DE102022212772A1 - Method for controlling an injector, control unit - Google Patents

Abstract

The invention relates to a method for controlling an injector (1), in particular a fuel injector, in which the lifting movement of a nozzle needle (2) for opening and closing at least one injection opening (3) is controlled via the pressure in a control chamber (4) and the pressure in the control chamber (4) is measured with the aid of a pressure sensor (5) integrated in the injector (1). According to the invention, the sensor signal of the pressure sensor (5) and/or time data derived from the sensor signal of the pressure sensor (5) and characteristic of the lifting movement of the nozzle needle (3) is/are fed into an artificial neural network (19) and the current injection quantity (Qe) is determined with the aid of the artificial neural network (19).The invention further relates to a control device (14) for carrying out steps of the method.

Description

The invention relates to a method for controlling an injector, in particular a fuel injector for injecting fuel under high pressure into a combustion chamber of an internal combustion engine.

Furthermore, the invention relates to a control device for carrying out steps of the method.

State of the art

Fuel injectors with a nozzle needle arranged in a pressure chamber so that it can be moved longitudinally for opening and closing at least one injection opening are known from the prior art. The movements of the nozzle needle are controlled servo-hydraulically, i.e. via the pressure in a control chamber, which exerts a hydraulic closing force on the nozzle needle. The pressure in the control chamber is controlled via a control valve that regulates the pressure in the control chamber electromagnetically or with the help of a piezo actuator. The control valve can be controlled very precisely by a control unit. However, there is a time delay between the control current and the actual movement of the nozzle needle. For precise control, it is therefore important to know the exact point in time at which the nozzle needle moves and injection begins in order to adjust the control current for controlling the control valve if necessary.

The movement of the nozzle needle can be detected, for example, via the pressure curve in the control chamber. In the EN 10 2015 207 307 A1 A fuel injector with a servo-hydraulically controlled nozzle needle and a pressure sensor is proposed for this purpose. The pressure sensor can be used to detect pressure changes in the control chamber or in a pressure chamber hydraulically connected to the control chamber.

The temporal progression of the pressure in the control chamber of a servo-hydraulic injector provides precise information about the movement of the nozzle needle. By using algorithms for signal analysis, characteristic injection times such as the start of injection, the end of injection or the needle opening time can be determined. These algorithms typically use combinations of digital filtering, threshold value methods and numerical derivation.

When operating an internal combustion engine, injection rate models are usually used to regulate the injected fuel quantity with greater accuracy and to optimize the combustion parameters. Due to the lower computing power, 0D models are preferred over 1D models based on flow simulations, for example. The models usually use a highly simplified representation of the injection rate curve, for example in the form of a trapezoid. However, the measured actual curve shows deviations from the model, so that the model's prediction of the injected quantity is inaccurate.

In order to increase accuracy, it has already been proposed to use the absolute pressure or pressure differences from the sensor signal of a pressure sensor integrated in the injector to estimate the injection quantity. However, wear processes change the sensor signal amplitude of the pressure sensor over the service life of the injector, so that this change influences the accuracy of the estimate. In addition, if the maximum injection rate is calculated using a fixed linear correlation to the pressure drop, this correlation changes over time due to drift/wear of the pressure sensor, which leads to a systematic error in the injection quantity estimate.

The present invention is therefore concerned with the task of increasing the accuracy in estimating the injection quantity.

To solve the problem, the method with the features of claim 1 is proposed. Advantageous further developments of the invention can be found in the subclaims. In addition, a control device for carrying out steps of the method is specified.

Disclosure of the invention

In the proposed method for controlling an injector, in particular a fuel injector, the stroke movement of a nozzle needle for opening and closing at least one injection opening is controlled via the pressure in a control chamber and the pressure in the control chamber is measured using a pressure sensor integrated in the injector. According to the invention, the sensor signal of the pressure sensor and/or time data derived from the sensor signal of the pressure sensor that are characteristic of the stroke movement of the nozzle needle is/are fed into an artificial neural network and the current injection quantity is determined using the artificial neural network.

With the help of an artificial neural network that has been trained beforehand, the injection quantity can be estimated very precisely. A trained, fixed An artificial neural network is used that is not further trained during operation.

To train the artificial neural network, both the sensor signal from the pressure sensor and time data derived from the sensor signal that are characteristic of the stroke movement of the nozzle needle are used. The data used for training preferably covers the entire wear that can be expected over the service life. The actual wear state of the injector can therefore be taken into account when estimating the injection quantity.

In principle, any artificial neural network can be used in the process. The artificial neural network is preferably implemented using a control unit on which all parameters describing the network are stored. Which type of artificial neural network and which topology is used depends in particular on the respective injector and/or control unit design.

Based on defined input parameters, the current state of wear can be detected and taken into account when estimating the injection quantity. Based on the estimated injection quantity, the control of the injector, in particular the control duration, can then be regulated individually for each injector - preferably with the help of the control unit.

The injection quantity estimation carried out using the artificial neural network takes place during operation of the injector, preferably over the entire service life of the injector. Since the proposed method allows the age or current state of wear of the injector to be taken into account, the injection quantity can be estimated over the service life of the injector with very high accuracy.

In addition to correcting the injector control, the proposed method can also be used to predict the injector's wear conditions over its service life ("predictive diagnosis"). In this case, the method can be used to identify when the injector needs to be repaired or replaced.

When implementing the proposed method, the current injection quantity determined with the aid of the artificial neural network is preferably compared with a target value and, in the event of a deviation, the control of the injector is corrected. When making the correction, the control duration in particular can be changed or adjusted. The comparison is preferably carried out with the aid of the control unit so that the target value required for the comparison is stored in the control unit. If necessary, the control can then be corrected with the aid of the control unit.

In order to rule out possible errors, particularly model errors, an average value is preferably calculated from several estimates of an injection point, i.e. at the same system pressure and the same injection duration. The forecast value with the greatest deviation from the calculated average value is then discarded. Alternatively or in addition, a plausibility check can be carried out, in particular with the help of an injection rate model.

• - Beginn der Öffnungsbewegung der Düsennadel (NOS),
• - Ende der Öffnungsbewegung der Düsennadel (NOE),
• - Beginn der Schließbewegung der Düsennadel (NCS) und
• - Ende der Schließbewegung der Düsennadel (NCE).

When implementing the proposed method, the following time data are preferably derived from the sensor signal of the pressure sensor and fed into the artificial neural network:

• - Start of the opening movement of the nozzle needle (NOS),
• - End of the opening movement of the nozzle needle (NOE),
• - Start of the closing movement of the nozzle needle (NCS) and
• - End of the closing movement of the nozzle needle (NCE).

The time data mentioned above are all characteristic of the stroke movement of the injector's nozzle needle, so that they can be used as a basis for estimating the injection quantity. This is especially true for an injector with an upper stop for the nozzle needle, so that the stop defines the position of the nozzle needle when it is fully opened. The time data derived from the sensor signal of the pressure sensor can be fed into the artificial neural network as an alternative or in addition to the sensor signal of the pressure sensor. The artificial neural network has then been trained beforehand using this data.

In addition, other operating parameters, such as system pressure, fuel temperature and other fuel properties, can be taken into account when estimating the injection quantity.

The time data NOS, NOE, NCS and NCE are preferably derived from the sensor signal of the pressure sensor using a detection algorithm. The detection algorithm can use derivatives and/or averages of the sensor signal. In addition, the sensor signal can be pre-processed in advance, in particular filtered, in which case for example using a low-pass filter. The appropriately pre-processed time data is then fed into the artificial neural network to estimate the injection quantity.

In a further development of the invention, it is proposed that an injection quantity estimate be carried out using injection rate modeling in parallel to the injection quantity estimate using the artificial neural network. The injection quantity determined using the injection rate model can then be used for a plausibility check. In this way - as already mentioned - the accuracy of the injection quantity estimate can be optimized, for example a signal or model error can be excluded.

When estimating the injection quantity using injection rate modeling, a change in the maximum injection rate is preferably determined from the time data NOS, NOE, NCS and NCE. In this way, changes in the maximum injection rate over time that are caused by aging/wear of the injector are recorded.

It was found that correlations exist between the time data NOS, NOE, NCS, NCE, the maximum injection rate and the nozzle needle lift. These can be identified using conventional methods with linear equations and/or by using artificial intelligence and used to determine the maximum injection rate. It was also found that these correlations are independent of other wear parameters, such as the wear of the nozzle needle seat and/or changes in the flow conditions in the area of a throttle plate delimiting the control chamber. The correlations are preferably determined experimentally in advance and stored, preferably in the control unit for controlling the injector.

The injection rate model can be parameterized and corrected using the time and flow data derived directly and indirectly from the pressure sensor signal. For example, a corrected injection rate curve can be calculated by mathematically transforming the time data derived from the pressure sensor signal.

A method for correcting an injection rate model using time and flow data derived directly or indirectly from the sensor signal of a pressure sensor is the subject of an earlier, as yet unpublished application by the same applicant, so that it will not be discussed in more detail in the context of the present application.

The corrected injection rate model can then be used to calculate the current injection quantity, preferably by integrating the corrected injection rate curve. The calculated current injection quantity is then compared with the current injection quantity determined using the artificial neural network for plausibility purposes. After the comparison, the estimated injection quantity is compared with the target value and, if necessary, the control, in particular the control duration, of the injector is adjusted accordingly.

Advantageously, in the proposed method, the sensor signal of the pressure sensor is not only used to estimate the injection quantity, but also to predict injector wear and the cause of the wear. This can also be determined by evaluating the sensor signal of the pressure sensor, since the signal differs depending on the cause of the wear. The cause could be, for example, wear of the nozzle or wear of the throttle plate of the injector. In this way, a prediction can be made as to when the injector needs to be repaired or replaced.

Furthermore, a control device is proposed which is set up to carry out steps of a method according to the invention. Preferably, software is installed on the control device which implements the artificial neural network. Furthermore, preferably, a target value is stored in the control device for comparing the injection quantity estimated according to the method, so that a correction of the control of the injector can be made if necessary.

The control unit can also store an injection rate model that is based on the measurement data of a nominal injector and is continuously corrected in relation to the current actual state during operation by evaluating the sensor signal from the pressure sensor. For this purpose, a detection algorithm can be stored in the control unit, with the help of which the relevant time and flow data can be derived directly or indirectly from the sensor signal. If the evaluation reveals drift or wear, the injection rate model can be corrected accordingly with the help of the control unit.

• 1 ein Flussdiagramm zur Darstellung eines möglichen Ablaufs eines erfindungsgemäßen Verfahrens,
• 2 a) einen schematischen Längsschnitt durch einen Injektor, der nach dem erfindungsgemäßen Verfahren ansteuerbar ist, und b) einen vergrößerten Ausschnitt des Injektors, und
• 3 verschiedene Diagramme zur Darstellung des zeitlichen Verlaufs a) des Stromverlaufs, b) des Düsennadelhubs, c) der Einspritzrate und d) des Sensorsignals des Drucksensors.

The method according to the invention and its advantages are explained in more detail below with reference to the accompanying drawings. The drawings show:

• 1 a flow chart showing a possible sequence of a method according to the invention,
• 2 a) a schematic longitudinal section through an injector which can be controlled according to the method according to the invention, and b) an enlarged section of the injector, and
• 3 Various diagrams to illustrate the temporal progression of a) the current curve, b) the nozzle needle lift, c) the injection rate and d) the sensor signal of the pressure sensor.

Detailed description of the drawings

The 1 The method according to the invention shown serves to control an injector 1 with a servo-hydraulically controlled nozzle needle 2 and a pressure sensor 5 for measuring the control pressure applied to the nozzle needle 2. Such an injector 1 is shown by way of example in the 2 shown. 2a) shows the complete injector 1, 2 B) only the area of the nozzle and the control valve.

The Indian 2 The injector 1 shown has a nozzle needle 2 which is accommodated in a nozzle body 6 so as to be movable in stroke and for opening and closing at least one injection opening 3. The end of the nozzle needle 2 facing away from the injection opening 3 delimits a control chamber 4 which is hydraulically connected to a pressure chamber 8 via a channel 7. The pressure chamber 8 is formed in a throttle plate 9 and is delimited by a membrane 10 which rests on the throttle plate 9 and behind which a pressure sensor 5 is arranged so that pressure changes occurring in the pressure chamber 8 can be detected with the aid of the pressure sensor 5. The pressure sensor 5 is integrated in a valve plate 11, which in turn is rested on by a valve body 12 of a control valve 13. The actuation of the control valve 13 leads to a change in the pressure in the control chamber 4, which exerts a hydraulic force on the nozzle needle 2 in the closing direction. The pressure in the control chamber 4 thus controls the lifting movement of the nozzle needle 2. Information that is characteristic of the lifting movement of the nozzle needle 2 can therefore be derived from the sensor signal of the pressure sensor 5.

In the 3 Examples of relationships between nozzle needle lift ( 3b) ) and sensor signal of pressure sensor 5 ( 3d )). The sensor signal allows both the beginning of the nozzle needle movement when opening (time NOS) and the end of the nozzle needle movement when opening (time NOE) to be identified. Similarly, the beginning (NCS) and the end (NCE) of the nozzle needle movement when closing can be derived from the sensor signal. The times NOS, NOE, NCS and NCE result in the 3c ) with a maximum of the injection rate Qmax. However, the injection rate curve shown in the 3a) The current curve shown occurs because there is a time delay between the control current and the movement of the nozzle needle 2.

As exemplified in the 1 As shown, the pressure in the control chamber 4 above the nozzle needle 2 is recorded with the aid of the pressure sensor 5 integrated in the injector 1 and the sensor signal of the pressure sensor 5 is made available to a control unit 14 via a signal line 15. Software is installed on the control unit 14 which implements an artificial neural network 19 for estimating the current injection quantity Qe. The sensor signal of the pressure sensor 5 and/or the time data NOS, NOE, NCS, NCE derived from the sensor signal are fed into the artificial neural network 19. The artificial neural network 19 then estimates the current injection quantity Qe from the sensor signal and/or from the time data derived from it.

To derive the time data, the control unit 14 carries out step A. In step A, the sensor signal is first preprocessed with a low-pass filter 17. The time data NOS, NOE, NCS, NCE are then derived from the sensor signal using a detection algorithm 18 stored in the control unit 14. The detection algorithm 18 mainly uses derivations and averages of the sensor signal to determine the relevant points in time.

The times NOS, NOE, NCS and NCE derived from the sensor signal can also be - as exemplified in the 1 shown - to check the plausibility of the injection quantity Qe estimated with the aid of the artificial neural network 19. For this purpose, the steps BD are carried out with the aid of the control unit 14.

Step B is used for drift detection. To detect a change in the injection behavior of injector 1 over the running time, also known as drift, the time data NOS, NOE, NCS, NCE derived from the sensor signal in step A are saved at different system pressures and/or current durations in order to be able to calculate the change in the maximum injection rate Qmax. Correlations between the time data NOS, NOE, NCS, NCE, the maximum injection rate Qmax and the nozzle needle lift are used for the calculation. These correlations are determined experimentally in advance.

For this purpose, the needle closing time t _closing = NCE-NCS can be investigated at two different system pressures in an operating range in which the nozzle needle opens up to an upper stop. The experiment is preferably designed in such a way that relevant variations in needle stroke and nozzles flow can be measured. The relationship for the two system pressures can then be determined using equation (1). The further apart the two system pressures are, the more accurately the nozzle needle lift and the maximum injection rate Qmax can be determined. This is because the coefficients a and b are also far apart when the system pressures are far apart and thus the difference in slope of the two straight lines according to equation (1) is greater. $$t_{clOsinG}^{psys}=a_{psys}\ast Hub+b_{psys}\ast Q_{max}^{pnOm}+c_{psys}$$

kann nach der Gleichung (2) bestimmt werden. Für die Bestimmung werden vorzugsweise im Steuergerät 14 gespeicherte Zeitdaten bei 0-km mit den aktuell gemessenen Zeitdaten laut der Gleichung (3) verwendet. $$Q_{max,ist}^{pnom}=Q_{max\mathrm{,0}-km}^{pnom}+\mathrm{\Delta }{Q}_{max}^{pnom}$$

$Q_{max,ist}^{pnom}$

das Maximum der Einspritzrate des geregelten Injektors bei Nominaldruck p_nom.
$Q_{max\mathrm{,0}-km}^{pnom}$

wird mit der 0-km Messung des geregelten Injektors berechnet. $$\mathrm{\Delta }{t}_{closing}^{psys}=a_{psys}\ast \mathrm{\Delta }Hub+b_{psys}\ast \mathrm{\Delta }{Q}_{max}^{pnom}$$

In equation (1), a, b and c are experimentally determined specific constants at a system pressure p _sys . The change in the injector flow $\mathrm{\Delta }{Q}_{max}^{pnOm}$

can be determined according to equation (2). For the determination, time data at 0 km stored in the control unit 14 are preferably used with the currently measured time data according to equation (3). $$Q_{max,ist}^{pnOm}=Q_{max\mathrm{,0}-km}^{pnOm}+\mathrm{\Delta }{Q}_{max}^{pnOm}$$

$Q_{max,ist}^{pnOm}$

the maximum injection rate of the controlled injector at nominal pressure p _nom .
$Q_{max\mathrm{,0}-km}^{pnOm}$

is calculated using the 0-km measurement of the controlled injector. $$\mathrm{\Delta }{t}_{clOsinG}^{psys}=a_{psys}\ast \mathrm{\Delta }Hub+b_{psys}\ast \mathrm{\Delta }{Q}_{max}^{pnOm}$$

das heißt die Änderung des Maximums der Einspritzrate des Injektors bei nominalem Druck bestimmt werden. Das Ergebnis kann dann in Schritt C (siehe 1) als Eingabewert für ein Einspritzratenmodell verwendet werden. Mit Hilfe des Einspritzratenmodells kann dann ebenfalls eine Einspritzmenge Qe geschätzt werden.With these correlations, $\mathrm{\Delta }{Q}_{max}^{pnOm},$

that is, the change in the maximum injection rate of the injector at nominal pressure can be determined. The result can then be used in step C (see 1 ) can be used as an input value for an injection rate model. The injection rate model can then also be used to estimate an injection quantity Qe.

In step D (see 1 ) the plausibility check is then carried out. For this purpose, the injection quantity Qe derived from the injection rate model is compared with the injection quantity Qe that was estimated using the artificial neural network 19.

In step E, the estimated and possibly plausible injection quantity Qe is then compared with a target value Qe,target stored in the control unit 14. If the comparison results in a deviation ΔQe, the electrical control duration of the injector 1 can be changed. For this purpose, a correspondingly adjusted control current is passed from the control unit 14 to the injector 1 via a control line 16.

QUOTES INCLUDED IN THE DESCRIPTION

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

Cited patent literature

• DE 102015207307 A1 [0004]

Claims (10)

Method for controlling an injector (1), in particular a fuel injector, in which the lifting movement of a nozzle needle (2) for opening and closing at least one injection opening (3) is controlled via the pressure in a control chamber (4) and the pressure in the control chamber (4) is measured with the aid of a pressure sensor (5) integrated in the injector (1), characterized in that the sensor signal of the pressure sensor (5) and/or time data derived from the sensor signal of the pressure sensor (5) and characteristic of the lifting movement of the nozzle needle (3) is/are fed into an artificial neural network (19) and the current injection quantity (Qe) is determined with the aid of the artificial neural network (19). Procedure according to Claim 1 , characterized in that the current injection quantity (Qe) determined with the aid of the artificial neural network (19) is compared with a target value (Qe,target) and in the event of a deviation (ΔQe) a correction of the control of the injector (1) is carried out. Procedure according to Claim 1 or 2 , characterized in that the following time data are derived from the sensor signal of the pressure sensor (5) and fed into the artificial neural network (19): - start of the opening movement of the nozzle needle (NOS), - end of the opening movement of the nozzle needle (NOE), - start of the closing movement of the nozzle needle (NCS) and - end of the closing movement of the nozzle needle (NCE). Procedure according to Claim 3 , characterized in that the time data (NOS, NOE, NCS, NCE) are derived from the sensor signal of the pressure sensor (5) with the aid of a detection algorithm (18) which uses derivatives and/or averages of the sensor signal. Procedure according to Claim 3 or 4 , characterized in that a change in the maximum of the injection rate (ΔQmax,actual) is determined from the time data (NOS, NOE, NCS, NCE). Procedure according to Claim 5 , characterized in that to determine the maximum injection rate (Qmax), previously experimentally determined and stored correlations are used which exist between the time data (NOS, NOE, NCS, NCE), the maximum injection rate (Qmax) and the stroke of the nozzle needle (2). Method according to one of the preceding claims, characterized in that a corrected course of the injection rate is calculated by mathematical transformation of the time data derived from the sensor signal of the pressure sensor (5). Procedure according to Claim 7 , characterized in that in the mathematical transformation of the time data only the time data (NOS, NOE, NCS, NCE) derived from the sensor signal of the pressure sensor (5) are used. Method according to one of the preceding claims, characterized in that the corrected injection rate model is used to calculate the current injection quantity (Qe), preferably by integrating the corrected course of the injection rate, and the calculated current injection quantity (Qe) is compared for plausibility with the current injection quantity (Qe) determined with the aid of the artificial neural network (19). Control device (14) which is designed to carry out steps of a method according to one of the preceding claims, wherein preferably software which implements the artificial neural network (19) is installed on the control device (14).

Images