Controller, control method, and program
By using machine learning generated prediction models in the controller, combining the measurement results of liquid quantity and affecting parameters, adjusting the working conditions of the filling machine, the problem of low liquid quantity deviation control efficiency is solved, and higher prediction accuracy and production are achieved efficiency.
Patent Information
- Application Number
- CN202380035649.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-05-31
- Filing Date
- 2023-05-29
- Publication Date
- 2024-12-03
AI Technical Summary
In the prior art, the deviation control of the liquid quantity is limited, resulting in a decrease in production efficiency.
Using a controller including a acquisition unit, a prediction unit and an adjustment unit, by obtaining the parameters and measurement results of the liquid quantity, using a prediction model generated by machine learning to make predictions, and adjusting the working conditions of the filling machine based on the predicted value, to Reduce the deviation of liquid volume.
It improves the prediction accuracy of liquid volume, significantly reduces the deviation of liquid volume in the container, and improves production efficiency.
Smart Images

Figure CN119072451A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to controllers, control methods, and procedures. Background Technology
[0002] Beverages such as beer, tea, and water are shipped in containers filled with liquid. The amount of liquid in the container is controlled within a certain range. Containers filled with liquid outside this range are judged as defective products. If the number of defective products increases, production efficiency decreases. Therefore, Japanese Patent Application Publication No. 2001-240004 (Patent Document 1) discloses a liquid filling device that controls the filling valve using a new set value calculated based on the measured value when there is a certain difference between the measured value of the liquid in the container and a preset value.
[0003] Existing technical documents
[0004] Patent documents
[0005] Patent Document 1: Japanese Patent Application Publication No. 2001-240004 Summary of the Invention
[0006] The problem that the invention aims to solve
[0007] In the technology disclosed in Patent Document 1, the operating conditions of the filling valve are not changed until the measured value differs from the preset value to a certain extent. Therefore, the reduction in liquid volume deviation is limited. Consequently, the reduction in production efficiency is also limited.
[0008] This disclosure was made in view of the above-mentioned problems, and its purpose is to provide a controller, control method, and program that can reduce the deviation of the liquid volume in a container.
[0009] Methods for solving problems
[0010] One aspect of this disclosure relates to a controller that controls a filling machine that fills a container with liquid. The controller includes an acquisition unit, a prediction unit, and an adjustment unit. The acquisition unit acquires first data and second data, the first data representing the measurement result of one or more parameters affecting the amount of liquid in the container, and the second data representing the measurement result of the amount of liquid in the container. The prediction unit inputs the first data and the second data into a prediction model generated using machine learning with one or more training datasets, thereby obtaining a predicted value for the amount of liquid in the container, wherein the one or more training datasets represent at least one actual value of the parameters and the actual amount of liquid in the container. The adjustment unit adjusts the operating conditions of the filling machine based on the predicted value.
[0011] According to the disclosure, the prediction model accepts not only second data representing the measurement results of the liquid volume as input, but also first data representing the measurement results of more than one parameter affecting the liquid volume in the container, and outputs a predicted value of the liquid volume in the container to be filled in the future. Therefore, the prediction model has high prediction accuracy. As a result, by adjusting the operating conditions of the filling machine based on the predicted value, the deviation of the liquid volume in the container can be reduced.
[0012] In the above disclosure, each of the more than one training datasets represents the average of more than one parameter during a first period, the average amount of liquid in a container filled with liquid during the first period, and the average amount of liquid in a container filled with liquid in a second period following the first period. The first data represents the average of more than one parameter during an object period having the same time length as the first period. The second data represents the average amount of liquid in a container filled with liquid during the object period.
[0013] The measurement results of more than one parameter and the liquid volume contain noise components. However, according to the above disclosure, since a first data representing the average of each of the more than one parameter during the object period and a second data representing the average liquid volume in the container filled with liquid during the object period are used, the influence of the variation of the noise components contained in the measurement results of more than one parameter and the liquid volume can be suppressed. As a result, the prediction accuracy of the prediction model is further improved, and the deviation of the liquid volume in the container is further reduced.
[0014] In the above disclosure, the more than one parameter includes at least one of the following: parameters related to the operation of the filling machine, parameters related to the external environment of the filling machine, and parameters related to the properties of the liquid. According to the above disclosure, the prediction accuracy of the prediction model is further improved.
[0015] In the above disclosure, as a machine learning method, a tree-structured learning algorithm can be used, for example.
[0016] The control method for controlling a filling machine that fills a container with liquid according to one aspect of this disclosure comprises steps 1 to 3. Step 1 involves obtaining first data and second data, wherein the first data represents the measurement result of one or more parameters affecting the amount of liquid in the container, and the second data represents the measurement result of the amount of liquid in the container. Step 2 involves inputting the first data and the second data into a prediction model generated using machine learning with one or more training datasets, thereby obtaining a predicted value for the amount of liquid in the container, wherein the one or more training datasets represent at least one actual value for one or more parameters and the actual amount of liquid in the container. Step 3 involves adjusting the operating conditions of the filling machine based on the predicted value.
[0017] A program according to one aspect of the present disclosure causes a computer to execute the above-described control method. According to these disclosures, it is also possible to reduce the deviation in the amount of liquid in the container.
[0018] Advantages of the Invention
[0019] According to the present disclosure, it is possible to reduce the deviation in the amount of liquid in the container. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 is a schematic diagram showing a system according to an embodiment.
[0021] Figure 2 is a diagram showing an example of the structure of a filling device.
[0022] Figure 3 is a schematic cross-sectional view showing an example of the structure of a filling valve. <00Here is an example illustrating a scenario in which the present invention is applied. Figure 1 This is a schematic diagram illustrating the system of an implementation. Figure 1 The system 1 shown includes a controller 100, a learning device 200, a display device 300, and a filling device 400. The controller 100 and the learning device 200 are connected to the filling device 400 in a communicative manner via a fieldbus 4.
[0035] The filling device 400 fills a container with liquid. For example, the liquid includes beer, effervescent beverages other than beer, fruit juice, tea, water, alcoholic beverages other than beer, condiments, oils, chemicals, pharmaceuticals, cosmetics, etc.
[0036] The filling equipment 400 includes a filling machine 410, a monitoring panel 420, and a liquid volume checking device 430.
[0037] The filling machine 410 fills a container with liquid. A monitoring panel 420 monitors the external environment of the filling machine 410, the state of the liquid, etc. A liquid level checking device 430 measures the amount of liquid in the container. The amount of liquid in the container is sometimes referred to as the "filling amount".
[0038] The learning device 200 operates during the learning phase, generating a prediction model 112 using machine learning with one or more training datasets. These training datasets represent the actual conditions of at least one parameter affecting the amount of liquid in the container and the actual amount of liquid in the container. The learning phase is set during the trial operation of the filling device 400, or for a certain period after the filling device 400 is started.
[0039] One or more parameters, including parameters related to the operation of the filling machine 410, parameters related to the external environment of the filling machine 410, and parameters related to the characteristics of the liquid, are obtained from the filling machine 410 and the monitoring panel 420. The real-time liquid volume in the container is obtained from the liquid volume monitoring device 430. The prediction model 112 generated during the learning phase is provided to the controller 100.
[0040] In the application phase following the learning phase, the controller 100 performs the following steps (1) to (3) as a process for controlling the filling machine 410.
[0041] Step (1): Obtain the first data representing the measurement result of one or more of the above parameters and the second data representing the measurement result of the liquid volume in the container.
[0042] Step (2): By inputting the first data and the second data into the prediction model 112, the predicted value of the liquid volume in the container is obtained.
[0043] Step (3): Adjust the working conditions of filling machine 410 based on the predicted value.
[0044] The display device 300 is connected to the controller 100 and displays a screen representing the control status based on the controller 100.
[0045] In the system 1 according to the present embodiment, the predicted value of the liquid volume is obtained using the prediction model 112, and the operating conditions of the filling machine 410 are adjusted based on the predicted value. Thereby, the deviation of the liquid volume in the container is reduced.
[0046] <B. Filling Equipment>
[0047] Figure 2 is a diagram showing an example of the structure of the filling equipment. As Figure 2 shown, the filling equipment 400 includes a filling machine 410, a monitoring panel 420, a liquid volume inspection device 430, and a cleaning machine 440.
[0048] The cleaning machine 440 cleans the container 2. The container 2 is, for example, a tank, a bottle, a PET (polyethylene terephthalate) bottle, a barrel, a paper package, or a bag. The container 2 cleaned by the cleaning machine 440 is conveyed to the filling machine 410.
[0049] The filling machine 410 includes: a rotating member 414 that is driven to rotate; a plurality of filling valves 412 arranged at equal intervals on the circumference of the rotating member 414; and a control panel 416 that serves as the control panel of the filling machine 410.
[0050] Figure 3 is a schematic cross-sectional view showing an example of the structure of the filling valve. The filling valve 412 has a cylinder 451, a flow meter 452, a liquid delivery pipe 453, a feed tank 454, and a nozzle portion 455. The feed tank 454 and the nozzle portion 455 are connected by the liquid delivery pipe 453. A flow meter 452 is disposed midway in the liquid delivery pipe 453. The feed tank 454 stores the liquid supplied from a buffer tank (not shown) via the liquid delivery pipe 456. The nozzle portion 455 has a flow path for allowing the liquid supplied from the feed tank 454 via the liquid delivery pipe 453 to flow to the filling port, and a liquid valve 457 disposed in the flow path.
[0051] The control panel 416 (refer to Figure 2 ) opens and closes the liquid valve 457. Thereby, the liquid is filled into the container 2 held by the container holding portion 458 from the filling port. The control panel 416 closes the liquid valve 457 at a timing when the measurement amount based on the flow meter 452 reaches a preset target flow rate.
[0052] The control panel 416 can receive a correction value from the controller 100. When the control panel 416 receives the correction value, it corrects the target flow rate by the amount of the correction value.
[0053] The liquid-filled container 2 is discharged from the container holding part 458 into a can supply chain (not shown) and conveyed to a sealing device (not shown), and is sealed by a can lid, a crown cap, a lid, or other methods corresponding to the container.
[0054] The control panel 416 outputs values of parameters related to the operation of the filling machine 410. Parameters related to the operation of the filling machine 410 include, for example, the measured quantity based on the flowmeter 452, the time difference from when the liquid valve 457 is opened until the flowmeter 452 starts counting ("liquid valve opening delay time"), the time difference from when the liquid valve 457 is closed until the flowmeter 452 ends counting ("liquid valve closing delay time"), and the like.
[0055] Figure 2 The illustrated monitoring panel 420 monitors the external environment of the filling machine 410, the state of the liquid, and the like. Specifically, the monitoring panel 420 measures parameters related to the external environment of the filling machine 410, parameters related to the characteristics of the liquid, and the like, and outputs the measurement results to the controller 100. Parameters related to the external environment of the filling machine 410 include the external air temperature, the temperature of the rinser water used by the rinsing machine 440, and the like. Parameters related to the characteristics of the liquid include the liquid temperature and the like.
[0056] The liquid quantity inspection device 430 measures the quantity of liquid in the liquid-filled container 2. The liquid quantity inspection device 430 outputs the measurement results of the liquid quantity for each container 2. The liquid quantity is represented by volume, weight, liquid level height, and the like. As the liquid quantity inspection device 430, a conventionally known inspection device can be applied. For example, the liquid quantity inspection device 430 is composed of an X-ray linear sensor, a digital camera, a video camera, a linear sensor camera, a weight checker, and the like. When the container 2 is an opaque can, the liquid quantity inspection device 430 is composed of an X-ray linear sensor. For example, the liquid quantity inspection device 430 outputs the difference between the liquid level height in the container 2 measured by the X-ray linear sensor and a predetermined reference height as the liquid quantity. When the container 2 is transparent like a bottle, the liquid quantity inspection device 430 can also measure the liquid quantity using images obtained by a digital camera, a video camera, or the like.
[0057] After a time Δt has elapsed since the filling machine 410 fills the container 2 with liquid, the liquid quantity inspection device 430 measures the liquid quantity of the container 2.
[0058] <C. Hardware Structure of the Controller>
[0059] Figure 4This is a block diagram illustrating an example of the hardware structure of the controller according to an embodiment. The controller 100 is, for example, composed of a PLC (Programmable Logic Controller), an IPC (Industrial Personal Computer), etc. Figure 4 As shown, the controller 100 includes processors such as CPU (Central Processing Unit) and MPU (Micro-Processing Unit) 101, chipset 102, main memory 103, memory 104, fieldbus controller 105, network controller 106, USB controller 107, and memory card interface 108.
[0060] The processor 101 reads various programs stored in the memory 104, expands and executes them in the main memory 103, thereby realizing control operations for controlling the controlled object. The chipset 102 controls data transmission between the processor 101 and various components.
[0061] The control program 111 and the prediction model 112 generated by the learning device 200 are stored in the memory 104. The control program 111 contains a set of commands for performing operations such as collecting various data from the filling device 400, controlling the filling machine 410, and outputting data to the filling device 400.
[0062] The fieldbus controller 105 controls the data exchange between itself and the filling device 400, which is the object of control, via the fieldbus 4.
[0063] The network controller 106 controls the data exchange between various devices via a host network.
[0064] USB controller 107 controls data exchange with external devices (e.g., display device 300) connected via USB.
[0065] The memory card interface 108 is configured to be able to load and unload the memory card 120, write data to the memory card 120, and read various data (control program 111, etc.) from the memory card 120.
[0066] Figure 4The structure example of providing the required processing by executing a program with the processor 101 is shown. However, one or more dedicated hardware circuits (e.g., ASIC or FPGA, etc.) can also be used to implement a part or all of the provided processing. Alternatively, hardware following a general architecture (e.g., an industrial personal computer based on a general personal computer) can also be used to implement the main part of the controller 100. In this case, virtualization technology can also be used to execute multiple OSs with different uses in parallel, and execute the required applications on each OS. Additionally, a database management system, which is software for database management, can be stored in the memory 104, and this database management system can be utilized for information processing in the controller 100.
[0067] <D. Hardware Structure of the Learning Device>
[0068] Figure 5 It is a block diagram showing an example of the hardware structure of the learning device of the embodiment. The learning device 200 is constituted by, for example, an IPC, a general computer device, etc. As Figure 5 shown, the learning device 200 includes a processor 201 such as a CPU and MPU, a chipset 202, a main memory 203, a memory 204, a field bus controller 205, a network controller 206, a USB controller 207, and a memory card interface 208.
[0069] The processor 201 reads out various programs stored in the memory 204, expands and executes them in the main memory 203, thereby implementing control operations for controlling a control object. The chipset 202 controls data transfer, etc. between the processor 201 and each component.
[0070] A collection program 211 for collecting various data from the filling device 400 and a model generation program 212 are stored in the memory 204.
[0071] The field bus controller 205 controls data exchange with the filling device 400 via the field bus 4.
[0072] The network controller 206 controls data exchange with various devices via the network. The USB controller 207 controls data exchange with an external device connected via USB.
[0073] The memory card interface 208 is configured to be able to load and unload the memory card 220, and be able to write data to the memory card 220 and read out various data from the memory card 220.
[0074] Figure 5The structural example of providing the required processing by executing a program by the processor 201 is shown. However, one or more dedicated hardware circuits (e.g., ASIC or FPGA, etc.) can also be used to implement a part or all of the provided processing. Alternatively, hardware following a general architecture (e.g., an industrial personal computer based on a general personal computer) can also be used to implement the main part of the learning device 200. In this case, virtualization technology can also be used to execute multiple OSs with different uses in parallel, and execute the required applications on each OS.
[0075] <E. Functional Structure of Learning Device>
[0076] Figure 6 is a diagram showing an example of the functional structure of the learning device. As Figure 6 shown, the learning device 200 includes a collection unit 21 and a model generation unit 22. The collection unit 21 is implemented by executing a collection program 211 by the processor 201 as Figure 5 shown. The model generation unit 22 is implemented by executing a model generation program 212 by the processor 201.
[0077] The collection unit 21 collects, for each container, the values of one or more parameters that affect the liquid amount in the container 2 and the value of the liquid amount in the container 2 from the filling device 400.
[0078] The model generation unit 22 generates a prediction model 112 through machine learning using one or more training data sets, and the one or more training data sets represent the actual situations of one or more parameters and the liquid amount in the container 2.
[0079] The model generation unit 22 extracts one or more training data sets from the values of one or more parameters and the liquid amount of each container 2 collected by the collection unit 21.
[0080] Figure 7 is a diagram showing the data collected by the collection unit. As [[ID=]24] Figure 7 shown, the collection unit 21 collects, for each of the multiple filling valves 412, the values of one or more parameters related to the filling valve 412 from the filling machine 410 and the monitoring panel 420. The collection unit 21 collects, for each of the multiple filling valves 412, the value of the liquid amount of the container 2 filled with liquid by the filling valve 412 from the liquid amount inspection device 430. As referred to Figure 2 Figure 2As explained, the time from when the filling machine 410 fills the container 2 with liquid until the liquid volume in the container 2 is measured is Δt. That is, for a given container 2, the time difference between the timing of collecting the values of one or more parameters when the container 2 is filled with liquid and the timing of collecting the value of the liquid volume in the container 2 is Δt. Therefore, as a preprocessing step for extracting the training dataset, the model generation unit 22 performs a process that associates the values of one or more parameters with the value of the liquid volume for each container 2.
[0081] Figure 8 This is a diagram showing the training dataset. Training datasets are generated separately for each of the multiple filling valves 412. The training datasets represent explanatory variables and target variables. The explanatory variables include: the average of one or more parameters related to the corresponding filling valve 412 during a first period Ta with a certain time length T1; and the average amount of liquid filled into the container 2 using the corresponding filling valve 412 during the first period Ta. The target variable includes the average amount of liquid filled into the container 2 using the corresponding filling valve 412 during a second period Tb (with a time length T2) after a predetermined time T3 elapsed from the end time of the first period Ta. The model generation unit 22 extracts one or more training datasets with different start times for the first period Ta. Furthermore, the model generation unit 22 can extract multiple training datasets either in a way that a portion of the first period Ta of a certain training dataset overlaps with the first period Ta of other training datasets, or in a way that the first periods Ta do not overlap.
[0082] The time lengths T1, T2, and T3 are appropriately set according to the filling equipment 400, the type of liquid, and the type of container 2. Time length T1 is, for example, 10 seconds, 30 seconds, 1 minute, 5 minutes, 10 minutes, 30 minutes, 1 hour, etc. Time length T2 is the sum of time length T1 and time T3. Time T3 is longer than the time required to determine the correction value for the target flow rate and to respond to the filling machine 410, for example, 30 seconds, 45 seconds, 1 minute, etc.
[0083] The model generation unit 22 generates a prediction model 112 by performing machine learning using one or more training datasets. The model generation unit 22 generates a single prediction model 112 shared by multiple filling valves 412. Alternatively, the model generation unit 22 may generate a separate prediction model 112 for each of the multiple filling valves 412.
[0084] The generated prediction model 112 is sent to the controller 100. For example, the model generation unit 22 may also send the generated prediction model 112 to the controller 100 via the network controller 206 or the USB controller 207. Alternatively, the prediction model 112 may be saved in the memory card 220 via the memory card interface 208. In this case, the user can install the prediction model 112 on the controller 100 by connecting the memory card 220 to the controller 100.
[0085] The prediction model 112 receives the input of the first data representing the measurement results of one or more parameters affecting the liquid volume in the container 2 and the second data representing the measurement results of the liquid volume in the container 2, and outputs a predicted value of the liquid volume in the container 2 to be filled with liquid in the future.
[0086] As the learning algorithm, well-known algorithms such as support vector regression, deep learning, and tree-structured learning algorithms (e.g., random forest) can be adopted. An algorithm with good prediction accuracy for using the prediction model 112 can be used. The prediction accuracy is represented by, for example, the mean square error of the difference between the predicted value and the measured value.
[0087] The prediction accuracy was confirmed for the filling device 400 using beer as the liquid. The prediction accuracy of the prediction model 112 generated by machine learning using the random forest as the tree-structured learning algorithm is superior to that when using support vector regression and deep learning. Therefore, it is preferable to use a tree-structured learning algorithm (e.g., random forest).
[0088] In the case of machine learning using random forest, the capacity of the prediction model 112 depends on the number of trees and the depth of the trees. Therefore, machine learning is performed by restricting the number of trees and the depth of the trees to reduce the capacity of the prediction model 112. As a result, the time required for predicting the liquid volume using the prediction model 112 is shortened.
[0089] <F. Functional Structure of the Controller>
[0090] Figure 9 It is a diagram showing an example of the functional structure of the controller. As Figure 9 shown, the controller 100 includes an acquisition unit 10, a prediction unit 13, an adjustment unit 14, and a screen data generation unit 17. The acquisition unit 10, the prediction unit 13, the adjustment unit 14, and the screen data generation unit 17 are implemented by Figure 4 the processor 101 executing the control program 111 as shown.
[0091] The acquisition unit 10 acquires first data and second data. The first data represents the measurement result of one or more parameters affecting the amount of liquid in the container 2, and the second data represents the measurement result of the amount of liquid in the container 2. The acquisition unit 10 acquires the first data and second data for each of the multiple filling valves 412. That is, the first data represents the measurement result of one or more parameters affecting the amount of liquid in the container 2 filled with liquid using the corresponding filling valve 412. The second data represents the measurement result of the amount of liquid in the container 2 filled with liquid using the corresponding filling valve 412. Figure 9 As shown, the acquisition unit 10 includes a collection unit 11 and an averaging unit 12.
[0092] The collection unit 11 collects measurements of one or more parameters affecting the amount of liquid in each container 2 from the filling device 400, as well as measurements of the amount of liquid in each container 2.
[0093] The averaging unit 12 calculates the average of one or more parameters related to each of the plurality of filling valves 412 for each target period, and calculates the average amount of liquid in the container 2 filled with liquid using the filling valve 412 during the target period. The target period has... Figure 8 The duration T1 of the first period Ta is shown.
[0094] For each of the multiple filling valves 412, the averaging unit 12 outputs first data as the average of one or more parameters during the target period, and outputs second data as the average of the amount of liquid in the container 2 filled with liquid during the target period.
[0095] For each of the multiple filling valves 412, the prediction unit 13 inputs the first data and the second data into the prediction model 112, thereby calculating the predicted value of the liquid volume in the container 2. Specifically, the prediction unit 13 calculates the predicted value of the liquid volume in the container 2 to be filled in the future during a correction period after time T3 has elapsed. The correction period has Figure 8 The time length T2 of the second period Tb is shown.
[0096] The adjustment unit 14 adjusts the operating conditions of the filling machine 410 based on predicted values for each of the multiple filling valves 412. Specifically, the adjustment unit 14 adjusts the operating conditions of the filling machine 410 based on predicted values to bring the liquid volume in the container 2 close to a target value. In other words, the adjustment unit 14 adjusts the operating conditions of the filling machine 410 based on predicted values to keep the liquid volume in the container 2 constant. Figure 9 As shown, the adjustment unit 14 includes a deviation calculation unit 15 and a correction value output unit 16.
[0097] The deviation calculation unit 15 calculates the deviation between the predicted value and a pre-determined target value. The correction value output unit 16 calculates a correction value for the target flow rate to make the deviation approach 0. The correction value output unit 16 outputs the calculated correction value to the controller 100.
[0098] For example, the correction value output unit 16 may calculate the correction value using a function that takes the deviation as an explanatory variable and the correction value as a target variable. This function is pre-created based on experiments or simulations.
[0099] The screen data generation unit 17 generates screen data representing the trends of the first data and the second data acquired by the acquisition unit 10, the trend of the predicted value obtained by the prediction unit 13, etc. The generated screen data is sent to the display device 300 via the USB controller 107. As a result, the screen represented by the screen data is displayed on the display device 300.
[0100] <G. Processing Example of the Controller>
[0101] Figure 10 is a diagram for explaining the processing of the controller. As Figure 10 shown, the controller 100 repeatedly executes the same processing for each correction cycle. The time length of the correction cycle is the same as the time length T2 of the second period Tb shown in Figure 8 shown. In the Figure 10 shown example, the correction cycle starts at time t1, and the next correction cycle starts at time t2 which is after a time length T2 has elapsed since time t1.
[0102] The processor 101 acquires the first data, which represents the average of one or more parameters collected during an object period from the start timing of the correction cycle until a time length T1 has elapsed. Moreover, the processor 101 acquires the second data, which represents the measurement result of the liquid amount in the container 2 filled with liquid during the object period. As explained in reference to Figure 2 the time from when the liquid is filled into the container 2 by the filling machine 410 until the liquid amount in the container 2 is measured is Δt. Therefore, the time required to acquire the first data and the second data is the sum of the time length T1 and Δt. The processor 101 acquires the first data and the second data for each of the plurality of filling valves 412.
[0103] Next, the processor 101 obtains a predicted value of the liquid amount in the container 2 filled with liquid in the next correction cycle by inputting the first data and the second data into the prediction model 112. Further, in the case where the prediction model 112 is generated for each filling valve 412, the processor 101 inputs the first data and the second data corresponding to the filling valve 412 into the prediction model 112 corresponding to the filling valve 412 for each of the plurality of filling valves 412. Moreover, the processor 101 calculates a correction value corresponding to the deviation between the predicted value and a predetermined target value. The processor 101 calculates a correction value for each of the plurality of filling valves 412.
[0104] Then, the processor 101 stands by until the output timing of the correction value is reached. The output timing of the correction value is a time point that is a predetermined time backward from the start timing of the next correction cycle. This predetermined time is determined in consideration of the time required for the output process of the correction value and the reflection process of the correction value in the filling machine 410.
[0105] When the output timing of the correction value is reached, the processor 101 outputs the calculated correction value to the filling machine 410. The processor 101 associates the valve ID identifying each filling valve 412 with the correction value calculated for the filling valve 412 and outputs it to the filling machine 410. The control panel 416 of the filling machine 410 corrects the target flow rate for each of the plurality of filling valves 412 according to the correction value received from the controller 100.
[0106] <H. Deviation of liquid amount>
[0107] Figure 11 is a diagram showing the deviation of the liquid amount. In Figure 11 the upper part shows the frequency distribution of the liquid amount when the operation of the adjustment unit 14 is turned off, that is, the frequency distribution of the liquid amount when the target flow rate is not corrected. In Figure 11 the lower part shows the frequency distribution of the liquid amount when the operation of the adjustment unit 14 is turned on, that is, the frequency distribution of the liquid amount when the target flow rate is corrected according to the predicted value. As Figure 11 shown, by correcting the target flow rate according to the predicted value, the deviation of the liquid amount is reduced.
[0108] <I. Example of screen>
[0109] Figure 12 is a diagram showing an example of a screen displayed on the display device based on the screen data generated by the screen data generation unit. In Figure 12 it shows Figure 8 the screen 50 when the time lengths T1 and T2 shown are 1 minute and 2 minutes, respectively. That is, the time of the correction cycle is 2 minutes (120 seconds). As Figure 12As shown, screen 50 includes chart 51, selection bars 52a and 52b, buttons 53a, 53b and 58, status indicator lights 54, display bar 55, and switches 56 and 57.
[0110] Enter the valve number of the selected filling valve 412 among multiple filling valves 412 in selection fields 52a and 52b.
[0111] Chart 51 contains four line segments 51a-51d. The horizontal axis of Chart 51 represents time. Line segments 51a-51d are lines connecting the points of each calibration cycle. Each point on line segment 51a represents the target value of the liquid volume set for the corresponding calibration cycle (referred to as the "target liquid volume" in the chart). Each point on line segment 51b represents the average measurement result of the liquid volume of multiple containers 2 filled with liquid using all filling valves 412 during the target period of the corresponding calibration cycle. Each point on line segment 51c represents the average measurement result of the liquid volume of multiple containers 2 filled with liquid using the filling valves 412 input to selection bar 52a during the target period of the corresponding calibration cycle. Each point on line segment 51d represents the average measurement result of the liquid volume of multiple containers 2 filled with liquid using the filling valves 412 input to selection bar 52b during the target period of the corresponding calibration cycle. Furthermore, in Figure 12 In the example shown, the liquid volume represents the difference between the liquid level in container 2 and the reference height.
[0112] By entering the valve number of the desired filling valve 412 in selection fields 52a and 52b, the user can confirm the change in the amount of liquid in the container 2 filled with liquid using the filling valve 412 in Figure 51.
[0113] In response to the pressing of button 53a, the screen data generation unit 17 generates detailed screen data containing information about the filling valve 412 associated with the valve number entered into selection bar 52a. In response to the pressing of button 53b, the screen data generation unit 17 generates detailed screen data containing information about the filling valve 412 associated with the valve number entered into selection bar 52b. The detailed screen data, for example, for the corresponding filling valve 412, shows the shift in values of parameters indicating operation, the average shift in the amount of liquid filled in the multiple containers 2 during the calibration cycle, the shift in predicted values, and the shift in calibration values, etc.
[0114] The status indicator light group 54 includes multiple lights indicating the operating status and communication connection status of the filling equipment 400. The display bar 55 displays the type of filling machine 410, the target value of the liquid volume (target liquid volume), and the calibration cycle.
[0115] The switch 56 is operated to switch between the effective and ineffective operations of the acquisition unit 10 and the prediction unit 13. By switching the switch 56 to the open state, the acquisition unit 10 and the prediction unit 13 start operating.
[0116] The switch 57 is operated to switch between the effective and ineffective operations of the output operation of the correction value based on the adjustment unit 14. By switching the switch 57 to the open state, the adjustment unit 14 starts operating.
[0117] The screen data generation unit 17 generates screen data including a detailed trend graph related to the liquid amount of the container 2 in response to the button 58 being pressed. The trend graph shows, for example, the change in the measurement result of the liquid amount, the change in the predicted value, the change in the correction value, and the change in one or more parameters affecting the liquid amount of the container 2 for a plurality of filling valves 412.
[0118] The user can grasp the status of the filling device 400 in detail by confirming the screens displayed by operating the operation screen 50 and various buttons.
[0119] <J. Advantages>
[0120] As described above, the controller 100 of the present embodiment controls the filling machine 410 that fills the liquid into the container 2. The controller 100 includes an acquisition unit 10, a prediction unit 13, and an adjustment unit 14. The acquisition unit 10 acquires the first data and the second data, where the first data represents the measurement result of one or more parameters affecting the liquid amount in the container 2, and the second data represents the measurement result of the liquid amount in the container 2. The prediction unit 13 inputs the first data and the second data into the prediction model 112 generated by machine learning using one or more training data sets, thereby obtaining a predicted value of the liquid amount in the container 2 to which the liquid will be filled in the future. The one or more training data sets represent at least the actual situation of one or more parameters and the actual situation of the liquid amount in the container 2. The adjustment unit 14 adjusts the operating conditions of the filling machine 410 based on the predicted value to make the liquid amount in the container 2 constant.
[0121] Thus, the predicted value of the liquid amount is obtained using the prediction model 112, and the operating conditions of the filling machine 410 are adjusted based on the predicted value to make the liquid amount in the container constant. The prediction model 112 not only accepts the input of the second data representing the measurement result of the liquid amount, but also accepts the input of the first data representing the measurement result of one or more parameters affecting the liquid amount in the container 2. Therefore, the prediction accuracy of the prediction model 112 is high. As a result, the deviation of the liquid amount in the container 2 can be reduced.
[0122] Each of the one or more training data sets represents the average of each of the one or more parameters within the first period Ta, the average of the liquid volume within the container 2 filled with liquid during the first period Ta, and the average of the liquid volume within the container 2 filled with liquid during the second period Tb that is later than the first period Ta. The first data represents the average of each of the one or more parameters within an object period having the same time length T1 as the first period Ta. The second data represents the average of the liquid volume within the container 2 filled with liquid during the object period.
[0123] Noise components are included in the measurement results of the one or more parameters and the liquid volume. By using the first data representing the average of each of the one or more parameters within the object period and the second data representing the average of the liquid volume within the container 2 filled with liquid during the object period, the influence of the variation of the noise components is suppressed. As a result, the prediction accuracy of the prediction model 112 is further improved. Consequently, the deviation of the liquid volume within the container 2 can be further reduced.
[0124] <K. Variation Example>
[0125] Alternatively, after generating the prediction model 112, the learning device 200 may disconnect from the fieldbus 4.
[0126] The learning device 200 may also generate the prediction model 112 for each variety of the filling machine 410, the liquid, and the container 2. The multiple generated prediction models 112 are installed in the controller 100. In this case, the prediction unit 13 of the controller 100 selects the prediction model 112 corresponding to the varieties of the filling machine 410, the liquid, and the container 2 from the multiple prediction models 112, and uses the selected prediction model 112 to predict the liquid volume of the container 2 to be filled with liquid in the future.
[0127] In addition, the learning device 200 may generate a common prediction model 112 regardless of the variety of the liquid. In this case, the learning device 200 may generate the prediction model 112 by machine learning using multiple training data sets when using multiple varieties of liquids. As a result, the controller 100 can use the common prediction model 112 regardless of the variety of the liquid. Consequently, in the memory 104 of the controller 100, the capacity for storing the prediction model 112 can be reduced.
[0128] In the above description, the learning device 200 includes the collection program 211. However, the learning device 200 may not include the collection program 211, but retrieve data from a database of an external hardware device, thereby generating one or more training data sets.
[0129] In a database of an externally constructed hardware device, in the learning stage, values of one or more parameters collected from the filling machine 410 and the monitoring panel 420 for each container 2 and values of the liquid volume of each container 2 collected from the liquid volume inspection device 430 are accumulated. Accumulation of data in the database is performed by an external hardware device. Alternatively, accumulation of data in the database may also be performed by the controller 100.
[0130] In the above description, the controller 100 is constituted by one device including the processor 101. However, the controller 100 may also be constituted by a plurality of devices (e.g., PLC, IPC) each including a processor. In this case, Figure 4 the shown control program 111 is divided into a plurality of partial programs, and the plurality of partial programs are respectively installed in corresponding devices among the plurality of devices.
[0131] For example, the controller 100 may also be constituted by one PLC and one IPC. The PLC performs: acquiring first data and second data; sending the first data and the second data to the IPC and acquiring a predicted value from the IPC; and adjusting the operating conditions of the filling machine 410 based on the predicted value. The IPC includes a prediction model 112, and by inputting the first data and the second data received from the PLC into the prediction model 112, outputs a predicted value of the liquid volume in the container into which liquid will be filled in the future.
[0132] In the above description, the learning device 200 is separate from the controller 100. However, the controller 100 may also have the function of the learning device 200 and generate the prediction model 112.
[0133] <I. Note>
[0134] As described above, the present embodiment includes the following disclosure.
[0135] (Structure 1)
[0136] A controller (100) that controls a filling machine (410) for filling a container (2) with a liquid, wherein the controller (100) includes:
[0137] an acquisition unit (10, 101) that acquires first data and second data, the first data representing a measurement result of one or more parameters that affect the liquid volume in the container (2), and the second data representing a measurement result of the liquid volume in the container (2);
[0138] The prediction unit (13, 101) inputs the first data and the second data into a prediction model (112) generated using machine learning with one or more training datasets, thereby determining a predicted value for the amount of liquid in the container (2), wherein the one or more training datasets at least represent the actual values of the one or more parameters and the actual amount of liquid in the container (2); and
[0139] The adjustment unit (14, 101) adjusts the operating conditions of the filling machine based on the predicted value.
[0140] (Structure 2)
[0141] According to the controller (100) of structure 1, each of the more than one training datasets represents the average of the more than one parameter during the first period (Ta), the average amount of liquid in the container (2) filled with liquid during the first period (Ta), and the average amount of liquid in the container (2) filled with liquid during the second period (Tb) after the first period (Ta).
[0142] The first data refers to the average of one or more parameters over an object period having the same time length as the first period (Ta).
[0143] The second data represents the average amount of liquid in the container (2) during the period in which the liquid was filled.
[0144] (Structure 3)
[0145] According to the controller (100) of structure 1 or 2, the more than one parameter includes at least one of the following: a parameter related to the operation of the filling machine (410), a parameter related to the external environment of the filling machine (410), and a parameter related to the properties of the liquid.
[0146] (Structure 4)
[0147] The controller (100) according to any one of structures 1 to 3, wherein the machine learning uses a tree-structured learning algorithm.
[0148] (Structure 5)
[0149] A control method for controlling a filling machine (410) that fills a container (2) with liquid, wherein the control method comprises the following steps:
[0150] Obtain first data and second data, wherein the first data represents the measurement result of one or more parameters affecting the amount of liquid in the container (2), and the second data represents the measurement result of the amount of liquid in the container (2);
[0151] The first data and the second data are input into a prediction model (112) generated using machine learning with more than one training dataset, thereby obtaining a predicted value for the amount of liquid in the container (2), wherein the more than one training dataset represents at least the actual situation of the more than one parameter and the actual situation of the amount of liquid in the container (2); and
[0152] The operating conditions of the filling machine (410) are adjusted based on the predicted values.
[0153] (Structure 6)
[0154] A program for controlling a filling machine (410) that fills a container (2) with liquid, the program causing a computer to perform the following steps:
[0155] Obtain first data and second data, wherein the first data represents the measurement result of one or more parameters affecting the amount of liquid in the container (2), and the second data represents the measurement result of the amount of liquid in the container (2);
[0156] The first data and the second data are input into a prediction model (112) generated using machine learning with more than one training dataset, thereby obtaining a predicted value for the amount of liquid in the container (2), wherein the more than one training dataset represents at least the actual situation of the more than one parameter and the actual situation of the amount of liquid in the container (2); and
[0157] The operating conditions of the filling machine (410) are adjusted based on the predicted values.
[0158] (Structure 7)
[0159] According to any one of the structures 1 to 4, the controller (100) adjusts the operating conditions of the filling machine (410) so that the amount of liquid in the container (2) is close to the target value.
[0160] (Composition 8)
[0161] According to any one of structures 1 to 4, the controller (100) wherein the adjustment unit (14, 101) adjusts the operating conditions of the filling machine (410) to keep the amount of liquid in the container (2) constant.
[0162] Embodiments of the present invention have been described, but should be considered as illustrative rather than restrictive in all respects. The scope of the invention is defined by the claims and is intended to include all modifications within the scope and equivalent meaning of the claims.
[0163] Label Explanation
[0164] 1: System; 2: Container; 4: Fieldbus; 10: Acquisition Unit; 11, 21: Collection Unit; 12: Averaging Unit; 13: Prediction Unit; 14: Adjustment Unit; 15: Deviation Calculation Unit; 16: Correction Value Output Unit; 17: Provision Unit; 22: Model Generation Unit; 50: Screen; 51: Chart; 51a-51d: Polyline; 52a, 52b: Selection Bar; 53a, 53b, 58-62: Button; 54: Status Indicator Light Group; 55: Display Bar; 56, 57: Switch; 100: Controller; 101, 201: Processor; 102, 202: Chipset; 103, 203: Main Memory; 104, 204: Memory; 105, 205: Fieldbus Controller; 1 06, 206: Network controller; 107, 207: USB controller; 108, 208: Memory card interface; 111: Control program; 112: Predictive model; 120, 220: Memory card; 200: Learning device; 211: Collection program; 212: Model generation program; 300: Display device; 400: Filling equipment; 410: Filling machine; 412: Filling valve; 414: Rotating component; 416: Control panel; 420: Monitoring panel; 430: Liquid volume checking device; 440: Cleaning machine; 451: Cylinder; 452: Flow meter; 453, 456: Liquid delivery pipe; 454: Feed tank; 455: Nozzle; 457: Liquid valve; 458: Container handle.
Claims
1. A controller for controlling a filling machine for filling a container with a liquid, wherein: The controller has: an acquisition unit that acquires first data and second data, the first data indicating a measurement result of one or more parameters that affect the amount of liquid in the container, and the second data indicating a measurement result of the amount of liquid in the container; a prediction unit that inputs the first data and the second data into a prediction model generated by machine learning using one or more training data sets, thereby obtaining a predicted value of the amount of liquid in the container, wherein the one or more training data sets at least represent actual conditions of the one or more parameters and actual conditions of the amount of liquid in the container; as well as An adjustment unit adjusts an operating condition of the filling machine based on the predicted value.
2. The controller according to claim 1, wherein: The one or more training data sets each represent an average of the one or more parameters in a first period, an average of the amount of liquid in the container filled with the liquid in the first period, and an average of the amount of liquid in the container filled with the liquid in a second period later than the first period, The first data represents an average of each of the one or more parameters in a target period having the same time length as the first period. The second data indicates an average of the amount of liquid in the container filled with the liquid during the target period.
3. The controller according to claim 1 or 2, wherein: The one or more parameters include at least one of a parameter related to the operation of the filling machine, a parameter related to the external environment of the filling machine, and a parameter related to the characteristics of the liquid.
4. The controller according to any one of claims 1 to 3, wherein: The machine learning uses a tree-structured learning algorithm.
5. A control method for controlling a filling machine for filling a liquid into a container, wherein: The control method comprises the following steps: acquiring first data and second data, wherein the first data represents a measurement result of one or more parameters affecting the amount of liquid in the container, and the second data represents a measurement result of the amount of liquid in the container; Inputting the first data and the second data into a prediction model generated by machine learning using one or more training data sets, thereby obtaining a predicted value of the amount of liquid in the container, wherein the one or more training data sets at least represent actual conditions of the one or more parameters and actual conditions of the amount of liquid in the container; and An operating condition of the filling machine is adjusted based on the predicted value.
6. A program for controlling a filling machine for filling a container with a liquid, the program causing a computer to execute the following steps: acquiring first data and second data, wherein the first data represents a measurement result of one or more parameters affecting the amount of liquid in the container, and the second data represents a measurement result of the amount of liquid in the container; Inputting the first data and the second data into a prediction model generated by machine learning using one or more training data sets, thereby obtaining a predicted value of the amount of liquid in the container, wherein the one or more training data sets at least represent actual conditions of the one or more parameters and actual conditions of the amount of liquid in the container; and An operating condition of the filling machine is adjusted based on the predicted value.
Citation Information
Patent Citations
Filling quantity regulating method in liquid filling apparatus
JP2001240004A
Method and device for filling target containers
CN101625252A
Pressure filling device
JP1990290236A
Liquid filling apparatus and liquid filling method
JP2013001417A
System and method for automated adjustment of a container filling machine
WO2019048051A1
