Method for monitoring and modifying a training intelligent algorithm for setting operating parameters of a

Abstract

The invention relates to a method for monitoring and modifying an intelligent algorithm (1) for setting an operating parameter (4) of a container handling machine (5), the method comprising: determining a current reward (13) from a comparison of a set value (9) of a container characteristic with a measured actual value (8) of the container characteristic, an expected reward (16) is determined from a comparison of the setpoint with an expected simulated actual value (10) of the container characteristic, where the expected simulated actual value is based on the intelligent algorithm (1), the current reward (13) and the expected reward (14) are automatically analyzed (19), and at least one manipulated variable (2) of the intelligent algorithm is modified (21) based on a result of the automatic analysis to set an operating parameter of the container handler. Further, the present invention relates to an apparatus comprising a computer-readable storage device having stored thereon instructions that, when executed by a processor of the apparatus, cause it to perform the method.

Description

Technical Field

[0001] The present invention relates to a method for monitoring and modifying intelligent algorithms for setting operating parameters of a container processor, according to the independent claims, and to apparatus for performing the method. Background Technology

[0002] US 2022 / 0147876 A1 discloses a method for providing an interpretable agent for estimating an interpretable reward function. The method includes detecting an observation space comprising one or more states and one or more actions, and modeling the observation space as multiple interpretable state-action pairs. The method also includes forming one or more interpretable models comprising a simulated environment and a reward function, and returning at least one interpretation for at least one of the one or more interpretable models corresponding to at least one state and at least one reward.

[0003] To configure container processor parameters, monitoring can be provided on the machine, such as through a programmable logic controller (SPS; also known as a Programmable Logic Controller (PLC)) or edge devices. The operation and performance of artificial intelligence (KI; also known as Artificial Intelligence (AI)) can be monitored. Monitoring can deliver messages or information about problems and weaknesses of the container processor. When anomalies occur, human operators can analyze the data, make decisions, and initiate appropriate actions when inappropriate behavior occurs. Summary of the Invention

[0004] The purpose of this invention is to provide a method for monitoring and modifying intelligent algorithms for setting operating parameters of a container processor, and an apparatus for performing the method, which allows the intelligent algorithm to self-heal under changing framework conditions.

[0005] Solution

[0006] This objective is achieved by a method for monitoring and modifying intelligent algorithms for setting operating parameters of a container processor, as described in the independent claim, and by means of performing the method. Further embodiments are disclosed in the dependent claims.

[0007] The method according to the present invention for monitoring and modifying an intelligent algorithm for setting operating parameters of a container processor includes: determining a current reward based on a comparison of a set value of a container characteristic with a measured actual value of the container characteristic; determining an expected reward based on a comparison of the set value with an expected simulated actual value of the container characteristic, wherein the expected simulated actual value is based on the intelligent algorithm; automatically analyzing the current reward and the expected reward; and modifying at least one manipulated variable of the intelligent algorithm based on the result of the automatic analysis to set the operating parameters of the container processor.

[0008] Container handling equipment may include blow molding machines for manufacturing PET containers from preforms, filling equipment for filling products into containers, sealing devices for sealing containers filled with products, pasteurization devices for pasteurizing containers filled with products, labeling machines for labeling containers, and / or cleaning devices for cleaning containers, etc.

[0009] Operating parameters may include heating settings for heating the preform and / or blow molding pressure for forming a container from the preform in a blow mold (e.g., in a blow molding machine), product filling speed and / or filling temperature (e.g., in a filling machine), closure application pressure and / or closure application speed (e.g., in a closure device), pasteurization temperature and / or residence time below pasteurization temperature (e.g., in a pasteurization device), operating speed (e.g., in a labeling machine), heating temperature of the alkali bath and / or residence time of the container in the alkali bath (e.g., in a cleaning device).

[0010] Setting operating parameters can include setting a single operating parameter or setting multiple operating parameters. Multiple operating parameters can be independent of each other, at least partially related to each other, or related to each other.

[0011] Besides the operating parameters of the container handler, other frame conditions can also affect the containers processed in the container handler. For example, the ambient temperature (i.e., frame conditions) during operation in the container handler may have a first temperature in a first operating cycle and a second temperature in a second operating cycle; for example, the difference between the first and second temperatures may be 30% or greater, or it may be smaller. Humidity also affects the results of the container handler. For example, the wall thickness of each batch of preforms (i.e., frame conditions) may vary, thus affecting the wall thickness of the container to be achieved, and for example, for one batch, the desired wall thickness (set value) of the container cannot be achieved by the container handler under the main operating parameters.

[0012] Containers may include bottles or jars made of materials such as PET, glass, or fibrous materials (e.g., wood fibers or other biodegradable fibers).

[0013] The settings for container characteristics can include the container wall thickness or wall thickness distribution, the number of pasteurization units of the product contained in the container, the location of the label on the container, and / or the location of the closure on the container.

[0014] The actual values ​​of container characteristics can be measured as follows: the wall thickness or wall thickness distribution of the container, the number of pasteurization units of the product contained in the container, the location of the label on the container, and / or the location of the closure on the container, etc.

[0015] The expected simulated actual values ​​can be obtained by simulating the process of a container processing machine, wherein the method may include appropriate container processing (e.g., blow molding a container from a preform in a blow mold, the blow mold being heated to a certain temperature and formed by blow molding pressure), and wherein manipulated variables available through intelligent algorithms may be used.

[0016] By comparing the current reward with the expected reward, it can be concluded whether an intelligent algorithm with the provided manipulative variables can adjust the container processor by setting operating parameters so that the measured actual value corresponds to the set value to a predetermined degree.

[0017] For example, such conclusions can be reached by automatically analyzing current and expected rewards. Based on the results of this automatic analysis used to set the operating parameters of the container processor, at least one manipulated variable of the intelligent algorithm can then be optionally modified.

[0018] The method may also include measuring the actual values ​​of the container characteristics to obtain the measured actual values.

[0019] The method may also include simulating actual values ​​to obtain expected simulated actual values. Simulation may include simulating the process of a container processor, wherein the process may include corresponding container processing, and wherein manipulated variables available through intelligent algorithms may be used.

[0020] Automated analysis may include determining that the difference between the current reward and the expected reward is less than a limit, and modifications may include automatically expanding the dataset of the Design of Experiments (DOE) run by automatically launching training runs and automatically deploying intelligent algorithms using further setpoints. For example, the limit could be 50.

[0021] For example, during the operation of a container processor, it can be determined that the current reward decreases or adjusts to a low level during operation. For instance, during operation, the current reward might decrease from a maximum of 250 to below 100, or adjust to a low level below 100. This could mean that the intelligent algorithm cannot provide better results based on at least one of its available manipulated variables. The simulated actual value is expected to meet the objective. The difference between the current reward and the expected reward may be large. Increasing the dataset of the DOE used by the intelligent algorithm may be meaningful. The training dataset of the intelligent algorithm can be further expanded to include setpoints. New training runs can be launched and the intelligent algorithm deployed.

[0022] Automated analysis may include determining that the difference between the current reward and the expected reward is greater than a limit, and / or that the difference between the setpoint and the expected simulated actual value is less than another limit. Modifications may include initiating a new training run using the setpoint of the new experimental design and automatically deploying intelligent algorithms to complete the new experimental design run. For example, the limit could be 50. Another limit could be the transmittance difference. The order of magnitude of the other limit could be given, for example, by the maximum deviation from the setpoint during the DOE run.

[0023] For example, during the operation of a container processor, it can be determined that the current reward decreases or adjusts to a low level during operation. For instance, during operation, the current reward might decrease from a maximum of 250 to below 100, or adjust from a maximum of 250 to a low level below 100. The difference between the actual reward and the expected reward can be large. The expected reward might be good, which could mean that the difference between the setpoint and the expected simulated actual value might be small. The expected reward might be good when the value is greater than 180 out of 250. The difference between the setpoint and the expected simulated actual value can be a difference in transmittance. This order of magnitude can be given, for example, by a small deviation from the setpoint during the DOE run, such as within 10% of the maximum deviation. It might make sense to run a complete new DOE run using all setpoints (e.g., 126 pieces). Based on the current general conditions on the container processor, new training of the intelligent algorithm may be required. Therefore, by launching a new training run and deploying the intelligent algorithm, a new training dataset can be generated at all setpoints.

[0024] Automated analysis may include determining that the difference between the current reward and the expected reward is less than a limit value, and that at least one of the manipulated variables is at the limit of its value range. Modification may include automatically expanding the value range of at least one manipulated variable and automatically adjusting for extrapolation through at least one manipulated variable. For example, the limit value could be 50.

[0025] For example, during the operation of a container processor, it can be determined that the current reward decreases or adjusts to a low level during operation. For instance, during operation, the current reward might decrease from a maximum of 250 to below 100, or adjust from a maximum of 250 to a low level below 100. This could mean that the intelligent algorithm cannot provide better results based on at least one of its available manipulated variables. The expected simulated actual value can be realistic. Allowing a wider range of values ​​for at least one manipulated variable might be meaningful. The settings can be obtained through extrapolation of the manipulated variables.

[0026] The determination of current rewards and expected rewards can be performed during the operation of the container processor. Therefore, the temporal behavior of current and expected rewards can be analyzed. Similarly, the temporal processes of measuring actual values ​​and expecting simulated actual values ​​can be analyzed.

[0027] Automated analysis may include determining that both the current reward and the expected reward are greater than or less than a limit, and that the difference between the setpoint and the expected simulated actual value is less than or greater than another limit. Modification may include the automatic shifting of the basic setpoint defined by the basic manipulated variable, and the automatic assignment of at least one new manipulated variable with an offset to the intelligent algorithm. The other limit may be the difference in transmittance. This order of magnitude may be given, for example, by a small deviation from the setpoint during DOE operation, such as within 10% of the maximum deviation or, for example, within 60% of the maximum deviation.

[0028] This occurs when an intelligent algorithm encounters the limits of the range of values ​​for the manipulated variable. The current reward may be poor, and the measured actual value may differ significantly from the set value. The expected reward may also be poor. The current reward can be less than 100 and at most 250. The measured actual value could be poor transmittance. This order of magnitude can be given, for example, by a small deviation from the set value during DOE operation, such as 60% of the maximum deviation. The expected reward may be less than 100 (at most 250).

[0029] The basic setpoint can be or includes a portion of the setpoints used by the DOE.

[0030] The method may also include automatically calculating the offset. Alternatively, the offset may be manually defined, for example by a human operator of the container processor.

[0031] Automatic offset calculation can include using the limits of the range of values ​​of the basic setpoint as the new basic setpoint.

[0032] In addition, error messages from the container processor can be taken into account during automated analysis.

[0033] Error messages may include faults such as a container bursting in a blow molding machine, a container bursting in a filling machine, or a container cracking white in a container handling machine.

[0034] Automated analysis may include determining that the number of error messages has increased and the current reward exceeds a limit, and modifications may include using new operating parameters and obtaining new manipulated variables from them, as well as automatically transmitting the new manipulated variables to the intelligent algorithm. For example, the limit could be 180 out of 250.

[0035] For example, while an unfavorable material distribution may lead to good actual measurements, errors can still occur, such as too much material at the bottom of the container and too little material at the neck, or vice versa.

[0036] Automated analysis may include determining that the number of error messages has increased and the current reward exceeds a limit, and modifications may include using new operating parameters and initiating a new training run using a new experimental design run setpoint, and automatically deploying intelligent algorithms to complete the new experimental design run. For example, the limit could be 180 out of 250.

[0037] An apparatus is also provided, comprising a computer-readable storage device having instructions stored thereon that, when executed by a processor of the apparatus, cause it to perform the methods described above or further below. Attached Figure Description

[0038] For better understanding and explanation, the accompanying drawings exemplarily illustrate various aspects and / or embodiments of the invention. Wherein:

[0039] Figure 1 A block diagram is shown to describe a method for monitoring and modifying intelligent algorithms used to set operating parameters of a container processor. Detailed Implementation

[0040] Figure 1 A block diagram is shown describing a method for monitoring and modifying the operating parameters 4 of a container processor 5. The intelligent algorithm 1 can determine the setting 3 of one or more operating parameters 4 of the container processor 5 by means of manipulating variables 2. A container 6 can be processed in the container processor 5. A measured actual value 8 of a container characteristic can be determined from the container 6 by measurement 7. The set value 9 of this container characteristic can be known.

[0041] By inputting the actual value 8 (input 11) and the set value 9 (input 12), the current reward 13 can be determined based on the comparison between the set value 9 and the measured actual value 8.

[0042] The expected simulated actual value 10 can be obtained by simulating the process of container processor 5, wherein the process may include corresponding container processing, and wherein the manipulated variable 2 available from smart algorithm 1 can be used.

[0043] By inputting 14 to the setpoint 9 and inputting 15 to the expected simulated actual value 10, the expected reward 16 can be determined by comparing the setpoint 9 with the expected simulated actual value 10.

[0044] Input 17 for the current reward 13 and input 18 for the expected reward 16 can be performed in the automatic analysis 19. Additionally, one or more error messages 20 can be input into the automatic analysis 19. The current reward 13 and expected reward 16, along with one or more possible error messages 20, can be analyzed in the automatic analysis 19.

[0045] Based on the results of automatic analysis 19, at least one of the manipulation variables 2 of intelligent algorithm 1 can be modified for setting the operating parameters 4 of container processor 5.

[0046] For example, container processing machine 5 can be a blow molding machine, in which a preform is formed into a container 6 by blow molding under blow pressure in a heated blow mold. The operating parameters 4 of the blow molding machine can be heating settings for heating the preform and blow molding pressure. Using these operating parameters 4, the container 5 can now be formed in the blow molding machine, where the container characteristic to be measured can be the wall thickness. By measuring the wall thickness of container 5, a measured actual value 8 of the wall thickness is obtained. For containers manufactured using a blow molding machine, a set value 9 of the wall thickness can be specified. Furthermore, an expected simulated actual value 10 of the wall thickness can be determined. If all the operating parameters 4 of the blow molding machine, the assumed thickness of the preform, and, for example, the assumed environmental parameters of the blow molding machine, are theoretically appropriate, then this value can correspond to the theoretical value of the wall thickness that the container should theoretically have.

[0047] The current reward 13 can be determined by comparing the set value 9 of the wall thickness with the actual measured value 8 of the wall thickness. The expected reward 16 can be determined by comparing the set value 9 of the wall thickness with the expected simulated actual value 10 of the wall thickness.

[0048] An automated analysis 19 of the current reward 13 and the expected reward 14 can be performed so that, based on the results of the automated analysis 19, at least one manipulated variable 2 of the intelligent algorithm 1 can be modified 21 for setting 3 of the operating parameter 4, i.e., for setting 3 of the heating temperature and / or blowing pressure of the blow molding machine. If the assumed wall thickness of the preform varies, for example, in the case of different batches and / or if environmental parameters change significantly during the operation of the blow molding machine, it may be necessary to modify at least one manipulated variable 21.

[0049] In addition, one or more error messages 20 from the blow molding machine can be entered into the automatic analysis 19, for example, if the blow molding process is not performed correctly.

[0050] In addition, the automatic analysis 19 can also initiate the modification 22 of the intelligent algorithm 1, in which new training and deployment replace the intelligent algorithm 1 with an optimized algorithm.

Claims

1. A method for monitoring and modifying an intelligent algorithm (1) for setting (3) operating parameters (4) of a container processor (5), wherein, The method includes: -The current reward (13) is determined by comparing the set value (9) of the container characteristic with the measured actual value (8) of the container characteristic. - The expected reward (16) is determined by comparing the set value (9) with the expected simulated actual value (10) of the container characteristics, wherein the expected simulated actual value (10) is based on the intelligent algorithm (1). - Automatically analyze (19) the current reward (13) and the expected reward (14). - Based on the results of the automatic analysis, modify at least one manipulation variable (2) of the intelligent algorithm (1) to set the operating parameters (4) of the container processor (5).

2. The method according to claim 1 further includes measuring (7) the actual value of the container characteristic to obtain the measured actual value (8).

3. The method according to claim 1 or 2 further includes simulating actual values ​​to obtain the expected simulated actual value (10).

4. The method according to any one of claims 1 to 3, wherein, The automatic analysis (19) includes determining that the difference between the current reward (13) and the expected reward (16) is less than a limit value, and wherein the modification (21) includes automatically expanding the dataset of the experimental design run by automatically starting the training run and automatically deploying the intelligent algorithm (1) by using further setpoints.

5. The method according to any one of claims 1 to 3, wherein, The automatic analysis (19) includes determining that the difference between the current reward (13) and the expected reward (16) is greater than a limit value, and the difference between the set value (9) and the expected simulated actual value (10) is less than another limit value, wherein the modification (21) includes completing the new experimental design run by starting a new training run using a set point of a new experimental design run and automatically deploying the intelligent algorithm (1).

6. The method according to any one of claims 1 to 3, wherein, The automatic analysis (19) includes determining that the difference between the current reward (13) and the expected reward (16) is less than a limit value, and that at least one of the manipulated variables (2) is at the limit of its value range, and wherein the modification (21) includes automatically expanding the value range of the at least one manipulated variable (2) and automatically adjusting by extrapolation of the at least one manipulated variable (2).

7. The method according to any one of claims 1 to 3, wherein, The automatic analysis (19) includes determining that the current reward (13) and the expected reward (16) are both greater than or less than a limit value, and that the difference between the set value (9) and the expected simulated actual value (10) is less than or greater than another limit value, and wherein the modification (21) includes the automatic shift of the basic set point defined by the basic manipulated variable, and the automatic assignment of at least one new manipulated variable (2) with an offset to the smart algorithm (1).

8. The method of claim 7, further comprising automatically calculating the offset.

9. The method according to claim 8, wherein, The automatic calculation of the offset includes using the limit of the value range of the basic setpoint as the new basic setpoint.

10. The method according to any one of claims 1 to 9, wherein, It also includes taking into account error messages (20) from the container processor (5) during the automatic analysis (19).

11. The method according to claim 10, wherein, The automatic analysis (19) includes determining that the number of error messages (20) has increased and that the current reward (13) is greater than the limit, and wherein the modification (21) includes using a new operating parameter (4) and obtaining a new manipulation variable (2) therefrom, and automatically transmitting the new manipulation variable (2) to the smart algorithm (1).

12. The method according to claim 10, wherein, The automatic analysis (19) includes determining that the number of error messages (20) has increased and that the current reward (13) is greater than the limit value, and wherein the modification (21) includes using new operating parameters (4) and starting a new training run by using a new experimental design run setpoint and automatically deploying the intelligent algorithm (1) to complete the new experimental design run.

13. An apparatus comprising a computer-readable storage means having instructions stored thereon that, when executed by a processor of the apparatus, cause it to perform the method according to any one of claims 1 to 12.

Images