detailed Statement about possible causes to be hit. The measuring chain as such seems in this case to be okay. However, there is a significant temperature drift at the sensor, e.g. mounted on one in the immediate vicinity heated lambda probe are returned can. As a result, the actually interesting exhaust gas temperature is not played correctly.

Digital resolution:

In the control of the standard deviation, the analysis of the digital resolution is preferably again necessary. This can be attributed to the so-called digitization noise, which causes a significantly higher standard deviation in unfavorable position of the analog signal to the digital values than the analog signal actually shows. This is in 9 illustrated. The maximum possible increase in Standard deviation is preferably at a half digital step size.

This phenomenon is common to be found in pressure sensors. For pressure transducers of, for example, 0-10 bar and AD cards with, for example, less than 12 bits may be the standard deviation already clearly above the desired Limits for the noise behavior are.

All Limit values, ie the minimum and maximum values, the allowed standard deviation as well as the maximum allowable drift are preferably in a database before and are based more preferably on the measurement of various Test benches. There the measuring devices Even with different companies are always very similar, is a subsequent modification This data is usually hardly necessary.

exhaust gas temperatures are used e.g. typically measured with NiCrNi thermocouples, whereas oil, Water and intake air temperatures preferably with PT100 thermocouples be measured. Even with the rest measuring instruments is usually a very similar one Measuring equipment at the various companies.

Steady-state recognition

The Program was developed under the constraint that the extra effort for the test personnel should be very low. But physical models have the Peculiarity of including many boundary conditions such as heat capacities, geometries, Volumes, etc. However, this is in direct contradiction to the required low parameterization effort.

A solution results if only the stationary parts of the equations be considered, which usually simplifies this considerably.

When Example is the carbon balance mentioned here, which in a later section treated in more detail becomes. For example, it becomes a system boundary around the entire engine drawn. Carbon becomes, for example, over the fuel and in small Quantities over fed the fresh air. About the Exhaust pipe leaves the carbon is usually the engine in different gaseous Links.

Around Carbon mass flows At a system boundary to be able to calculate, are usually the total mass flows in the corresponding Cross-section necessary.

The carbon mass flow due to the CO2 emission in the exhaust pipe is calculated, for example

in the stationary Engine running corresponds to the exhaust mass of the sum of fresh air and Fuel mass flow, so it is preferably present as a measured value.

in the Transient operation, on the other hand, would have to the exhaust gas mass e.g. also be calculated from these sizes. Indeed would additional approaches such as. filling and emptying method or characteristic method, to get to the current exhaust gas mass flow. This requires but already in the simplest case, all gas tube lengths and Volumes in the intake and exhaust area. Further increases the number of input variables by stroke volume and engine speed.

simultaneously is an identification of the transmission behavior the exhaust, air and fuel measurement required a time synchronization to reach the values. A calculation can therefore only always done in retrospect, if the engine may already stationary Operation has reached.

The additional complexity the models, the much higher Computational effort and the much more complex parameterization of Trying to justify this approach in view of only slightly higher Not useful. But also the increase the number of input variables a later one Error isolation more difficult, which is another major disadvantage is to be valued. Any size entering a model is also potential Cause of discrepancies within this model. For this Basically, models with few input sizes are preferred.

For the most plausibility considerations was for the reasons mentioned dispensed with the dynamic modeling. The corresponding stationary valid Models need Therefore, preferably be switched off in transient operation.

through Therefore, a persistence detection must preferably be between dynamic and stationary Engine run are distinguished. It is a tool preferably by means of a mathematical method (gradient formation, System identification, final value calculation) this necessary distinction meets.

The Procedure is, for example, in the operating conditions after load point change (Md, n, alpha) and parameter variation (AGR, ZZP ...) used. In addition to the transient operation with setpoint changes must preferably Also, the duration of the cold start can be detected as the validity different empirical models requires a warm engine. These are, for example, oil pressure or exhaust models.

Basics to the measurement process

In normal engine surveys for the purpose of base application and analysis of exhaust behavior, the engine is preferably measured at a plurality of stationary points. 10 illustrates the reaction of any measured variable with respect to a parameter or load change (jump of speed, injection quantity, ZZP or similar).

Of the Degree of persistence is on the test bench either by the test stand driver or monitored by a timed procedure. Since the stationary value for Time of measurement is not known, is preferred the change of the measured value, i. the gradient that the measurement data course describes.

options the steady state assessment

Basically come a set of methods by which determines the stationarity degree can be.

The Simplest consists of comparing averages, which consist of two groups be formed by measurement data. The change between these points can be easily converted into a straight line slope a what as a direct measure of stationarity can be viewed.

The same applies to a linear interpolation of the last n measured values. Again, you can the gradient is read directly from the straight line equation and be matched with a limit. The calculation can be very just after minimizing the square deviation from model to measured data.

The slope is here too

Both Although methods are easy to apply and in terms of computer load very suitable, which is why they are partly used, however is a statement about the gradient only for Values possible, that have been in the past for a while On the current value no statement can be made.

Of course, it is now possible to view only very short time windows of a few seconds in order to come very close to the current gradient. However, this leads to very uncertain statements, since the ever-present noise, small oscillations and even the digital resolution take too much place in the gradient calculation. For a practically useful statement is as a minimum to set a period of preferably 30 seconds. Consequently, at the earliest, for example, 30 seconds after stationarity has been reached, it can be detected. For stage durations in the range of 1.5-2 minutes, this time delay is already relatively large and should be avoided if possible.

• – Durch die Verwendung aller Daten wird die Sicherheit der Aussage gesteigert, da Rauschen und kleine Oszillationen einen kleineren Stellenwert erhalten.
• – Durch die Systemidentifikation und die Kenntnis des aktuellen Gradienten kann eine Zeitersparnis erreicht werden.

Furthermore, a method with system identification is recommended in which preferably all the available data material is used. Thereby two advantages can be achieved:

• - The use of all data increases the certainty of the statement, since noise and small oscillations get a smaller value.
• - By the system identification and the knowledge of the current gradient, a time saving can be achieved.

A another possibility in the system identification is given by that with the knowledge the model parameter is also the final value for a very long step duration is fixed and thus a direct comparison between actual value and Final value can be done.

Of the last approach, a calculation and comparison with the model end value includes, should not be continued at this point. The knowledge of the gradient is sufficient for this application.

theoretical Consideration of the transmission behavior

Basically, the transmission behavior of the system engine measuring device can be represented as follows:
An input quantity U may be an ignition point, a rotational speed, an injection quantity or any other input variable on the engine. A change of such a size causes in most output variables such as exhaust gas temperature, exhaust gas emission, torque, etc., a corresponding change according to the motor transfer behavior Gs. This is followed by another transmission behavior Gm of the measuring device.

Mostly in addition to interactions between the output variables, as in 12 shown.

To the Example like the input quantity as the injection quantity Serve a diesel engine whose increase in an increase in exhaust gas temperature entails. The conditions in the combustion chamber are very fast on, i. the temperature increases very fast and significantly.

simultaneously heat but also the components such as combustion chamber walls, manifold and turbine housing, so that the heat transfer decreases and the temperature in the exhaust system continues to rise. This Process is much slower in nature and can last for minutes extend. Overlay it So different processes with different time behavior.

In order to obtain a simple system identification, can. the series-connected transfer functions are combined, so that a parallel connection of in the simplest case two transfer functions remains, as in 13 shown.

The Approximability of such a function is exemplified by the charge air temperature being represented. Again, the simplest case comes, namely the Parallel connection of two PT1 links, used.

The quality of the model approach can be shown by any temperature curves, as in 14 shown. Gerde these are very good indicators of stability and steadiness of the engine.

But even faster measurement techniques with dominant noise can be described with this PDT ₂ -Glied. However, the difference to the much easier to calculate PT1 approximation is not so clear. As an example, in 15 the response of the turbocharger speed to a load jump shown.

Well to recognize is the big one Scattering of the signal, among other things, on large speed fluctuations going back within a working game. The modeling allows but Again, the decision whether the channel is a significant gradient describes or not.

Important But it is also here that with one methodology all channels are preferred Equally well described must be input errors and Parametrieraufwände to avoid.

If the model parameters are present, the stability of the current point can be assessed. The gradient results by means of

The Determination of the model parameters is no longer analytical in this case solvable by means of a closed equation, since it is a nonlinear Equation with two unknowns T1 and T2 in the exponent.

through a gradient method, it is possible to online the function the square deviation between measured data and target function so that the model parameters are preferably close to zero Real time.

A very simple method can be used since the error function preferably has no local minimums. The error function of 16 corresponds to the charge air temperature already shown. It can be seen that the minimum can be found, for example, in the combination of a fast and a slow time constant. Of course, the variables can be used as a mirror image, which is why, for example, two minima occur here.

For the key variables exhaust gas temperature, Temperature after intercooler, oil and water temperature as well as boost pressure, the gradient is preferably calculated continuously. If the value falls below a defined value, these channels then become preferably referred to as stable. When old key sizes stable are further preferably carried out the activation of only stationary valid methods.

Error detection (FD Fault detection)

The Error detection preferably includes everything needed to detect Inconsistencies within the measurement data contributes. This does not include yet the consideration, what cause these inconsistencies have caused.

The detection itself succeeds, for example, by observing all or part of the process variables. Depending on the combination of inputs and outputs and the strategy of the test, the in 17 define defined subgroups.

in the Here are some examples of the above points before then, by way of example, some models from the worked out Software will be presented.

Control directly measurable signals:

These Subgroup preferably forms an essential building block to Detection of sensor failures. Depending on the engine type are for the common ones channels preferably limits defined, which also in transient operation no over- or may fall below. Due to the wide variability In internal combustion engines, these limits are preferably very high generously laid out, that's why only failures can be detected. A poor calibration of a measured variable can thus rarely be detected become.

On a gradient monitoring or a signal prediction using polynomial approaches omitted at this point, over there the system engine brake and also the objective of the survey too little information is available.

Analysis of signal models:

Depending on the location of the measurement, the analysis of the individual signals can be a very interesting source of information Condition of either the engine or the sensor. Thus, by means of Fourier transformation, the frequencies of the motor and, for example, disturbing network frequency can be selected and evaluated. Even a disturbance of a single cylinder can be detected by the signal analysis of the exhaust gas pressure pulsations or the lambda signal. The corresponding error cases can then be analyzed on the basis of the amplitudes in the various higher orders of the motor base frequency.

Analysis of process models

Compared to the models described so far provide process models to the first Time an analytical redundancy of different values. So it can Direct measurement and calculation variables with each other be compared, what the error sensitivity to the Significantly raises the limit value assessment.

process models are in this case calculation rules, which are an output variable from different Determine input variables.

The function type in general is: Y = f (X, P, F)) [7]

In which X is the group of input quantities, P on or several model parameters and F an arbitrary disturbance or Sensor error represents.

Parameter Estimation aim a particular model in its temporal behavior analyze and from there on the model parameters such as stiffness, Volumes, resistances etc. close to be able to. A change this parameter may already be an error Problematic are these methods, since most errors already arrive with the first measurement point. That is, the model is learning it Mistake as correct and sets, for example, a 3-fold too high Intake manifold volume as basic parameter. This error will eventually only found if an input is randomly repaired and now the parameters no longer match.

Better suitable models are those in which the model parameters from the beginning to be present. These are called parity models. In these can be calculated from the residuals, i. the difference values between model and measurement, be concluded on erroneous input variables.

Preferably limited on models in stationary points procedurally a whole set of physically based bills, in particular in the field of exhaust gas composition, air mass and fuel consumption provide very accurate results, although based on a modeling of Instationäranteile is preferably omitted.

Next the single channel analysis by means of signal model or limit value control and the formation of process models is the fourth option the calculation of parameters too without exact knowledge of motor and load unit possible. These Provide parameters both valuable information about the goodness of Combustion process as well over the state of various engine subsystems such as compressors, Intercooler or turbine.

• – direkte Vergleiche zwischen Messgrößen
• – empirische, messdatenbasierte Polynommodelle oder neuronale Netze
• – physikalisch basierte Bilanzgleichungen (Masse und Energie)
• – physikalisch basierte ideale Vergleichsprozesse

The motor part models used can preferably be subdivided further, for example

• - direct comparisons between measured variables
• Empirical, measurement-data-based polynomial models or neural networks
• - Physically based balance equations (mass and energy)
• - physically based ideal comparison processes

exemplary The following are three different models with very different Model structures are presented.

1) carbon balance

The carbon balance indicates how much carbon is being supplied to the engine via the system boundary and how much the engine is giving off. In 18 the corresponding carbon paths are shown simplified. About the air is usually a small amount of carbon supplied due to the naturally present CO2 content. However, the share that is significant in the high-load range is preferable the fuel. Carbon is also preferably removed via various gaseous components in the exhaust gas.

The proportions of air and fuel can be derived directly from corresponding measured quantities. However, the measured values of the exhaust gas measuring system are preferably not the true concentrations in the exhaust pipe of the engine, but the exhaust gas has passed on the way to the analyzer, for example, a gas cooler in which the water formed during combustion was condensed out, as in 19 shown. However, this does not happen with all exhaust gas components, as some of them can also be analyzed directly from the raw exhaust gas (HC, NOx).

It is therefore preferably first a conversion of the dry measured exhaust gas quantities into the actually existing ones Raw gas values necessary.

For this Although there are finished formulas from the emissions legislation, however, these formulas are always numerical equations with some physically based additional terms, but usually not on alternative Fuel types or special combustion conditions designed are (biodiesel, GTL, Optimax or incomplete burns at lambda <1).

There it's about tracking down this small misalignments goes, these neglects can not be accepted. It therefore became a requirement Humidity correction with the input variables lambda, fuel composition and exhaust values compiled.

Generally valid moisture correction

Basically, the Gas masses, apart from water vapor, preferably before and after the radiator remain.

The Measured values of the gas analyzers represent, for example, room volumes the above following relationship and knowledge of the molecular weights in parts by weight can be transformed.

Of the Weight fraction multiplied by the total mass flow represents the Mass flow of a gas component, which after the mass conservation on the gas cooler must remain constant (assumption: It is preferably no gas in the condensate over).

Basically, the following transformation is preferably used for the conversion of dry measured values to moist raw exhaust gas: y i_feucht = p i_trocken · Corr_tr_fe [10] With

Around the four unknown sizes of the It is possible to determine molar masses and gas masses before and after radiators preferably the following approach is chosen.

Under the requirement that the entire exhaust gas mass dried in a cooler will, lets at least indicate the moist exhaust gas mass directly. Later will make it clear that each partial flow also fulfills the same equations.

The moist exhaust gas mass is preferably the sum of all supplied mass flows. m Abgas_feucht = m air + m fuel [12]

For the dry Exhaust mass must now preferably first Assumptions are made as to how much water was produced during combustion is. A hydrogen balance, for example, provides information about this.

hydrogen is supplied via the in the fuel present proportion, hydrogen is removed in the water itself, in molecular form as well as in hydrocarbons. in the Case of hydrocarbons is very often the replacement size C3H8 (Propane), because usually also calibrates the hydrocarbon analyzers become. The hydrogen content from the humidity is at this Place neglected.

Around independent of the mass measurement to be let The equation is based on the stoichiometric air requirement rewrite and dissolve after the molecular weight.

The proportion of free hydrogen can preferably be determined by means of the chemical equilibrium of the water gas reaction. Carbon dioxide (CO ₂ ) and hydrogen (H ₂ ) react mutually to carbon monoxide (CO) and water (H ₂ O). Especially at high temperatures and sub-stoichiometric engine operation, this relationship is no longer negligible. CO 2 + H 2 ↔ CO + H 2 O [16]

To this reaction leaves preferably define an equilibrium constant, the example for Gasoline engines at Lambda <1 for example, takes values around 3.5. In addition, this reaction hardly plays another role. For example, for diesel engines with lambda values <1, the same Invoice to be used. The influence decreases due to The very low CO values preferably strong, so it arises no appreciable calculation error.

The molar mass of the moist exhaust gas can be represented as follows:

womit sich zum Beispiel eine zweite Gleichung mit den Unbekannten H₂O und der Molmasse bestimmen lässt.At the same time:

which allows, for example, to determine a second equation with the unknown H ₂ O and the molecular weight.

If, for example, the hydrogen content is replaced by the relationships of the water gas reaction and the molar masses are then equated, then the water vapor content in the raw exhaust gas can be stated. For the sake of clarity, some of the molar mass differences are already shown as numerical values

From Formula 17 and 21 follows, for example:

Analogously to equation 20, equation 22 follows:

Also the input variables for Calculation of the dried exhaust mass flow is now preferred certainly.

By This illustration will also show the independence of humidity correction clearly from the exhaust mass. For example, become the four unknowns used in the formula of the correction value, so shorten the exhaust gas mass flows out. The only other unknown remains Lambda, but that can be determined without problems preferably from the exhaust gas sizes.

The The derivation shown preferably uses the measured exhaust gas concentrations for the Calculation of dry and wet molecular weights. The molecular weight of the dry exhaust gas, however, preferably requires for their calculation dry exhaust gas values, the molar mass of the moist exhaust gas, of course, too Values of the moist exhaust gas.

For this purpose, an iterative calculation step is preferably necessary. For example, the calculated correction value performs a preliminary humidity correction so that all concentrations are both humid and dry. In a second step, it is then possible to specify a new, more accurate correction value. Another iteration is then no longer visible in the quality of the result. The derived version will be compared with two other methods, as in 20 presented in order to give an impression of the resulting deviations, the ultimate reason for their own calculation method.

Either the empirical approach as well as the correction formula according to the EU exhaust emission legislation are numerical equations, in which neither the incomplete combustion still the fuel composition is received. The influences, the incomplete Combustion is most evident in the accuracy of the carbon balance noticeable.

modeling

After this now preferably those which are directly at the system boundaries Sizes were calculated (Air and fuel mass measurement do not require any corrections), can balance over the entire engine are set up.

The Carbon masses introduced into the engine are subdivided into Example in the proportions from the air, the oil and the fuel.

Air:

Carbon is transported in the air, for example, by the existing CO ₂ .

in the Part load range of diesel engines, this proportion can be a few percent from the fuel carbon content That is why neglecting the required accuracy inadmissible is.

Fuel:

Up to 89% of the fuel weight is due to the carbon. The fuel data therefore plays a decisive role in addition to a precise measurement of consumption. C_Fuel = C fuel · m fuel [27]

The Values for different fuels are preferably used as default values in Program before, when using alternative fuels is but on preferably a fuel analysis and the entry of the current Values needed.

In the table of 21 the areas of the composition are presented by means of various analyzes within the project period. If no current fuel analysis is available, the confidence level of the carbon balance must preferably be widened by, for example, approximately 1.5%.

CO ₂ :

The gaseous Ingredients can very simple with knowledge of the molar masses of proportions by weight be converted. Multiplied by the exhaust gas mass flow results then the carbon mass flow of a particular gas component.

CO and HC:

The carbon flow, resulting from the CO and HC emission, is calculated analogously to the CO ₂ calculation.

For example, the carbon balance can be written as follows:

Of the percent difference between added and removed carbon is preferably the error threshold, which leads to the triggering of an error message. Responsible for the Violation of the limit can until then all sizes involved.

following naturally turns for every Test method the question how big the Residues may be before an error message should occur. This is preferably an analysis of systematic error necessary. The systematic simplifications, some of which are also consciously accepted, are intended for the carbon balance briefly discussed become. First, there is a potential soot emission to call in the engine. Carbon is here in solid form in the exhaust gas emitted. In the presence of soot particles are also stored always some hydrocarbons on the soot particles. The particles However, preferably before the Abgasmessleitungen for example mechanically deposited in filters to prevent contamination of the whole Exhaust gas measuring section to avoid. They therefore dive into the exhaust gas values not on their carbon content is therefore usually not in the equations considered. A possibility is the consideration a blackening index or opacity measurement. Preferably based on a correlation between blackening index or soot emission the influence on the residual carbon balance should be estimated. It is believed that a blackening number of, for example 3 is measured, a value that certainly already regarded as an extreme is. In normal operation of modern diesel engines, values of 2 hardly exceeded. Only for DoE test runs is the operating range is extended so that values above 3 occur can.

In connection with the soot meter AVL 425s, for example, there is a relationship between the soot concentration RK and the blackening number SZ of the following form: RK = 12.22 · SZ · e 0.38 · SZ [35]

The soot concentration, however, preferably refers to the mass in mg / m ³ , so that initially reasonable assumptions about the exhaust gas volume must be made. The exhaust gas temperature in the vicinity of the measuring location (downstream of the turbine housing) is preferably approximately 700 ° C. in the case of these operating ranges rich in diesel engines. The pressure can be approximated by turbine with ambient pressure, by means of the ideal gas equation can thus be determined and the flow rate:

The exhaust gas mass may preferably be estimated by means of the stoichiometric air requirement, assuming that the lambda should be 1. Any other lambda values do not lead to such high blackening numbers anyway. m. exhaust = m. air + m. fuel = m. fuel · (Λ · L Stö + 1) [37]

Substituting the values and refers the soot mass to the fuel mass in the carbon mass,

So For example, this results in a model deviation of 0.6%.

When second inaccuracy is the oil consumption to call. The oil pan provides carbon storage in terms of carbon footprint be entered in the carbon and discharged can. The more important case is certainly to be seen in the burning of oil, especially in the field of mechanical endurance test stands. The engines will be there long times under high loads, which makes them quite strong Wear subject. Oil consumption of in mathematical terms, for example, 1 1/1000 km are quite common. Will a mean fuel economy of for example 10 1/100 km for a medium car diesel engine assumed, which is quite justified at the present high loads is, so is the carbon content of the oil for example, 1% of the fuel fraction. This is preferred assuming the same carbon content in oil and diesel fuel.

By combustion chamber wall wetting it is possible that fuel in the engine oil intercalated or stored. This effect may preferably occur at very long or late conveyor angles when the injection jets strike the relatively cool cylinder walls. The fuel then preferably passes through the piston rings and passes unburned into the oil pan. In this case, the CO ₂ emission drops below the expected value and an error occurs. However, this is also desirable because regardless of the interpretation of these results, a very critical engine operating condition has occurred requiring the attention of test personnel.

Further Inaccuracies can be seen in the determination of the moisture correction, but at more careful Calculation has no significant influence.

Otherwise Preferably, the influences from the measurement inaccuracy add up and the digitization of all measuring devices involved in this systematic error added. Because the analysis of the entire measuring chain not possible in practice is, is preferably further based on empirical values, the error threshold finally Are defined.

For the carbon balance is called that, in addition to the largest possible systematic error of for example 1.6% still for example approx. Add 3.5% measurement error, resulting in an error threshold of Example 5% is achieved. If this error threshold is exceeded, so at least one erroneous input is assumed.

Finally, be in 22 an example from the evaluation part of the diagnostic software shown, in which the respective model quality is displayed. By way of example, the added and removed carbon mass for a load cut of a 1.8 l turbo-charged gasoline engine is shown here. Despite being enriched to lambda values of, for example, 0.75 in the right hand area of the graph, the model deviation may preferably be kept well below 5%.

2) exhaust gas temperature model

in the Unlike the physical models, as with the carbon balance representative it is possible with data-based models, very much more complex relationships display. This can be particularly beneficial if not all model parameters or special measuring sensors are available.

The following exhaust temperature model, shown in 23 , based on this idea.

Next a physically based total energy balance for determination the exhaust gas temperature was preferably additionally an empirical approach implemented, which even with a smaller number of measuring channels, a review of the Exhaust gas temperature allowed. The missing input variables were further preferably stored in the form of empirical knowledge.

The System limit is preferably directly around the cylinder unit placed, i. it is the air in the intake and the exhaust gas temperature in the calculation. The achievable temperature in the exhaust gas results preferably from the supplied Energy (fuel), the discharged Performance, the amount of heat removed and the amount and condition of the intake air.

In a first step is further preferably calculated which Residual enthalpy remains after deduction of the mechanical work in the gas.

through an empirical term is now further preferably calculated how much of this energy remains through heat transfer in the engine and how much energy is available to the exhaust gas. This empirical Term thus includes the energy conversion through the heat transfer. This size is metrological very difficult to grasp, an empirical representation helps here to a much simpler view.

The used polynomial parameters preferably originate from a regression calculation over several different diesel engines. Again, there is a problem this diagnostic task. The models must preferably one as possible cover a wide range from small cars to heavy truck models.

One Polynomial model for Only one engine can be easily predicted on forecasts of accuracy Example less ° C to be brought. A model only for single engines, however, needed preferably a data management that currently not on test benches provided.

• – Mittlere Kolbengeschwindigkeit (kg)
• – Verhältniszahl aus Zylindermasse/Zylindervolumen (m Vol)
• – Lambda (lam)
• – Effektiver Mitteldruck (pme)
• – Förderbeginn (FB)

The input variables for determining the heat transfer are, for example:

• - Average piston speed (kg)
• - Ratio of cylinder mass / cylinder volume (m Vol)
• Lambda (lam)
• - Effective medium pressure (pme)
• - Start of funding (FB)

und den beispielhaften Parametern Proportion = α1 + α2 · kg + α3 · kg 2 + α4 · kg 3 + α5 · pme + α6 · pme 2 + α7 · pme 3 + α8 · kg · pme ... + α9 · λ + α10 · λ 2 + α11 · kg · λ + α12 · λ · pme + α13 · λ 2 · Pme + α14 · ME_FB + α15 · ME_FB 2 ... + α16 · m_V [40]

and the exemplary parameters

With the proportion of residual enthalpy remaining in the exhaust gas, it is further preferable to calculate the exhaust gas temperature. For this purpose, however, the preferably specific heat capacity of the exhaust gas is necessary, but just has the exhaust gas temperature as an input variable. Therefore, an iteration is necessary that calculates a preliminary exhaust gas temperature based on an estimated heat capacity cp of, for example, 1030 J / kgK, and a corrected exhaust gas temperature with correct cp in the next step. This cp corresponds to the specific heat capacity of air, ie the chemical composition of the exhaust gas is preferably disregarded. Only the temperature influence is stored in the polynomial. However, such inaccuracies in modeling can preferably be mapped in the empirical part, so that no major errors result.

For example, the available diesel engines is in 24 the comparison between model and measurement is plotted.

The Spectrum of measured engines is enough for diesel engines, for example the lowest car performance class up to commercial vehicle engine for heavy Tractors.

One such universal Model assumes that the input variables are preferably exclusively specific Sizes or ratios include that for independent of each engine type defined by its size can be. Due to the structure of the calculation, the polynomial value is also a ratio preferably always between 0 and 1, i. Again, the engine size is not one. Only in the combination of the polynomial value with physical approaches it is possible to calculate a motor-specific temperature.

The mean model deviation, also referred to as prediction accuracy, is for example 29 °, the coefficient of determination R ² is for example 0.91.

As in the carbon balance must also be preferred for this model an error threshold is defined, if exceeded an error message he follows. Naturally The procedure involves a number of systematic errors.

First of all The model does not correspond to the actual physical conditions that to the expression lead to a certain temperature. Even if the added and removed Enthalpia and achievements still to be properly evaluated, exists but in the heat transfer a clear uncertainty.

This depends on from the combustion chamber geometry, i. for example, from the heat-absorbing surface, from Turbulence degree of the cylinder internal flow as well as of the temporal and spatial Wall and gas temperature. All these sizes are usually not on the test bench before or can at all not be measured. The universality can still be achieved become; Since diesel engines are all very similar in geometry, the conditions so are quite comparable.

The purely statistical values for the model evaluation could with appropriate parameter selection be much better. However, here was not on a minimization the mean error, but the maximum deviation between model and reading was preferably by a corresponding Algorithm minimized. This worsened the top mentioned values, but for error detection thus preferably results in narrower limits, in which the exhaust gas temperature must be.

The maximum deviation over all engines are at 98 °, for example. However, this is preferably true only a few measuring points whose correctness is not finally clarified could.

Therefore For example, the confidence limit has been set to, for example, 100 °. Values above are at least conspicuous and require preferably by the tester driver. For the illustrated Learning data means that, for example, about 1% of all measuring points showy would be declared. The conservative interpretation justified through the demand for minimal over-detection rates. That is, it should rather miss a mistake be as the test stand driver is overloaded with error messages, which are rather contentious after reviewing the data. Decide here ultimately the requirements of the users.

3) Check the turbocharger

After now almost all diesel engines in the passenger car and commercial vehicle sector preferably with a door bolsters are equipped and gasoline engines in the wake of downsizing more and more often receive a turbocharger, a specially adapted to turbocharger diagnosis becomes more important.

There the charger type on the test bench preferably as known and can be assumed in the automotive companies the loader pads are still preferably in digital form, was in this context a special diagnosis of the turbocharger and created the sensors concerned him. The main problem is as well as the automated implementation of the very few digital ones Values in a higher quality and diagnostic way and way.

In principle, four could relationships To be defined:

For the compressor are preferably two maps. The first preferably relates the total pressure ratio via compressor, the reduced air mass flow and the reduced supercharger speed to each other. In addition, it is also possible to represent an efficiency map for the same input variables pressure ratio and air mass flow. The quantities are all not directly comparable to measured data, but are preferably in standardized form. The connections are in 25 clarified. This allows the comparison even under changed environmental conditions. The total sizes, characterized by a small t in the index, refer to the state variables in which the gas was ideally brought to rest.

Similar relationships can be defined for the turbine. The mass flow rate is hereby preferably replaced by the throughput parameter. This is a dimensioned size, which in addition to "kg / s", for example, also the units "K ^ 0.5 / bar" contains. The vividness suffers of course something. Also, the pressure ratio and the speed are preferably defined on the turbine side in a different form. The connections are, however, in 26 shown. In the case of a VTG turbine, the variable turbine geometry is preferably added as a further family parameter. For reasons of clarity, however, this presentation is omitted.

The Data material on the test bench first is present, preferably describes the characteristics and fields shown only in a few bases.

to Preparation, a separate, upstream program was developed the data preferably from a standardized Excelfile read in, processed and stored in a neural network. At this point is not intended to represent the complete theory of neural networks Please refer to the relevant literature.

This neural network is preferably automatic in online application read in and evaluated. Special value had to turn this on minimizing the extra Effort to be laid. The once digitized maps can with simplest means every test stand be given.

Parallel For this purpose, the generated network should preferably also for the simulation of Turbochargers in engine models enabled which is why special emphasis is placed on comprehensive interpolation of the map was laid. This is necessary as it is about tracking down Measurement errors goes. That is, even in the case of the error, it is preferable a correct calculation of the map data must be made. The normal one Operating range can be quite abandoned, which is a good Extrapolation required. Furthermore is expected that, for example, diesel engines in the future with Throttles and widely varying back pressures are working (particle filter problem), and therefore also to expect an expansion of the operating area is.

For the calculation of neural networks, it is preferable to take some steps, which are discussed below:
Taking into account only the measuring points made available by the manufacturer, the entire operating range of the loader is not reproduced correctly.

In addition, the use of standardized interpolation mechanisms causes problems that lead to false results between the interpolation points. In 27 is such a survey shown graphically. The interpolation is preferably done by Matlab by means of linear or polynomial approaches, but all do not provide a satisfactory result. The reason is that the interpolation with the surrounding interpolation points leads to deformations in the map due to too few data points, the phy sikalisch are not meaningful.

When solution offers, for example, based on physical conditions increase number of support points whose result is preferably a data set for training neural networks.

When first approach to the solution This problem is preferably first the values within a Speed line interpolated by means of polynomial approaches, as defined for this purpose and enough interpolation points available.

In Further stages are more preferably these speed lines extrapolated to the right and left, using different assumptions to be hit.

In the right part of the map, for example, a sharp drop in the speed line is required because preferably sets the speed of sound in the narrowest cross section of the supercharger. As a result, no further increase in the mass flow rate can be achieved even with further de-throttling of the compressor (stuffing limit). This relationship is usually indicated in the basic surveys. Taking preferably the last range of the speed line and extrapolated by means of a polynomial second degree, this approach is sufficiently well taken into account. This also applies to the efficiency under the boundary condition that, for example, it becomes zero at Π _v = 1.

This approach can certainly be critically questioned. The efficiency of the compressor, shown in 28 , is in some works at a total pressure ratio of 1, that is occupied in the absence of pressure build-up, with positive values. Other studies indicate that the efficiency at Π _v = 1 is not defined, as neither an isentropic nor a real state change from p1 to p2 can be defined.

If one looks more closely at the formula of isentropic efficiency, it becomes clear that the value must at least go to zero. The counter preferably automatically approaches zero, since Π _v = 1, but by heat conduction the denominator is at the same time preferably slightly larger than 0, ie the efficiency can be calculated independently of the interpretation of this value.

The heat conduction is synonymous for that responsible for the review of the Efficiency preferably remains limited to total pressure ratios greater than 1.3.

Of the Value of, for example, 1.3 is here an empirical value above which the influences the heat conduction preferably sufficiently behind the influences of the temperature change withdraw by pressure build-up.

In the underlying map areas, the efficiency in dependence the turbine temperature preferably vary by up to 12%. A Efficiency diagnosis is therefore in these areas without prior Measurement of the entire system is not possible.

The assumption that the efficiency becomes 0, for example, at Π _v = 1, is nevertheless important in forcing the quality of the extrapolation in these ranges to physically meaningful values.

Of the left part of the map is limited by the surge line, the separates the stable from the unstable operating range. Especially with acceleration the engine drops a large Part of the operating points very close to this limit, so why Here preferably an extrapolation should take place. It exists thus the possibility at least the more harmless form of pumping, the so-called "surge", with.

For this are preferably approaches More and Greizer in conjunction with the theory of radial Equilibrium related.

In 29 is a typical speed line shown. It starts preferably with the positive on the right Mass flows. The subsequent drop is the unstable region, which is preferably traversed periodically in the case of the surge or the deep surge, ie the classic compressor pumping with reversal of the flow direction.

For the diagnosis is hereby preferably an extrapolation up to mass flow rates of 0 sufficient. Negative mass flows fall anyway through the isolated viewing of the air mass signal.

wobei p_r0 zum Beispiel mittels des radialen Gleichgewichts ermittelt werden kann. Das radiale Gleichgewicht führt gerade zu dem statischen Druck, der sich aufgrund der Zentripetalkräfte bei Durchflüssen von 0 bei einer bestimmten Drehzahl des Verdichters einstellt. Damit liegt neben den Messwerten aus dem Kennfeld vorzugsweise auch der y-Achsenabschnitt physikalisch begründet fest.An extrapolation to the left of the surge limit can preferably be carried out by means of a third degree polynomial according to the following relationship:

where p _r0 can be determined, for example, by means of the radial equilibrium. The radial equilibrium leads just to the static pressure, which occurs due to the centripetal forces at flows of 0 at a certain speed of the compressor. Thus, in addition to the measured values from the characteristic diagram, the y-axis section is also preferably determined physically.

Integrated the above equation and eliminates the density by means of the ideal one Gas equation, we obtain preferably the pressure at an air mass flow of 0.

The Radii of the compressor wheel can in most cases preferably be taken directly from the compressor designation. Otherwise, preferably a further measurement must be made.

Of the Efficiency, however, is preferably by means of the dimensionless Torque coefficient determined, plotted over the Flow coefficient can be extrapolated linearly.

Of the Torque coefficient is preferably

umgeformt werden kann. Wird dieses Drehmoment mittels Formel 45 dimensionslos über dem Durchflusskoeffizienten aufgetragen, so kann vorzugsweise eine lineare Interpolation erfolgen. Der Durchflusskoeffizient selbst wird weiter vorzugsweise mittels

berechnet. Das Ergebnis dieser Vorgehensweise ist in 30 für eine Drehzahllinie ersichtlich. Anhand des extrapolierten dimensionslosen Drehmoments kann auf den Wirkungsgrad zurückgeschlossen werden. Zwischen Nulldurchgang des Massenstroms und Stopfgrenze sind jetzt vorzugsweise folglich Drehzahl- als auch Wirkungsgrade bestimmt, so dass ausreichend Lerndaten zum Trainieren eines neuronalen Netzes zur Verfügung stehen. Das Ergebnis ist eine flächendeckend gute interpolierte Information über das Verdichterverhalten auch in den Regionen, die in den Basisdaten noch nicht bestimmt waren, wie in 31 dargestellt. Mittels dieser teils physikalischen, teils empirischen Extrapolation kann der große Nachteil der Neuronetze, nämlich die mangelhafte Extrapolation über die Lerndaten hinaus, behoben werden. Für eine Diagnosesoftware ist dies besonders wichtig, da es durch nicht bestimmte Bereiche oder durch falsch extrapolierte Werte zu Programmschwierigkeiten kommen kann.At the same time:

can be transformed. If this torque is dimensionlessly plotted above the flow coefficient by means of formula 45, preferably a linear interpolation can take place. The flow coefficient itself is further preferably by means of

calculated. The result of this procedure is in 30 for a speed line visible. Based on the extrapolated dimensionless torque can be deduced on the efficiency. Between zero crossing of the mass flow and the stuffing limit, speed and efficiencies are now preferably determined, so that sufficient learning data is available for training a neural network. The result is a comprehensive good interpolated information about the compressor behavior even in the regions that were not yet determined in the basic data, as in 31 shown. By means of this partly physical, partly empirical extrapolation, the great disadvantage of the neural networks, namely the lack of extrapolation beyond the learning data, can be eliminated. This is especially important for a diagnostic software, as non-specific areas or incorrectly extrapolated values can lead to program difficulties.

Of the The network type used is preferably a two-way forward network Levels, due to the solution algorithm is referred to as a feed-forward backpropagation network.

These networks preferably each have two input variables, namely the total pressure ratio and the reduced air mass. For example, the output variable is either the reduced supercharger speed or the isentropic efficiency, as in 32 shown.

In The two networks preferably have no physical knowledge or a certain logic filed, it is more about a pure data storage. The advantage with correctly selected number of neurons is the very smooth interpolation area. Polynommodelle tend to overshoot even when storing the efficiency between the speed lines, something avoided with this approach could be.

Each neuron includes 33 a transfer function that significantly influences the behavior of the neuron. The transfer function of the first two hidden layers is preferably of the "tan-sigmoid" type, that of the output neuron is more preferably "purelin" (see also US Pat 34 ). Here is a brief explanation of the processes on a single neuron necessary:
Within the neurons of the first two hidden layers, the information is preferably weighted, added up, offset with an offset value and then passed on between -1 and 1 depending on the function value of the transfer function. Values above or below are preferably not permitted. In the output neuron, preferably all input information is added together, ie the output value can assume any values.

The quality of the map comparison can preferably be based on 35 , Result of a commercial vehicle diesel engine measurement, can be assessed.

Of the Efficiency is the more sensitive part, but he can with accuracy of, for example, +/- 6%. The Loading speed may preferably be specified to within 3.6%, for example become.

For the diagnosis are preferably error thresholds deviations from the efficiency of for example 10% and deviations from the loading speed of the Example 5% approved.

Becomes at higher value Measuring equipment a larger value For accuracy, these accuracy specifications are preferred easily editable by means of an Excel spreadsheet.

For the turbine let yourself preferably with similar ones Means generate another neural network for the relationship between Speed, throughput parameters and pressure ratio.

The Consideration of the efficiency can on the turbine side means the normal measurement technology and equipment but not successful carried out become. The reasons are, for example, in the high dynamics of gas sizes in the exhaust manifold with the associated problems in metrology.

When the exhaust valve opens, a very high pressure gradient always sets in, followed by a significantly slower pressure reduction, as in 36 shown.

There The pressure sensors of the piezoresistive design usually temperature sensitive they are preferably over longer Pressure pipes connected to the measuring point. The flow resistance but these measuring lines is preferably approximately proportional to the square the flow velocity. Therefore, the pressure wave directed to the sensor becomes higher resistance opposite than the outflowing pressure wave. Although it turns out preferably directly on the sensor a pulsating Gas movement, however, is due to the flow resistances of the Mean value always smaller than in the manifold pipe itself.

Similar Inaccuracies occur in the temperature measurement in the manifold area on. The during the outflow After opening the exhaust valves present hot exhaust gases have on the sensor only a short residence time. A larger residence time is the gas quantities reserved, which is nearing "outlet closes "leave the cylinder. However, some parts are purge air (valve overlap), which is characterized by significantly lower temperatures. Both Effects lead to a seemingly higher Efficiency of the turbine, especially in the partial load easily values of for example 300%. These purely mathematical values of course, describe not the correct, physical facts.

A correct measurement of pressure and temperature preferably requires a much more elaborate Measurement setup, which certainly can not be implemented consistently on the test bench. Out For this reason it is necessary to deal with the deficiencies set out and preferably an excessive efficiency permit. A limit analysis of the efficiency at higher turbine pressure ratios is sufficient in this case, however, to make appropriate mistakes detect.

Error isolation (FI Fault isolation)

The present fault detection methods are based on four pillars. The First is also the simplest and often used limit monitoring or monitoring. Here are preferably limit values for different engine types the common ones Measured variables determined. If these are exceeded during operation, then a Error message.

The second pillar form inequalities, the simple relationships between different ones To press or allow temperatures. These are preferably direct comparisons between different ones Measurands for which one defined larger / smaller relationship can be defined.

When Third, systems of equations are used which preferably determine Sub-processes or fundamental Characteristics of the engine brake system. Because the equation systems due to simplifications in the modeling not quite error-free are preferably an individual tolerance band is necessary, that must be exceeded to output error messages.

The fourth pillar consists of engine submodels, preferably based on previous knowledge stored in data stores and compared online with the system behavior can be (For example, maps of turbochargers or empirical models for exhaust gas concentrations).

all Methods in common is the generation of error symptoms. This are for example binary results, whether one or the other equation (inequality) is wrong or Not. In the physical models (for example balance equations) is a common symptom also a residuum defined by sign and size. In the efficiency considerations or the calculations of system parameters (Friction) are changes these sizes preferably itself as a symptom.

The Error isolation now has the task of different in the methods analyze the error symptoms and the actual causes to find these symptoms.

The Requirements for this For example, one is possible simple and fast calculation, a high degree of flexibility with the most diverse Combinations of analysis methods and the best possible isolation of the faulty channels.

• – Eine Isolierung der Fehler in einer Weise, dass der Prüfstandsfahrer zu konkreten Reparaturen an einem bestimmten Messgerät aufgerufen wird, erscheint nicht realistisch.
• – Vielmehr sollte der Prüfstandsfahrer vorzugsweise darauf hingewiesen werden, welche der Kanäle nicht stimmig zueinander sind oder welche Schlüsselgrößen gerade besonders auffällig sind.

On the basis of practical experience with the test bench diagnosis, the following points can be recorded as boundary conditions:

• - Isolating the faults in such a way that the tester driver is called to make specific repairs on a specific measuring device does not seem realistic.
• Rather, the test stand driver should preferably be advised which of the channels is not are consistent with each other or which key sizes are particularly noticeable.

Logical probability based Fault isolation procedures

The previously analyzed methods, all based on pattern recognition or fault trees are based, have the advantage in rigid structures and known Systems to work well. In case of insufficient system information or but they provide less input information useful results. Also the extensibility of these methods are narrow limits, since the subsequent modification of fault trees or Patterns are difficult to realize.

• – Welche Eingangsgrößen liegen den fehlerhaften Methoden zu Grunde?
• – Welchen Einfluss haben diese Größen innerhalb dieser Methoden?
• – Wie müssten die Eingangsgrößen beschaffen sein, damit kein Fehler auftritt?
• – Wie oft ist eine bestimmte Größe innerhalb der fehlerhaften Methoden verwendet worden?
• – Wie oft ist eine bestimmte Größe in den fehlerfreien, aber benutzten Methoden aufgetreten?

Now, in a highly flexible process, all the questions that an engineer would use in analytical procedures should be used. These include:

• - Which input quantities underlie the faulty methods?
• - What influence do these variables have within these methods?
• - How should the input quantities be designed so that no errors occur?
• - How many times has a certain size been used within the faulty methods?
• - How many times has a certain size occurred in the error-free but used methods?

These Questions can independently of which measuring channels or which methods are currently being set present in the current data record. There is therefore unlimited flexibility.

In addition, will assumed that preferably only encoder errors or sensor errors occur. If you look at the list of the 17 most common mistakes, that is this conclusion obvious. The topic is simplified considerably, there no consideration more on the many electrical, operating and component faults more must be taken.

in the If in doubt, it is then preferably the test stand driver, for example from the isolation result "pmi, Md or consumption error "on a defective indexing or alternatively a defective load cell close. But this is quite quickly realizable, if the appropriate Notes are available.

Estimation of error number

anticipating should be here first will be presented how to estimate the minimum number of causes can. This then results in a somewhat different isolation strategy.

requirement are preferably exclusive both errors in the measuring devices as well as a minimal number of causes.

in the Program is preferably for Each channel and method notes whether a channel has been used and if he was involved in faulty methods. From this knowledge Further preferably, a line-up can be formed, each one faulty method is compared with the input variables used.

Now Preferably, the measured variable is picked out, the most common occurs in the faulty methods. This is more preferable first declared as the first and most likely cause of error. The affected Methods are preferably chosen from the list of incorrect methods painted. So there remain the models where the size is not has been received.

There A minimal number of causes is assumed, too Preferably these are the most common of these remaining methods selected variable used. This is the supposed second cause of the error. The procedure will further preferably continued until no faulty Method is more present.

In the present case, shown in 37 , so at least two different causes of error must be present. Depending on whether only one or more variables are responsible for the error, preferably some different isolation strategies are used.

Isolation above limit value and probabilities

In motor operation, with all its inaccuracies and cyclical fluctuations, most models will output a residual, ie the expected value is never hit exactly. For this reason, a tolerance threshold is preferably defined for such models. It indicates how large the deviation between the key variables (eg comparison of measured and calculated value) may be. (Calculated value - measured value) / calculated value = ± tolerance threshold [%] [53]

Conversely, it is possible to calculate backwards what the measured data should look like in order to reach the error limit. Min / Max value = calculated value · (1 ± tolerance threshold) [54]

Since a deviation upwards and downwards is possible in each case, two values can be calculated for each input variable (measured value) that fulfill this condition. In the case of non-injured models, the measured value considered is of course between the minimum and maximum values. In 38 The minimum and maximum values of 10 different methods for a load point for error-free measured data are shown. The limit values in this example preferably relate only to the air mass, but it would be possible to generate such an image for each measured variable.

Of the Measured value does not conflict with any limit value, the error thresholds be for all models complied, so no error is displayed. It can be seen that all models are different a statement about can meet the expected air mass. For models in which The air mass is received only very low, are the back calculated Sizes very much far from the right value (Model 8). Conversely, others can Models very accurately predict the correct air mass (Model 5).

If a model not met is, so at least one input quantity is above the calculated maximum value or below the calculated minimum value.

Based on the effects of a false air mass signal, the relationship should be clarified. (In this case, a reduction in air mass, for example, by 15%.) Of course, the limits for the air mass have not changed, since in their calculation, the air mass itself is not received. This results in the 39 illustrated picture.

The Models 1 through 5 and 9 provide error messages because of the tolerance range not stable with the available measured quantities is. However, it is for the air mass a value possible which returns all models back to the allowed thresholds. For others Measurands that can go into the same models, should also be considered this diagram.

It is finally conceivable that also by a correction of for example fuel consumption, Exhaust emissions or speed the models can be faultless. It but it turns out that this is not the case in most cases. For this Sizes that can not cause the error, can not change the Value a error-free model system can be generated. There is always contradictions between the thresholds, so as a true source of error only the air mass left over remains.

In 40 the limits of all fuel consumption methods are shown for the same wrong air mass. Some methods are violated, ie the measured value is not within the limits. With a correction of the measured value, the conditions for some methods can be improved, but it is not possible with any consumption signal to make all methods error-free.

Of the The conclusion must be that the consumption alone is not the cause of these disagreements can be.

Remains assuming that preferably a minimum number of Sizes for the discrepancies is responsible, so consumption can not cause the error be.

A similar picture is shown preferentially for all other measurands, as in 41 for example still shown for CO ₂ .

In sum, from the 39 to 41 be concluded that the air mass flow must be the cause of the faulty models. It can furthermore preferably be specified with which air mass value the models become error-free, ie the value can be corrected.

Now Still the question arises, as so far preferably due to further preferably determined by visual observation analytically with the help of the computer can be implemented.

aim is an evaluation key, the one at big Error probability "p" for example the Value 1 reached, with very little probability, for example takes the value 0.1. This value, of a real probability very similar is preferably assigned to each potentially faulty channel become. The value of 0 is preferably avoided because the information that the channel is potentially involved, should not be lost. Furthermore, there is also the assumption about a minimum number of errors, which is why no 100% reliable statement can be generated.

• – Wenn der Messwert zwischen Minimal- und Maximalwert liegt, dann ist die Fehlerwahrscheinlichkeit sehr klein, zum Beispiel (p1 = 0,1).
• – Wenn der Messwert kleiner als der Maximalwert ist, aber kein Minimalwert vorhanden ist, dann spricht das für unsichere Modelle und damit für eine größere Fehlerwahrscheinlichkeit. Ein Fehler kann allerdings noch nicht damit nachgewiesen werden, zum Beispiel (p1 = 0,5).
• – Das gleiche gilt für den Fall, dass der Messwert größer als der Minimalwert ist, jedoch kein Maximalwert berechnet werden kann.
• – Wenn der Minimalwert größer ist als der Maximalwert, dann kann der Kanal alleine nicht Ursache der Unstimmigkeiten sein, zum Beispiel (p = 0,1).
• – Der Minimalwert ist kleiner als der Maximalwert, der Messwert liegt aber nicht dazwischen. Dies deutet auf einen direkten Fehler im betreffenden Kanal hin, zum Beispiel (p1 = 1).

The following case distinctions are preferably made:

• - If the measured value lies between minimum and maximum value, then the error probability is very small, for example (p1 = 0.1).
• - If the measured value is smaller than the maximum value but there is no minimum value, this indicates unsafe models and thus a greater probability of error. However, an error can not yet be proven, for example (p1 = 0.5).
• - The same applies in the case that the measured value is greater than the minimum value, but no maximum value can be calculated.
• If the minimum value is greater than the maximum value, then the channel alone can not be the cause of the discrepancies, for example (p = 0.1).
• - The minimum value is smaller than the maximum value, but the measured value is not in between. This indicates a direct error in the channel concerned, for example (p1 = 1).

Indeed This consideration represents a very favorable case. Frequently stand for one Measured variable not many or only one method available. The isolation strategy works in principle the same way, but the thresholds and isolation options then very limited because of the lack of uniqueness. It then usually gives you several options how the models can be converted to the faultless state.

In the case of multiple inaccuracies, ie at least two faulty measured variables, the fault isolation described above can not work properly. The reason for this is the interaction of the measured variables in the calculation of the limit values. As an example, here is the same load point as mentioned in the previous example, shown in 42 , In addition to the air mass now also a small error in the CO ₂ concentration is installed. The consequence of this is that now the previously unique pattern in the air mass is distorted, and precisely in the methods that also have CO ₂ as an input variable. A correction value can no longer be displayed for the air mass; the fault isolation fails.

It However, it should be noted that most fault isolation methods fail in this regard. Otherwise, the errors should preferably filed with all interactions as expert knowledge (patterns, fault trees) what, in such a complex and variable system test bench is not to be implemented in a reasonable time frame.

Out For this reason, the presented method is preferably only then related, if the discrepancies very likely to only one faulty size is due. At the same time and also at several potentially wrong sizes more preferably on a probabilistic and logic-based methodology resorted.

adoption is preferably first, that a minimal number of causes to the faulty models leads. The probability of a channel, cause of discrepancies within the faulty one Being models hangs certainly from the frequency the size inside of these methods. It can be defined a key figure p2, the a probability is at least very similar.

By means of the knowledge about the influences of the individual sizes within the methods can preferably the significance of this factor can be significantly improved. The basis is, for example, a weighting of the input variables between 1 and 3. A 1 means, for example, only a small influence, a 2 means a medium, a 3 a large influence. This weighting is preferably assigned a sign, depending on whether the input quantities must be too large or too small to obtain the calculated residual. These weights reflect the expert knowledge of a test bench engineer on how large the influence of different parameters on the individual models. This assistance can then preferably be communicated online to the inexperienced test stand driver in order to assist him in troubleshooting.

An example:

Of the Comparison of lambda air / fuel to lambda probe signal results for example, a residuum that over the method-specific Tolleranzschwelle lies.

The limit can only be reached if

The Weighting is preferred for all sizes uniformly big, since all sizes directly to enter into the result.

Another weighting results for the air mass, for example in the oxygen balance, which has the following structure (a, b, c, d, e are arbitrary additional terms): Residual = O entry - O exit = m. air mass · (A + b) - (m. air mass + m. fuel ) · (C + d + e) [57]

wobei k die Häufigkeit des Messkanals i innerhalb der fehlerhaften Modelle ist.It can be seen very quickly that the residual does not change to the same extent as the air mass, since the error in the air mass compensates to a large extent. For example, the air mass receives the weighting +1, wherein the sign due to the lack of uniqueness preferably has to be calculated analytically or more preferably numerically. If all sizes are assigned the defined weightings, the evaluation can be carried out in the following form:

where k is the frequency of the measurement channel i within the faulty models.

Now is preferably in addition The information utilized if the symptoms were within the faulty one Methods are consistent or contradictions occur. Will for example a channel of method 1 is assigned a weighting value of -3 and of method 2 with +3, the corresponding channel can hardly cause be the mistake.

It but it should be referred again: the value of 1 is usually after the definition of probability a "safe" event. In this case it is not so. Even if in this formula the value of 1 is not exceeded can be, then constellations are conceivable in which a Measured value reaches the value p = 1, but still not cause the error is. On the other hand, for example, is the value from p = 0 actually the "impossible" event, i. Variables, which do not occur in the faulty methods, too Do not be cause of the error symptoms. Also channels whose weights are neutralize each other (for example, +3 and -3), can not cause the error be.

For the estimation of whether a measuring channel is the cause of an error, the following consideration may preferably be made:
Although a measurement channel is relatively common in the faulty methods, but also in the wrong lerischen methods still occurs very often, then the probability sinks again, to be the cause. Therefore, it is preferable to define a measure that relates the use in the erroneous methods to the total number of uses.

The program proposes, with only one expected cause, to calculate and combine preferably all three factors. The overall score is therefore preferably p _ges = p1 * p2 * p3. For several causes where the analytical correction values fail, p1 is more preferably set to one, thus not included in the calculation.

implementation

The Implementation of all mentioned aspects to an overall program requires still a lot of detail, summarized in some key words here shall be.

connection to the test bench

If the engine type has not changed after the last attempt, the program preferably does not require any information other than the Names and units of the measurement channels as well as the readings themselves with a data rate of for example 1 Hz. The program is therefore preferably a pure observer.

If For example, errors in the exhaust system are detected, so may preferably an automated calibration or flushing of the system are triggered. Also the fuel consumption measurement can with appropriate error messages preferably automatically calibrated or restarted. These thoughts are becoming increasingly important as the proportion of staff-free Test stands in Automatic operation is steadily increasing.

The Software is preferably directly into the measurement data acquisition system integrated, because a much higher Communication effort between diagnostic software and test bench is needed.

necessary sizes

Next the basic engine data such as stroke, bore, compression and number of cylinders is preferably still the input required by which type of motor it is about. For example, differences are made between Otto and Diesel engines and, for example, after charged or uncharged Engines. The maximum speed is preferably the only variable that indicating the character of the engine. It serves to estimate various Geometries. Because this information anyway in the test bench software So there is no additional input required.

For the exhaust gas measuring devices is preferably an indication necessary, which makes a distinction between dry and wet Concentrations possible. This information can not be generated and requires a manual Input. Likewise behaves It preferably with the fuel data. However, it is too consider, that this input preferably takes place once. Neither the emissions towers yet the fuel grade changes at significant times, but mostly stay over Weeks or months constant.

by virtue of logical connections can correct the pressures be made. For Physical methods is always interesting only one absolute pressure. Many pressures however, are preferably measured only relative to the environment.

Here helps to observe the pressures in the stationary Functional test. As already stated, the result of the value comparison is at engine standstill on the one hand a very accurate ambient air pressure, on the other hand, it is preferably obtained at the same time, the information as to which pressures still require correction.

When Manual input therefore only the parameters for humidity correction, an indication of the engine type and fuel data is left. Of the Time required preferably less than one minute if the routine is appropriate.

parameterization the test task

Remains the task after the parameterization of the test task itself. For everyone Attempt must be defined, which methods are with the given ones Measuring channels at all realize and which methods are suitable for the specified engine type suitable.

These Question was the test personnel decreased. For a list of 83 methods and comparisons, the can be extended at any time, this task can be performed by the test personnel not to be expected. A factual confrontation with the Formulas sets as well a sound knowledge of the topic ahead, which certainly partially present, but not to the previous fields of activity belonging to the appropriate staff.

Therefore is the program preferably a logic in the form of its own configuration program included, preferably independently searches for the correct channel assignments for each individual method. If a method matches the engine type and the measurement data, then it is also preferably automatically activated.

methods not with the measured variables can be realized remain preferably still inactive and interfere with the further testing operation Not.

simultaneously It is clear that a sophisticated error detection with only one preferably high number of active methods can work. Therefore, the goes Logic so far, preferably in the absence of different sizes automatically search for replacement sizes.

An example:

wobei v aus der Kontinuitätsgleichung gewonnen werden kann. In diese geht dann wiederum eine Luftmasse ein.Calculating the pressure loss via the intercooler (LKK) requires an air mass flow

where v can be obtained from the equation of continuity. Into this in turn enters an air mass.

• – Die Luftmasse als Messgröße,
• – die Luftmasse wird berechnet aus einem Lambdasondenwert und dem Kraftstoffverbrauch,
• – die Luftmasse wird aus den Abgasgrößen und dem Kraftstoffverbrauch berechnet.

For the analysis of the intercooler it is now completely unimportant, from which the used air mass is generated. For the example mentioned, for example, three alternative assignments are possible.

• - The air mass as a measured variable,
• - the air mass is calculated from a lambda probe value and the fuel consumption,
• - The air mass is calculated from the exhaust gas quantities and the fuel consumption.

Of the Fuel consumption can turn, for example, either from a external measuring device or the engine's own ECU.

These Work, the selection of measured quantities and alternatively replacement sizes, can preferably only the computer itself reliably perform. It must continue to ensure that preferably also the fault isolation with the corresponding Information is supplied. Otherwise, the calculation is the probability values impossible.

So it happens that the measured air mass is no longer potential Error source is, but the composition of the exhaust gas, and that in a pressure loss model over LKK. Thus, it is preferable to create a network of background information, without which the program can not work reliably.

Of course it stays leave it to the tester driver, to deactivate individual methods by hand, i. he remains most important Instance on the test bench. This in turn is very important for the introduction of such a system.

Regarding the Parametrieaufwandes thus the goal has been achieved.

output

The Test diagnostics, which was developed under the aspect of minimal extra effort for the tester driver, preferably allows only very limited operating options.

To the booting of all computers and the Einachalten and calibration the measuring devices preferably takes place first the static functional test with a cold engine, the so-called cold test. The results of this Tests can preferably entered directly in the message window on the same surface become. In case of error the test stand driver has this preferably further to acknowledge or the corresponding error to fix.

The The "Warm test" key preferably starts the online diagnostic method, more preferably either the Steady state detection and fault isolation.

in the ongoing test operation is first once to avoid a flood of data, which unnecessarily burden the test bench driver. Of the many generated sizes is first once exclusively of interest, whether or not there is an error.

in the Error case must preferably also be comprehensible, which methods have detected the error, how big this is and what causes responsible for could be. This information can However, sometimes only be evaluated by knowledgeable staff, why they are preferably in a second, deeper level are stored.

At the It is still at the end of the test or for a more detailed analysis during the measurement of great help, to be able to look at the method-specific results. For this is preferably a separate graphics evaluation necessary, the one fast and clear Output of single methods allows. The Calculated variables of Methods, residuals or other important key variables are more preferable than all Applied measuring points. With that you can very fast drifting gauges or jumps within of the test, even if they are within tolerance lie and did not cause an error message.

Of Further can preferably at the end of the measurement, automatically statistical graphics to failure frequencies, Detection rates of the methods or error overviews are created.

kick an error, i. the corresponding signal lamp switches on red, there is the possibility a closer analysis.

For example, the "Details" button can be activated to enable access to the individual methods, as in 43 shown.

In This level of detail is preferably initially some edge information issued for the test. This includes in addition to the number of identified channels the number of active methods, the underlying engine type as well as the current fuel data. This will be very fast It is clear whether the error messages are parameterization errors or is true measurement or component faults.

among them There is preferably a message window in which all errors Single methods with a short description of errors are listed. The accomplished test stand driver I can quickly tell where the problem is.

Directly besides this, preferably the analytically generated error isolation result shown. On the basis of the color code it is immediately clear which size for the error responsible for. For example, have red deposited channels a high probability of error, green deposited channels are Although involved in the faulty methods, it is very likely not cause of discrepancies.

When Further sources of information are preferably one to the user Method Brief information with graphical and text-based assistance. In order to may be under circumstances hard to understand Error messages set out transparently. Furthermore, it is preferable a graphical evaluation option to disposal, which, however, requires its own surface.

The graphics analysis preferably shows the gene there depending on the selected method result values. This is usually for example the comparison between incoming and outgoing energy and the like. This representation is preferably accompanied by the size of the residuals and the selected error thresholds.

to Documentation or requesting help, for example responsible test engineer, can the Images are preferably printed automatically. The idea behind it is the following, that the information is the responsible one Engineer packed tight can be sent and this then in case of doubt about the further test progress decides.

Next the first very striking issue in level 1 are the test bench driver and the test engineer thus also given a series of more in-depth information, in most cases for a quick check of the Data quality to lead.

capacity of the procedure

To test the quality of the diagnosis, for example, two measurement points that were demonstrably error-free were synthetically documented with errors of different size and sign. Two test datasets were created, one of a gasoline engine, shown in 44 another from a diesel engine, pictured in 45 Detection rates define the ratio of detected errors to existing errors

The isolation rate tells you whether the right cause is isolated when a fault is found

Of the Sensortotalausfall is in this regard a special case, since whose proof is trivial. Make detection as well as isolation no problems here. Therefore, all results are without Defects (o.F.) shown.

The result are the data in a medium load point on a commercial vehicle diesel engine, also on a passenger car gasoline engine of the latest design.

Both Motors are common with all channels equipped, which preferably also occur in the database. Unmeasured channels were retrospective calculated and added, to one as possible comprehensive picture of performance to obtain.

The Variation consisted in this consideration for each measuring channel in one percent change after up as well as down. At error sizes of, for example, 5% in most models, the error thresholds preferably not yet exceeded, which is why the detection rate is still very low here. For example 15% mistakes speak most balance sheet models as well as the empirical ones Models preferentially, causing a sharp increase in the detection rate to be recorded. The further increase is more preferable relatively flat. The reason is that in this area the very simple structured inequalities for the control of To press and temperatures are added, but only in one direction can prove a mistake. A steep rise is preferably only again for very large errors to expect, for example, if the limits exceeded from the database become. Then detection and isolation increase, for example 100% on.

The Isolation rate is preferably always over 80% for errors> 15% and thus to be considered good. It still has to be considered that the results for the limit analysis as well as the probability analysis preferably approximately are the same, although the computational effort of both methods clearly different.

The advantage of the more complicated calculation is that the deeper error analysis with the calculated limits is much more helpful. However, the interpretation of these values so far requires an expert who is familiar with the physics of models and thus interpret the values White.

Of the test operator receives as a result, preferably a list of measured variables, all of which potentially cause could be the fault. Each measurand is preferred either red or green deposited, depending on how big the Error probability of the corresponding size.

Of the Advantage of the presented method is the flexibility with which any combination of analysis methods and measuring channel combinations can be evaluated. With sufficiently large Errors and sufficient model number, the isolation rate is good. To the detection rate noticeable to increase, brings a higher resolution Contrary to first guesses not much, since most mistakes are not Component failure, but actually instrumentation errors are. These can also be determined by single work game analyzes etc. not better recognize as by mean value considerations.

With the method according to the invention Is it possible, to detect a whole series of engine test bench faults and thus the quality to reproducibly increase the measurement data. Sensitize the error messages the test stand driver, the goodness and data accuracy more intense.

Next empirical and logical checks in the Motor standstill, preferably already first inaccuracies in many important measuring channels can uncover more preferably, above all, assures a plurality of individual checks running engine for it, that the most disturbing Error in everyday testing can be discovered faster as before.

By the small amount of system information is suitable for diagnosis the data only basic relationships, based on chemical and physical relationships. For example, empirical models and neural networks can indeed in a few cases be used, but their field of application remains preferably rather the already fully measured engine. In case of desired universality it falls Accuracy very fast in less promising regions with which the physically based models gain in importance.

Also For example, if the isolation mechanism is only faulty Measurement signals and just does not start from component defects, so offers but by the generated resources and computational an effective Assistance by which the actual mistakes are discovered can.

• – Die Kommunikation mit der Messdatenerfassung fällt deutlich leichter.
• – Die Beschaffung von Informationen, sei es über AK-Protokoll oder andere Schnittstellen, ist damit deutlich einfacher umzusetzen.
• – Durch den Wegfall einer limitierenden Schnittstelle kann eine höhere Datenerfassungsrate analysiert werden, womit die Diagnose der Rohsignale aufgrund statistisch basierter Diagnoseverfahren einen höheren Aussagewert erhält.
• – Im Hinblick auf den vollautomatischen Prüfstand wird immer häufiger die Forderung nach automatisierten Reaktionen der Diagnosesoftware auf entsprechende Fehlermeldungen laut. Die Ideen reichen von versendeten Benachrichtigungen über Kalibrierungen, Korrekturen und Neustart des Tests, basierend auf logische Schlüsse des Diagnosesystems.

It is also conceivable that the modules are not provided on a separate test computer, but that, for example, the diagnosis is integrated in the measurement data acquisition computer. Thereby several advantages can be achieved:

• - Communication with the measurement data acquisition is much easier.
• - The procurement of information, be it via AK protocol or other interfaces, is thus much easier to implement.
• - By eliminating a limiting interface, a higher data acquisition rate can be analyzed, whereby the diagnosis of raw signals due to statistically based diagnostic procedures receives a higher value statement.
• With regard to the fully automatic test stand, the demand for automated reactions of the diagnostic software to corresponding error messages is becoming more and more frequent. Ideas range from sending notifications about calibrations, corrections, and restarting the test based on logical conclusions from the diagnostic system.

Claims (8)

Device for determining the measuring accuracy of a measuring device, in particular of a motor test stand, comprising at least one arithmetic unit, which is suitable for detecting measurement data of the measuring device, characterized in that a plurality of modules for reference measurement of the measured data is provided, by means of which the measured data are comparable to reference values , wherein at least a first module for carrying out a standstill examination before the measuring device is put into operation, a second module for carrying out a preliminary measurement during commissioning of the measuring device for determining the reproducibility of the measured data and a third module for carrying out a main measurement during the ongoing operation of the measuring device for online Determination of the plausibility of the measured data is formed. Apparatus according to claim 1, characterized in that a fourth module for performing a Re-measurement after completion of the operation of the measuring device for offline determination of the plausibility of the measured data is provided. Device according to one of the claims, characterized characterized in that another module for performing a Scatter band measurement during the commissioning of the measuring device for the evaluation of the measured data via a long period of time is provided. Device according to one of the claims, characterized characterized in that the reference measurements are automatically by means of electronic data transfer to a pre-selectable one Receiver can be transmitted are. Method for determining the measuring accuracy of a Measuring device, in particular a motor test bench, comprising at least a computing unit that is suitable for the acquisition of measurement data of the measuring device is characterized in that a plurality of modules for Reference measurement of the measured data is provided by means of which the Measured data are compared with reference values, at least a first module a standstill inspection before commissioning the measuring device, a second module a Vormessung during the Commissioning of the measuring device to determine the reproducibility the measurement data and a third module a main measurement during the ongoing operation of the measuring device for the online determination of the plausibility perform the measurement data. Method according to claim 5, characterized in that that a fourth module for performing a final measurement after Termination of the operation of the measuring device for offline determination the plausibility of Measurement data is provided. Method according to one of claims 5 or 6, characterized that another module for performing a scatter band measurement while the commissioning of the measuring device for the evaluation of the measured data via a long period of time is provided. Method according to one of claims 5 to 7, characterized that the reference measurements automatically by means of electronic data transmission to a previously selected one Receivers are transmitted.