Method and device for ascertaining a closure point in time of an injector of an internal combustion engine with the aid of a machine learning system

Abstract

A computer-implemented method for ascertaining a closure point in time of an injector of an internal combustion engine using a classifier. The method includes: ascertaining a time series of input signals, each corresponding to a point in time within the time series, and each characterizing a deformation of the injector; ascertaining a plurality of first values using the classifier based on the time series, in each case a first value corresponding to a point in time of the time series, and the first value characterizing a probability that the closure point in time of the injector matches the point in time; ascertaining a plurality of second values, each being a sum of neighboring first values, of a first value and the first value, the second value corresponding to the point in time to which the first value corresponds; ascertaining the closure point in time based on the largest second value.

Claims

1. A computer-implemented method for ascertaining a closure point in time of an injector of an internal combustion engine with the aid of a classifier, the method comprising the following steps: ascertaining a time series of input signals, each of the input signals corresponding to a point in time within the time series, and each of the input signals characterizing a deformation of the injector; ascertaining a plurality of first values using the classifier based on the time series of input signals, each of the first values corresponding to a point in time of the time series, and each of the first values characterizing a probability that the closure point in time of the injector matches the point in time corresponding to the first value; ascertaining a plurality of second values, each second value of the second values being a sum of neighboring first values of a first value of the first values and the first value, the neighboring first values being ascertained based on the points in time corresponding to the first values, and the second value corresponding to the point in time to which the first value corresponds; and ascertaining the closure point in time based on a largest second value of the plurality of second values.

2. The method as recited in claim 1, wherein the input signals are ascertained using a piezo sensor.

3. The method as recited in claim 1, wherein the classifier includes a neural network using which the plurality of first values is ascertained.

4. The method as recited in claim 1, wherein in the ascertaining of the plurality of second values, each second value is ascertained using a one-dimensional discrete convolution.

5. The method as recited in claim 1, wherein in the ascertaining of the plurality of second values, a predefined first number of preceding first values of the first value, a predefined second number of subsequent first values of the first value, and the first value form the neighboring first values.

6. The method as recited in claim 1 wherein in the ascertaining of the closure point in time, a point in time that corresponds to the largest second value is ascertained as the closure point in time.

7. The method as recited in claim 1, wherein the internal combustion engine is activated based on the ascertained closure point in time.

8. The method as recited in claim 1, further comprising: training the classifier, the classifier being trained in such a way that for a time series of input signals of an injector, the classifier ascertains whether or not a particular point in time of the time series characterizes a closure point in time of the injector.

9. A control system configured to ascertain a closure point in time of an injector of an internal combustion engine with the aid of a classifier, the control system configured to: ascertain a time series of input signals, each of the input signals corresponding to a point in time within the time series, and each of the input signals characterizing a deformation of the injector; ascertain a plurality of first values using the classifier based on the time series of input signals, each of the first values corresponding to a point in time of the time series, and each of the first values characterizing a probability that the closure point in time of the injector matches the point in time corresponding to the first value; ascertain a plurality of second values, each second value of the second values being a sum of neighboring first values of a first value of the first values and the first value, the neighboring first values being ascertained based on the points in time corresponding to the first values, and the second value corresponding to the point in time to which the first value corresponds; and ascertain the closure point in time based on a largest second value of the plurality of second values.

10. A non-transitory machine-readable memory medium on which is stored a computer program for ascertaining a closure point in time of an injector of an internal combustion engine with the aid of a classifier, the computer program, when executed by a processor, causing the processor to perform the following steps: ascertaining a time series of input signals, each of the input signals corresponding to a point in time within the time series, and each of the input signals characterizing a deformation of the injector; ascertaining a plurality of first values using the classifier based on the time series of input signals, each of the first values corresponding to a point in time of the time series, and each of the first values characterizing a probability that the closure point in time of the injector matches the point in time corresponding to the first value; ascertaining a plurality of second values, each second value of the second values being a sum of neighboring first values of a first value of the first values and the first value, the neighboring first values being ascertained based on the points in time corresponding to the first values, and the second value corresponding to the point in time to which the first value corresponds; and ascertaining the closure point in time based on a largest second value of the plurality of second values.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) FIG. 1 schematically shows a classifier, in accordance with an example embodiment of the present invention.

(2) FIG. 2 schematically shows a design of a control system for activating an injector with the aid of the classifier, in accordance with an example embodiment of the present invention.

(3) FIG. 3 schematically shows a training system for training the classifier, in accordance with an example embodiment of the present invention.

DETAILED DESCRIPTION OF EXAMPLE EMBODIMENTS

(4) FIG. 1 shows a classifier 60. In the exemplary embodiment, classifier 60 includes a neural network 61 for classifying a time series x of input signals. For each input signal, a point in time also exists. The point in time may be either an absolute point in time, for example the time of day or the number of elapsed seconds after a predefined point in time, or a relative point in time within the time series. In alternative exemplary embodiments, it is also possible to use some other machine learning model, for example a support vector machine, a random forest classifier, or a Gaussian process, for classifying the time series.

(5) Neural network 61 ascertains a plurality of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n based on time series x, a first value p.sub.1 p.sub.2, p.sup.3, p.sub.n being ascertained by neural network 61 for each input signal of time series x. A first value n p.sub.1, p.sub.2, p.sub.3, p.sub.n in each case characterizes a probability that the point in time of the input signal corresponding to first value p.sub.1, p.sub.2, p.sub.3, p.sub.n is a closure point in time of an injector. For this purpose, the neural network preferably outputs a vector 63 of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n, first values p.sub.1, p.sub.2, p.sub.3, p.sub.n being ascertained from an output layer of neural network 61. In the exemplary embodiment, the output layer uses a sigmoid function as an activation function. In alternative exemplary embodiments, it is also possible to use a softmax function as an activation function, or to use no activation function.

(6) The arrangement of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n in vector 63 is preferably selected according to a sequence of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n, the sequence being selected based on the points in time corresponding to first values p.sub.1, p.sub.2, p.sub.3, p.sub.n. For example, it is possible for an ascending index of the vector components to characterize a continuation of the points in time corresponding to first values p.sub.1, p.sub.2, p.sub.3, p.sub.n.

(7) Vector 63 is supplied to a one-dimensional discrete convolution function which ascertains a plurality of second values z.sub.1, z.sub.2, z.sub.3, z.sub.n. Prior to the convolution, vector 63 is preferably filled (“padded”) with zeroes 65 in such a way that the convolution ascertains as many second values z.sub.1, z.sub.2, z.sub.3, z.sub.n as the number of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n that exist in vector 63. Alternatively, however, it is also possible for the convolution to take into account only first values p.sub.1, p.sub.2, p.sub.3, p.sub.n that are present in vector 63, and therefore fewer second values z.sub.1, z.sub.2, z.sub.3, Z.sub.n to be ascertained than the number of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n that exist. In the exemplary embodiment, the convolution in each case includes three first values p.sub.1, p.sub.2, p.sub.3, p.sub.n.

(8) A second value z.sub.1, z.sub.2, z.sub.3, z.sub.n may be understood in such a way that it characterizes the sum of a neighborhood of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n. In particular, a second value z.sub.1, z.sub.2, z.sub.3, z.sub.n may be understood in such a way that, with regard to the plurality of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n, it has a reference value to which it corresponds. The convolution is preferably based on an uneven number of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n, a second value z.sub.1, z.sub.2, z.sub.3, z.sub.n having first value p.sub.1, p.sub.2, p.sub.3, p.sub.n as a reference value, which within the meaning of the sorting of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n is the middle element. The reference element may be understood in such a way that it defines a point in time that corresponds to second value z.sub.1, z.sub.2, z.sub.3, z.sub.n. In other words, the reference element determines the point in time within the time series with which second value z.sub.1, z.sub.2, z.sub.3, z.sub.n correlates.

(9) It is also possible for the convolution to be based on an even number of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n. In this case, a second value z.sub.1, z.sub.2, z.sub.3, z.sub.n may preferably have two reference values, namely, the two middle values within the meaning of the sorting of first values p.sub.1, p.sub.2, p.sub.3, p.sub.n. The point in time corresponding to second value z.sub.1, z.sub.2, z.sub.3, z.sub.n may then, for example, be a point in time situated between the point in time of the first of the two reference values, and the point in time of the second of the reference values, preferably the point in time that forms the middle between both points in time.

(10) Second values z.sub.1, z.sub.2, z.sub.3, z.sub.n may then be provided in an output signal y. Alternatively or additionally, it is possible for a point in time to be provided in output signal y as an ascertained closure time, the point in time that corresponds to the largest of second values z.sub.1, z.sub.2, z.sub.3, z.sub.n being assumed as the point in time.

(11) FIG. 2 shows a control system 40 for controlling an injector 20 of an internal combustion engine. A deformation of injector 20 is detected at preferably regular time intervals with the aid of a sensor 30. In the exemplary embodiment, sensor 30 is a piezo sensor. In alternative exemplary embodiments, other sensors 30 for ascertaining a deformation of the injector are also possible, for example sensors 30 based on a strain gauge.

(12) A measurement S ascertained by sensor 30 is transferred to control system 40. Control system 40 thus receives a sequence of measurements S. Control system 40 ascertains activation signals A therefrom, which are transferred to a control unit 10 of injector 20.

(13) Control system 40 receives the sequence of measurements S of sensor 30 in a receiving unit 50, which converts the sequence of measurements S into a time series x of input signals. The time series may be ascertained, for example, using a selection of past measurements and present measurement S. Alternatively, it is possible for the time series to include in each case a predefined number of past measurements and present measurement S. In other words, time series x is ascertained as a function of sensor signal S. Time series x of input signals is supplied to classifier 60.

(14) Classifier 60 is preferably parameterized by parameters ϕ, which are stored in a parameter memory P and provided by same.

(15) Based on time series x, classifier 60 ascertains an output signal y. Output signal y is supplied to an optional conversion unit 80, which ascertains therefrom activation signals A that are supplied to control unit 10 of injector 20 in order to appropriately activate control injector 20.

(16) Control unit 10 receives activation signals A, is appropriately activated, and carries out a corresponding action. Control unit 10 may include a control logic system which is not necessarily structurally integrated, and which ascertains from activation signal A a second activation signal via which injector 20 is then activated.

(17) In further specific embodiments of the present invention, control system 40 includes sensor 30. In yet further specific embodiments, control system 40 alternatively or additionally includes control unit 10 as well.

(18) In further preferred specific embodiments of the present invention, control system 40 includes at least one processor 45, and at least one machine-readable memory medium 46 on which instructions are stored which, when executed on the at least one processor 45, prompt control system 40 to carry out the method according to the present invention.

(19) In alternative specific embodiments of the present invention, as an alternative or in addition to control unit 10 it is provided that at least one further device 10a is activated with the aid of activation signal A. Device 10a may be, for example, a pump of a common rail system to which injector 20 belongs. Alternatively or additionally, it is possible for the device to be a control unit of the internal combustion engine. Alternatively or additionally, it is also possible for device 10a to be a display unit with the aid of which the information concerning the classification ascertained by classifier 60 may be appropriately displayed to a person, for example a driver or a mechanic.

(20) FIG. 3 shows one exemplary embodiment of a training system 140 for training classifier 60 of control system 40 with the aid of a training data set T. Training data set T includes a plurality of time series x.sub.i of input signals that are used for training classifier 60, training data set T also including for each time series x.sub.i a desired output signal y.sub.i that corresponds to time series x.sub.i and characterizes a classification of time series x.sub.i.

(21) For the training, a training data unit 150 accesses a computer-implemented database St.sub.2, database St.sub.2 providing training data set T. Training data unit 150 ascertains from training data set T, preferably randomly, at least one time series x.sub.i and desired output signal y.sub.i corresponding to time series x.sub.i, and transfers time series x.sub.i to classifier 60. Classifier 60 ascertains an output signal ŷ.sub.i based on input signal x.sub.i.

(22) Desired output signal y.sub.i and ascertained output signal ŷ.sub.i are transferred to a changing unit 180.

(23) Based on desired output signal y.sub.i and ascertained output signal ŷ.sub.i, changing unit 180 then determines new parameters Φ′ for classifier 60. For this purpose, changing unit 180 compares desired output signal y.sub.i and ascertained output signal ŷ.sub.i with the aid of a loss function. The loss function ascertains a first loss value, which characterizes the extent of the deviation of ascertained output signal ŷ.sub.i from desired output signal y.sub.i. In the exemplary embodiment, a negative log likelihood function is selected as a loss function. Other loss functions are also possible in alternative exemplary embodiments.

(24) Changing unit 180 ascertains new parameters Φ′ based on the first loss value. In the exemplary embodiment, this takes place with the aid of a gradient descent method, preferably stochastic gradient descent, Adam, or AdamW. In alternative exemplary embodiments, new parameters Φ′ may also be ascertained with the aid of an evolutionary algorithm.

(25) Ascertained new parameters Φ′ are stored in a model parameter memory St.sub.1. Ascertained new parameters Φ′ are preferably provided as parameters Φ to classifier 60.

(26) In further preferred exemplary embodiments, the described training is iteratively repeated for a predefined number of iteration steps, or is iteratively repeated until the first loss value falls below a predefined threshold value. Alternatively or additionally, it is also possible for the training to be ended when an average first loss value with regard to a test data set or validation data set falls below a predefined threshold value. In at least one of the iterations, new parameters Φ′ determined in a previous iteration are used as parameters Φ of classifier 60.

(27) Furthermore, training system 140 may include at least one processor 145, and at least one machine-readable memory medium 146 that contains commands which, when executed by processor 145, prompt training system 140 to carry out a training method according to one of the aspects of the present invention.

(28) The term “computer” encompasses arbitrary devices for processing predefinable computation rules. These computation rules may be present in the form of software, or in the form of hardware, or also in a mixed form made up of software and hardware.

(29) In general, a plurality may be understood to be indexed; i.e., a unique index is assigned to each element of the plurality, preferably by assigning consecutive integers to the elements contained in the plurality. When a plurality includes N elements, where N is the number of elements in the plurality, the integers from 1 to N are preferably assigned to the elements.