DE102020206786A1 - Method for determining coking of an injection valve - Google Patents

Abstract

A method for determining a first variable characterizing coking of an injection valve, in particular for an internal combustion engine, in particular a motor vehicle, comprising the following steps: Providing several operating variables of the injection valve, determining the first variable by means of an, in particular artificial, neural network based on the several operating variables .

Description

State of the art

The disclosure relates to a method for determining a first variable characterizing coking of an injection valve, in particular for an internal combustion engine, in particular of a motor vehicle.

The disclosure also relates to an apparatus for carrying out such a method.

Disclosure of the invention

Exemplary embodiments relate to a method for determining a first variable characterizing coking of an injection valve, in particular for an internal combustion engine, in particular a motor vehicle, having the following steps: Providing several operating variables of the injection valve, determining the first variable by means of an, in particular artificial, neuronal Network based on the multiple operating parameters. As a result, the first variable can be determined efficiently as a measure of the coking of the injection valve.

In further exemplary embodiments, it is provided that the neural network is designed as a multi-layer perceptron, in particular a multilayer perceptron.

In further exemplary embodiments it is provided that the several operating variables have at least one of the following elements: a) fuel pressure, b) needle reversal point, c) injection duration. For example, the fuel pressure characterizes the pressure of the fuel to be injected by means of the injection valve. In some embodiments, the fuel can be provided, for example, by means of a high-pressure fuel reservoir (“rail” or “common rail”), the fuel pressure then corresponding, for example, to the rail pressure. For example, the injection duration characterizes a period of time during which a fuel injection is carried out by means of the injection valve. For example, the needle reversal point characterizes the point in time during a fuel injection at which a valve needle or nozzle needle of the injection valve reverses its direction of movement.

In further exemplary embodiments it is provided that the neural network has at least one input layer, one output layer, and at least one (hidden) layer arranged between the input layer and the output layer.

In further exemplary embodiments, it is provided that one or the input layer of the neural network has at least three input neurons, wherein in particular one of the plurality of operating variables is fed to each of the three input neurons, and / or wherein one or the output layer of the neural network has an output neuron, especially for the output of the first size.

In further exemplary embodiments it is provided that at least one or the at least one layer, in particular hidden, arranged between the input layer and the output layer has at least 20 neurons, in particular at least 40 neurons.

In further exemplary embodiments it is provided that the neural network has at least three (hidden) layers arranged between the input layer and the output layer, in particular at least one of the at least three layers arranged between the input layer and the output layer, 40 or more neurons, in particular 60 or more neurons.

In further exemplary embodiments, it is provided that the neural network uses an activation function of the ReLU type, which enables efficient evaluation even on devices with limited computing resources such as embedded systems, e.g. control units, in particular for motor vehicles.

In further exemplary embodiments, it is provided that at least some of the plurality of operating variables are normalized. In further exemplary embodiments, this can also take place in particular in the training method described below.

Further exemplary embodiments relate to a method for training an, in particular artificial, neural network, wherein the neural network can be used in particular for a method according to the embodiments, the method for training having the following steps: providing training data, dividing the training data into learning data and test data, with the splitting in particular taking place on a random or pseudo-random basis, training of the neural network based on the learning data and / or the test data.

Further exemplary embodiments relate to a device for carrying out the method according to the embodiments.

Further exemplary embodiments relate to a computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the method according to the embodiments.

Further exemplary embodiments relate to a computer program, comprising instructions which, when the program is executed by a computer, cause the computer to execute the method according to the embodiments.

Further exemplary embodiments relate to a data carrier signal which the computer program transmits and / or characterizes according to the embodiments.

Further exemplary embodiments relate to a use of the method according to the embodiments and / or the device according to the embodiments and / or the computer-readable storage medium according to the embodiments and / or the computer program according to the embodiments and / or the data carrier signal according to the embodiments for at least one of the the following elements: a) Determination of a coking of an injection valve, in particular for an internal combustion engine, in particular of a motor vehicle, characterizing first variable, b) detection of coking of an injection valve, c) reduction of the amount of data required to determine coking of an injection valve, in particular for a Internal combustion engine, in particular of a motor vehicle, characterizing first variable.

Further features, possible applications and advantages of the invention emerge from the following description of exemplary embodiments of the invention, which are shown in the figures of the drawing. All of the features described or shown form the subject matter of the invention individually or in any combination, regardless of their summary in the claims or their reference, and regardless of their wording or representation in the description or in the drawing.

• 1 schematisch ein vereinfachtes Blockdiagramm eines Einspritzventils gemäß beispielhaften Ausführungsformen in einem Zielsystem,
• 2A schematisch ein vereinfachtes Flussdiagramm gemäß weiteren beispielhaften Ausführungsformen,
• 2B schematisch ein vereinfachtes Flussdiagramm gemäß weiteren beispielhaften Ausführungsformen,
• 3 schematisch ein neuronales Netzwerk gemäß weiteren beispielhaften Ausführungsformen,
• 4 schematisch ein vereinfachtes Blockdiagramm einer Vorrichtung gemäß weiteren beispielhaften Ausführungsformen,
• 5 schematisch Aspekte eines Einspritzventils gemäß weiteren beispielhaften Ausführungsformen,
• 6 schematisch Zeitverläufe gemäß weiteren beispielhaften Ausführungsformen,
• 7 schematisch ein Diagramm einer simulierten Verkokung gemäß weiteren beispielhaften Ausführungsformen,
• 8 schematisch eine Verteilung der simulierten Verkokung gemäß 7,
• 9 Zusammenhänge simulierter Größen gemäß weiteren beispielhaften Ausführungsformen,
• 10 Zusammenhänge simulierter Größen mit einer Verkokung gemäß weiteren beispielhaften Ausführungsformen,
• 11 schematisch einen Lernfortschritt bei nicht normierten Eingangsgrößen gemäß weiteren beispielhaften Ausführungsformen,
• 12 schematisch einen Lernfortschritt bei normierten Eingangsgrößen gemäß weiteren beispielhaften Ausführungsformen, und
• 13 schematisch einen Abbildungsfehler eines neuronalen Netzwerks auf einem Testdatensatz gemäß weiteren bevorzugten Ausführungsformen.

In the drawing shows:

• 1 schematically a simplified block diagram of an injection valve according to exemplary embodiments in a target system,
• 2A schematically a simplified flowchart according to further exemplary embodiments,
• 2 B schematically a simplified flowchart according to further exemplary embodiments,
• 3 schematically a neural network according to further exemplary embodiments,
• 4th schematically a simplified block diagram of a device according to further exemplary embodiments,
• 5 schematically aspects of an injection valve according to further exemplary embodiments,
• 6th schematic time courses according to further exemplary embodiments,
• 7th schematically a diagram of a simulated coking according to further exemplary embodiments,
• 8th schematically shows a distribution of the simulated coking according to FIG 7th ,
• 9 Relationships between simulated variables according to further exemplary embodiments,
• 10 Correlations of simulated variables with coking according to further exemplary embodiments,
• 11th schematically, a learning progress with non-standardized input variables according to further exemplary embodiments,
• 12th schematically a learning progress with standardized input variables according to further exemplary embodiments, and
• 13th schematically shows a mapping defect of a neural network on a test data set according to further preferred embodiments.

1 schematically shows a simplified block diagram of an injection valve 10 according to exemplary embodiments in a target system 20th , which in the present case is an internal combustion engine, for example 20th of a motor vehicle 1 acts. For example, is the injector 10 designed as a solenoid valve and intended to deliver fuel into a combustion chamber (not shown) of the internal combustion engine 20th inject. The injection valve can do this 10 a valve needle or nozzle needle 12th have, see the schematic representation according to 5 , with the magnification 11th an end region of the valve needle 12th and a nozzle opening 13th shows. The nozzle opening 13th can be at least partially clogged at least occasionally by combustion products of the injected fuel and the like (“coking”), see the reference symbol 14th whereby the injection of fuel by means of the injector 10 is influenced in a manner that is undesirable per se.

6th shows schematically in diagram A an exemplary time curve of a control signal (“injection pattern”) for the injection valve 10 , in diagram B with the same scaling of the time axis t, a needle lift of the injection needle resulting therefrom 12th , and in diagram C, also with the same scaling of the time axis t, an associated injection rate. The curve C1 from diagram B indicates an ideal needle lift of an ideal injection valve, whereas the curve C2 symbolizes a real needle stroke, for example as a result of coking 14th with the real injector 10 adjusts. The curve C3 from diagram C indicates an ideal injection rate of the ideal injection valve, whereas the curve C4 a real injection rate of the real injector 10 symbolizes how it is, for example, due to coking 14th adjusts.

It can be seen that the coking 14th the injection rate is reduced and the injection duration is increased, which, for example, has a negative impact on the emissions behavior of the internal combustion engine 20th ( 1 ) can affect.

Further exemplary embodiments relate to a method for determining coking 14th an injector 10 , especially for an internal combustion engine 20th , in particular a motor vehicle 1 , characterizing first size G1 , see. 2A comprising the following steps: Deploy 100 several operating variables BG1, BG2, BG3 of the injection valve 10 , Determine 110 the first size G1 by means of an, in particular artificial, neural network based on the multiple operating variables BG1, BG2, BG3. This allows the first size G1 efficient as a measure of coking 14th of the injector 10 be determined.

In further exemplary embodiments, it is provided that the neural network is designed as a multi-layer perceptron, in particular a multilayer perceptron. 3 shows a structure of the neural network 30th according to exemplary embodiments.

In further exemplary embodiments it is provided that the several operating variables BG1, BG2, BG3 have at least one of the following elements: a) fuel pressure, b) needle reversal point, c) injection duration.

For example, the fuel pressure characterizes the pressure by means of the injection valve 10 fuel to be injected. In some embodiments, the fuel can be provided, for example, by means of a high-pressure fuel reservoir (“rail” or “common rail”), the fuel pressure then corresponding, for example, to the rail pressure.

For example, the injection duration characterizes a period of time during which a fuel injection by means of the injection valve 10 is performed.

For example, the needle reversal point characterizes the point in time during a fuel injection at which a valve needle or nozzle needle 12th ( 5 ) of the injector 10 reverses their direction of movement.

In further exemplary embodiments it is provided that the neural network 30th ( 3 ) at least one input layer 31 has an output layer 37 , and at least one between the input layer 31 and the output layer 37 arranged (hidden), layer 33 , 35 .

In further exemplary embodiments it is provided that one or the input layer 31 of the neural network 30th at least three input neurons N1 , N2 , N3 having, in particular, each of the three input neurons N1 , N2 , N3 in each case one of the several operating variables BG1, BG2, BG3 is supplied.

In further exemplary embodiments it is provided that the output layer 37 of the neural network 30th has an output neuron NA, in particular for outputting the first variable G1 .

In further exemplary embodiments it is provided that at least one or the at least one layer arranged between the input layer and the output layer has at least 20 neurons, in particular at least 40 neurons.

In further exemplary embodiments it is provided that the neural network 30th at least three between the input layer 31 and the output layer 37 having arranged layers, wherein in particular at least one of the at least three layers arranged between the input layer and the output layer has 40 or more neurons, in particular 60 or more neurons.

In further exemplary embodiments it is provided that the neural network 30th an activation function of the ReLU type is used, which enables efficient evaluation even on devices with limited computing resources such as embedded systems, for example control devices, in particular for motor vehicles.

In further exemplary embodiments, the network 30th the input layer 31 on, an activation layer 32 , a hidden one layer 33 , an activation layer 34 , a hidden layer 35 , an activation layer 36 , and the output layer 37 .

In further exemplary embodiments, it is provided that at least some of the multiple operating variables BG1, BG2, BG3 are normalized. This can be done in the case of further exemplary embodiments, in particular also in the case of the one below with reference to FIG 2 B described training procedure.

Further exemplary embodiments, cf. 2 B , relate to a method for training an, in particular an artificial, neural network 30th ( 3 ), being the neural network 30th can be used in particular for a method according to the embodiments, the method having the following steps for training: providing 150 of training data TrD, splitting 160 of the training data TrD into learning data LD and test data TD, in particular the splitting 160 random or pseudo-random, training 170 of the neural network 30th based on the learning data LD and / or the test data TD. For example, 75 percent of the training data TrD are used as learning data LD and 25 percent of the training data as test data TD.

In further exemplary embodiments, when training the neural network 30th For example, a third of the learning data LD can be used as validation data.

In further exemplary embodiments, overfitting (“overfitting”) can be avoided, for example, by using an “early stopping” method.

In further exemplary embodiments, the training data TrD can, for example, by means of a simulation of the injection valve 10 be determined.

Further exemplary embodiments, cf. 4th , refer to a device 200 to carry out the procedure (see e.g. 2A and or 2 B) according to the embodiments.

The device 200 has, for example: one at least one computation kernel 202a having computing device (“computer”) 202, one of the computing device 202 associated storage device 204 for at least temporary storage of at least one of the following elements: a) data DAT, b) computer program PRG, in particular for executing a method according to the embodiments.

In further exemplary embodiments, the computer program PRG can also characterize at least one application that is at least temporarily on the computing device 202 is carried out, e.g. to control the injection valve 10 or from components (e.g. solenoid actuator (not shown) on the valve needle 12th can work) of this.

In further preferred embodiments, the storage device 204 volatile memory 204a (e.g. main memory (RAM)) and / or a non-volatile memory 204b (e.g. Flash EEPROM).

In further exemplary embodiments, the computing device 202 at least one of the following elements or is designed as at least one of these elements: microprocessor (µP), microcontroller (µC), application-specific integrated circuit (ASIC), system on chip (SoC), programmable logic module (e.g. FPGA, field programmable gate array ), Hardware circuit, graphics processor (GPU), accelerator circuit for evaluating / executing the neural network 30th , or any combination of these.

Further exemplary embodiments relate to a computer-readable storage medium SM, comprising instructions PRG which, when executed by a computer 202 cause them to carry out the method according to the embodiments.

Further exemplary embodiments relate to a computer program PRG, comprising instructions that are used when the program is executed by a computer 202 cause them to carry out the method according to the embodiments.

Further exemplary embodiments relate to a data carrier signal DCS, which the computer program PRG characterizes and / or transmits according to the embodiments. The data carrier signal DCS is, for example, via an optional data interface 206 the device 200 receivable.

In further exemplary embodiments, the data interface 206 also for receiving at least one of the operating variables BG1, BG2, BG3 and / or for outputting the first variable G1 be used.

Further exemplary embodiments relate to a use of the method according to the embodiments and / or the device according to the embodiments and / or the computer-readable storage medium according to the embodiments and / or the Computer program according to the embodiments and / or the data carrier signal according to the embodiments for at least one of the following elements: a) determining a first variable characterizing coking of an injection valve, in particular for an internal combustion engine, in particular of a motor vehicle, b) detection of coking of an injection valve, c ) Reduction of the amount of data required for the determination of a first variable characterizing coking of an injection valve, in particular for an internal combustion engine, in particular of a motor vehicle.

The principle according to the embodiments can advantageously replace or supplement conventional methods for determining coking and, in some embodiments, can be carried out with comparatively little or even negligible data input, whereby it can be, for example, more robust and / or faster than some conventional methods.

In further exemplary embodiments, the method can, for example, according to 2A during operation of the internal combustion engine 20th be carried out in the field, for example by a control unit for the internal combustion engine 20th trained device 200 ( 4th ).

In further exemplary embodiments, the training method, cf. 2 B , also by one as a control unit for the internal combustion engine 20th trained device 200 are executed. In further exemplary embodiments, the training method, cf. 2 B , alternatively or in addition also by a not as a control unit for the internal combustion engine 20th trained device 200 be executed, for example a stationary computing device, which may have a greater computing power than the control unit.

1. a) Kraftstoffdruck, z.B. Raildruck. Als Stimulus ist z.B. der diskrete Wertebereich von sechs Raildruckstufen verwendbar,
2. b) Nadelumkehrpunkt: Stimulus ist z.B. der Wertebereich zwischen dem minimalen und maximalen sinnvollen Wert für die jeweilige Druckstufe, variiert in diskreten Stufen,
3. c) Einspritzdauer: Stimulus ist z.B. der Wertebereich basierend auf dem Wert des Nadelumkehrpunktes mit einem Offset zum Nadelumkehrpunkt und Variation bis zu einem vorgegebenen Maximalwert in diskreten Stufen.

In further exemplary embodiments, a method for determining coking can be simulated using actual vehicle data, with variables being varied that have a known, in particular direct, relationship to coking of an injection valve, the variables in particular having at least one of the following elements:

1. a) Fuel pressure, e.g. rail pressure. For example, the discrete value range of six rail pressure levels can be used as a stimulus,
2. b) Needle reversal point: Stimulus is e.g. the range of values between the minimum and maximum meaningful value for the respective pressure level, varies in discrete levels,
3. c) Injection duration: The stimulus is, for example, the range of values based on the value of the needle reversal point with an offset to the needle reversal point and variation up to a predetermined maximum value in discrete steps.

In further exemplary embodiments, the result of the simulation can correspond, for example, to an instantaneous unfiltered coking value. In further exemplary embodiments, coking values that are not physically meaningful can, for example, be excluded from the simulated data record. This leads to the example in 7th The diagram shown, which illustrates a simulated coking, and in which scanning values ("Samples") are plotted on the horizontal axis, which are dependent on the fuel pressure or rail pressure, with a measure for the coking being plotted on the vertical axis. 8th shows an example of a distribution of the simulated coking according to FIG 7th .

9 shows, by way of example, a relationship between the simulated variables in accordance with further exemplary embodiments, where series R1 and column S1 is assigned to the rail pressure, where series R2 and column S2 is assigned to the needle reversal point, where row R3 and column S3 is assigned to the injection duration, and where series R4 and column S4 is associated with coking.

10 shows, by way of example, a relationship between the simulated variables in accordance with further exemplary embodiments, where series R1 and column S1 is assigned to the rail pressure, where series R2 and column S2 is assigned to the needle reversal point, where row R3 and column S3 is assigned to the injection duration. The different data of the individual diagrams, see e.g. the curves C5 , C6 correspond to different values for coking.

11th shows an example of a learning progress when training an artificial neural network with non-standardized input variables, with a loss plotted over epochs.

12th shows an example of a learning progress when training an artificial neural network with standardized input variables, with a loss plotted over epochs. It can be seen that the use of standardized input variables accelerates learning progress and makes it more robust.

13th shows resulting mapping errors R according to exemplary embodiments of a trained neural network on a test data set. A percentage of coking is plotted on the vertical axis, over a number of samples on the horizontal axis.

In further exemplary embodiments, as an alternative or in addition to the three operating variables BG1, BG2, BG3 mentioned above by way of example, one or more further variables of the injection valve can be used 10 and / or the target system 20th used to based on it by means of the neural network 30th the first size G1 to investigate. As a result, the method can possibly be made even more precise and / or more robust.

In further exemplary embodiments, at least some of the input variables BG1, BG2, BG3 and / or the output variable G1 be filtered, for example by means of averaging over several values or using a PT1 filter, for example (linear time-invariant transmission element with proportional transmission behavior and first-order delay).

In further exemplary embodiments, the neural network 30th repeatedly, in particular periodically, evaluated to the first size G1 to investigate. In further exemplary embodiments, a period can be, for example, several seconds or minutes, in particular also several hours or even days. Since coking is a comparatively slow process, these exemplary values for the period offer an accuracy that is sufficient for many applications. Corresponding computing time resources of the device are also advantageous as a result 200 saved.

Claims (15)

Method for determining a first variable (G1) characterizing coking of an injection valve (10), in particular for an internal combustion engine (20), in particular a motor vehicle (1), comprising the following steps: Providing (100) several operating variables (BG1, BG2, BG3) of the injection valve (10), determining (110) the first variable (G1) by means of an, in particular artificial, neural network (30) based on the several operating variables (BG1, BG2, BG3). Procedure according to Claim 1 , the neural network (30) being designed as a multilayer perceptron, in particular a multilayer perceptron. Method according to at least one of the preceding claims, wherein the several operating variables (BG1, BG2, BG3) have at least one of the following elements: a) fuel pressure, b) needle reversal point, c) injection duration. Method according to at least one of the preceding claims, wherein the neural network (30) has at least one input layer (31), an output layer (37), and at least one layer (33, 35) arranged between the input layer (31) and the output layer (37) ). Method according to at least one of the preceding claims, wherein one or the input layer (31) of the neural network (30) has at least three input neurons, in particular one of the several operating variables (BG1, BG2, BG3) being fed to each of the three input neurons, and / or wherein one or the output layer (37) of the neural network (30) has an output neuron, in particular for outputting the first variable (G1). Method according to at least one of the preceding claims, wherein at least one or the at least one layer (33, 35) arranged between the input layer (31) and the output layer (37) has at least 20 neurons, in particular at least 40 neurons. Method according to at least one of the preceding claims, wherein the neural network (30) has at least three layers (33, 35) arranged between the input layer (31) and the output layer (37), in particular at least one of the at least three between the input layer (31 ) and the output layer (37) arranged layers (33, 35), 40 or more neurons, in particular 60 or more neurons. Method according to at least one of the preceding claims, wherein the neural network (30) uses an activation function of the ReLU type. Method according to at least one of the preceding claims, wherein at least some of the plurality of operating parameters (BG1, BG2, BG3) are normalized. Method for training an, in particular artificial, neural network (30), wherein the neural network (30) can be used in particular for a method according to at least one of the preceding claims, wherein the method for training has the following steps: providing (150) training data (TrD), splitting (160) the training data (TD) into learning data (LD) and test data (TD), in particular the splitting (160) being random or pseudo-random, training (170) the neural network (30) based on the learning data (LD) and / or the test data (TD). Device (200) for carrying out the method according to at least one of the preceding claims. Computer-readable storage medium (SM), comprising instructions (PRG) which, when executed by a computer (202), cause the computer (202) to perform the method according to at least one of Claims 1 until 10 to execute. Computer program (PRG), comprising instructions which, when the program (PRG) is executed by a computer (202), cause the computer (202) to execute the method according to at least one of Claims 1 until 10 to execute. Data carrier signal (DCS), which the computer program (PRG) according to Claim 13 transmits and / or characterizes. Use of the method according to at least one of the Claims 1 until 10 and / or the device (200) Claim 11 and / or the computer-readable storage medium (SM) according to Claim 12 and / or the computer program (PRG) Claim 13 and / or the data carrier signal (DCS) Claim 14 for at least one of the following elements: a) determining a first variable (G1) characterizing coking of an injection valve (10), in particular for an internal combustion engine (20), in particular of a motor vehicle (1), b) detection of coking of an injection valve (10) ), c) Reduction of the amount of data required to determine a coking of an injection valve (10), in particular for an internal combustion engine (20), in particular a motor vehicle (1), characterizing first variable (G1).

Images