Configuration method for a monitoring device for improved quality forecasting of workpieces

EP4630888A1Pending Publication Date: 2025-10-15SIEMENS AG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
EP2024702249
Authority / Receiving Office
EP · EP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-01-30
Filing Date
2024-01-17
Publication Date
2025-10-15

AI Technical Summary

Technical Problem

Configuring monitoring devices for production systems with increasing sensor data complexity and the need for flexible, quick setup and reliable quality prediction in manufacturing processes is challenging, especially in cyclical production processes where workpieces are repeatedly processed.

Method used

A method that records process variables as time series, aggregates them into lower-volume statistical variables, combines these with quality parameters to form workpiece data sets, and uses machine learning to identify prediction variables, allowing for automatic configuration of monitoring devices without pre-known connections, thereby reducing data complexity and enabling quick quality prediction.

Benefits of technology

This method significantly reduces data volume and complexity, allowing for efficient prediction variable identification and quick configuration of monitoring devices, enabling reliable quality forecasting and flexible production system adaptation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure EP2024051015_08082024_PF_FP
    Figure EP2024051015_08082024_PF_FP
Patent Text Reader

Abstract

The invention relates to a method (100) for configuring a monitoring device (30), which is designed for monitoring a production process (15) in a production system (10) for quality forecasting. A process variable (16) is detected and, on this basis, a plurality of aggregation variables (45) are determined. A quality parameter (22) is also detected. Multiple aggregation variables (45) and the quality parameter (22) are combined to form a workpiece data set (49). In addition, a respective interdependence (46) between a respective aggregation variable (45) and the quality parameter (22) is determined and a forecasting variable (54) is determined from among the aggregation variables (45) based on the determined interdependence (46). The forecasting variable (54) is specified as the variable to be monitored of the production process (15) for the operation of the monitoring device (30). The invention also relates to a computer program product (60) with which a method of this type (100) can be carried out. The invention also relates to a monitoring method (200) for a production process (15) in a production system (10). The invention further relates to a production system (10) provided with a monitoring device (30) of this type.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] Description

[0002] Configuration method for a monitoring device for improved quality prediction of workpieces

[0003] The invention relates to a method for configuring a monitoring device belonging to a production plant. Furthermore, the invention relates to a monitoring method for a manufacturing process. Likewise, the invention relates to a computer program product for at least partially implementing the method for predicting quality results. Furthermore, the invention relates to a correspondingly configured monitoring device and a production plant having such a monitoring device.

[0004] International application WO 2022 / 063702 A1 discloses a method for quality control of a mass-produced product, in which parameters of the manufacturing process are recorded. The product is then subjected to an inspection, and on this basis a quality assessment is carried out. The recorded parameters and the quality assessment are fed into a machine learning model. Based on the recorded parameters, a quality forecast is carried out and checked for agreement with or deviation from the quality assessment. In this way, machine learning is carried out, through which the machine learning model is trained to recognize the causes of quality problems.

[0005] Production plants have a growing number of sensors, resulting in an increasing amount of data available for monitoring. This makes the proper configuration of a monitoring device for such production plants more complex and time-consuming. There is also a need for flexible production plants that can be quickly converted to other production processes and reliably deliver high product quality after a short start-up time. There is a need for a way to configure monitoring devices for production plants quickly and largely automatically.

[0006] The object is achieved by a method according to the invention for configuring a monitoring device which is part of a production plant. The monitoring device is designed to monitor a production process which is carried out by means of the production plant. The production process can be a cyclical production process in which workpieces are repeatedly machined in the same way. The workpiece can be a blank, a machined blank or a component to be repaired. The method is based on the assumption that the monitoring device is in a basic state in which it is at least suitable for receiving inputs by means of which a process variable to be monitored as a predictive variable can be specified.Furthermore, the monitoring device in its basic state can be free of known or suspected relationships that influence a desired quality of the workpiece, i.e. free of a corresponding configuration.

[0007] The method comprises a first step in which at least one predeterminable process variable is recorded on a component of the production plant. The recording of the predeterminable process variable takes place during a pass through the production process on a workpiece. The process variable can be a quantity that is measurable on the respective component, for example the electrical current strength of the heating elements or the temperature in the oven during a hardening process. The at least one process variable is a time-varying quantity, so that the process variable recorded in the first step is designed as a time series, i.e. as a plurality of data points that can be sorted chronologically, for example using a timestamp. The process variable is at least temporarily or at least partially stored in the first step and thus made available for further steps.The process variable can be specified by a user, a table and / or an algorithm, for example an expert knowledge database.

[0008] The method further comprises a second step in which a predeterminable quality parameter of the workpiece is recorded. The quality parameter can be a physical quantity that is present directly on the workpiece and is measurable. For example, a material property of the workpiece or a deviation of a material property from a nominal value can be a quality parameter. Among other things, such a material property can be a hardness or porosity of the workpiece. A desired property of the workpiece can be quantified using the quality parameter. The quality parameter can be specified by a user, a table and / or an algorithm, for example an expert knowledge database. In addition, the method comprises a third step in which aggregation variables are determined based on the at least one process variable from the first step.The aggregation variables each represent a value derived from the process variable configured as a time series. In particular, the aggregation variables can be statistical values ​​of the at least one process variable. At least one aggregation variable can have a data volume that is smaller than the data volume of the underlying process variable. The third step can be performed after the first step or at least partially simultaneously with the first step.

[0009] The method further comprises a fourth step in which the aggregation variables are combined with the quality parameter recorded in the second step, thereby forming a workpiece data record. The aggregation variables which belong to the same workpiece are combined with the quality parameter. The workpiece data record therefore comprises a plurality of aggregation variables specific to the workpiece, for which a connection with the associated quality parameter is conceivable. Furthermore, the workpiece data record can be free of the process variable on the basis of which the aggregation variables stored therein are determined. The fourth step can be carried out after the third step or at least partly simultaneously with the third step.Furthermore, the first to fourth steps in the method according to the invention are performed multiple times, thus providing a plurality of workpiece data sets, for example, for several different workpieces. The plurality of workpiece data sets is provided for a subsequent fifth step.

[0010] In the fifth step of the method according to the invention, a dependency is determined between a plurality of the aggregation variables and the quality parameter. On the basis of a determined dependency, one of the aggregation variables is recognized as a prediction variable. The recognition of the respective dependencies can be carried out using a machine learning model. Consequently, in the fifth step it is recognized between which individual aggregation variable or multiple aggregation variables and the quality parameter a meaningful relationship exists. This relationship is characterized by the at least one recognized dependency and can be quantified. The fifth step can be carried out using a machine learning model, for example with a decision tree, random forest, gradient boost, XGB boost or neural network.The method further comprises a sixth step in which the predictive variable identified in the fifth step is specified to the monitoring device as the variable of the production process to be monitored. For this purpose, for example, a designation of the predictive variable can be transmitted to the monitoring device. The configuration of the monitoring device is at least partially set by the predictive variable.

[0011] In the method according to the invention, a plurality of aggregation variables are derived from the process variable designed as a time series. Time series are extensive in terms of data volume, so that processing them into aggregation variables represents data compression. The determination of the aggregation variables can therefore essentially be carried out alongside the production process. The workpiece data set created in the method according to the invention is small in terms of data volume, which significantly reduces the number of dimensions and the complexity of the machine learning model. Even with an increased number of workpiece data sets, their evaluation in the fifth step, i.e. the determination of dependencies, the underlying data volume is relatively small, which in turn allows the dependencies to be determined more quickly.The invention is based, among other things, on the surprising discovery that aggregation variables are at least sufficiently meaningful as predictive variables for a quality parameter. The recognition of an aggregation variable as a predictive variable does not require the specification of known or suspected relationships within the production process and can therefore be carried out automatically. The method is suitable for quickly generating and specifying meaningful predictive variables in a production plant, which enables a forecast of quality results for machined workpieces.

[0012] In one embodiment of the claimed method, data points belonging to an idle time action of the corresponding component are determined on the basis of the time stamp or the value of the predefinable process variable, i.e. time stamps or values ​​of individual data points of the process variable. An idle time action is any operation of the component in which no productive activity of the corresponding production process takes place, for example during an idle time movement of a machine tool in another process phase or during idling. Such data points belonging to an idle time action are removed. As a result, the process variable on the basis of which the aggregation variables are determined only contains data points that characterize the corresponding production process. The generation of misleading aggregation variables is thus reduced or prevented.Alternatively or additionally, the at least one process variable used in the third step can be checked using another process variable, thus identifying data points that belong to a non-productive action. The data points identified in this way are also removed. This type of filtering allows for the creation of aggregation variables with increased significance.

[0013] In a further embodiment of the claimed method, the fifth step, i.e. determining the dependencies, is carried out using a machine learning model. Such a machine learning model can be designed, for example, as a decision tree, random forest, gradient boost, XGB boost or neural network. A variety of machine learning models are suitable for quickly and reliably determining dependencies in data sets such as the tool data sets. The machine learning model used in the fifth step can therefore be replaced separately. This makes the claimed method easily suitable for using future machine learning models and exploiting their technical potential.

[0014] Furthermore, the machine learning model can be trained using a plurality of workpiece data sets provided according to the first, second, third and fourth steps. Such workpiece data sets can, for example, be obtained by a configuration method on another production plant, or from previous runs of the claimed method. The claimed method is therefore suitable for creating workpiece data sets suitable for further training of the machine learning model during operation of the production plant, i.e. during the production process. Such further workpiece data sets are respectively generated for further workpieces. Such further training can take place on a hardware platform separate from the production plant. The machine learning model is thus updated by a further trained version during the production process.The performance of the claimed method can thus be increased automatically.

[0015] Furthermore, in the claimed method, one of the aggregation variables can be identified as a predictive variable if there is a quantifiable relationship, for example a linear or non-linear dependency, between the corresponding aggregation variable and the quality parameter, which is characterized by a dependency coefficient. If the dependency coefficient of the corresponding dependency exceeds a predefinable threshold value in terms of amount. The predefinable threshold value can be used to set the strength below which the existing quantifiable relationship is ignored in the claimed method. The threshold value can therefore be used to specify how many aggregation variables are to be expected as predictive variables. The fewer predictive variables that are specified for the monitoring device in the sixth step, the more efficiently the production process can be monitored.The threshold value also makes it possible to quantify the influence of a predictor variable on the quality parameter.

[0016] Furthermore, at least one of the aggregation variables can be designed as a maximum, minimum, arithmetic mean, geometric mean, or standard deviation of the associated process variable. These represent quantities that can be easily determined from a time series, which can be stored in a data-saving manner, and thus offer increased compression compared to the underlying process variable. Efficient algorithms are available for such statistical functions, for example in so-called function libraries, which allow the respective aggregation variable to be determined quickly. Alternatively or additionally, the aggregation variable can also be any other statistical quantity derived from the process variable. In a further embodiment of the claimed method, at least one of the aggregation variables can be designed as a set of coefficients of a comparison function, for example a fitting function.The set of coefficients can comprise one or more coefficients. The comparison function can, for example, be a function by which a temporal profile of at least one process variable is approximated at least in sections. The comparison function, and thus also the associated set of coefficients, can be determined using an evaluation unit that is assigned to a component that is designed to act directly or indirectly on workpieces. Alternatively or additionally, the comparison function, and thus its set of coefficients, can be determined using the monitoring device. The set of coefficients of the comparison function also makes it possible to reproduce a complex temporal profile of the process variables in a data-efficient manner. Accordingly, complex relationships can also be taken into account when determining the at least one prediction variable with reduced memory and computing requirements.This makes it possible to determine particularly meaningful predictive variables. Alternatively or additionally, at least one of the aggregation variables can be designed as a comparison value for a model output value of a process model or a sub-process model. The process model or the sub-process model can be designed as a physical model of the production process or sub-production process. As an example, a linear or exponential model with corresponding coefficient sets can be used to depict the course of the process variables in the corresponding process phase during the manufacture of the workpieces. Process models or sub-process models represent realistic representations of the underlying production process that can be used by the claimed method.Furthermore alternatively or additionally, the at least one aggregation variable can be designed as a process interruption indication which is determined based on the process variable which is recorded in the first step. The process interruption indication can indicate a duration of an interruption or a number of interruptions in the production process, a start and end time of the interruption and / or a frequency indication of interruptions. A process interruption indication is suitable for indicating whether and to what extent a decline in product quality is to be expected. For example, an interim failure of a heating element in a hardening furnace can have different effects on the hardening process depending on the position in the temperature profile being used. Overall, the claimed method is suitable for determining aggregation variables using different types of data, i.e. process variables.This allows correlations in the production process to be captured in the form of predictive variables that are not visible to the user. The monitoring device can thus be set to a predictive variable that has increased significance with respect to a quality parameter.

[0017] Furthermore, the first to sixth steps can be performed repeatedly during the production process. This allows workpiece data sets to be obtained from a large number of workpieces, which can then be used to further train the machine learning model used in the fifth step. This further increases the overall performance of the monitoring device.

[0018] The underlying problem is also solved by a quality forecasting method according to the invention for a production process which is carried out by means of a production plant. The production plant comprises at least one component with which the production process is carried out, for example a furnace for hardening. The quality forecasting method comprises a first step in which the production plant is made available in an active operating state in which the production process to be monitored is running. In particular, a workpiece can be machined. Furthermore, in the first step at least one process variable is recorded. Furthermore, the quality forecasting method comprises a second step in which based on the recorded process variable at least one prediction variable is determined which can be designed as an aggregation variable derived from the process variable.The claimed quality forecasting method also includes a third step in which a forecast value for a quality parameter of a workpiece is determined using the at least one forecast variable. This is an expected quality parameter. In the third step, the determined forecast value is compared with an associated, predeterminable target value or at least a predeterminable tolerance limit, i.e. a difference between them is quantified. Furthermore, the quality forecasting method has a fourth step in which a warning is issued to a user and / or a data interface if an improper condition of the production plant is identified based on the comparison in the third step.According to the invention, the at least one predictive variable determined in the second step is predetermined by a method for configuring the monitoring device according to one of the embodiments outlined above. Such a quality prediction method provides targeted and reliable detection of a decline in the quality of the workpieces being processed in the monitored production process.

[0019] In one embodiment of the claimed quality forecasting method, its third step is carried out while its first step is being carried out. Accordingly, a process variable is recorded during machining of the workpiece and the forecast value is determined based on a subset of the data points belonging to the corresponding production process, for example since the start of machining of the workpiece. The claimed quality forecasting method therefore determines the forecast value for the quality parameter before the underlying machining of the workpiece is completed, and this value is compared with the associated predeterminable target value or the predeterminable tolerance limit. This allows an early quality forecast, and accordingly an early reaction to impending quality losses, particularly in the case of long machining processes, for example hardening in a hardening furnace.

[0020] Alternatively, at least the third step of the claimed quality forecasting method can be carried out once the first step has been completed. This means that all data points generated during processing of the corresponding workpiece are used to determine the forecast value. This avoids inaccurate forecast values. This applies in particular to forecast variables, i.e. an aggregation variable that only arises from the dynamics of the underlying process variable. This can, for example, be an arithmetic mean of a temperature profile that is still at ambient temperature at the beginning of the first step.

[0021] Whether the third step is performed during the first step or after the first step can be specified by a user, for example.

[0022] The described object is also achieved by a computer program product according to the invention. The computer program product is designed to receive at least one process variable and to process the process variable. By processing the process variable, a predictive variable can be determined. According to the invention, the computer program product is designed to carry out a method for configuring a monitoring device according to one of the embodiments set out above. The computer program product can be designed to be executed on a control unit which belongs to the production plant at which the process variable which can be received and processed by the computer program product is recorded.Furthermore, the computer program product can be monolithic, i.e. can be executed on a single hardware platform, for example a programmable logic controller (PLC for short), a host computer and / or an operator station that is coupled to the production plant or belongs to it. Alternatively, the computer program product can be modular and comprise a plurality of subprograms that can be executed on separate hardware platforms and that interact via data exchange in order to implement the functionality of the claimed method. For example, the claimed computer program product can be executable in the form of subprograms on different computers in a computer cloud. The computer program product can in particular be designed as software for an edge device or a computer cloud.Furthermore, the computer program product can be at least partially hard-wired, for example, as a chip, integrated circuit, or FPGA. Alternatively or additionally, the computer program product can be at least partially embodied as software.

[0023] Furthermore, the underlying object is achieved by a monitoring device according to the invention which is designed to monitor a production process which is carried out by means of a production plant. The monitoring device is suitable for receiving at least one process variable and for determining at least one predictive variable on the basis of this. According to the invention, the predictive variable is designed by means of a method for configuring a monitoring device according to one of the embodiments presented above. Alternatively or additionally, the monitoring device is designed to carry out a monitoring method according to one of the embodiments described above. Such a monitoring device allows the associated production plant to be set up quickly and ensures a reliable quality forecast for the workpieces machined therewith.

[0024] The object outlined above is also achieved by a production system according to the invention, which comprises a plurality of components, each of which is designed to machine a workpiece, in particular during a production process. The production system also comprises a monitoring device connected to the components. According to the invention, the monitoring device is designed according to one of the embodiments presented above. Such production systems offer a reliable forecast for the quality of the workpieces machined therewith.

[0025] The invention is explained in more detail below with reference to individual embodiments in figures. The figures are to be read as complementary to one another in that identical reference numerals in different figures have the same technical meaning. The features of the individual embodiments can also be combined with one another. Furthermore, the embodiments shown in the figures can be combined with the features outlined above. They show in detail:

[0026] FIG 1 schematically shows a first embodiment of the claimed configuration method in a first stage;

[0027] FIG 2 schematically shows a second embodiment of the claimed configuration method during its first or second step;

[0028] FIG 3 schematically shows the first embodiment of the claimed configuration method in a second stage;

[0029] FIG 4 schematically shows the first embodiment of the claimed configuration method in a third stage;

[0030] FIG 5 schematically shows a first embodiment of the claimed monitoring method.

[0031] FIG 1 shows a first embodiment of the claimed method 100 for configuring a monitoring device 30 in a first stage. The monitoring device 30 belongs to a production plant 10 on which a production process 15 can be carried out in which workpieces 20 are machined. For this purpose, the production plant 10 has a plurality of components 12 which are designed to directly or indirectly machine the workpieces 20. The production process 15 is designed as a cyclical production process such that the components 12 essentially repeatedly carry out the same processing 13 for each workpiece 20. The workpieces 20 are machined successively station by station. The components 12 can be designed, for example, as drive motors for tools.The components 12 of the production plant 10 are communicatively connected to the monitoring device 30 and a control unit 40 of the production plant 10 for transmitting measurement signals 35. The control unit 40 and the monitoring device 30 are coupled to one another via a communicative data connection 36. Furthermore, a process model 42 of the underlying production process 15, which includes a digital twin 44 of the production plant 10, runs on the control unit 40.

[0032] A processing 13 of a workpiece 20 carried out by a component 12 is characterized by at least one process variable 16 which can be recorded in the claimed method 100. The process variable 16 is a measurable quantity which forms a time series 18 during a processing 13 of the respective workpiece 20. A recorded process variable 20 comprises a plurality of data points 19 which form the time series 18 along a horizontal time axis 17. Accordingly, a time stamp is present for each of the data points 19. The value of the process variable 20 is symbolized in FIG 1 on a vertical value axis 14. As a result of the processing 13 of the workpieces 20, a physical quantity is changed in these, which is measurable and represents a quality parameter 22. The quality parameter 22 can, for example, be a dimension of the workpiece 20 which is to be manufactured to a specified size.

[0033] The method 100 comprises a first step 110 in which a predeterminable process variable 16 is detected which is present at one of the components 13, for example an electrical current at a drive motor. The process variable 16 is detected on the component 12 during a processing 13 of the respective workpiece 20 by a detection means not shown in detail. The detected process variable 16 can be transmitted to the control unit 40 in the first step 110 in the form of a measurement signal 35. After the first step 110 or partly during the first step 110, a second step 120 is carried out in the claimed method 100. In this step, a predeterminable quality parameter 22 is detected on the respective workpiece 20 which is processed in the first step 110. For this purpose, the production plant 10 is provided with suitable detection means not shown in detail.

[0034] By means of the claimed method 100, at least one aggregation variable 45 is to be determined on the basis of the process variable 16, which aggregation variable is to be specified as a prediction variable 54 for the monitoring device 30, i.e. is to be set in its configuration. For this purpose, a third, fourth and fifth step 130, 140, 150 is carried out by means of the control unit 40, which are described in more detail in the following figures. The at least one prediction variable 54 determined in the fifth step 150 is specified to the monitoring device 30 in a sixth step 160. Alternatively or additionally, at least one of the steps 130, 140, 150 can also be carried out by a different hardware platform than the control unit 40.

[0035] Given predetermined prediction variables 54, the monitoring device 30 is suitable for carrying out a monitoring method 200 in which a forecast value 58 is determined which corresponds to a quality parameter 22 expected for the respective workpiece 20. By comparing it with a predefinable target value 55, the monitoring method 200 determines whether an improper condition of the production system 10 exists. Depending on this, a warning 33 can be output to a user via a user interface 32, for example a visual display or an acoustic signal. Alternatively or additionally, the warning 33 can also be output to a data interface 34. The control unit 40 is equipped with a computer program product 60 which is designed to carry out at least one of the steps 110, 120, 130 shown in FIG. 1.A second embodiment of the claimed method 100 is shown in FIG 2 during the first or third step 130. There is a process variable 16 recorded in a first step 110, as shown for example in FIG 1, which is in the form of a time series 18. The process variable 16 comprises a plurality of data points 19 which are arranged along a horizontal time axis 17. The value of the respective data points 19 is shown in FIG 2 by a vertical magnitude axis 14. An adjustable threshold value 38 is shown which is derived from a further process variable 16 (not shown). Data points 19 are recognized as data points 19 which belong to an active phase 37 of the associated component 12, and thus of the production process 15.Data points 19 below the threshold value 38 are recognized as data points 19 that belong to an idle time action of the associated component 12, and thus to the production process. Such idle time actions form inactive phases 39 of the production process 15 or of the associated component 12 between the active phases 37. For further processing of the process variable 16, data points 19 that belong to the idle time action are removed and thus ignored in the further method 100. This ensures that the claimed method 100 avoids processing data points 19 that belong to an incorrect process phase and thus cannot lead to the recognition of a prediction variable 54. The active phases 37 each correspond to process phases of the production process 15 in which a workpiece 20 is machined.The second embodiment shown in FIG. 2 can be combined with the first embodiment of the method 100, as shown in FIG. 1. The steps shown in FIG. 2 can be at least partially carried out by means of a computer program product 60 (not shown in detail), which can be executed on a control unit 40 of the production plant 10.

[0036] The first embodiment of the claimed method 100 is shown schematically in FIG 3 in a second stage, which follows the first stage shown in FIG 1. A third step 130 of the method 100 is carried out, in which the process variable 16 recorded in the first step 110 is present in the form of a time series 18. In this third step, a plurality of aggregation variables 45 are determined based on the process variable 16. The aggregation variables 45 are determined for data points 19 of a process phase in which a workpiece 20 is machined in a second step 120. The aggregation variables 45 represent a processing of the corresponding process variable 16 and each have a smaller data volume than the process variable 16 itself. The aggregation variables 45 may also have a smaller data volume than the corresponding process variable 16 .The aggregation variables 16 can be, for example, a maximum, a minimum, an arithmetic mean, a geometric mean, and / or a standard deviation of the process variable 16, which are determined based on its data points 19. The different aggregation variables 45 can be predefined in the claimed method 100 or selected by an algorithm or a user.

[0037] The determined aggregation variables 45 are linked, i.e. combined, in a fourth step 140 with the quality parameter 22, which is recorded in the first step 110, as shown for example in FIG. 1. By combining 47 the aggregation variables 45 with the quality parameter 22, a workpiece data record 49 is formed. The workpiece data record 49 can also have a smaller data volume than the underlying process variable 16. The workpiece data record 49 generated in the fourth step 140 is made available for a subsequent fifth step 150. The control unit 40 of the production plant 10 is equipped with a computer program product 60 which is designed to carry out at least one of the steps 130, 140 shown in FIG. 3. A third stage of the first embodiment of the claimed method 100 is shown schematically in FIG. 4. The third stage follows the second stage shown in FIG 3 .The third stage according to FIG 4 assumes that the first, second, third and fourth steps 110, 120, 130, 140 have been carried out several times and that a plurality of workpiece data sets 49 is available. Each of the workpiece data sets 49 comprises a plurality of aggregation variables 45, which are each linked to a quality parameter 22. The linking, i.e. the combining 47, in the respective workpiece data set 49 takes place in the fourth step 140 of the method 100. The third stage includes a fifth step 150 of the method 100, in which the workpiece data sets 49 are processed by a machine learning model 50, which is designed, for example, as a neural network 52. In the fifth step 150, the machine learning model 50 determines at least one dependency 46 between the aggregation variables 45 of the processed workpiece data sets 49 and the associated quality parameter 22.The dependency 46 can be quantified using a dependency coefficient (not shown in detail). Based on the dependency 46 identified in the fifth step 150, the associated aggregation variable 45 is identified and selected as a prediction variable 54. The selection of the prediction variable 54 from among the aggregation variables 45 can be based, among other things, on the degree of the dependency 46, i.e., the associated dependency coefficient. The selection of the aggregation variable 45 as a prediction variable 54 can also be performed by the machine learning model 50.

[0038] The detected prediction variable 54 is transmitted to the monitoring device 30 in a sixth step 160 and is thus specified as a variable to be monitored for operation of the production plant 10. The monitoring device 30 is accordingly configured by the claimed method 100. The fifth and / or sixth step 150, 160 is carried out in the claimed method 100 by the control unit 40. The control unit 40 is equipped with a computer program product 60 which is designed to carry out at least one of the steps 150, 160 shown in FIG. 4. Alternatively or additionally, at least one of the steps 150, 160 can also be carried out by a different hardware platform than the control unit 40.

[0039] A first embodiment of the claimed monitoring method 200 is schematically shown in FIG 5. The monitoring method 200 is carried out by means of a monitoring device 30 of a production plant 10, which is at least partially configured by a claimed method 100. The monitoring method 200 comprises a first step 210, in which the production plant 10 is provided in an active operating state and at least one process variable 16 is recorded. During the active operating state, at least one workpiece 20 is machined by the production plant 10. This is followed by a second step 220, in which the process variable 16 is further processed and thus an aggregation variable 45 is determined. The aggregation variable 45 is specified, i.e. set, by the configuration of the monitoring device 30 as a prediction variable 54 and thus as a variable to be monitored.The prediction variable 54 is predetermined by the claimed method 100 for configuring the monitoring device 30. This is followed by a third step 230 in which a prediction value 58 is determined based on the determined prediction variable 54. The prediction value 58 represents an expected quality parameter 22 for the corresponding workpiece 20, which is machined in the active operating state during the first step 210. The prediction value 58 is compared with a predefinable target value 58. Based on the comparison carried out in this way, a branch 235 of the monitoring method 200 results. If it is determined that the prediction value 58 meets the requirements defined by the target value 55, a proper state of the production plant 10 is recognized. Accordingly, a return 250 to the first step 210 of the monitoring method 200 takes place.If an improper condition of the production plant 10 is detected by comparing the target value 55 with the forecast value 58, a fourth step 240 follows. In the fourth step 240, a warning 33 is issued via a user interface 32 and / or a data interface 34. The monitoring method 200 then reaches a final state 300.

Claims

Patent claims 1. Method (100) for configuring a monitoring device (30) which is designed to monitor a production process (15) in a production plant (10) and which is in a basic state, comprising the steps of: a) detecting at least one predeterminable process variable (16) on a component (12) of the production plant (10), wherein the process variable (16) is designed as a time series (18); b) detecting a predeterminable quality parameter (22) of a workpiece (20) which has been processed by the production plant (10); c) determining aggregation variables (45) based on the process variable (16) from step a); d) combining a plurality of the aggregation variables (45) and the quality parameter (22) in a workpiece data set (49);wherein a plurality of workpiece data sets (49) are provided according to steps a) to d) for a plurality of different workpieces (20), and e) determining a dependency (46) between a respective aggregation variable (45) and the quality parameter (22) and recognizing a prediction variable (54) among the aggregation variables (45) on the basis of the determined dependencies (46); f) specifying the prediction variable (54) for the operation of the monitoring device (30) as a variable of the production process (15) to be monitored, wherein at least one of the aggregation variables (45) is designed as a set of coefficients of a comparison function; 2. Method (100) according to claim 1, characterized in that in step a) based on a time stamp of the predeterminable process variable (16) and / or a further predeterminable process variable (16) data points (19) are determined and removed which belong to an idle time action of the component (12) of the production plant (10).

3. Method (100) according to claim 1 or 2, characterized in that step e) is carried out by means of a machine learning algorithm (50).

4. Method (100) according to claim 3, characterized in that the machine learning algorithm (50) is trained on the basis of a plurality of workpiece data sets (22) provided according to steps a) to d).

5. Method (100) according to one of claims 1 to 4, characterized in that one of the aggregation variables (45) is recognized as a prediction variable (54) if there is a dependency between the corresponding aggregation variable (54) and the quality parameter (45) with a dependency coefficient which exceeds a predeterminable threshold value in terms of amount.

6. Method (100) according to one of claims 1 to 5, characterized in that at least one of the aggregation variables (45) is designed as a maximum, minimum, arithmetic mean, geometric mean, or standard deviation of the associated process variable (16).

7. Method (100) according to one of claims 1 to 6, characterized in that at least one further one of the aggregation variables (45) is designed as a comparison value to a model output value of a process model (42) or sub-process model, as a derivation of a predeterminable order of the process variable (16), or as a process interruption indication which is determined on the basis of the process variable (16).

8. Method (100) according to one of claims 1 to 7, characterized in that steps a) to f) are carried out repeatedly accompanying the production process (15).

9. Quality forecasting method (200) for a production process (15) which is carried out by means of a production plant (10) with at least one component (12), comprising the steps: a) providing the production plant (10) in an active operating state and detecting at least one process variable (16); b) determining at least one prediction variable (54) based on the detected process variable (16); c) determining a prediction value (58) based on the prediction variable (54) for a quality parameter (22) of a workpiece (22) and comparing the prediction value (58) with an associated predefinable target value (55) or a predefinable tolerance limit; d) issuing a warning (33) to a user and / or a data interface (34) if an improper condition of the production plant (10) is detected based on the comparison in step c);characterized in that the prediction variable (54) is predetermined by means of a method (100) according to one of claims 1 to 8; 10. Quality prediction method (200) according to claim 8, characterized in that at least step c) is carried out while step a) is carried out or that at least step c) is carried out when step a) is completed.

11. Computer program product (60) for receiving at least one process variable (16) and processing the at least one process variable (45) to determine at least one prediction variable (54), characterized in that the computer program product (60) is designed to at least partially carry out a method (100) according to one of claims 1 to 8.

12. Monitoring device (30) for monitoring a production process (15) on a production plant (10), which is designed to receive at least one process variable (45) and to determine at least one prediction variable (54) based on the process variable (45), characterized in that the at least one prediction variable (54) is predetermined by means of a method (100) according to one of claims 1 to 8.

13. Production plant (10), comprising a plurality of components (12), each of which is designed to machine a workpiece (20), and a monitoring device (30) connected to the components (12), characterized in that the monitoring device (30) is designed according to claim 12.