DE102016206627A1 - Method and device for measuring a system to be tested - Google Patents

Abstract

The invention relates to a method for measuring a technical system (2), in particular an internal combustion engine for a motor vehicle, for the creation of a system model of the technical system (2), wherein the surveying with a number of measuring points (X) is performed, one at a time Obtain value of at least one output variable (y), comprising the following steps: - determining (S2) confidence quantities for a set of provided measurement points based on a given classification model, wherein the confidence quantity at a measurement point indicates a modeled allowance of an operating point that is due to the operation of the technical system at the measuring point; - selecting (S4) a measuring point from the set of measuring points, the selected measuring point having an associated confidence quantity indicating a predetermined proximity to an operating limit indicating a limit between permissible and non-permissible operating states; - performing (S5) a survey of the technical system (2) at the selected measurement point to determine the at least one output variable; and - updating (S6) the classification model with the measuring point and an indication dependent on the at least one output variable, which indicates whether the operating point of the technical system set by the measurement of the selected measuring point is admissible.

Description

Technical area

The invention relates to test methods, and in particular to methods for providing measuring points with which a technical system to be tested can be measured or with which suitable system models can be provided. In particular, the present invention relates to methods for determining measurement points near system boundaries.

State of the art

When measuring a technical system with measuring points, it is necessary to apply the measuring points in such a way that as many combinations of values of input variables as possible, in particular for different excitation modes, ie. H. including combinations of gradients of the input variables, so that a spatially and dynamically filling assignment of the input data space with measuring points is obtained. The values of the output obtained during the measurement at the measurement points can serve as training data for the creation of a data-based non-parametric function model.

For the technical system to be modeled, for. As an Otto engine, it is particularly important that the physical limitations of the technical system in the measurement, for example on an engine test stand, are not violated in order to avoid a risk or damage to the technical system. During the measurement, the measuring technology monitors the technical system in such a way that combinations of values of the input variables that are hazardous to the system are detected before the unit is damaged.

The information on the course of an operating limit, which represents the transition from permissible value combinations of the input variables (values of the input variables within an admissible operating range) to system-hazardous value combinations of the input variables (values of the input variables outside a permissible operating range) in the input data space, results from previous methods for measuring of technical systems automatically. Often, however, there is a desire in practice to obtain a specific set of measurement points near the operating limit. In general, it makes sense to know the course of the operating limit as accurately as possible. The measuring points at the operating limit make it possible to verify this limit and to increase the accuracy of a corresponding classification model for the selection of measuring points as lying within or outside the permissible operating range. With the help of selected measuring points, a modeling of the technical system and the parameterization of setpoint characteristic maps can be carried out in an improved manner, since the optimal operating points are often close to the limit of the permissible operating range.

It is therefore desirable to provide a way to perform a survey of a technical system exclusively with measurement points in the vicinity of an estimated limit of the operating limit.

Disclosure of the invention

According to the invention, a method for measuring a technical system with a set of measuring points according to claim 1 and the corresponding device according to the independent claim are provided.

Further embodiments are provided in the dependent claims.

• – Bestimmen von Konfidenzgrößen für eine Menge von bereitgestellten Messpunkten anhand eines vorgegebenen Klassifikationsmodells, wobei die Konfidenzgröße an einem Messpunkt eine modellierte Zulässigkeit eines Betriebspunktes angibt, der sich durch das Betreiben des technischen Systems an dem Messpunkt ergibt;
• – Auswählen eines Messpunkts aus der Menge von Messpunkten, wobei der ausgewählte Messpunkt eine zugeordnete Konfidenzgröße aufweist, die eine vorgegebene Nähe zu einer Betriebsgrenze angibt, die eine Grenze zwischen zulässigen und nicht-zulässigen Betriebszuständen angibt;
• – Durchführen eines Vermessens des technischen Systems an dem ausgewählten Messpunkt, um die mindestens eine Ausgangsgröße zu bestimmen; und
• – Aktualisieren des Klassifikationsmodells mit dem Messpunkt und einer von der mindestens einen Ausgangsgröße abhängigen Angabe, die angibt, ob der durch die Vermessung des ausgewählten Messpunkts eingestellte Betriebspunkt des technischen Systems zulässig ist.

According to a first aspect, a method for measuring a technical system, in particular an internal combustion engine for a motor vehicle, is provided for the creation of a system model of the technical system, wherein the surveying is carried out with a number of measuring points in order to obtain in each case a value of at least one output variable , The method comprises the following steps:

• Determining confidence quantities for a set of provided measurement points based on a predetermined classification model, wherein the confidence quantity at a measurement point indicates a modeled permissibility of an operating point resulting from the operation of the technical system at the measurement point;
• Selecting a measurement point from the set of measurement points, the selected measurement point having an associated confidence quantity indicative of a predetermined proximity to an operational boundary indicating a boundary between allowed and non-permitted operating conditions;
• - performing a survey of the technical system at the selected measurement point to determine the at least one output variable; and
• Updating the classification model with the measuring point and an indication dependent on the at least one output variable, which indicates whether the operating point of the technical system set by the measurement of the selected measuring point is admissible.

An idea of the above method is measuring points for a technical system to be provided at or near an operating limit between allowable and non-permissible operating points of the technical system. This is achieved in that after finding a measuring point close to the operating limit, further measuring points are selected from a set of measuring points that are provided as far as possible from the measured measuring points and within a range around a modeled course of the operating limit or in the vicinity the operating limit lie. Thus, measuring points can be provided and measured, which run along the operating boundary or in an area near the operating boundary.

The fact that a measuring point is close to the operating limit can be determined on the basis of a confidence quantity. The confidence quantity is determined as a model prediction or as a prediction according to a classification model created with the previously measured measurement points and gives a certainty of the operation of the technical system when the corresponding measurement point is applied to the technical system or a degree of admissibility of the operating point set by the measurement points of the technical system. An estimate of how close a measuring point is to the operating limit can be made on the basis of the confidence level, eg. By checking whether the confidence quantity lies within a predetermined value range.

As a result, the above method provides measurement points and associated output values of one or more output variables along the operating boundary, whereby the boundary curve is known as accurately as possible and the actual course of the operating boundary can be modeled faster and more accurately with fewer measurement points. Nominal value maps can be parameterized with the aid of the measured measuring points, since optimum operating points are often close to the operating limit.

Furthermore, the measuring point can be selected from the set of measuring points provided as the measuring point which has the predetermined proximity to the operating limit and has a minimum distance, in particular the greatest possible distance, from the measuring points already measured.

According to an embodiment, the method for measuring a number of measurement points can be performed repeatedly and the classification model updated with each measured measurement point, wherein the set of measurement points provided is renewed after the measurement of each measurement point or after the measurement of a predetermined number of measurement points.

It can be provided that the method for measuring a number of measuring points is repeatedly performed and the classification model is updated only after a predetermined number of measured measuring points.

In particular, the amount of the measuring points provided can be randomly distributed and, in particular, space-filling determined for an input data space of the measuring points.

It can be provided that the classification model is created by a k-nearest neighbor method, a variable kernel density estimation method, an SVM method or a Gaussian process classification algorithm.

According to one embodiment, the classification model can be designed so that the predetermined proximity to the operating limit can be defined by a value range of the confidence quantity.

According to a further aspect, an apparatus, in particular a computing unit, is provided, which is designed to carry out the above method.

Brief description of the drawings

Embodiments are explained below with reference to the accompanying drawings. Show it:

1 a representation of a test system for measuring a technical system;

2 a flowchart illustrating a method for measuring a technical system with selected measuring points;

3 a representation of a two-dimensional input data space with measuring points and a course of an operating limit.

Description of embodiments

zusammengefasst sind und somit einen Messpunkt x bilden. Zudem gilt für jede der d Eingangsgrößen ein zulässiger Wertebereich. Weiterhin resultiert die Ansteuerung des technischen Systems 2 in einer oder mehreren Ausgangsgrößen y, die an den Messpunkten X gemessen werden. 1 shows a schematic representation of a test or testing system 1 for measuring a technical system 2 is trained. A technical system 2 For example, it may be an internal combustion engine of a motor vehicle or a subsystem thereof. A measuring unit 3 controls the technical system 2 with a sequence of measuring points X at which to specific operating points of the technical system 2 to lead. However, the measuring points X usually comprise a number d of a plurality of input variables which are present in an input variable vector x ε

are combined and thus form a measuring point x. In addition, an admissible one applies for each of the d input variables Value range. Furthermore, the control of the technical system results 2 in one or more output quantities y, which are measured at the measuring points X.

As a rule, to complete the measurement of the technical system 2 the measuring points X are varied over a large range within the permissible value ranges in order to achieve the most space-occupying possible occupancy of the input data space by the measuring points.

zu den Eingangspunkten X ∊

The measuring points together with the corresponding values of the output quantity form data points D _i = (y, X) with the outputs y ε

to the entry points X ε

Furthermore, permissible value ranges also exist for at least some of the output quantities y, so that an operating limit can also be defined by one or more of the output variables.

In 2 FIG. 3 is a flow chart illustrating a method of measuring a technical system with measurement points near an operating limit.

In step S1, a set of randomly distributed measuring points is provided, which occupy the input data space, which may be defined by the permitted value ranges of the input variables, as space-filling as possible. For generating the set of measurement points x _k ∈ X _k , a quasi-random distribution algorithm can be used.

In step S2, a confidence quantity φ (x _k ) corresponding to a classification model is subsequently determined for each measurement point from the provided quantity of measurement points. The continuous confidence quantity needed for the classification model describes the probability of belonging to a permitted or inadmissible operating range, which the classifier predicts for a point from the set of measurement points. The classification model is created during the measurement process. The classification model, which has not yet been trained with measurement points initially, yields, as a confidence variable for each of the measurement points of the provided set of measurement points, a neutral value of the confidence quantity without informative value with respect to a position of the measurement point with respect to an operation boundary. For example, the confidence quantity may assume a value between 0 and 1, where 0 indicates a position of the measuring point outside the operating range and 1 a position of the measuring point within the operating range. Initially, a confidence level of 0.5 can be assigned to all measurement points of the provided set of measurement points.

The operating limit indicates a course of a boundary between a range of operating points permissible according to predetermined admissibility conditions and non-permissible operating points. The conditions of admissibility may exclude areas of operation in which the technical system assumes an inadmissible, dangerous to the system or the environment, hazardous and / or unstable operating conditions. The operating limit may be a limit of a range determined by a respective range of values of the input quantities and resulting outputs.

Alternatively, initially an already pre-trained classification model can be provided, which is specified in the course of measuring further measurement points as described below.

The classification model provides the confidence quantity as a model size for each measurement point as a measure of the actual or predicted allowance of the application of that measurement point to the technical system. Ie. the confidence quantity indicates to what extent (with what probability) it is to be expected according to the classification model or can be predicted that when the measuring point is applied to the technical system, a permissible operating state is accepted according to the admissibility conditions or an admissible or permitted system response occurs. The admissibility of the measuring point is thus determined by the fact that the system response of the technical system complies with predetermined conditions, eg. B. with regard to a settability of a stationary operating point (no oscillation), with respect to value limits for state variables or the like.

In step S3, a set of candidate points is selected from the provided set of measurement points. The candidate points correspond to those measurement points for which a predetermined condition for the confidence quantity applies. The predetermined condition defines confidence quantities indicating that the measurement point is within an environmental range of the operating limit. For example, the predetermined condition may indicate that the confidence quantity is in a range between a minimum φ _min and a maximum value φ _max , for example, in a range between φ _min = 0.4 and φ _max = 0.6, for example, a confidence quantity with a value of 0.5 defines a position of a measuring point exactly at the operating limit. The candidate points are defined X _e = {x _k ∈ X _k : φ _min <φ (x _k ) <φ _max }

The minimum value and the maximum value of the confidence quantity φ _min , φ _max can be _calculated in Essentially arbitrary. However, the range of values for the confidence quantity should be chosen to be sufficiently small, in order in particular to select those measuring points which are arranged very close to the operating limit.

From the candidate points now selected, one of the candidate points x _{opt is} selected in step S4. In this case, the candidate point x _{opt can be selected} with the maximum nearest neighbor distance to all previously measured points X _m :

This makes it possible for the successively measured measuring points to move as close as possible to the operating limit. To calculate the maximum nearest-neighbor distance, the distances between each of the candidate points x ∈ X _e and the already measured points x _m ∈ X _{m are} calculated. For each candidate point, the minimum distance is determined, and then the candidate point with the largest minimum distance is selected.

In step S5, the selected candidate point x _opt thus found is set on the technical system and thus measured. The values of the corresponding assigned output variables are recorded.

In step S6, the classifier is trained with the measurement points and the output quantity.

Subsequently, it is checked in step S7 whether an abort condition is fulfilled. This can be, for example, a predetermined number of measuring points to be measured, a predetermined total measuring time or a quality criterion for the operating limit determined by the classification model. If the abort condition is fulfilled, the method is ended, otherwise the process returns to step S1.

Alternatively, it is also possible to jump to step S2, so that the amount of measurement points provided does not have to be determined again.

According to one embodiment, provision may be made for the classification model not to be updated or trained according to step S6 after each individual measurement with the selected candidate point x _opt , but first for a specific set of measurements to be carried out with successively selected candidate points x _opt and only then for the classifier to be trained becomes. So u. U. be provided that all candidate points x _{opt are} selected without updating the classifier.

In 3 is an example of a two-dimensional input data space with the input variables E1, E2 shown. In the first pass, only the input data space bounded by the dashed horizontal and vertical lines L1, L2, L3, L4 marking boundaries of value ranges of inputs is known. Furthermore, in the course of carrying out the method, the measured measuring points are represented as crosses. The minimum value φ _{min of} the confidence quantity and the maximum value φ _{max of} the confidence quantity are indicated by the contour lines (shown in dashed lines) and the operating limit is indicated as a solid contour line with, for example, φ = 0.5. The selected candidate points are shown as the bold points within the dashed contour lines. The next measurement point to be measured from this set, which has the largest distance to the previously measured measuring points, is shown encircled.

As a classification algorithm for the creation of the classification model, various algorithms from the field of machine learning can be used. The classification algorithm is preferably selected such that it can already be used with a large number of input variables, in particular more than five, even with a very small number of measuring points. Furthermore, the classification algorithm should after measuring a correspondingly next selected measurement point within a short time, i. H. can be updated with little computational effort. Furthermore, the classifier should provide a continuous confidence quantity, which in particular can assume a value range between 0 and 1. In this case, "0" can indicate a non-permissible measuring point and "1" an admissible measuring point.

Possible classification algorithms may include a k-nearest neighbor method, a variable kernel density estimation method, a support vector machines (SVM), a gaussian process classification algorithm, and the like. These classification algorithms make it possible, based on a measuring point and an indication of whether the concern of the measuring point has led to the technical system to a permissible or impermissible operating state of the technical system to be specified.

In an internal combustion engine as a technical system that is under scrutiny 1 is to be measured, a permissibility or inadmissibility of a measuring point can be determined by the operability of the internal combustion engine. Further criteria may be fuel consumption, pollutant emissions or the like. The corresponding degree of admissibility can be obtained by evaluating the output variable determined by measuring a measuring point.

Claims (10)

Method for measuring a technical system ( 2 ), in particular an internal combustion engine for a motor vehicle, for the creation of a system model of the technical system ( 2 ), wherein the measuring is performed with a number of measuring points (X) in order to obtain in each case a value of at least one output variable (y), comprising the following steps: - determining (S2) confidence quantities for a set of provided measuring points on the basis of a predetermined classification model wherein the confidence quantity at a measuring point indicates a modeled permissibility of an operating point resulting from the operation of the technical system at the measuring point; - selecting (S4) at least one measurement point from the set of measurement points, the selected measurement point having an associated confidence quantity indicating a predetermined proximity to an operational boundary indicating a boundary between allowable and non-permitted operating conditions; Performing (S5) a survey of the technical system ( 2 ) at the selected measurement point to determine the at least one output variable; and - updating (S6) the classification model with the measuring point and an indication dependent on the at least one output variable, which indicates whether the operating point of the technical system set by the measurement of the selected measuring point is admissible. The method of claim 1, wherein the measuring point is selected from the set of provided measuring points as the measuring point, which has the predetermined proximity to the operating limit and a minimum distance, in particular a maximum distance, to the already measured measuring points. The method of claim 1 or 2, wherein the method for measuring a number of measuring points is repeatedly performed and the classification model is updated with each measured measuring point, wherein the amount of the provided measuring points renewed after the measurement of each measuring point or after the measurement of a predetermined number of measuring points becomes. Method according to one of claims 1 to 3, wherein the method for measuring a number of measuring points is repeatedly performed and the classification model is updated after a predetermined number of measured measuring points. Method according to one of claims 1 to 4, wherein the amount of the provided measuring points distributed randomly and in particular space filling for an input data space of the measuring points is determined. Method according to one of claims 1 to 5, wherein the classification model by a k-nearest neighbor method, a variable kernel density estimation method, an SVM method or a Gaussian process classification algorithm is created. Method according to one of claims 1 to 6, wherein the classification model is designed so that the predetermined proximity to the operating limit is defined by a range of values of the confidence quantity. Device, in particular a computing unit, which is designed to carry out the method according to one of Claims 1 to 7. Computer program, which is set up to carry out all the steps of a method according to one of Claims 1 to 7. A machine-readable storage medium on which a computer program according to claim 9 is stored.

Images