Method and device for determining a closing point in time of an injector of an internal combustion engine

Abstract

A computer-implemented method for determining a closing time point of an injector of an internal combustion engine by means of a classifier, wherein the method comprises the steps of: determining a time series of input signals, wherein the input signals each correspond to a time point within the time series and each characterize a deformation of the injector; determining, by means of the classifier, a plurality of first values based on the time series of input signals, wherein each first value corresponds to a time point of the time series and the first value characterizes a probability that the closing time point of the injector coincides with the time point; determining a plurality of second values, wherein a second value is a sum of an adjacent first value and the first value, respectively, wherein the second value corresponds to the time point to which the first value corresponds; determining the closing time point based on a maximum second value of the plurality of second values.

Description

Technical Field

[0001] The present invention relates to a method for determining the shut-off time of an injector in an internal combustion engine, a control system, a training system, a computer program, and a machine-readable storage medium. Background Technology

[0002] A method for determining the shut-off time of a fuel injector by means of a sensing device having a piezoelectric element is known from DE 10 2014 219 242 A1.

[0003] Advantages of the present invention

[0004] In internal combustion engines, such as diesel or gasoline engines with direct injection, direct fuel injection allows for advantageous operating characteristics. One challenge is to control the combustion process as precisely as possible to improve the engine's operating characteristics, particularly regarding fuel consumption, efficiency, emissions, and operational smoothness.

[0005] Therefore, it is important that the injectors (including nozzles or valves) of the internal combustion engine be operated so that the amount of fuel to be injected can be dispensed with high repeatability. The injector may, for example, have an electromagnetic or piezoelectric actuator that can manipulate the injector's valve needle to lift the needle from its seat and open the injector's outlet orifice to discharge fuel into the combustion chamber.

[0006] Due to structural differences, structural tolerances, and / or different operating conditions, such as temperature, fuel pressure, or fuel viscosity, there is uncertainty in determining the exact shut-off point of the injector, that is, the point at which the injector needle closes and no more fuel enters the combustion chamber via the injector.

[0007] The method having the features of independent claim 1 has the advantage that the shut-off time of the injector in the internal combustion engine can be determined more precisely. Advantageously, this allows for a more precise determination of the amount of fuel released by the injector during injection. This allows for improved control of the internal combustion engine because the amount of fuel can be measured more accurately. Advantageously, this improves the operating characteristics of the internal combustion engine, particularly those relating to fuel consumption, efficiency, pollutant emissions, and operational smoothness.

[0008] The advantage of this method is that it uses a special classifier that can classify the exact shut-off time of the injector with great precision. Summary of the Invention

[0009] In a first aspect, the present invention relates to a computer-implemented method for determining the shut-off time of an injector in an internal combustion engine by means of a classifier, wherein the method comprises the following steps:

[0010] • Determine the time series of the input signals, wherein these input signals correspond to time points within the time series and each input signal characterizes the deformation of the injector;

[0011] • Using a classifier, multiple first values ​​are determined based on the time series of the input signal, where each first value corresponds to a time point in the time series, and the first value represents the probability that the shut-off time of the injector coincides with that time point;

[0012] • Determine multiple second values, wherein each second value is the sum of a first value and its adjacent first values, wherein the adjacent first values ​​are determined based on the time points corresponding to these first values, and the second value corresponds to the time point corresponding to the first value;

[0013] • The shutdown time point is determined based on the largest of the plurality of second values.

[0014] This method can be understood as determining the precise shut-off time of the injector based on its deformation. The inventors discovered that this method for determining the shut-off time can pinpoint the exact shut-off time.

[0015] The injector can be, in particular, a diesel or gasoline engine injector with direct injection.

[0016] To determine this time series, measurements can be taken at preferably periodic time points using suitable sensors capable of determining the deformation of the injector, wherein each measurement can be used as an input signal for the time series. In particular, this allows each input signal of the time series to be assigned a time point, such as a clock time or a relative time point of the input signal within the time series.

[0017] It is also possible to use segments or subsets of other time series as the time series.

[0018] In an advantageous implementation, it is possible to determine these input signals by means of a piezoelectric sensor.

[0019] This is advantageous because piezoelectric sensors can achieve very precise measurements of deformation because they are insensitive to electromagnetic radiation or fields and have a high, fixed frequency. This allows for very precise determination of the injector's deformation, leading to a more accurate determination of the shut-off time.

[0020] The time series is transmitted to a classifier. The classifier can be understood as determining whether a time point is a shutdown point for each time point depicted by the time series. Advantageously, the classifier is shown the measurement context by transmitting the input signal of the time series, based on which it can determine whether a shutdown point is a shutdown point for the corresponding time point. The classifier determines the plurality of first values ​​based on the time series. The first values ​​can be understood as corresponding to the time points represented by the time series. In particular, it is possible that a first value is determined for each input signal of the time series, wherein the first value represents the probability that the time point at which the input signal is determined by the sensor is a shutdown point.

[0021] The classifier may include a machine learning model, which can be used to determine the first values. Alternatively or additionally, these first values ​​may be determined based on a rule-based model of the classifier.

[0022] In an advantageous implementation, the classifier may include a neural network, by means of which the plurality of first values ​​are determined.

[0023] The inventors were able to discover that neural network-based classifiers can advantageously achieve the highest accuracy when determining the shutdown point based on the time series of the input signal.

[0024] These first values ​​may have already been used to determine the shutdown time point. This is especially problematic when the distribution of the determined first values ​​is multimodal, leading to an imprecise determination of the shutdown time point. Therefore, this method advantageously uses the plurality of second values ​​to estimate the shutdown time point. In particular, it is conceivable that for each first value, there exists a second value representing the probability that the time point corresponding to that first value is the shutdown time point, wherein the second value is determined based on the first value and another first value adjacent to it.

[0025] Therefore, the second value can be understood as representing the probability that a given time point is a closing time point based on multiple probabilities around that time point. Thus, the second value can be particularly understood as representing a probability measure.

[0026] By determining the shutdown time point based on probability measures, the accuracy of the shutdown time point determination can be improved. In particular, this can suppress possible inaccuracies or numerical outliers within these initial values, which improves the accuracy of the determination.

[0027] In particular, it is possible that in the step of determining the plurality of second values, a predetermined first number of previous first values ​​and a predetermined second number of subsequent first values ​​of the first value form adjacent first values.

[0028] The order of these first values ​​can be determined, in particular, based on the time points corresponding to these first values ​​in the time series. Specifically, these first values ​​can be ordered according to their corresponding time points. Then, for each first value, its predecessor and successor are determined based on this order. It is also possible, in particular, that the neighborhood, i.e., the set of first values ​​preceding and following the first value, can be determined based on this order.

[0029] In the case of the first value at a given time point in the marginal region of the time series, it is possible that only subsequent first values ​​or only previous first values ​​are used as the neighborhood.

[0030] In an advantageous implementation, it is possible that each of the plurality of second values ​​is determined by means of a one-dimensional discrete convolution in the step of determining the plurality of second values.

[0031] This is advantageous because there are dedicated hardware and software for discrete convolutions, which can speed up the method. This is also advantageous because the method is typically repeated frequently to determine the ejector's shut-off time, allowing for a time-efficient design.

[0032] It is also possible that, in the step of determining the shutdown time point, the time point corresponding to the largest second value is determined as the shutdown time point.

[0033] In another preferred embodiment, it is possible to control the internal combustion engine based on the determined shut-off time.

[0034] By precisely determining the shut-off time, the internal combustion engine can be controlled with great precision. Advantageously, this can improve the engine's fuel consumption, efficiency, emissions, and / or smoothness of operation.

[0035] In a preferred embodiment of the method, it is possible that the method additionally includes training a classifier, wherein the classifier is trained to determine, for a time series of the injector's input signal, whether a corresponding time point in the time series represents the injector's shut-off time point.

[0036] In particular, the classifier can be trained under supervision, that is, the classifier can be trained to minimize the error between the output signal determined by the classifier and the desired output signal.

[0037] The training dataset required for this training can preferably be determined on a measuring station. In particular, by corresponding measurements within the measuring station, the flow of fuel from the injector can be determined, and thus the precise shut-off time of the injector can be determined with respect to the time series. The shut-off time thus determined can then be used as the desired shut-off time that the classifier should predict for the time series. Attached Figure Description

[0038] Embodiments of the present invention will then be described in more detail with reference to the accompanying drawings. In the drawings:

[0039] Figure 1 The classifier is illustrated schematically;

[0040] Figure 2 The schematic diagram illustrates the construction of a control system for manipulating an injector by means of a classifier;

[0041] Figure 3 The training system used to train the classifier is illustrated schematically. Detailed Implementation

[0042] Figure 1 A classifier (60) is shown. In an embodiment, the classifier (60) includes a neural network (61) for classifying a time series (x) of the input signal. Furthermore, for each input signal, there is a time point. This time point can be either an absolute time point, such as clock time or the number of seconds elapsed since a predetermined time point, or a relative time point within the time series. In alternative embodiments, it is also possible to use other machine learning models, such as a Support Vector Machine, a Random Forest Classifier, or a Gaussian Process, to classify the time series.

[0043] The neural network (61) determines multiple first values ​​(p1, p2, p3, p4, p5) based on the time series (x). n ), where a neural network (61) determines a first value (p1, p2, p3, p) for each input signal of the time series (x). n Here, the first value (p1, p2, p3, p) n ) respectively represent the first value (p1, p2, p3, p n The probability that the input signal's time point corresponds to the injector's closing time point is determined by the input signal's time point. Therefore, the neural network preferably outputs a first value (p1, p2, p3, p...). nThe vector (63), where these first values ​​(p1, p2, p3, p) n The activation function is determined by the output layer of the neural network (60). In this embodiment, the output layer uses the Sigmoid function as the activation function. In alternative embodiments, it is also possible to use the Softmax function as the activation function or not use any activation function.

[0044] These first values ​​(p1, p2, p3, p) n The arrangement of the vector (63) is preferably according to these first values ​​(p1, p2, p3, p...). n The selection is based on the order of these first values ​​(p1, p2, p3, p...). n The order is selected based on the corresponding time points. For example, it might be: the incrementing index of the vector components represents the first values ​​(p1, p2, p3, p...). n (The continuation of the corresponding time point)

[0045] The vector (63) is fed into a one-dimensional discrete convolution function, which determines multiple second values ​​(z1, z2, z3, z4). n Preferably, vector (63) is padded with zeros (65) before convolution, such that the first values ​​(p1, p2, p3, p4) present in vector (63) are determined by the convolution. n The same number of second values ​​(z1, z2, z3, z) n However, alternatively, it is also possible that the convolution only considers the first values ​​(p1, p2, p3, p) present in vector (63). n ), and thus compared with the first value (p1, p2, p3, p) that exists. n The second value (z1, z2, z3, z) is less. n The values ​​are determined. In this embodiment, the convolution includes three first values ​​(p1, p2, p3, p...). n ).

[0046] The second value (z1, z2, z3, z n This can be understood as the second value representing the first value (p1, p2, p3, p...). n The sum of the neighborhood of (z1, z2, z3, z). n In particular, this can be understood as the second value relating to the plurality of first values ​​(p1, p2, p3, p...). n The second value corresponds to a reference value. Preferably, the convolution is based on an odd number of first values ​​(p1, p2, p3, p...). n ), where the second value (z1, z2, z3, z n) has these first values ​​(p1, p2, p3, p n In terms of sorting, it is the first value of the middle element (p1, p2, p3, p) n (z1, z2, z3, z4) is used as a reference value. A reference element can be understood as defining the boundary between the second value (z1, z2, z3, z4) and the reference value (z5, z6). n The reference element determines the time point corresponding to the second value (z1, z2, z3, z) within this time series. n The relevant time points.

[0047] Alternatively, it can be conceivable that the convolution is based on an even number of first values ​​(p1, p2, p3, p...). n In this case, the second value (z1, z2, z3, z) n Preferably, it can have two reference values, namely, for these first values ​​(p1, p2, p3, p...). n Regarding the sorting of z1, z2, z3 ... n The corresponding time point can be, for example, the time point between the time point of the first reference value and the time point of the second reference value, preferably the time point in the middle of the two time points.

[0048] Next, a second value (z1, z2, z3, z4) can be provided in the output signal (y). n Alternatively or additionally, it is also possible to provide a time point in the output signal (y) as the determined off time, where the second values ​​(z1, z2, z3, z) are used. n The time point corresponding to the second largest value in () is taken as the time point.

[0049] Figure 2 A control system (40) for controlling the injector (20) of an internal combustion engine is shown. Preferably, the deformation of the injector (20) is detected at regular time intervals by means of a sensor (30). In this embodiment, the sensor (30) is a piezoelectric sensor. In alternative embodiments, other sensors (30) can also be used to determine the deformation of the injector, such as strain gauge-based sensors (30).

[0050] The measurements (S) determined by the sensor (30) are transmitted to the control system (40). Therefore, the control system (40) receives the sequence of measurements (S). Based on this, the control system (40) determines the control signals (A), which are transmitted to the control unit (10) of the injector (20).

[0051] The control system (40) receives a sequence of measurements (S) from the sensor (30) in a receiving unit (50), which converts the sequence of measurements (S) into a time series (x) of the input signal. This time series can be determined, for example, by selecting between measurements that occurred in the past and measurements (S) that are currently in use. Alternatively, it is conceivable that the time series includes a predetermined number of measurements that occurred in the past and measurements (S) that are currently in use. In other words, the time series (x) is determined based on the sensor signals (S). The time series (x) of the input signal is then fed to a classifier (60).

[0052] The classifier (60) is preferably parameterized by parameters (φ), which are stored in and provided by a parameter memory (P).

[0053] The classifier (60) determines the output signal (y) based on the time series (x). The output signal (y) is sent to an optional modification unit (80) which determines control signals (A) accordingly. These control signals are sent to the control unit (10) of the injector (20) so as to control the injector (20) accordingly.

[0054] The control unit (10) receives the control signal (A), is correspondingly controlled, and performs the corresponding action. In this case, the control unit (10) may include (not necessarily structurally integrated) control logic that determines a second control signal for controlling the injector (20) based on the control signal (A).

[0055] In other embodiments, the control system (40) includes a sensor (30). In still other embodiments, alternatively or additionally, the control system (40) also includes a control unit (10).

[0056] In other preferred embodiments, the control system (40) includes at least one processor (45) and at least one machine-readable storage medium (46) on which commands are stored, which, when executed on the at least one processor (45), cause the control system (40) to perform the method according to the invention.

[0057] In alternative embodiments, instead of the control unit (10) or other than the control unit, it is specified that at least one other device (10a) is operated by means of a control signal (A). This device (10a) could, for example, be a pump of the common-rail system to which the injector (20) belongs. Alternatively or additionally, it is conceivable that the device is a control device for an internal combustion engine. Alternatively or additionally, it is also conceivable that the device (10a) is a display unit by means of which information corresponding to the classification determined by the classifier (60) can be displayed to a person (e.g., a driver or mechanic).

[0058] Figure 3 An embodiment of a training system (140) for training a classifier (60) of a control system (40) using a training dataset (T) is shown. The training dataset (T) includes multiple time series (x) of the input signal. i These time series were used to train a classifier (60), where the training dataset (T) was also used for each time series (x). i All include the desired output signal (y) i The output signal and the time series (x) i ) corresponds to and characterizes the time series (x) i ) classification.

[0059] For training purposes, the training data unit (150) accesses a computer-implemented database (St2), which provides a training dataset (T). The training data unit (150) preferably randomly determines at least one time series (x) based on the training dataset (T). i ) and the time series (x) i The desired output signal (y) corresponding to ) i ), and this time series (x) i The signal is transmitted to the classifier (60). The classifier (60) processes the input signal (x) based on the input signal (x). i To determine the output signal .

[0060] Desired output signal (y) i and the determined output signal It is transferred to the modification unit (180).

[0061] Next, based on the desired output signal (y) i and the determined output signal The modified unit (180) determines the new parameters for the classifier (60). Therefore, the modified unit (180) uses a loss function to adjust the desired output signal (y) i ) and the determined output signal A comparison is made. The loss function determines a first loss value, which characterizes the determined output signal. With the desired output signal (y) i The degree of deviation from the target value is considered. In this embodiment, the negative log-likelihood function is chosen as the loss function. In alternative embodiments, other loss functions are also conceivable.

[0062] The modified unit (180) determines these new parameters based on the first loss value. In this embodiment, this is achieved using gradient descent, preferably stochastic gradient descent, Adam, or AdamW. In alternative embodiments, these new parameters can also be determined using evolutionary algorithms. .

[0063] The determined new parameters The new parameters are stored in the model parameter memory (St1). Preferably, the determined new parameters... As a parameter It is provided to the classifier (60).

[0064] In other preferred embodiments, the described training is iteratively repeated for a predetermined number of iterations, or iteratively repeated until a first loss value falls below a predetermined threshold. Alternatively or additionally, it is also conceivable that the training terminates when the average first loss value on the test or validation dataset falls below a predetermined threshold. In at least one of these iterations, new parameters determined in previous iterations are... The parameters used as classifier (60) .

[0065] Furthermore, the training system (140) may include at least one processor (145) and at least one machine-readable storage medium (146) containing instructions that, when executed by the processor (145), cause the training system (140) to implement the training method according to any of these aspects of the invention.

[0066] The term "computer" includes any device used to run computational rules that can be given in advance. These computational rules can exist in the form of software, hardware, or a combination of both.

[0067] Typically, "multiple" can be understood as indexed, meaning that a unique index is assigned to each element in the multiple, preferably by assigning a unique index to each element in the multiple by assigning consecutive integers to the elements contained in the multiple. Preferably, if "multiple" includes N There are elements, among which N If the number of elements in the plurality is given, then these elements are assigned from 1 to... N Integers.

Claims

1. A computer-implemented method for determining the shut-off time of an injector (20) of an internal combustion engine by means of a classifier (60), wherein the method comprises the following steps: Determine the time series (x) of the input signals, wherein these input signals correspond to time points within the time series and respectively characterize the deformation of the injector (20); Using this classifier (60), multiple first values ​​(p1, p2, p3, p4, p5) are determined based on the time series (x) of the input signal. n ), where each first value (p1, p2, p3, p) n All of these correspond to the time points of the time series (x), and the first value (p1, p2, p3, p...) corresponds to the time points of the time series (x). n ) represents the probability that the shut-off time of the injector (20) coincides with this time point; Determine multiple second values ​​(z1, z2, z3, z) n ), where the second value (z1, z2, z3, z n The values ​​are the first values ​​(p1, p2, p3, p). n The first adjacent values ​​(p1, p2, p3, p) n ) and the first value (p1, p2, p3, p n The sum of the first values ​​(p1, p2, p3, p) of each pair, where the first adjacent values ​​are (p1, p2, p3, p4). n Based on these first values ​​(p1, p2, p3, p...) n The second value (z1, z2, z3, z) is determined by the corresponding time point, and the second value (z1, z2, z3, z4) is determined by the corresponding time point. n ) and the first value (p1, p2, p3, p n The corresponding time point; Based on the aforementioned multiple second values ​​(z1, z2, z3, z... n The second largest value in ) (z1, z2, z3, z n This determines the shutdown time.

2. The method of claim 1, wherein these input signals are determined by means of a piezoelectric sensor (30).

3. The method according to claim 1 or 2, wherein the classifier (60) comprises a neural network (61) by means of which the plurality of first values ​​(p1, p2, p3, p4) are determined. n ).

4. The method according to claim 1 or 2, wherein in determining the plurality of second values ​​(z1, z2, z3, z... n In the step of ), each second value (z1, z2, z3, z) is determined by means of one-dimensional discrete convolution. n ).

5. The method according to claim 1 or 2, wherein in determining the plurality of second values ​​(z1, z2, z3, z... n In the steps of ), the first value (p1, p2, p3, p) n The first number of the predefined first values ​​(p1, p2, p3, p) n ) and the first value (p1, p2, p3, p n The predefined second number of subsequent first values ​​(p1, p2, p3, p) n ) and the first value (p1, p2, p3, p n ) form adjacent first values ​​(p1, p2, p3, p) n ).

6. The method according to claim 1 or 2, wherein in the step of determining the shutdown time point, the largest second value (z1, z2, z3, z4) is compared with the second value (z5, z6, z7, z8, z9, z1, z1, z1, z2, z1, z1, z2, z1, z2, z3 ...2, z1, z2, z3, n The corresponding time point is determined as the closing time point.

7. The method of claim 1 or 2, wherein the internal combustion engine is controlled based on the determined shut-off time.

8. The method of claim 1 or 2, wherein the method additionally includes training the classifier (60), wherein the classifier is trained such that the classifier is trained for the time series (x) of the input signal of the injector (20). i Determine the time series (x) i Does the corresponding time point represent the closing time point of the injector (20)? 9. A control system (40) configured to implement the method according to claim 7.

10. A training system (140) configured to implement the method according to claim 8.

11. A computer program product comprising a computer program configured to perform the method according to any one of claims 1 to 8 when the computer program is executed by a processor (45, 145).

12. A machine-readable storage medium (46, 146) having a computer program stored thereon, the computer program being configured to perform the method according to any one of claims 1 to 8 when executed by a processor (45, 145).

Images