DE102012104885B4 - Method for the error-free operation of a production machine - Google Patents

Abstract

Method for operating a manufacturing machine in which, during ongoing production, process values from sensors assigned to the manufacturing machine are recorded with the aid of a process monitoring unit and a check is made for each workpiece that is manufactured or processed by means of the manufacturing machine whether a process point defined by the process values is within a processing area ( B), which excludes all spatial points within a value space (Y) whose process values lead to reject parts, characterized in that for this purpose it is checked whether the spatial point is in a processing window (V), preferably at least largely completely within the processing area (B) ), the limits of which have been determined in advance with the aid of a classification process and where, as part of pre-production, the processing window (V) is determined by first assigning its shot number uniquely to each workpiece and adding the W part is identified as a good part or as a reject part with a reason for rejecting or rejecting the cause, according to which the processing window (V) is determined using V: = V0- S - A with the aid of a classification of the process points belonging to good parts and the process points belonging to reject parts, with V0 being a pre-determined subset the process values is an output processing window, A is the union of scrap quantities Ajist, the boundary hypersurface to the good parts is determined with a classification method and S is a safety range, preferably with a width between 0 and a target quality, most preferably between 0 and 6σ, to the scrap quantities and that as a classification method the support vector machines method is used with the hinge loss or the logistic loss and that a target process point is defined, with a process control intervening in the ongoing production of a workpiece with the aim of reaching the target process point en if a deviation from the target process point is to be expected on the basis of the measured values collected by the process monitoring unit

Description

The present invention relates to a method for operating a manufacturing machine, in which process values from sensors assigned to the manufacturing machine are recorded during ongoing production with the aid of a process monitoring unit, and for each workpiece that is manufactured or processed by the manufacturing machine, it is checked whether a workpiece defined by the process values Process point lies within a processing area which excludes all spatial points within a value range whose process values lead to rejects.

Such a procedure is already from the DE 102010037552 A1 previously known. This document teaches that you can eliminate defects and rejects in manufacturing machines with the help of a process computer. If the process values are adequately measured, this is achieved using a 5-stage procedure, which consists of the steps of definition, measurement, classification, improvement and prediction. With this method, the setting data are changed after an error has occurred.

This method can only be used in traceable production with a continuous improvement process. The usual production is not traceable and does not have a continuous improvement process, but follows a sequence of samples, series start-up, series production and spare parts production.

In 1947, Norbert Wiener had a basic idea which the application made use of, among other things. Norbert Wiener called the process "cybernetic control". He used the term cybernetic control after "Cybernetes", a Greek helmsman of a sailing ship who has to prevent the boat from crashing on the cliffs in wind and weather. Cybernetes also means “guide”. The pilot has knowledge of the routes in the sea. Just as the pilot ensures that a ship does not run aground, the cybernetic control system should ensure that the measured process data remain in the processing area and that good parts are created.

The research on cybernetic control remained unsuccessful because it was not possible to precisely determine the processing area. By processing area we mean the set of all process points at which good parts are created. There is good reason to believe that it is impossible to accurately determine the processing area in typical manufacturing machines, such as an injection molding machine.

In injection molding, for example, the quality is determined by approx. 80 machine sizes and 2 process curves, the cavity pressure curve and the cavity temperature curve. The number of extreme points increases exponentially with each dimension. In order to be able to determine the limits of the processing area exactly, you have to measure extremely precisely. With 80 measured variables, however, the error rate increases; at least one of these 80 measuring devices generally does not measure accurately. The exact determination of the processing area is therefore considered impossible.

There was a second flaw in cybernetic research. The classification of the processing area was seen as a mapping f: Y -> Z of the process values in Z, the 2-element quantity {0.1} or scrap and good. However, the committee has an internal structure, the committee reasons or the causes of the committee. The causes of rejects are easy to classify. The situation is well comparable to medicine, a doctor can find it difficult to diagnose whether someone is healthy, but a diagnosis for a specific disease is routine.

In the book Hastie T .; Tibshirani R .; Friedmann J .: The Elements of Statistical Learning; Springer 2009 carried out a mathematical analysis of the cognitive ability in high-dimensional spaces. An important finding of these professors from Stanford was that mathematical procedures based on the concept of proximity no longer work in high-dimensional spaces, but that simple concepts such as separating hyperplanes also work in high-dimensional space. This means that the processing area cannot be determined, but locally separating hyperplane or separating manifolds can be determined.

The DE 198 01 881 C1 relates to an injection molding process. A tolerance band is determined which enables the reject rate to be reduced to the actual bad parts even in the event of strong fluctuations in the value of the parameter to be monitored.

The DE 10 2004 063 388 A1 relates to the adaptive classification of workpieces (e.g. vibration analysis-based material testing). A neural network (NN) with classifiers is used for classification, which classify according to quality characteristics according to good or bad.

The DE 10 2010 037 552 A1 relates to error handling on production machines due to small variations in the production process.

The WO 2012/048 700 A1 describes a process computer optimized operation of a manufacturing machine. In particular, there should be a division into good and reject. Based on the Six Sigma method, which also includes technical specifications, computational methods are used to determine relationships between technical process parameters.

Against this background, the present invention is based on the object of creating a method for error-free operation of a manufacturing machine which works as a cybernetic control.

This problem is solved by the attached claim 1. Advantageous embodiments of the invention are mentioned in the subclaims and in the following description with further alternatives.

From the above considerations, the idea arose that with a reasonable amount V , which is to be called the processing window, and which only consists of points at which good parts arise, a cybernetic control could be built. V is also exactly the amount that you are looking for for process monitoring. The key to solving cybernetic control is process monitoring technology, is the processing window. Such a processing window enables process monitoring and cybernetic control of the manufacturing machine.

Fitters in plastics processing, material suppliers and the IKV Aachen have developed the technology of a processing window, a subset of the processing area that is easier to describe.

The concept of the processing window is also described in the book Johannaber, F .; Michaeli, W .: Injection Molding Manual. Hanser, Munich 2004. In this book, Prof. Michaeli assumes that the processing window is determined manually. Prof. Michaeli defines: "A processing window ... includes areas between values at which plastics can be processed into usable injection-molded parts."

The number of vertices of a polyhedron increases exponentially with the dimension, so a manual empirical determination is not possible.

The main approaches to automatically improve quality either use sensors, e.g. after Gruber, J .: Process control in thermoplastic injection molding based on cavity pressure. Dissertation RWTH Aachen 2005 - ISBN: 3-86130-843-6; or statistical test planning, eg according to Gierth, MM: Methods and aids for process-related quality assurance in the injection molding of thermoplastics. Dissertation RWTH Aachen 1992 - ISBN: 3-86073-073-8 .

Quality control theory regards variations as the cause of rejects and tries to limit the variations with test plans. This is also possible with modern injection molding machines. Test plans set an upper and lower limit for each measured value, so one tries to use a cuboid instead of the non-determinable processing area. However, a cuboid is poorly suited for the subject of quality. Everyone knows that if a steak is cooked too hot, the roasting time has to be shorter. If the roasting time is longer, the temperature must be lower. The processing area is inclined.

It is similar with injection molding, where there are, for example, upper and lower limits for processing temperature, dwell time and screw speed. If two of these readings rise, the third must fall, otherwise scrap occurs. Here, too, the processing area is inclined. The approach of a test plan simplifies too much, instead one needs a properly described processing window for process monitoring in plastics processing.

The only successful way to date to eliminate errors comprehensively is the so-called Six Sigma DMAIC process. The abbreviation “DMAIC” stands for “Define, Measure, Analyze, Improve, Control”.

In the first step, the definition, the reasons for errors and rejects are defined from the customer's point of view. The number of errors, sorted by reason, is recorded manually and statistically evaluated. It is important to eliminate the errors that occur frequently.

During the measurement, all relevant data that could possibly cause the error are recorded at selected points.

The recorded measured values are statistically examined as part of the analysis by a specialist in order to establish a connection between the occurrence of the errors and variations in the measured values. The improvement process uses the analysis results to determine a better target process point.

The SIX SIGMA procedure was implemented in DE 10 2010 037 552 A1 mathematized and automated. In the following we call the procedure “Automated SIX SIGMA” or A6σ for short.

wobei Y die Menge der Prozesswerte ist und Z die Menge mit den Zahlen 0 und 1, d h. {0,1}.As with SIX SIGMA, reasons for errors are also defined and measured with A6σ. A classification is then made for each error reason recorded. A classification is a mapping at A6σ $$f:Y\to Z$$

where Y is the set of process values and Z is the set with the numbers 0 and 1, i.e. {0.1}.

0 here symbolizes rejects, 1 symbolizes good parts. But one should not think that the archetype of 1 meant the processing area. The meaning is something completely different.

The easiest way to understand the meaning is to select linear SVM as the classification method, one of the best methods. In the case of linear SVM, the mapping is determined by a separating hyperplane. This separating hyperplane is the optimal separating hyperplane between good parts and parts with the examined reason for rejects. This hyperplane can best be imagined as the tangential plane of the processing area, the tangential plane being spanned at the point where the scrap arises. The separation between good and reject is therefore determined locally.

The tangent plane has a normal vector that points into the interior of the processing area. In analogy to the tangential plane, A6σ uses the normal vector of the separating hyperplane to get inside the processing area. A6σ determines a new target process point by shifting the old target process point to the distance 6σ of the separating hyperplane with the help of the normal vector. As usual, σ is the statistical standard deviation of the process data. In the last step, you are looking for a setting data record that can be used for production in the vicinity of the target process point.

SIX SIGMA and A6σ do not deal with the question of cybernetic control and the processing area, but are limited to examining the limits of the processing area locally and using these results to arrive at a better target process point and a better setting.

• Startverarbeitungsfenster V₀ bestimmen Ausschussgründe definieren
• Messen
• Ausschussbereich A klassifizieren
• Sicherheitsbereich S bestimmen

So it is a big surprise that there is a simple procedure to determine the processing window in 5 steps:

• Start processing window V ₀ Determine Define Rejection Reasons
• measure up
• Committee area A. classify
• security area S. determine

wobei
V₀ als vorab bestimmte Teilmenge der Prozesswerte ein Ausgangsverarbeitungsfenster ist,
A die Vereinigung von Ausschussmengen A_j ist, deren Grenzhyperfläche zu den guten Teilen mit einem Klassifikationsverfahren bestimmt wird,
S ein Sicherheitsbereich zu den Ausschussmengen ist.The processing window V now determined as $$\text{V}:={\text{V}}_0-\text{S.}-\text{A.}$$

in which
V ₀ as a pre-determined subset of the process values is an output processing window,
A. is the union of scrap _{sets A j} , whose boundary hypersurface to the good parts is determined using a classification method,
S. is a safety area to the scrap amounts.

The method described above is explained in more detail below with the aid of an exemplary embodiment.

• 1 einen in einen Ausschussbereich und einen Verarbeitungsbereich geteilten Prozesswertebereich,
• 2 den Wertebereich gemäß 1 mit einem dem Verarbeitungsbereich zugeordneten Startverarbeitungsfenster,
• 3 den Wertebereich gemäß 1 mit den trennenden Hyperflächen (hier Linien) zu den Ausschussbereichen,
• 4 den Wertebereich gemäß 3 mit einem Sicherheitsbereich und dem Verarbeitungsfenster, sowie
• 5 ein Detail des Verarbeitungsfensters gemäß 4 zur Veranschaulichung der Bestimmung eines adaptiven Sollprozesspunkts.

Show it

• 1 a process value area divided into a reject area and a processing area,
• 2 the range of values according to 1 with a start processing window assigned to the processing area,
• 3 the range of values according to 1 with the separating hypersurfaces (here lines) to the reject areas,
• 4th the range of values according to 3 with a security area and the processing window, as well
• 5 a detail of the processing window according to FIG 4th to illustrate the determination of an adaptive target process point.

1 Now visualizes the invention using the example of plastics processing in an injection molding process, with only processing temperature T and length of stay t are shown as dimensions. The processing area B. , the number of all process points at which good parts are produced cannot be determined, but the material manufacturer's documents contain information such as: Processing temperature: 220 to 260 ° C Dwell time 4 to 6 min. Peripheral screw speed 0.1 to 0.3 m / s

Such information is now stored in material databases and serves as the starting point for a sample. When the specified processing temperature is exceeded T or too long a dwell time t you get into a decomposition area Az, in which the material of the workpiece is burned. If the temperature falls below the processing temperature, there is a solidification area A _e achieved in which the material is no longer sufficiently flowable for processing in the injection molding machine considered as an example. If the specified dwell times are not reached t becomes an inhomogeneity area A _i reached.

2 complements the processing area B. therefore initially a start processing window V ₀ . Under a startup processing window V ₀ we mean a closed (possibly spurious) subset of a sub-space of the process space that contains at least one process point at which good parts are created.

A start processing window can be determined automatically. In plastics processing, for example, linear SVM can be used to classify all known reasons for rejects across all parts of a material that have already been recorded. In machining, you can start with a target process point with a cuboid around the target process point. However, such a start processing window becomes the boundary hypersurfaces of the processing area B. not correctly map the causes of rejects.

In the process, it is first necessary to define when an error occurs which makes a workpiece a reject part and when a workpiece is to be regarded as a good part. A distinction should be made between the reasons for rejects and, if necessary, the reject locations.

When sampling or the start of series production, the setter can assign a reject message to the associated shot number. A test database can thus be created. The prerequisite for this is that the machine is adequately equipped with sensors and that the measured process values are saved shot by shot.

i

Schussnummer bzw. Losnummer

y_i

Vektor der Prozesswerte des i-ten Schuss

Y

Raum der Prozesswerte

j

Nummer des Ausschussgrunds

A_j

j-ter Ausschussbereich

Z

Menge mit den Zahlen 0,1

f_j

charakteristische Funktion von A_j

V₀

Startverarbeitungsfenster

S_j

Sicherheitsbereich zum j-ten Ausschussgrund

V

Verarbeitungsfenster

The following notation will be used in the following:

i

Shot number or lot number

y _i

Vector of the process values of the i-th shot

Y

Process values room

j

Number of the rejection reason

A _j

jth committee area

Z

Quantity with the numbers 0.1

f _j

characteristic function of A _j

V ₀

Start processing window

S _j

Safety area for the jth reason for rejects

V

Processing window

A test database is created during sampling and the start of series production $$\mathrm{D.}=\left\{y_1,y_2,...,y_n\right\}$$

The setter also knows whether the part is bad or good.

The characteristic function of the reject area A _j has the property $$y_i\in {\mathrm{A.}}_j\iff f_j\left(y_i\right)=0$$

The task of the classification is _{to approximate the functions f j} as well as possible. Mathematically, the deviation is measured with a loss function $$\mathrm{L.}:Y\times Z\times \mathrm{R.}\to \mathrm{R.}$$

There are many good classification methods, but SVM is characterized by a special mathematical elegance. Good classification methods provide damping that mitigates the negative effects of measurement errors. SVM has the additional property that SVM delivers explainable results. An outer processing window determined with linear SVM V _a is in 3 shown.

The parameterization of the SVM calculation takes place in such a way that all reject parts as well as some good parts can lie outside the area separated from the hyperplanes.

wobei nunmehr A für eine Matrix und $\overrightarrow{b}$

für einen konstanten Vektor steht.In this case, a convex processing polyhedron is created as the outer processing window. The polyhedron equation applies $$\mathrm{A.}\cdot \overrightarrow{x}<=\overrightarrow{b}$$

where now A. for a matrix and $\overrightarrow{b}$

stands for a constant vector.

Scrap results from variations in process values, so that the process values are outside the processing window due to the variation. Therefore you have to move the target process point far enough into the processing window. Any method that improves the existing target process item (or target process item) by adding a target process item further inside the processing area is permissible. A simple procedure is recommended for cybernetic control: vertical shifting to a SIX SIGMA distance.

A simple way to quality in parts and alternating flow production can be described by an equation: $${\overrightarrow{y}}_{sOll}:={\overrightarrow{y}}_{ist}+\left({\sigma }_{sOll}-{\sigma }_{ist}\right)\cdot {\overrightarrow{\nu }}_{\text{exp}lain}$$

The vectors mentioned are vectors of the process space, i.e. n-tuples of process values.

handelt es sich um den Sollprozesswertvektor mit Sollqualität, bei dem Vektor ${\overrightarrow{y}}_{ist}$

um den Istprozessvektor als Mittelwert. σ_soll ist ferner die Sollqualität, angegeben in Standardabweichung, bei SIX SIGMA natürlich 6, und σ_ist die Istqualität, ebenfalls angegeben in Standardabweichung. Der Explainvektor ${\overrightarrow{\nu }}_{\text{exp}lain}$

hat 3 Bedeutungen gleichzeitig : Er steht für die Richtung größter Qualitätsänderung, den Normalenvektor auf der Grenzfläche des Verarbeitungsfensters und das Ergebnis von Explain bei einer SVM-Klassifikation. Eine entsprechende Gleichung gilt nicht nur für diskrete Prozesswerte, sondern auch für Prozessfunktionen. Auf den ersten Blick denkt man, dass der Normalenvektor auf der Grenzfläche des Verarbeitungsfensters immer die beste Lösung ist. Leider ist das nicht der Fall. Ein Verarbeitungsfenster kann wie ein Canyon aussehen, da ist es manchmal besser, schräg zu gehen. Auch kann es sein, dass der Canyon nicht 12σ breit ist, so dass die Schrittweite $$\left({\sigma }_{soll}-{\sigma }_{ist}\right)$$

nicht möglich ist. In diesem Fall ergibt sich eine verallgemeinerte Qualitätsgleichung $${\overrightarrow{y}}_{soll}:={\overrightarrow{y}}_{ist}+\alpha \cdot \left({\overrightarrow{\nu }}_{\text{exp}lain}+{\overrightarrow{\nu }}_{translation}\right)$$

wobei $$0<\alpha <\left({\sigma }_{soll}-{\sigma }_{ist}\right)$$

und ${\overrightarrow{\nu }}_{translation}$

eine zum Explainvektor senkrechte Translation.With the vector ${\overrightarrow{y}}_{sOll}$

it is the setpoint process value vector with setpoint quality, the vector ${\overrightarrow{y}}_{ist}$

around the actual process vector as a mean value. σ _{is to} further the desired quality level, expressed in standard deviation, SIX SIGMA course 6, and σ _is the Istqualität also indicated in standard deviation. The explain vector ${\overrightarrow{\nu }}_{\text{exp}lain}$

has 3 meanings at the same time: It stands for the direction of the greatest change in quality, the normal vector on the interface of the processing window and the result of explain in an SVM classification. A corresponding equation applies not only to discrete process values, but also to process functions. At first glance, one thinks that the normal vector on the interface of the processing window is always the best solution. Unfortunately, this is not the case. A processing window can look like a canyon, so sometimes it is better to go at an angle. It can also be that the canyon is not 12σ wide, so the step size is $$\left({\sigma }_{sOll}-{\sigma }_{ist}\right)$$

not possible. In this case a generalized quality equation results $${\overrightarrow{y}}_{sOll}:={\overrightarrow{y}}_{ist}+\alpha \cdot \left({\overrightarrow{\nu }}_{\text{exp}lain}+{\overrightarrow{\nu }}_{translatiOn}\right)$$

in which $$0<\alpha <\left({\sigma }_{sOll}-{\sigma }_{ist}\right)$$

and ${\overrightarrow{\nu }}_{translatiOn}$

a translation perpendicular to the explain vector.

One way to do this can be human intervention or a technical calculation.

besitzt. Als Beschreibung für das Verarbeitungsfenster V ergibt sich damit: $$V:=V_0-S-A$$

wobei V₀ das Startverarbeitungsfenster, S der Sicherheitsbereich und A der Ausschussbereich ist.This means that the processing window must have an additional security area which is $$0<\alpha <\left({\sigma }_{sOll}-{\sigma }_{ist}\right)$$

owns. As a description for the processing window V the result is: $$V:=V_0-\mathrm{S.}-\mathrm{A.}$$

in which V ₀ the startup processing window, S. the security area and A. the committee area is.

The task of cybernetic control is then to keep the process within the processing window.

The task of process monitoring is then to only accept the articles produced in the processing window as permissible. An explanatory process monitoring is created with linear SVM.

An expression

wird auch Kennzahl genannt. Das heißt, mit Hilfe von errechneten Kennzahlen kann man Gutteile und Ausschuss unterscheiden. $${\overrightarrow{\nu }}_{\text{exp}lain}\cdot \overrightarrow{x}$$

is also called a key figure. That means, with the help of calculated key figures, one can differentiate between good parts and rejects.

In some cases, key figures can be used to shorten the computation time of the classification. The classification is not based on the full process data stock, but the process data stock is first reduced with the help of key figures. The classification calculation is then carried out on the reduced database.

The vertical on the separating hypersurface calculated by SVM is called EXPLAIN and the values of EXPLAIN explain the causes of rejects.

The explain vector is not just a mathematical number, it explains the cause of the reject. Process variables, i.e. dimensions that cause rejects, have large coefficients in the explain vector. The coefficients of process variables that do not cause rejects are zero or at least small.

wobei ferner b die Materialkonstante ist.For each cause of rejects, linear SVM provides a separating hyperplane with the associated equation. The hyperplane is determined with the help of the explain vector and the vector of the process values $${\overrightarrow{\nu }}_{\text{exp}lain}\cdot \overrightarrow{x}\Leftarrow b$$

where furthermore b is the material constant.

wird eingefügt, wie in 4 dargestellt.So it is now clear how the processing window V should look like an additional security area of the width $$0<\alpha <\left({\sigma }_{sOll}-{\sigma }_{ist}\right)$$

is inserted as in 4th shown.

wobei V₀ das Startverarbeitungsfenster ist, S der Sicherheitsbereich und A der Ausschussbereich ist.This results in the solution for the processing window $$V:=V_0-\mathrm{S.}-\mathrm{A.},$$

in which V ₀ the startup processing window is S. the security area and A. the committee area is.

das Optimum bestimmen.After the processing window V is determined, the question arises as to which process point in V is the most cost-effective. In general, this is the point in V with the shortest cycle time. More generally, we can use that V is a convex set, so we can for any convex cost function $$\text{K}:\text{V}\to \text{R.}$$

determine the optimum.

1. a) Vorhersage einer fehlerfreien Einstellung durch Regression
2. b) Vorhersage mit Regelungstechnik oder Modell
3. c) Vorhersage durch kybernetische Steuerung
4. d) Halbautomatische Steuerung, bei der ein Einrichter auf Basis der Erklärung des Fehlers die Führungsgrößen selber setzt.

Manufacturing machine setters usually use two steps to correct errors. First you look for an explanation of the error, then you eliminate the error. If the error has been explained by the automatic procedure and an error-free target process point has been determined, there are now several options for continuing:

1. a) Prediction of an error-free setting by regression
2. b) Prediction with control technology or a model
3. c) Prediction through cybernetic control
4. d) Semi-automatic control in which a setter sets the reference values himself based on the explanation of the error.

The regression method is particularly useful when the processing window is wide. When using regression, it is necessary that the setting data and process values associated with a workpiece are clearly assigned to it via the shot number. With the other methods, the relationship is calculated from the model or set manually.

There are many regression methods with which one can determine a function f: X -> Y that assigns the expected mean process point to each set of adjustments. One of the best practices in a sense is SVM. The selection of the procedure depends on the technical boundary conditions and the dimensions occurring in the problem. During assembly, the number of dimensions increases and you should reduce the number of dimensions, e.g. with prior technical knowledge, cascade SVM or a main component analysis.

wobei $${\overrightarrow{y}}_{soll}:={\overrightarrow{y}}_{ist}+\left({\sigma }_{soll}-{\sigma }_{ist}\right)\cdot {\overrightarrow{\nu }}_{\text{exp}lain}$$

Wenn man eine Funktion $$f:=X\to Y$$

bestimmt hat, die jedem Einstelldatensatz die erwarteten mittleren Prozesswerte zuordnet, so kann man nun einen Einstelldatensatz x _∈ X bestimmen, so dass $$\Vert f\left(x\right)-\mu \Vert$$

klein oder minimal wird. Dazu gibt es mehrere rauschunempfindliche Näherungsverfahren.In the previous step of the procedure, a target process point µ was determined. One way to get there is by definition $$\mu :={\overrightarrow{y}}_{sOll}$$

in which $${\overrightarrow{y}}_{sOll}:={\overrightarrow{y}}_{ist}+\left({\sigma }_{sOll}-{\sigma }_{ist}\right)\cdot {\overrightarrow{\nu }}_{\text{exp}lain}$$

If you have a function $$f:=X\to Y$$

has determined which assigns the expected average process values to each setting data record, one can now determine a setting _{data record} x ∈ X so that $$\Vert f\left(x\right)-\mu \Vert$$

becomes small or minimal. There are several approximation methods that are insensitive to noise.

In recent times, mathematical modeling has spread more and more in the field of control and regulation technology. If the manufacturing machine has a calculated control system or the relationship between reference variables, manipulated variables and measured variables can be calculated or estimated with sufficient accuracy, this model-based prediction of the function f: X -> Y can be used to determine the desired setting data set.

5 finally visualizes the task in the partial aspect of the invention for realizing a cybernetic control.

So far, we have been able to use the quality equation to determine an improved target process point and thus improve the machine setting. The task of cybernetics was, however, during ongoing operation in the processing area B. to produce.

An adaptive process vector P _adp to be determined. The simple determination is only possible because we are working with a simple processing polyhedron and not with an incredibly complex interface of the processing area B. .

In the picture there is an auxiliary line through the expected process vector P _adult drawn, perpendicular to the dividing line between reject and good parts, i.e. a straight line with the direction of the explain vector.

This straight line meets the connecting line between a target process vector with target quality P _should and P _adp at right angles at an auxiliary point. We can determine this auxiliary point with the following equation: $$Hilfspunkt:=P_{erw}+\left(\left(P_{sOll}-P_{erw}\right)\cdot {\overrightarrow{\nu }}_{\text{exp}lain}\right)\cdot {\overrightarrow{\nu }}_{\text{exp}lain}$$

The four points P _should , P _adp , P _adult and H span isosceles triangles. So the proportions of the sides are the same. So the following applies: $$\frac{\left|P_{adp}-P_{sOll}\right|}{\left|P_{erw}-P_{sOll}\right|}=\frac{\left|P_{erw}-P_{sOll}\right|}{\left|H-P_{sOll}\right|}$$

wobei $$H:=P_{erw}+\left(\left(P_{soll}-P_{erw}\right)\cdot {\overrightarrow{\nu }}_{\text{exp}lain}\right)\cdot {\overrightarrow{\nu }}_{\text{exp}lain}$$

This results in the equation for the adaptive process point $$P_{adp}:=P_{sOll}+\frac{{\left|P_{erw}-P_{sOll}\right|}^2}{{\left|H-P_{sOll}\right|}^2}\cdot \left(H-P_{sOll}\right)$$

in which $$H:=P_{erw}+\left(\left(P_{sOll}-P_{erw}\right)\cdot {\overrightarrow{\nu }}_{\text{exp}lain}\right)\cdot {\overrightarrow{\nu }}_{\text{exp}lain}$$

These simple calculations assumed that it was best to move the adaptive process point away from the reject area with a vertical line, i.e. in the direction of the explain vector. In many cases, however, it is better to go at an angle with the adaptive process point inside the processing window V to push.

In the cockpit of an airplane today there are many instruments and many options for the pilot to intervene. The pilot can use the knowledge of the flight route and flight altitude calculations to steer better himself than without this support.

A normal production machine could now be controlled in a similar manner, provided there are sufficient measuring instruments and the method described is implemented.

The setter of a production machine needs sufficient information about the measurements, also about the previous cycles, for the sampling and the start of series production. It must be possible to report the rejects on a shot-related basis.

The classification provides the setter with an explanation of the cause of rejects; the accuracy of the result varies depending on the classification used. The setter should be able to intervene here.

The method described now provides suggestions for the target process point and the setting data record. Just as a pilot can change a calculated route, the set process point should also be adaptable by the setter.

Large parts of the flight can be carried out by autopilot, similarly in series production the machine can be controlled with the "cybernetic control" and the "explanatory process monitoring".

The method described above can be used in primary molding machines, processing machines and assembly machines; this applies in particular to injection molding machines with cavity pressure measurement.

List of reference symbols

Y

Process value space

A.

Committee area

B.

Processing area

V ₀

Start processing window

V _a

outer processing window

V

Processing window

S.

security area

A _i

Inhomogeneity area

A _e

Freezing area

A _z

Decomposition area

P _should

Target process value vector with target quality

P _adult

expected process vector

P _adp

adaptive process point

T

Processing temperature

t

Length of stay

Claims (11)

Method for operating a manufacturing machine in which, during ongoing production, process values from sensors assigned to the manufacturing machine are recorded with the aid of a process monitoring unit and a check is made for each workpiece that is manufactured or processed by means of the manufacturing machine whether a process point defined by the process values is within a processing area ( B), which excludes all spatial points within a value space (Y) whose process values lead to reject parts, characterized in that for this purpose it is checked whether the spatial point is in a processing window (V), preferably at least largely completely within the processing area (B) ), the limits of which have been determined in advance with the aid of a classification process and where, as part of pre-production, the processing window (V) is determined by first assigning its shot number uniquely to each workpiece and that Workpiece is identified as a good part or as a reject part with a reason for rejecting or rejecting the cause, after which the processing window (V) is determined using V: = V ₀ - S - A _{using a classification of the process points belonging to good parts and the process points belonging to reject parts, where V 0} as A pre-determined subset of the process values is an output _{processing window, A is the combination of rejects A j} , the boundary hypersurface of which to the good parts is determined using a classification method, and S is a safety range, preferably with a width between 0 and a target quality, most preferably between 0 and 6σ to the Is waste quantities and that the support vector machines method with the hinge loss or the logistic loss is used as a classification method and that a target process point is defined, with a process control with the aim of reaching the target process point in the ongoing production of a plant tücks can intervene if a deviation from the target process point is to be expected on the basis of the measured values collected by the process monitoring unit Procedure according to Claim 1 , characterized in that either a processing window (V) or several nested processing windows (Va, V) are defined, with a workpiece whose process point lies outside of all processing windows (Va, V) being rejected as a reject part. Method according to one of the Claims 1 or 2 , characterized in that in the event of an intervention by the process control and / or a new target process point is determined in advance by first determining the position of a central process point in relation to the limits of the processing window (V) and then determining a target process point which has a greater distance of the limits or has shorter cycle times. Procedure according to Claim 3 , characterized in that the target process point as the end point of a target process vector ${\overrightarrow{y}}_{sOll}$

shall be determined from the sum of the mean process vector pointing to the mean process point ${\overrightarrow{y}}_{ist}$

and the product of a distance value σ and the sum of an explain vector ${\overrightarrow{v}}_{explain}$

an SVM classification and a translation vector vtranslation perpendicular to the explain vector ${\overrightarrow{\nu }}_{translatiOn}$

, where the distance value σ is chosen between zero and the difference between the target and the actual quality in standard deviation, that is $${\overrightarrow{y}}_{sOll}:={\overrightarrow{y}}_{ist}+\alpha \left({\overrightarrow{v}}_{sOll}+{\overrightarrow{v}}_{translatiOn}\right),0<\alpha <\left({\sigma }_{sOll}-{\sigma }_{ist}\right)$$

Procedure according to Claim 1 , characterized in that an optimized target process point is determined by calculating that point within the processing window which has the shortest cycle time. Method according to one of the Claims 1 to 5 , characterized in that a setting data record belonging to a workpiece and its process values are clearly assigned to this via the shot number. Procedure according to Claim 6 , characterized in that a function f: X -> Y is determined locally or globally using a regression or prognosis method, which assigns the expected mean process values to each setting data record, an optimal function f preferably being determined by means of the SVM method and, in particular, additionally a Dimension reduction, for example by a principal component analysis or cascade SVM, is carried out. Procedure according to Claim 7 , characterized in that an adjustment data set x X is determined such that || f (x) - µ || is small or minimal, so that with this setting data set x no previously classified scrap can occur with a certainty of 6, with µ denoting the target process point. Method according to one of the Claims 6 to 8th , characterized in that the setting data set is determined with the aid of a model or transfer function of a control or regulation system, which is known or is determined by parameter estimation. Method according to one of the preceding claims, characterized in that during a current cycle an adaptive process point and an adaptive setting are determined by means of a prediction in such a way that the adaptive process point lies further inside the processing window (V) than an expected process point. Method according to one of the preceding claims, characterized in that the production machine is a primary forming machine, processing machine or assembly machine.

Images