CN113910562A

CN113910562A - Method for optimizing and/or operating at least one production process

Info

Publication number: CN113910562A
Application number: CN202110776444.4A
Authority: CN
Inventors: G·皮尔韦恩; J·福格内德尔; F·J·基利安
Original assignee: Engel Austria GmbH
Current assignee: Engel Austria GmbH
Priority date: 2020-07-10
Filing date: 2021-07-09
Publication date: 2022-01-11
Also published as: AT524001B1; AT524001A1; DE102021117513A1

Abstract

Method for optimizing and/or operating at least one production process, which is carried out by at least one production machine in a production plant for manufacturing at least one product, with the following steps: (a) recording-by means of a data recording unit-at least one set value, and/or-at least one value of at least one process variable and/or at least one value of at least one reference variable; (b) determining, by a computing unit, at least one computation setting by means of at least one rule, and/or at least one electronic message, in particular in the form of at least one action recommendation; (c) deciding by a decision unit and/or an operator through at least one operator interface whether at least one calculated setting of step (b) should be taken, and/or at least one action recommendation of step (b) followed, at least one rule of step (b) being created by a learning unit by means of at least one machine learning method while using training data of a plurality of production devices and/or production machines.

Description

Method for optimizing and/or operating at least one production process

Technical Field

The invention relates to a method for optimizing and/or operating at least one production process having the features of the preamble of claim 1 and to a feedback method having the features of the preamble of claim 28. Furthermore, the invention relates to a production plant according to claim 32, having means for carrying out the method of claim 1 and/or the feedback method of claim 28. The invention also relates to a computer program product according to claim 33.

In the disclosure of the present application, the word "method" is used as a shorthand for a method for optimizing and/or operating at least one production process. In particular, the word "method" does not refer to a feedback method. Similarly, "rules" should not be understood as shorthand forms of "feedback rules".

The production process can be a molding process, in particular an injection molding process, for example. Production machines similarly follow this terminology. The production process may be run continuously or periodically.

Background

Control of a production process typically requires a large number of set variables to be input into the process variables. Since these large numbers of variables can greatly affect each other during the production process, a rational setup that produces a high quality product, saves resources, and does not damage the production equipment is often difficult and can only be done manually by experienced operators.

The prior art includes expert systems to support operators in setting up the production process. However, these are based on the limited amount of data that is manually collected and processed by experts. Therefore, a large amount of measurement data of the sensors on the production equipment currently existing is not used.

Furthermore, the prior art includes, in addition to expert systems, the use of a large amount of process data and/or setup data of the production machine for machine learning of rules which, for example, can predict the quality of the product. Thus, for example, it can be tested whether a certain setting of the production machine is meaningful. Further, machine-learned rules can be used to automatically determine supervision limits (or monitoring limits). This is achieved, for example, in US7216005B2 by using a neural network. AT519491a1 discloses optimization of a simulation-based process optimization system. However, the training of rules is typically associated with a single production machine.

Furthermore, it is state of the art to send large amounts of data about the behavior of operators and/or process data of mass production processes to a central computer network. These data are often analyzed manually at the factory and help improve new generation production machines and/or production equipment. However, the setting of the existing production plant is not affected here.

Disclosure of Invention

The object of the present invention is to avoid the disadvantages of the prior art. In particular, an improved method, an improved feedback method, an improved production device and an improved computer program product shall be provided.

According to the invention, this object is achieved by a method having the features of claim 1, a feedback method according to claim 28, a production device according to claim 32 and a computer program product according to claim 33. Preferred embodiments of the invention are given in the dependent claims.

The method according to the invention for optimizing and/or operating at least one production process which is carried out by at least one production machine in a production plant for producing at least one product, wherein the production plant has at least one operator interface for inputting set values of at least one set variable, preferably at least one system configuration value of at least one system configuration variable is present in a memory, and particularly preferably at least one set value and/or at least one system configuration value is present as a classification value, has the following steps:

(a) recording by means of a data recording unit

At least one setting value of at least one setting variable, and/or

At least one value of a process variable of at least one production process and/or at least one value of at least one reference variable, which is determined from at least one value of a process variable, wherein the values mentioned in this step preferably exist as classification values;

(b) is determined by a computing unit by means of at least one rule

-at least one calculated setting, and/or

-at least one electronic message, in particular in the form of at least one action suggestion,

wherein the input data of the rule comprises the values recorded in step (a) and/or at least one system configuration value of at least one system configuration variable and/or a classification value of the mentioned values;

(c) deciding by the decision unit and/or the operator via the at least one operator interface whether or not it should be

Using at least one calculated setting from step (b), and/or

-following the at least one action recommendation from step (b).

According to the invention, it is provided that the at least one rule in step (b) is created by a learning unit by means of at least one machine learning method using training data of the mass production device and/or of the mass production machine.

In addition, therefore, action recommendations, parameterisations, setting recommendations, etc. of the production plant and/or production machine can be learned by means of the data of the mass production machine and can be provided directly to the operator interface and/or the production plant and/or the production machine.

Mass production facility and/or production machine means at least two, but preferably more than 100, particularly preferably more than 1000 production facilities and/or production machines.

It may be provided that the training data for creating the at least one rule comprises the following values:

at least one setting value of at least one setting variable, and/or

At least one value of at least one process variable, and/or

At least one value of at least one reference variable, and/or

At least one system configuration value of at least one system configuration variable, and/or

-at least one classification of the above-mentioned values, and/or

-an identifier of at least one of the variables and/or categories.

The identifier of a variable and/or a category is a number and/or a string of characters univocally assigned to said variable or category.

In one embodiment, it is provided that at least one value of at least one control variable is recorded and at least one value of at least one reference variable, in particular a supervision limit, and/or an electronic message, in particular an action recommendation, is determined from this value by means of a rule.

In one embodiment, it is provided that at least one value of at least one system configuration variable, for example a material of a specific product, is used as an input value in order to determine at least one value of at least one control variable and/or at least one value of at least one reference variable and/or at least one electronic message by means of a rule.

In one embodiment, it is provided that if all values of the setting variables required for starting up the production process are not defined, at least one missing value is determined in step (b) as the calculated value of the setting variable.

In one embodiment, it is provided that at least one value of at least one reference variable of at least one process variable of at least one production process is recorded and the value of the set variable is continuously optimized.

In one embodiment, it is provided that the values of the reference variables in step (a) are derived from a production process which is parameterized with the setting values present in step (a), in particular in this case the production process is run as an intermediate step directly before step (a) for a defined time and/or a defined number of cycles.

In one embodiment, it is provided that, at least in the case of a decision by the operator on at least one operator interface in step (c), the at least one calculated setting from step (b), preferably also its classification, and/or the at least one electronic message is visualized.

In one embodiment, it is provided that, in the case of a positive decision by the decision unit and/or the operator,

-using at least one calculated setting value and/or implementing an action recommendation, and/or in case of a negative decision made by the decision unit and/or the operator,

-retaining at least one old setting, and/or

-entering at least one new setting value on at least one operator interface by the decision unit and/or the operator.

In one embodiment, it is provided that, in the event of a change of the at least one setting value by the decision unit, the cause is displayed in the form of an electronic message on the at least one operator interface.

In one embodiment, it is provided that the set variable of the at least one production process comprises a control variable of the process variable and/or a monitoring limit of the process variable and/or a variable which determines the type of monitoring.

In one embodiment, it is provided that the system configuration variables include the following variables, which describe the properties of:

-the production facility is adapted to produce,

-the at least one production machine, in particular the tool of the at least one production machine,

material of the product, and/or

-a client.

For example, the area where the production equipment is located and/or the industry where the customer is located may be system configuration variables.

In one embodiment, it is provided that the following units are or can be placed in a data connection via a computer network:

-at least one production machine for producing a product,

-at least one operator interface for the operator,

-a data recording unit for recording the data,

-a decision-making unit for determining, based on the received information,

-a computing unit for computing the time-varying frequency of the received signal,

-the learning unit is adapted to learn,

-a production facility and at least one further production facility.

In one embodiment, it is provided that the production system has a connection device which is or can be connected to a computer network by means of a data transmission means, wherein the computer network comprises in particular an internal computer network arranged inside the production system and an external computer network arranged outside the production system, wherein the external computer network in particular connects the production system to at least one further production system. The connecting device may be configured as an edge device.

In one embodiment, the data recording unit permanently or temporarily stores data provided to it in the production facility, in the production machine and/or in the computer network.

In one embodiment, the learning unit executes at least one machine learning method on at least one external computer network to which a plurality of production devices are or can be connected.

In one embodiment, it is provided that the learning unit carries out the at least one machine learning method on the at least one connection device, to which a plurality of production machines are or can be connected via an internal computer network.

In one embodiment, it is provided that the training data of the learning unit are collected from a plurality of production machines in the at least one production facility, wherein the production machine parts are of different types.

In one embodiment, it is provided that the learning unit determines at least one rule for the question, wherein preferably at least one supervised machine learning method is used, wherein the machine learning method particularly preferably learns from training data with an associated answer to the question.

The training data can be present in different data structures, for example as tables and/or as databases and/or as lists.

The specific assignment of answers to input data for questioning can be achieved, for example, by arranging the training data in a data structure, for example, a table with rows and columns.

In one embodiment, it is provided that the learning unit can pass at least one rule of the first question on to the second question, in particular by: in a machine learning method, a rule pre-trained with a first question is trained with training data of a second question. In this regard, for example, one term in the literature is "migratory learning".

In one embodiment, it is provided that at least one rule is created for at least one instance of a system configuration class, wherein the at least one rule is trained, in particular, on questions specific to the at least one instance of the system configuration class.

In one embodiment, it is provided that the learning unit determines the at least one rule without asking questions, wherein preferably at least one unsupervised machine learning method is used.

In one embodiment, the machine learning method uses one of the following methods:

-a decision tree, in which the decision tree is,

-the neural network is a network of nodes,

-a look-up table for storing, in a memory,

-formula relationship (Formaler Zusammenhang),

dynamic models (random or model-based).

Using a "look-up table" approach, the training data may be written to a table and stored. Which can then be called as a rule.

The "formulaic relationship" method may mean, for example, calculating statistical variables, such as medians or averages, from data of mass production equipment and/or production machines.

A "dynamic model" approach may mean using a (preferably physical) model. In this case, for example, model parameters of the determined model can be learned and/or a suitable model can be selected by learning. Furthermore, the qualitative characteristics of the model can be learned by itself.

In one embodiment, it is provided that the rules are stored in the production facility, the production machine, the connection device and/or the computer network.

In one embodiment, it is provided that the classification of at least one value is carried out by a classification and evaluation unit upstream of step (a), wherein the classification and evaluation unit performs the following tasks:

-evaluating the data quality and discarding irrelevant data, in particular identifying anomalies,

-compressing and squeezing the data, and-compressing the data,

-creating metadata.

In one embodiment, it is provided that the classification and evaluation unit comprises at least one classification rule, which is created manually, in particular by means of expert knowledge, and/or by a second learning unit having at least one feature of the learning unit of at least one of the preceding embodiments.

In one embodiment, it is provided that the classification rules are stored in the production facility, on the production machine, on the connection device and/or in a computer network.

In the case of the method, which is carried out using at least one rule, the feedback method according to the invention is characterized in that the behavior values of the at least one behavior variable are collected by the data recording unit, which behavior values are trained as training data by the learning unit, wherein the feedback rules are used in particular for evaluating and/or further developing the method, in particular the rule.

In one embodiment of the feedback method, it is provided that the at least one behavior variable describes a behavior of the operator, for example a frequency of acceptance of the action proposal by the operator.

In one embodiment of the feedback method, it is provided that the operator is presented with questions, in particular with regard to the evaluation of the method, via the at least one operator interface, wherein the input of the operator relating to this is at least one behavior variable.

In one embodiment of the feedback method, it is provided that the at least one behavior variable describes a behavior of the rule and/or the method, for example a sensitivity of an output value of the rule to slight changes in an input value of the rule.

The production device according to the invention has means suitable for carrying out the method and/or the feedback method.

The computer program product according to the invention comprises instructions for causing the production device to carry out the method and/or the feedback method.

It should be noted that the method is applicable to cycle-based and continuous production processes. The process is therefore particularly suitable for implementation in a production plant comprising at least one injection molding machine and/or at least one plastic extruder.

Data transmission necessary for using data from mass production machines and/or production equipment may occur anonymously and/or non-anonymously.

The production plant has at least one production machine. The at least one production machine may have at least one peripheral device, which is also part of the production device. Furthermore, at least one operator interface is provided in the production facility. Control and supervision can be performed centrally, for example by a Manufacturing Execution System (MES).

The set-up variables are defined by an operator or a computer program, for example a method and/or a control algorithm for optimizing and/or operating a production process according to the invention.

Examples of set variables of a production process are, in particular, control variables and/or reference variables. For example, the control variable may be a reference variable

Or a variable determining the type of control, the instantaneous value of the reference variable corresponding to a target value. It may also refer to set variables of a control algorithm for the production process. For example, the reference variable may be a supervision limit for a process variable or a variable that determines the type of supervision.

Examples of setting variables of a method or a computer program, for example a method for optimizing and/or operating a production process according to the invention, are variables which determine which rule should be used. It may also be a setting variable of an expert system or of a control algorithm of a production machine.

The behavioral variables describe, for example, the behavior of the production process, method, or operator. As described variables, behavioral variables are not defined or set variables.

Process variables are physical measured variables of the production process or variables derived from them. The process variables describe the behavior of the production process and are therefore behavioral variables.

A reference variable or characteristic factor is a variable derived from one or more process variables. For example, the reference variable or characteristic factor may describe a property of a measurement curve of the process variable or a point in time at which the process variable takes a determined value, or for example a standard deviation of a number of past values of the process variable. Reference variables and characteristic factors are also behavior parameters.

The process variables and/or the reference variables may include quality variables, such as weight, dimensional stability, warpage and/or surface, in particular of components of the production machine and/or production plant. These can be measured directly and/or derived from process variables.

The operator's behavior variables record the operator's behavior. One example is the frequency of acceptance of action suggestions by the operator.

The behavior variables of the method, for example the method according to the invention for optimizing and/or operating a production process, for example the behavior of the recording rules. In this case, for example, the sensitivity of the output value of the rule to small changes in the input value of the rule can be recorded.

The system configuration variables are descriptive variables and are independent of setting variables and behavior variables. For example, they describe properties of a material, production machine, customer, tool, or geographic location. For example, the attributes of a production machine may be the machine type, while the attributes of a customer may be the industry in which he is located.

The values of the system configuration variables are therefore only changed when the configuration, for example, the tool, the customer, the production machine, etc., is changed, in particular they are not changed during steps (a), (b) and (c) of the method according to the invention and/or in steps (a), (b) and (c) of the method according to the invention or during the production process.

For example, parameter classification may combine process variables having the same unit from the same part of a production process and/or from the same area or component of a production machine.

For example, the system configuration classification may combine the type of production machine, the geographic area where the production machine/production equipment is located, or the industry of the customer.

The variables described can be classified, for example, according to the terminology in the present application into the following categories:

setting variables

Set variables of the production Process

o control variable

Reference variable o

Method/set variables of a computer program

o calculating the control variables of the unit/method;

behavioral variables

Behavioral variables of the production Process

o process variable

o reference variable or characteristic factor

Behavioral variables of the method/computer program

-a behavioral variable of the operator;

system configuration variables

In connection with production plants

-about the client.

Drawings

Embodiments of the invention are discussed with reference to the figures. Wherein:

figures 1 a-1 d show an embodiment of the method according to the invention,

figures 2a and 2b show the training of rules by a learning unit,

fig. 3a, 3b show a feedback method, a. user feedback, b. evaluation of the method,

figures 4 a-4 e show the network structure and the arrangement of the computing units,

figures 5a and 5b show a specific embodiment with rules in the form of a decision tree,

fig. 6a, 6b show the average values of the supervision limits of a plurality of injection molding machines.

Detailed Description

Fig. 1a-d show an embodiment of a method 7 for optimizing and/or operating a production process 911 according to the invention. Fig. 1a shows an exemplary embodiment in which an action recommendation based on the value of the setting variable 2 entered by the operator is to be displayed on at least one operator interface 93. Here, the operator has entered at least one value 211, in particular for a control variable 21 for controlling the production process 911. The at least one value 211 is recorded and temporarily stored by the data recording unit 71. The at least one value 211 is transmitted to the calculation unit 72. A plurality of sets of rules 76, here in particular two sets of rules, are stored in the calculation unit 72. The two sets of rules 76 each calculate different output data depending on the at least one value 211 of the control variable 21. The output data of these rules 76 are at least one calculated value (calculated value) 212 of a control variable 21 for controlling the production process 911 and at least one calculated value 222 of a reference variable 22 for supervising the production process 911. For example, the at least one calculated value 212 may be determined by a first rule 76 and the at least one calculated value 222 may be determined by a second rule 76. The calculated values are then sent to the at least one operator interface 93 and visualized for the operator, for example by means of an electronic message 5. Alternatively or additionally, the electronic message 5 can also be sent to the at least one operator interface as a regular output value, for example for outputting a warning or for user-friendly presentation to the operator of the at least one value 212 and/or 222 sent together.

Further, at least one system configuration value 301 of the system configuration variable 3 may be stored in the memory 711 in the data recording unit 71. This value may also be used as an input value for the rule 76.

For example, returning an electronic message 5 to the at least one operator interface 93 may be used to alert an operator to jeopardize the quality of the product or even to jeopardize poor settings of the production machine 91. Further, for a particular action recommendation 51, the return transmission of at least one calculated value 212 of the control variable 21 to at least one operator interface 93 may be used to change the set value of the control variable 21 to a better value, such as the at least one calculated value 212.

If the amount of material and the cooling time are set, for example, in an injection molding machine, the rules 76 can identify that the cooling time is too short relative to the amount of material and thus quality problems of the product, such as sink marks and warpage, are expected. The operator is then informed of an action recommendation 51 and, in particular, by means of the electronic message 5, the setting of the cooling time is increased accordingly, wherein a specific value or value range for the cooling time can also be specified.

Furthermore, the return of the at least one calculated value 212 of the control variable 21 and/or the calculated value 222 of the reference variable 22 to the at least one operator interface 93 may be used to suggest at least one value 212 and/or 222 to the control variable 21 and the reference variable 22 that has not been set. Thus, method 7 functions as a setup assistant.

If, for example, an operator uses a certain material, which is stored by the system configuration value 301 of the system configuration variable 3 assigned to this material, the missing setting values, for example at least one value 212 of the control variable 21 and/or at least one value 222 of the reference variable 22, can be automatically suggested by the rule 76.

The return of the at least one calculated value 222 of the reference variable 22 assigned to the process variable 11 of the production process 911 to the at least one operator interface 93 may be used, for example, to suggest supervisory limits for the process variable 11 to an operator.

The operator may, for example, set a target injection profile as the control variable 21 on at least one operator interface 93, which is transmitted to the calculation unit 72 via the data recording unit 71. From this target injection profile, the rule 76 then calculates a supervision limit for the process variable 11 of the molding process 911, wherein this supervision limit is the value 222 of the reference variable 22 assigned to the process variable 11. On at least one operator interface, the action recommendation 51 may then appear in the form of an electronic message 5 with the exemplary contents "customer having set a similar injection profile has set the following supervision limits for the process variable on the photomicrograph" together with a list of calculated values 222 of the reference variable 22. The operator then accepts, rejects or corrects these values according to the action advice 51.

Fig. 1b shows an alternative embodiment of the method according to the invention, in which, for data acquisition, one or more cycles of a production process 911 (or a continuous production process 911 for a specific time period) are carried out in accordance with values 211 of control variables 21 defined in at least one operator interface 93. For example, a cycle may be considered a test cycle (or a time period as a test run) prior to mass production. The values 211 of the controlled variables 21, for example from the at least one operator interface 93, and the values 111 of the process variables 11 assigned to the controlled variables 21 from the production process 911 are then collected by the data logging unit 71. These values are then transmitted together with the system configuration values 301 to the computing unit 72, which solves the different questions by means of in this case three rules 76.

For example, from the relationship between the target values and the actual values, a recommendation for optimizing the production process 911 and/or the control algorithm of the production process 911 can be displayed on the at least one operator interface 93, in particular together with a user-friendly dialog consisting of the short message 5.

At least one reference value 121 of the reference variable 12, for example a distribution (or scatter plot, degree of scatter, Streuma β es) of the values 111 of the process variable 11, can also be determined from a plurality of values 111 of the at least one process variable 11 (see also fig. 1 b). The scaled distribution of the values 111 of the process variable 11 may be used as an adaptive supervisory limit for the process variable 11 in the production process 911. In this case, it is advantageous to check again the temporary supervision limits given by the scaled distribution. This check and, if necessary, the value adjustment can be carried out by the calculation unit 72 by means of rules 76 created for this purpose. The calculation unit 72 then sends the adjusted supervision limit (as calculated reference value 222) back to the at least one operator interface 93 and/or directly back to the production process 911 (not shown in fig. 1 b).

Fig. 1c shows an embodiment of the method according to the invention for direct parameterization of a production process 911 or direct parameterization of a control algorithm of a production process 911 by means of a calculated value 212 of a control variable 21 or supervision of a production process 911 by means of a calculated value 222 of a reference variable 22.

An incomplete set of values for the variable 2, here control and

reference values

211, 221, is set by an operator input on at least one operator interface 93. The control and

reference values

211, 221 and the system configuration values 301 are forwarded to the calculation unit 72 via the data recording unit 71. The calculated control and

reference values

212, 222 are now a complete set of values for the set variable 2, which is suitable for parameterization of the production process 911 or for parameterization of the control algorithm of the production process 911. Instead of sending the output data of the rule 76 back to the at least one operator interface 93, the decision unit 73 may decide based on the calculated control and

reference values

212, 222 whether these values are to be passed to the production process 911. Furthermore, a full set of set values for the production process 911 or values for a control algorithm for the production process 911 may be optimized by the optimization method 7, for example by comparing target values and actual values of the process variables 11, as shown in fig. 1 b.

Fig. 1d shows a variant of the embodiment in fig. 1a, in which a classification and evaluation unit 74 is inserted between the data recording unit 71 and the calculation unit 74. This unit classifies, evaluates and selects data by classification rules 741. The unit therefore forwards the selected

values

213, 303 and the instance 41 of class 4 to the calculation unit 72. This contains rules 76 which process these values and classified instances as input data and output calculated

values

212, 222, calculated instances 42 of the classification 4 and, if appropriate, electronic messages 5, which are sent back to the at least one operator interface 93.

It is to be noted here that even without the classification and evaluation unit 74, the classified values can be used in the optimization cycle 7 and can be understood by the rules 76. In this case, the sorting is carried out before the method 7 is carried out, manually by an operator, automatically by another sorting unit and/or already at the factory.

Fig. 2a, b show an embodiment of the training of rules 76 by the learning unit 75. Fig. 2a shows a rule 76 which learns the missing setting values 202 of the production process 911 from existing setting values 201 by a large number of settings of other operators. For example, with the rule 76 trained in this way, a reference value 222 for the supervision limit can be determined from the operator-input value 211 of the control variable 21, as shown in fig. 1 a. In order to train the rules 76, the control values 211 and the reference values 221 are here entered by at least one operator on a plurality of operator interfaces 91 of a plurality of production machines 91. These are forwarded to the common learning unit 75 by the respective data recording unit 71 assigned to the production machine 91. In addition, the system configuration values 301 are also forwarded from each production machine 91 herein to the learning unit 75. To generate the rules, a table (or diagram) is now created, for example, which contains pairs of control values 211 and reference values 221, wherein the pairs originate from the production machine 91. This table may be used as a look-up table and the rules 76 have been formed. Furthermore, the table may be a training data set of a supervised machine learning method, for example for training a neural network. The rule 76 trained in this way can determine the reference value 222 by a query by means of the control value 221; rules 76 may thus determine appropriate supervision limits based on, for example, an injection profile. The questions of the rules 76 are specified here by the assignment of pairs of training data, i.e. by tables.

Fig. 2b shows the training of the rules 76 for different questions, i.e. in particular the setting values for identifying "bad", as shown in fig. 1 a. Here a set of control values 211 is used to control the production process 911. The control values 211 and the values 111 of the process variables are forwarded to the learning unit 75 and optionally also together with the system configuration values 301. Such data records (data items, Datensatz) training rules 76 from control values 211 and process values 111 allow the trained rules 76 to predict a process. If events exceeding the reference value 221 are also considered, the rules 76 may predict, evaluate, and issue warnings and value change recommendations to at least one operator interface 93.

Fig. 3a, b show schematic diagrams of the feedback method of method 7. Here, fig. 3a illustrates the training of the feedback rule 100 by operator feedback and/or operator behavior. Here, for example, an operator can evaluate the behavior of the method 7, which has been executed on the at least one operator interface 93, by means of the behavior variables 14 of the method 7. The values 141 of the behavioral variables 14 of the method 7 may then be used to train the feedback rules 100 in the learning unit 75. Feedback rules 100 may be used to evaluate and/or further develop rules 76. Furthermore, the operator's behavior when operating the production machine 91, the production plant 9 and/or the method 7 can be automatically recorded by means of the operator's behavior variables 13 and communicated to the learning unit 75 for training the feedback rules 100.

FIG. 3b shows the training of feedback rules 100 by observing the behavior of method 7 for the automatic parameterization of production process 911. Method 7 typically has rules 76 that are different from the feedback rules 100 in the learning unit 75. However, the feedback rules 100 may be based on the rules 76 of method 7 and trained for further development. The behavior of the method 7 is described by the behavior variable 14, the value 141 of which is transmitted to the learning unit 75. For example, the behavior variable 14 may indicate how sensitive the output value of the rule 76 reacts to small changes in the input value of the rule 76.

Furthermore, the control values 231 may be fed to the calculation unit 72 of the learning unit 75. The feedback rule 100 may thus infer the future behavior of the rule 76 from the control value 231 of the rule 76, e.g., in a trained state.

Fig. 4a-e show the positions of the devices and units. Fig. 4a shows a network structure with a plurality of production devices 9 connected via an external computer network 82. The internal computer network 81 is connected to the external computer network 82 by a connection device 92. A production machine 91, which can implement a production process 911 and has at least one operator interface 93, is located in the internal computer network 81 of the production facility 9. The at least one operator interface 93 may also be integrated in the internal computer network 81, for example in the form of a tablet or a smartphone, away from the production machine 91. All production machines 91 of the production plant 9 can also be controlled and monitored by the manufacturing execution system.

Fig. 4b-e show different possible locations of the units of the method 7 in the computer network 8. For the sake of simplicity of representation, only one production device 9 is shown here. In fig. 4b, the data recording unit 71, the calculation unit 72, the decision unit 73, the classification and evaluation unit 74 and the learning unit 75 are located in an external computer network 82. Nevertheless, it is preferred that all units except the learning unit 75 process data from only one production machine 91. In fig. 4c, all the above units are arranged in the connecting means 92. In fig. 4d, all units except the learning unit 75 are provided directly on the production machine 91. Instead, the learning unit 75 is provided in the connection device 92. Fig. 4e shows an arrangement in which the learning unit 75 and the evaluation and classification unit 74 are integrated in an external computer network 82, while the calculation unit 72 and the decision unit 73 are provided in the connection means 92. The data recording unit 71 is here arranged directly on the production machine 91.

It should be noted that the units shown in fig. 4b-e of the method 7 and the learning method may be arbitrarily set. In particular, one unit may also be provided in both the external computer network 82 and the internal computer network 81. In this sense, a "unit" is preferably a logical unit, but this does not exclude that it may also be a physical unit.

A specific embodiment of the injection molding process is shown in fig. 5a, b. The data recording unit 71 periodically records the following data:

the process variable 11:

actual value of o injection time

o injection curve

The control variables 21:

target value of the process variable 11

o standard of handover

The reference variables 22:

o minimum and maximum of current injection time

System configuration variable 3:

o type of material.

This data is transmitted to the connection device 92 or the external computer network 82. There, the data is classified by the classification and evaluation unit 74 into the following classification 5:

tool type

-handover type

-type of material processed.

Furthermore, the shape of the injection curve is evaluated by a classification and evaluation unit 74 in order to identify anomalies in advance.

To learn the rules 76, the raw data and the classified data of the learning unit 75 are used. Learning occurs in the external computer network 82. The resulting rule 76 is the decision tree shown in fig. 5 a.

Prior to carrying out the method 7, the completed rules 76 are loaded from the external computer network 82 onto the connection device 92 and are available there as OPC/UA services or as REST interfaces.

The rule 76 is continuously refined by continuous learning and is loaded from the external computer network 82 to the connection device 92 by continuous updating at certain time intervals and/or after a sufficiently large change in the learned rule 76, for which it can be used in the method 7.

When used as output data in method 7, the rules 76 in fig. 5a output the upper and lower supervision limits as a function of the actual value of the injection time. In fig. 5a, the upper monitoring limit is shown as a solid line, while the lower monitoring limit is shown as a dashed line. This function represents the reference variable 22 for the intelligent supervision of the production process 911, here the injection molding process. The input data for the rule 76 are class 5 "switch type" with example 51 "volume related" and "pressure related" and class 5 "processed material" with example 51 "PP" and "PE". This calculation by the calculation unit 72 takes place on the connection means 92.

The result of the at least one rule 76 is displayed as an action suggestion 51 in the form of a dialog on at least one operator interface 93, as is exemplarily shown in fig. 5b (the result of the decision tree in fig. 5a is not shown here). The operator may optionally test, accept or discard the action recommendation 51. Furthermore, after the operator has made the decision, an operator feedback dialog may be displayed, wherein the operator can evaluate the settings and/or action recommendations 51 of the test himself. The operator feedback may be fed to the feedback rules 100, e.g. according to fig. 3a, which rules 76 are used for improving the action suggestions 51 and/or for improving the rules 76, e.g. for improving the decision tree in fig. 5 a.

Fig. 6a, b show examples of the profile of the change in the torque DR (process variable 11) in newton meters (Nm) over a plurality of periods Z when three injection molding machines of identical construction are metered, which produce identical molded parts from identical materials. Furthermore, in fig. 6a, the mean DRM and the scaled distribution are shown, here six times the standard deviation σ. Here, the scaled distributions are the values 10Nm, 15Nm and 45 Nm.

In contrast, fig. 6b shows a value variation curve with the ordinate of the mean value DRM shifted accordingly. Furthermore, fig. 6b shows the median of the scaled distribution at 15 Nm.

The curve of the change in value of the process variable 11 and/or the reference value 121 determined therefrom, for example a scaled distribution, can be used, for example, for training the rule 76, which should check the adaptive supervision limit (as explained with respect to fig. 1 b). However, in general, a large number of injection molding machine values, in particular more than three, are used for this. The median can be "learned" from a scaled distribution of three (or typically a large number) injection molding machines.

In method 7, if this type of injection molding machine and this material are used, the median can be suggested as a supervision limit. Alternatively, as explained with reference to fig. 1b, adaptive monitoring limits can also be determined from the current production process 911, wherein these monitoring limits are checked in method 7 by rules 76. It may be checked, for example, whether the determined adaptive monitoring limit lies within a range allowed by the determination of a "learned" median around 15Nm, for example from 10Nm to 20 Nm. If necessary, the supervision limits can be adjusted to the permitted range.

List of reference numerals

1 behavioral variables

101 behavior value

11 Process variable

111 value of Process variable

12 reference variable

121 value of reference variable

13 operator's behavior variables

131 value of operator's behavior variable

14 behavioral variables of the method

141 method's value of a behavior variable

2 set variables

201 set value

202 calculated set value

21 controlled variable

211 values of control variables

212 calculated value of control variable

213 selected value of control variable

22 reference variable

221 value of reference variable

222 calculated value of reference variable

Selected values of 223 reference variables

23 calculating the control variables of the cell

231 calculating the value of the control variable of the cell

3 System configuration variables

301 values of system configuration variables

303 selected values of system configuration variables

4 classification of

41 examples of Classification

42 classified computed examples

5 electronic message

51 action suggestion

6 data transmission mechanism

7 method

71 data recording unit

711 memory

72 calculation unit

73 decision unit

74 classification and evaluation unit

741 Classification rules

75 learning unit

76 rules

8 computer network

81 internal computer network

82 external computer network

9 production facility

91 production machine

911 process of manufacture

912 control unit

92 connecting device

93 operator interface

95 memory

100 feedback rule

Claims

1. A method (7) for optimizing and/or operating at least one production process (911) which is implemented by at least one production machine (91) in a production plant (9) for manufacturing at least one product, wherein the production plant (9) has at least one operator interface (93) for inputting set values (201) of at least one set variable (2), preferably at least one system configuration value (311) of at least one system configuration variable (3) is present in a memory (711), and particularly preferably at least one set value (201) and/or at least one system configuration value (301) is present as at least one classification value, and the method (7) has the following steps:

(a) recording by a data recording unit (71)

-at least one setting value (201) of at least one setting variable (2), and/or

-at least one value (111) of at least one process variable (11) of at least one production process (911) and/or at least one value (121) of at least one reference variable (12) which is determined from at least one value (111) of a process variable (11), wherein the values mentioned in this step preferably exist as classification values;

(b) is determined by a calculation unit (72) by means of at least one rule (76)

-at least one calculated setting (202), and/or

-at least one electronic message (5), in particular in the form of at least one action suggestion (51),

wherein the input data of the rule (76) comprises the values recorded in step (a) and/or at least one system configuration value (301) of at least one system configuration variable (3) and/or classification values of the mentioned values;

(c) determining whether it should be determined by the decision unit (73) and/or the operator via the at least one operator interface (93)

-using the at least one calculated setting (202) from step (b), and/or

-following the at least one action recommendation (51) from step (b), characterized in that the at least one rule (76) from step (b) is created by the learning unit (75) by means of at least one machine learning method using training data of a plurality of production plants (9) and/or a plurality of production machines (91).

2. The method (7) according to claim 1, wherein the training data for creating the at least one rule (76) comprises the following values:

-at least one setting value (201) of at least one setting variable (20), and/or

-at least one value (111) of at least one process variable (11), and/or

-at least one value (121) of at least one reference variable (12), and/or

-at least one system configuration value (301) of at least one system configuration variable (3), and/or

-at least one classification of the above-mentioned values, and/or

-an identifier of at least one of the variables and/or categories.

3. Method (7) according to one of the preceding claims, characterized in that at least one value (211) of at least one control variable (21) is recorded and at least one value (222), in particular a supervision limit, of at least one reference variable (22) and/or an electronic message (5), in particular an action recommendation (51), is determined from this value (211) by means of the rule (76).

4. Method (7) according to one of the preceding claims, characterized in that at least one value (301) and/or an identifier of at least one system configuration variable (3), for example a material of a given product, is used as an input value for determining at least one value (212) of at least one control variable (21) and/or at least one value (222) of at least one reference variable (22) and/or at least one electronic message (5) by means of the rule (76).

5. The method (7) according to any of the preceding claims, characterized in that at least one missing value is determined in step (b) as the calculated value (202) of the setting variable (2) if not all values (201) of the setting variable (2) required for starting the production process (911) are defined.

6. A method (7) according to any of the preceding claims, characterized in that at least one value (121) of at least one reference variable (12) and/or process variable (11) of at least one production process (911) is recorded and the value of the set variable (2) is continuously optimized.

7. The method (7) according to any one of the preceding claims, characterized in that the value (121) of the reference variable (12) in step (a) is derived from a production process (911) which is parameterized with the setting values (201) present in step (a), in particular in that case the production process (911) is run as an intermediate step directly before step (a) for a determined time and/or for a determined number of cycles.

8. The method (7) according to any one of the preceding claims, characterized in that at least in case of a decision by an operator on the at least one operator interface (93) at least in step (c), at least one calculated setting (202) from step (b), preferably also its classification and/or the at least one electronic message (5), is visualized.

9. The method (7) according to any one of the preceding claims, wherein, in case of a positive decision by the decision unit (73) and/or the operator,

-using at least one calculated setting value (202) and/or implementing the action recommendation (51),

and/or in case a negative decision is made by the decision unit and/or the operator,

-retaining said at least one old setting (201), and/or

-entering at least one new setting (201) on at least one operator interface (93) by the decision unit (73) and/or an operator.

10. The method (7) according to claim 9, characterized in that in case the at least one setting value (201) is changed by the decision unit (73), a reason is displayed in the form of an electronic message on the at least one operator interface (93).

11. The method (7) according to any of the preceding claims, wherein the set variables (2) of the at least one production process (911) comprise control variables (21) of process variables (11) and/or supervision limits and/or variables determining the type of supervision.

12. The method (7) according to any of the preceding claims, wherein the system configuration variables (3) comprise variables describing the attributes of:

-the production plant (9),

-the at least one production machine (91), in particular the tools of the at least one production machine (91),

material of the product, and/or

-a client.

13. Method (7) according to any of the preceding claims, characterized in that the following units are or can be put into data connection by means of a computer network (8):

-at least one production machine (91),

-at least one operator interface (93),

-a data recording unit (71),

-a decision unit (73),

-a calculation unit (72),

-a learning unit (75),

-a production plant (9) and at least one further production plant (9).

14. Method (7) according to claim 13, characterized in that the production plant (9) has a connection device (92) which is connected or connectable with a computer network (8) by means of a data transmission means (6), wherein the computer network (8) comprises in particular an internal computer network (81) arranged inside the production plant (9) and an external computer network (82) arranged outside the production plant (9), wherein the external computer network (82) in particular connects the production plant (9) with at least one further production plant (9).

15. Method (7) according to any of claims 13 or 14, characterized in that the data recording unit (71) stores the data provided to it permanently or temporarily in the production plant (9), in a production machine (91) and/or in a computer network (8).

16. The method (7) according to any one of claims 14 or 15, wherein the learning unit (75) implements the at least one machine learning method on at least one external computer network (8) to which a plurality of production devices (9) are or can be connected.

17. Method (7) according to any one of claims 14 to 16, characterized in that said learning unit (75) implements said at least one machine learning method on said at least one connection device (92) to which a plurality of production machines (91) are or can be connected by means of an internal computer network (94).

18. The method (7) according to any of the preceding claims, wherein the training data of the learning unit (75) are collected from a plurality of production machines (91) in the at least one production plant (9), wherein the production machine (91) parts are of different types.

19. The method (7) according to one of the preceding claims, characterized in that the learning unit (75) determines at least one rule (76) for a question, wherein preferably at least one supervised machine learning method is used, wherein the machine learning method particularly preferably learns from training data with an associated answer to a question.

20. Method (7) according to claim 19, characterized in that the learning unit (75) is able to pass on at least one rule (76) of a first question to a second question, in particular by: in a machine learning method, a rule pre-trained with a first question is trained with training data of a second question (76).

21. The method (7) according to any one of the preceding claims, characterized in that at least one rule (76) is created for at least one instance (41) of a system configuration class (4), wherein the at least one rule (76) is trained in particular on questions specific to the at least one instance (41) of the system configuration class (4).

22. The method (7) according to any of the preceding claims, wherein the learning unit (75) determines at least one rule (76) without asking a question, wherein preferably at least one unsupervised machine learning method is used.

23. The method (7) according to any of the preceding claims, wherein the machine learning method uses one of the following methods:

-a decision tree, in which the decision tree is,

-the neural network is a network of nodes,

-a look-up table for storing, in a memory,

-the relationship of the formula,

dynamic models (random or model-based).

24. Method (7) according to any of claims 14 to 23, characterized in that the rules (76) are stored in a production facility (9), a production machine (91), a connection device (92) and/or a computer network (8).

25. Method (7) according to any one of the preceding claims, characterized in that the classification of the at least one value is carried out by a classification and evaluation unit (74) upstream of step (a), wherein the classification and evaluation unit (74) performs in particular the following tasks:

-evaluating the data quality and discarding irrelevant data, in particular identifying anomalies and/or outliers,

-compressing and squeezing the data, and-compressing the data,

-creating metadata.

26. The method (7) according to claim 25, characterized in that the classification and evaluation unit (74) comprises at least one classification rule (741) which is created manually, in particular by means of expert knowledge, and/or by a second learning unit having at least one feature of the learning unit (75) of at least one of the preceding claims.

27. Method (7) according to claim 26, characterized in that the classification rules (741) are stored in the production plant (9), on the production machine (91), on the connection device (92) and/or in the computer network (8).

28. Feedback method using a method (7) according to one of claims 1 to 27, wherein the method (7) is carried out using the at least one rule (76), characterized in that a behavior value (101) of at least one behavior variable (1) is collected by a data recording unit (71), which behavior value is trained as training data by a learning unit (75) as at least one feedback rule (100), wherein the at least one feedback rule (100) is in particular used for evaluating and/or further developing the method (7), in particular the at least one rule.

29. Feedback method according to claim 28, characterized in that the at least one behavior variable (1) describes the behavior of the operator, such as the frequency of acceptance of action suggestions (5) by the operator.

30. Feedback method according to any one of claims 28 or 29, characterized in that a question is posed to the operator via the at least one operator interface (91), in particular with regard to the evaluation of the method (7), wherein the input of the operator relating to this is at least one behavior variable (1).

31. Feedback method according to any of claims 28 to 30, characterized in that the at least one behavior variable (1) describes a behavior of the rule (76) and/or the method (7), such as a sensitivity of an output value of a rule (76) to small changes of an input value of the rule (76).

32. A production plant having a mechanism adapted to implement the method (7) according to any one of claims 1 to 27 and/or the feedback method according to any one of claims 28 to 31.

33. Computer program product comprising instructions for causing a production apparatus according to claim 32 to carry out the method (7) according to any one of claims 1 to 27 and/or the feedback method according to any one of claims 28 to 31.