WO2024075358A1 - Blow conditions adjustment device, machine learning device, inference device, information processing method, machine learning method and inference method - Google Patents

Blow conditions adjustment device, machine learning device, inference device, information processing method, machine learning method and inference method Download PDF

Info

Publication number
WO2024075358A1
WO2024075358A1 PCT/JP2023/025646 JP2023025646W WO2024075358A1 WO 2024075358 A1 WO2024075358 A1 WO 2024075358A1 JP 2023025646 W JP2023025646 W JP 2023025646W WO 2024075358 A1 WO2024075358 A1 WO 2024075358A1
Authority
WO
WIPO (PCT)
Prior art keywords
blow
information
conditions
target
condition
Prior art date
Application number
PCT/JP2023/025646
Other languages
French (fr)
Japanese (ja)
Inventor
洋一 田所
大紀 渡邉
Original Assignee
東洋製罐株式会社
東洋製罐グループホールディングス株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 東洋製罐株式会社, 東洋製罐グループホールディングス株式会社 filed Critical 東洋製罐株式会社
Publication of WO2024075358A1 publication Critical patent/WO2024075358A1/en

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B29—WORKING OF PLASTICS; WORKING OF SUBSTANCES IN A PLASTIC STATE IN GENERAL
    • B29C—SHAPING OR JOINING OF PLASTICS; SHAPING OF MATERIAL IN A PLASTIC STATE, NOT OTHERWISE PROVIDED FOR; AFTER-TREATMENT OF THE SHAPED PRODUCTS, e.g. REPAIRING
    • B29C49/00—Blow-moulding, i.e. blowing a preform or parison to a desired shape within a mould; Apparatus therefor
    • B29C49/08—Biaxial stretching during blow-moulding
    • B29C49/16—Biaxial stretching during blow-moulding using pressure difference for pre-stretching, e.g. pre-blowing
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B29—WORKING OF PLASTICS; WORKING OF SUBSTANCES IN A PLASTIC STATE IN GENERAL
    • B29C—SHAPING OR JOINING OF PLASTICS; SHAPING OF MATERIAL IN A PLASTIC STATE, NOT OTHERWISE PROVIDED FOR; AFTER-TREATMENT OF THE SHAPED PRODUCTS, e.g. REPAIRING
    • B29C49/00—Blow-moulding, i.e. blowing a preform or parison to a desired shape within a mould; Apparatus therefor
    • B29C49/42—Component parts, details or accessories; Auxiliary operations
    • B29C49/78—Measuring, controlling or regulating

Definitions

  • the present invention relates to a blow condition adjustment device, a machine learning device, an inference device, an information processing method, a machine learning method, and an inference method.
  • the amount of at least one parameter that controls the blow molding process is set based on a simulation model.
  • simulation models take into account the cross-sectional area of flow in the blow gas supply area, the flow resistance, the generated pressure, the volumetric flow of the blow gas, and the volume change (for example, Patent Document 1).
  • the present invention has been made to solve the above problems, and aims to provide a blow condition adjustment device, machine learning device, inference device, information processing method, machine learning method, and inference method that can easily determine target values for pre-blow conditions.
  • the blow condition adjustment device is a blow condition adjustment device that determines target pre-blow conditions, which are target values of one or more pre-blow conditions that are set in a pre-blow process, which is a process of introducing a pre-blow fluid into a preform, among the blow molding processes for blow molding a preform, and determines output data consisting of target pre-blow condition information, which is information regarding the target pre-blow conditions, based on input data consisting of pre-blow condition information, which is information regarding one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions.
  • target pre-blow conditions which are target values of one or more pre-blow conditions that are set in a pre-blow process, which is a process of introducing a pre-blow fluid into a preform, among the blow molding processes for blow molding a preform
  • output data consisting of target pre-blow condition information which is information regarding the target pre
  • blow condition adjustment device machine learning device, inference device, information processing method, machine learning method, and inference method of the present invention make it easy to determine the target value of the pre-blow condition.
  • FIG. 1 is a block diagram showing an example of a blow molding device according to a first embodiment.
  • FIG. FIG. 2 is a schematic configuration diagram showing an example of a molding unit for the blow-molded container according to the first embodiment.
  • 1 is a flowchart showing a flow of a blow molding process.
  • FIG. 13 is a diagram for explaining the amount of stretching.
  • FIG. 1 is a block diagram showing an example of a machine learning device according to a first embodiment.
  • FIG. 2 is a diagram illustrating an example of a learning model and learning data according to the first embodiment.
  • 1 is a flowchart illustrating an example of a machine learning routine executed by the machine learning device.
  • 1 is a block diagram showing an example of a blow condition adjustment device according to a first embodiment.
  • FIG. FIG. 9 is a functional explanatory diagram showing an example of a function of the blow condition adjustment device of FIG. 8 .
  • FIG. 2 is a hardware configuration diagram illustrating an example of a computer.
  • First Embodiment 1 is a block diagram showing an example of a blow molding apparatus according to a first embodiment.
  • the blow molding apparatus 2 includes, as its main components, a rotary unit 21, a plurality of molding units 22, and a control unit 23.
  • the rotary unit 21 has a rotary support and a rotation mechanism.
  • the rotary support is formed in a disk shape and supports a number of molding units 22 that are arranged at equal intervals around the circumference of the rotary support.
  • the rotation mechanism rotates the rotary support at a predetermined rotation speed.
  • the control unit 23 is electrically connected to the module group and the sensor group of the molding unit 22.
  • FIG. 1 shows a pre-blow valve 2242, a main blow valve 2243, and an exhaust valve 2244 as part of the module group, and a pressure sensor 2245 and a flow rate sensor 2246 as part of the sensor group. Other module groups and sensor groups are omitted from FIG. 1.
  • the control unit 23 is composed of, for example, a general-purpose or dedicated computer.
  • the control unit 23 has, as its main components, a control unit 230, a communication unit 231, an input unit 232, an output unit 233, and a memory unit 234.
  • the control unit 230 is configured, for example, by an arithmetic processing device or a sequencer.
  • the control unit 230 functions as a blow molding control unit 2300, a pressure monitoring unit 2301, and a flow rate monitoring unit 2302, for example, by executing a blow molding program 2340 stored in the memory unit 234.
  • the communication unit 231 is connected to a communication network and functions as a communication interface for transmitting and receiving various types of data between, for example, a terminal device used by a user of the blow molding device 2.
  • the input unit 232 accepts various input operations by the user of the blow molding device 2.
  • the output unit 233 functions as a user interface by outputting various types of information to the user through the display of a screen, the lighting of a signal tower, and the sounding of a buzzer.
  • the storage unit 234 stores various programs and data used in the operation of the blow molding device 2.
  • the programs include an operating system and a blow molding program 2340.
  • the data includes device setting information 2341.
  • the device setting information 2341 is information that can register various operating conditions when the blow molding device 2 executes the blow molding process, and is configured to be editable by the user via a display screen, for example.
  • the blow molding control unit 2300 operates the modules included in the molding unit 22.
  • FIG. 2 is a schematic diagram showing an example of a molding unit 22 for a blow molded container according to the first embodiment.
  • the molding unit 22 has a mold support mechanism 220, a seal support mechanism 221, a stretch rod 222, a stretch rod support mechanism 223, a blow fluid supply/discharge mechanism 224, and a temperature control mechanism 225.
  • the mold support mechanism 220 supports the mold 20 so that it can be opened and closed.
  • the seal support mechanism 221 supports the preform 3. When the mold 20 is closed by the mold support mechanism 220, it sandwiches the preform 3. This causes the preform 3 to be sealed within the mold 20.
  • the stretch rod 222 is arranged so that it can be inserted into the interior of the preform 3 from the opening of the preform 3.
  • the stretch rod support mechanism 223 supports the stretch rod 222 so that it can move back and forth.
  • the blow fluid supply/discharge unit 224 supplies or discharges blow fluid to the preform 3.
  • air is used as the blow fluid.
  • the temperature adjustment mechanism unit 225 adjusts the temperature of the mold 20. Note that in this embodiment, the blow fluid is described as being air, but it may be any gas other than air, or it may be a liquid.
  • the blow fluid supply/discharge section 224 has a main pipe 2240, three branch pipes 2241A, 2241B, and 2241C, a pre-blow valve 2242, a main blow valve 2243, an exhaust valve 2244, a pressure sensor 2245, and a flow rate sensor 2246.
  • the main pipe 2240 is connected to the stretch rod 222 via the seal support 221.
  • Three branch pipes 2241A, 2241B, and 2241C are branched off from the main pipe 2240.
  • the branch pipe 2241A is connected to a pre-blow air supply source (not shown).
  • the pre-blow air supply source is a supply source of pre-blow air as a pre-blow fluid.
  • the branch pipe 2241B is connected to a main blow air supply source (not shown).
  • the main blow air supply source is a supply source of high-pressure main blow air. The pressure of the main blow air supply source is higher than the pressure of the pre-blow air supply source.
  • the branch pipe 2241C is connected to an exhaust system (not shown).
  • the pre-blow valve 2242 is provided in the branch pipe 2241A.
  • the main blow valve 2243 is provided in the branch pipe 2241B.
  • the exhaust valve 2244 is provided in the branch pipe 2241C.
  • the pressure sensor 2245 is provided in the main pipe 2240.
  • the pressure sensor 2245 measures the pressure of the air supplied into the preform 3 at predetermined time intervals and outputs the result to the pressure monitoring unit 2301.
  • the flow rate sensor 2246 is provided in the main pipe 2240.
  • the flow rate sensor 2246 measures the flow rate of the air supplied into the preform 3 at predetermined time intervals and outputs the result to the flow rate monitoring unit 2302.
  • the specific configurations of the mold support mechanism 220, the seal support mechanism 221, and the stretch rod support mechanism 223 are omitted.
  • These mechanisms are configured by appropriately combining, for example, modules for generating driving force such as servo motors and cylinders, driving force transmission mechanisms such as linear guides, ball screws, gears, cams, belts, couplings, and bearings, and sensors such as linear sensors, encoder sensors, and limit sensors.
  • the specific configuration of the temperature adjustment mechanism 225 is omitted.
  • the temperature adjustment mechanism 225 is configured, for example, by appropriately combining a temperature adjustment module such as an electric heater and a sensor such as a temperature sensor.
  • the pressure sensor 2245 and the flow rate sensor 2246 may be provided in the seal support part 221 instead of in the main pipe 2240.
  • FIG. 3 is a flow chart showing the flow of the blow molding process.
  • the blow molding process is a process for performing a blow molding process, and for blow molding the preform 3 placed in the mold 20 to obtain a molded body of a blow molded container.
  • the blow molding process is carried out in the following order: setting the preform 3 (step S0), closing the mold 20 (step S1), stretching (step S2), starting the supply of pre-blow air (step S3), maintaining the pre-blow pressure (step S4), starting the supply of main blow air (step S5), maintaining the main blow pressure (step S6), starting the exhaust of blow air (step S7), opening the mold 20 (step S8), and removing the molded body (step S9).
  • step S0 the blow molding control unit 2300 causes the seal support unit 221 to support the preheated preform 3.
  • step S1 the blow molding control unit 2300 causes the mold support mechanism unit 220 to close the mold 20.
  • step S2 the blow molding control unit 2300 causes the stretch rod 222 to advance along the central axis of the preform 3, thereby stretching the preform 3.
  • step S3 the blow molding control unit 2300 opens the pre-blow valve 2242.
  • step S4 the blow molding control unit 2300 advances the stretch rod 222 while continuing to supply pre-blow air to maintain the pre-blow pressure.
  • step S5 the blow molding control unit 2300 closes the pre-blow valve 2242 to stop the supply of pre-blow air, and opens the main blow valve 2243 to increase the blow pressure.
  • step S6 the blow molding control unit 2300 maintains the blow pressure at the target pressure by continuing the supply of main blow air.
  • step S7 the blow molding control unit 2300 closes the main blow valve 2243 and opens the exhaust valve 2244 to discharge the blow fluid from the main pipe 2240 to the outside.
  • step S8 the blow molding control unit 2300 causes the mold support mechanism 220 to open the mold 20 while retracting the stretch rod 222 from inside the preform 3.
  • step S9 the blow molding control unit 2300 releases the state in which the molded body is fixed by the seal support unit 221. This makes the molded body removable from the seal support unit 221.
  • steps S0 to S4 use pre-blow air and are called the pre-blow process.
  • steps S5 to S9 use main blow air and are called the main blow process.
  • the pre-blow process is a blow molding process in which a stretch rod 222 is inserted into the heated preform 3 along the central axis of the preform 3 to stretch the preform 3 in the central axis direction and introduce air into the preform 3.
  • the pre-blow conditions in the pre-blow process include the air pressure as the blow fluid, the pressure maintenance period, the air flow rate when the air pressure is maintained, the stretch amount of the preform 3, and the stretch speed of the preform 3.
  • the pressure maintenance period is the period during which the air pressure is maintained, and corresponds to the period of step S4 in FIG. 3.
  • the stretch amount of the preform 3 is the amount by which the preform 3 stretches in the central axis direction.
  • the stretch speed of the preform 3 is the speed at which the preform 3 stretches in the central axis direction.
  • Figure 4 is a diagram for explaining the stretch amount.
  • the diagram on the left side of Figure 4 shows the relative positional relationship between the preform 3 and the stretch rod 222 in step S1 of Figure 3.
  • the diagram in the center of Figure 4 shows the state in step S2 of Figure 3 where the advancing stretch rod 222 hits the preform 3.
  • the diagram on the right side of Figure 4 shows the state in step S4 of Figure 3 when the stretch rod 222 stops.
  • the stretch amount Lx is the amount by which the preform 3 is stretched by inserting the stretch rod 222 into the preform 3.
  • the thickness of the shoulder portion of the blow molded container tends to be thicker than the bottom portion of the blow molded container.
  • the thickness distribution in the central axis direction of the blow molded container tends to be thicker in the shoulder portion than in the bottom portion.
  • the mass distribution in the central axis direction of the blow molded container tends to be heavier in the shoulder portion than in the bottom portion.
  • the thickness of the shoulder portion of the resulting blow-molded container tends to be thinner than the thickness of the bottom portion of the blow-molded container.
  • the thickness distribution in the central axis direction of the blow-molded container tends to be thicker at the bottom portion than at the shoulder portion.
  • the mass distribution in the central axis direction of the blow-molded container tends to be heavier at the bottom portion than at the shoulder portion.
  • FIG. 5 is a block diagram showing an example of a machine learning device according to the first embodiment.
  • the machine learning device 5 includes a control unit 50, a communication unit 51, a learning data storage unit 52, and a trained model storage unit 53.
  • the control unit 50 functions as a learning data acquisition unit 500 and a machine learning unit 501.
  • the communication unit 51 is connected to an external device via the network 7 and functions as a communication interface for sending and receiving various data.
  • the external device is, for example, the operator terminal device 1, the blow molding device 2, and the characteristic distribution measuring device 8.
  • the characteristic distribution measuring device 8 is a device that measures at least one of the wall thickness distribution and the mass distribution of the blow molded container.
  • the learning data acquisition unit 500 is connected to the worker terminal device 1, the blow molding device 2, and the characteristic distribution measurement device 8 via the communication unit 51 and the network 7, and acquires learning data 13 from at least one of the worker terminal device 1, the blow molding device 2, and the characteristic distribution measurement device 8.
  • the learning data 13 is composed of input data and output data.
  • the input data consists of pre-blow condition information and characteristic distribution information corresponding to the pre-blow condition information.
  • the output data consists of target pre-blow condition information.
  • the target pre-blow condition information is information on the target value of the pre-blow condition when the characteristic distribution of the blow molded container becomes the target characteristic distribution.
  • the characteristic distribution of the blow molded container is the mass distribution in the central axial direction of the blow molded container.
  • the characteristic distribution of the blow molded container may also be the wall thickness distribution in the central axial direction of the blow molded container.
  • the inventors' intensive research has revealed that the distribution of properties in the central axis direction of blow molded containers, i.e., the finished hollow container, has a significant impact on improving performance such as strength and transportability.
  • the characteristic distribution in the central axis direction of the blow molded container by bringing the characteristic distribution in the central axis direction of the blow molded container closer to a set characteristic distribution, it is possible to impart the necessary strength to each part of the blow molded container.
  • the characteristic distribution in the central axis direction of the blow molded container closer to a set characteristic distribution the overall weight distribution of the blow molded container and the shape of the bottle will be as designed, which can reduce the tipping of the blow molded container during transportation, etc., and reduce the occurrence of bottle bursting during the main blow due to uneven weight distribution.
  • the input data for the learning data 13 is composed of information on the characteristic distribution in the central axis direction, which has a significant effect on the performance of the blow-molded container, and information on the pre-blow conditions, which has a significant effect on the characteristic distribution.
  • the learning data 13 is data used as teacher data (training data), verification data, and test data in supervised learning.
  • the target pre-blow condition information is data used as a correct answer label in supervised learning.
  • the learning data storage unit 52 is a database that stores multiple sets of learning data 13 acquired by the learning data acquisition unit 500.
  • the specific configuration of the database that constitutes the learning data storage unit 52 is designed as appropriate.
  • the machine learning unit 501 performs machine learning using multiple sets of learning data 13 stored in the learning data storage unit 52. That is, the machine learning unit 501 inputs multiple sets of learning data 13 to the learning model 12, and generates a learned learning model 12 by having the learning model 12 learn the correlation between the input data included in the learning data 13, i.e., the pre-blow condition information and characteristic distribution information, and the output data, i.e., the target pre-blow condition information.
  • the trained model storage unit 53 is a database that stores the trained learning model 12 generated by the machine learning unit 501. Specifically, the trained learning model 12 is a set of adjusted weight parameters. The trained learning model 12 stored in the trained model storage unit 53 is provided to the actual system, for example, the blow molding device 2, via the network 7 or a recording medium. Note that, although the training data storage unit 52 and the trained model storage unit 53 are shown as separate storage units in FIG. 5, they may be configured as a single storage unit.
  • FIG. 6 is a diagram showing an example of the learning model 12 and learning data 13 according to the first embodiment.
  • the learning data 13 used for machine learning of the learning model 12 is composed of pre-blow condition information and characteristic distribution information.
  • the pre-blow condition information includes the pre-blow conditions of the blow molded container that is produced.
  • the pre-blow conditions included in the pre-blow condition information are the air pressure maintained in the pre-blow process, the pressure maintenance period in the pre-blow process, the air flow rate when the air pressure is maintained in the pre-blow process, the stretch amount of the preform 3, and the stretch speed of the preform 3.
  • the characteristic distribution information includes the evaluation results of the mass distribution in the central axis direction of the blow molded container molded by applying the above pre-blow conditions.
  • the mass distribution is obtained by measuring the actual blow molded container.
  • the evaluation results of the mass distribution can be obtained, for example, based on statistical values such as the average, variance, standard deviation, maximum, and minimum of mass measured multiple times at different positions along the central axis direction.
  • the characteristic distribution information may include the thickness distribution in the central axis direction of the blow molded container molded by applying the above pre-blow conditions.
  • the thickness distribution is obtained by measuring the blow molded container that has actually been molded.
  • the evaluation result of the thickness distribution can be obtained, for example, based on statistical values such as the average, variance, standard deviation, maximum, and minimum of the thickness measured multiple times at different positions along the central axis.
  • the characteristic distribution information may also include the evaluation result of the mass distribution and the evaluation result of the thickness distribution.
  • the mass distribution and the thickness distribution can also be sensed using an automatic measuring device.
  • the learning data acquisition unit 500 receives pre-blow condition information for the prototype blow-molded container from the blow molding device 2 or the worker terminal device 1, and also receives characteristic distribution information from the worker terminal device 1. In this way, the learning data acquisition unit 500 acquires learning data 13. In addition, the learning data acquisition unit 500 receives target pre-blow condition information as a correct label from the blow molding device 2 or the worker terminal device 1.
  • the target pre-blow conditions are based on, for example, the results of blow condition adjustments made by a skilled worker in the blow molding process.
  • the learning model 12 employs, for example, a neural network structure.
  • the learning model 12 includes an input layer 120, an intermediate layer 121, and an output layer 122.
  • a plurality of synapses are provided between the input layer 120, the intermediate layer 121, and the output layer 122, connecting a plurality of neurons, and each synapse is associated with a weight.
  • a group of weight parameters consisting of the weights of each synapse is adjusted by machine learning.
  • the input layer 120 has a number of neurons corresponding to the pre-blow condition information and characteristic distribution information as input data.
  • each value of the pre-blow condition information and characteristic distribution information is input to each neuron.
  • the output layer 122 has a number of neurons corresponding to the target pre-blow condition information as output data.
  • the output layer 122 outputs the prediction result of the target pre-blow condition information for the pre-blow condition information, i.e., the inference result, as output data.
  • FIG. 7 is a flowchart showing an example of a machine learning routine executed by the machine learning device 5.
  • the learning data acquisition unit 500 acquires multiple pieces of learning data 13 as preparation for starting machine learning, and stores the acquired learning data 13 in the learning data storage unit 52.
  • the number of pieces of learning data acquired for preparation may be set taking into consideration the inference accuracy required for the ultimately obtained learning model 12.
  • step S110 the machine learning unit 501 prepares a pre-learning learning model 12 in order to start machine learning.
  • the pre-learning learning model 12 prepared here is configured with the neural network model exemplified in FIG. 6. At this point, the weights of each synapse are set to their initial values.
  • step S120 the machine learning unit 501 acquires, for example, one set of training data 13 randomly from the multiple sets of training data 13 stored in the training data storage unit 52.
  • step S130 the machine learning unit 501 inputs the pre-blow condition information and characteristic distribution information (input data) contained in the set of learning data 13 to the input layer 120 of the prepared learning model 12 before or during learning.
  • target pre-blow condition information output data
  • this output data is data generated by the learning model 12 before or during learning
  • the output data output as an inference result indicates information different from the target pre-blow condition information (correct label) contained in the learning data 13.
  • step S140 the machine learning unit 501 compares the target pre-blow condition information (correct label) included in the set of learning data 13 acquired in step S120 with the target pre-blow condition information (output data) output from the output layer 123 as an inference result in step S130.
  • the machine learning unit 501 performs machine learning by implementing a process of adjusting the weight of each synapse, i.e., backpropagation, based on the comparison result between the correct label and the output data. In this way, the machine learning unit 501 causes the learning model 12 to learn the correlation between the pre-blow condition information and characteristic distribution information, and the target pre-blow condition information.
  • step S150 the machine learning unit 501 determines whether or not the learning end condition is met. For example, the machine learning unit 501 determines whether or not the learning end condition is met based on at least one of the evaluation value of the error function based on the correct label and the output data, and the remaining number of unlearned learning data 13 stored in the learning data storage unit 52.
  • step S150 the machine learning unit 501 performs the processes of steps S120 to S140 multiple times on the learning model 12 being trained, using untrained learning data 13. On the other hand, if the learning end condition is met in step S150, the machine learning unit 501 stores the generated trained learning model 12 (adjusted weight parameter group) in the trained model storage unit 53 in step S160, and temporarily ends this routine.
  • step S100 corresponds to the learning data storage process
  • steps S110 to S150 correspond to the machine learning process
  • step S160 corresponds to the trained model storage process.
  • FIG. 8 is a block diagram showing an example of a blow condition adjustment device according to the first embodiment.
  • the blow condition adjustment device 6 includes a control unit 60, a communication unit 61, and a trained model storage unit 62.
  • the control unit 60 functions as an information acquisition unit 600, an inference unit 601, and an output processing unit 602.
  • the communication unit 61 is connected to external devices via the network 7, and functions as a communication interface for sending and receiving various data.
  • the external devices are, for example, the operator terminal device 1, the blow molding device 2, and the characteristic distribution measuring device 8.
  • the information acquisition unit 600 is connected to an external device via the communication unit 61 and the network 7, and executes an information acquisition process to acquire input data from the external device. Specifically, the information acquisition unit 600 acquires pre-blow condition information as input data from the worker terminal device 1 or the blow molding device 2. The information acquisition unit 600 also acquires characteristic distribution information corresponding to the pre-blow condition information as input data from the worker terminal device 1 or the characteristic distribution measuring device 8.
  • the inference unit 601 inputs the pre-blow condition information acquired by the information acquisition unit 600 and the characteristic distribution information corresponding to the pre-blow condition information as input data to the learning model 12.
  • the inference unit 601 executes the inference process using the learning model 12 stored in the trained model storage unit 62.
  • the trained model storage unit 62 is a database that stores trained learning models 12 used in the inference unit 601.
  • the number of training models 12 stored in the trained model storage unit 62 is not limited to one, and multiple trained models with different conditions, such as the machine learning method, the shape of the blow molded container, and the material of the preform 3, may be stored and selectively used.
  • the trained model storage unit 62 may be replaced by a storage unit of an external computer, in which case the inference unit 601 only needs to access the external computer.
  • external computers include server-type computers and cloud-type computers.
  • the output processing unit 602 executes an output process to output output data including the target pre-blow conditions inferred by the inference unit 601 to an external device.
  • the output processing unit 602 may transmit the generated target pre-blow condition information to the worker terminal device 1 or to the blow molding device 2.
  • control unit 60 determines output data consisting of target pre-blow condition information based on input data consisting of pre-blow condition information and characteristic distribution information.
  • FIG. 9 is a functional explanatory diagram showing an example of the function of the blow condition adjustment device of FIG. 8.
  • the information acquisition unit 600 acquires pre-blow condition information and characteristic distribution information corresponding to the pre-blow condition information as input data, and inputs them to the inference unit 601.
  • the inference unit 601 inputs the pre-blow condition information and characteristic distribution information corresponding to the pre-blow condition information to the learning model 12, and generates target pre-blow condition information as an inference result by the learning model 12.
  • design pre-blow condition information and design characteristic distribution information are provided to the information acquisition unit 600.
  • the design pre-blow condition information is information about pre-blow conditions created based on the design information of the blow molded container.
  • the design characteristic distribution information is information about the characteristic distribution of the blow molded container molded by applying the pre-blow conditions created based on the design information.
  • FIG. 10 is a hardware configuration diagram showing an example of a computer.
  • the control unit 23, the machine learning device 5, and the blow condition adjustment device 6 are configured by a general-purpose or dedicated computer 900.
  • the computer 900 has as its main components a bus 910, a processor 912, a memory 914, an input device 916, an output device 917, a display device 918, a storage device 920, a communication I/F (interface) unit 922, an external device I/F unit 924, an I/O (input/output) device I/F unit 926, and a media input/output unit 928.
  • a bus 910 a bus 910
  • a processor 912 a memory 914
  • an input device 916 an output device 917
  • a display device 918 a display device 918
  • a storage device 920 a communication I/F (interface) unit 922
  • an external device I/F unit 924 an external device I/F unit 924
  • I/O (input/output) device I/F unit 926 an I/O (input/output) device I/F unit 926
  • media input/output unit 928 a media input/output unit
  • the processor 912 is composed of one or more arithmetic processing devices (CPU (Central Processing Unit), MPU (Micro Processing Unit), DSP (Digital Signal Processor), GPU (Graphics Processing Unit), etc.) and operates as a control unit that controls the entire computer 900.
  • the memory 914 stores various data and programs 930, and is composed of, for example, volatile memory (DRAM, SRAM, etc.) that functions as main memory, non-volatile memory (ROM), flash memory, etc.
  • the input device 916 is, for example, a keyboard, a mouse, a numeric keypad, an electronic pen, etc., and functions as an input unit.
  • the output device 917 is, for example, a sound (audio) output device, a vibration device, etc., and functions as an output unit.
  • the display device 918 is, for example, a liquid crystal display, an organic EL display, electronic paper, a projector, etc., and functions as an output unit.
  • the input device 916 and the display device 918 may be integrated, such as a touch panel display.
  • the storage device 920 is, for example, a HDD, an SSD (Solid State Drive), etc., and functions as a memory unit.
  • the storage device 920 stores various data necessary for the execution of the operating system and the program 930.
  • the communication I/F unit 922 is connected to a network 940 (which may be the same as the network 7 in FIG. 5) such as the Internet or an intranet by wire or wirelessly, and functions as a communication unit that transmits and receives data to and from other computers according to a predetermined communication standard.
  • the external device I/F unit 924 is connected to an external device 950 such as a camera, printer, scanner, or reader/writer by wire or wirelessly, and functions as a communication unit that transmits and receives data to and from the external device 950 according to a predetermined communication standard.
  • the I/O device I/F unit 926 is connected to an I/O device 960 such as various sensors and actuators, and functions as a communication unit that transmits and receives various signals and data, such as detection signals from sensors and control signals to actuators, between the I/O device 960.
  • the media input/output unit 928 is composed of a drive device such as a DVD drive or a CD drive, and reads and writes data to and from media (non-temporary storage media) 970 such as DVDs and CDs.
  • the processor 912 calls up the program 930 stored in the storage device 920 into the memory 914, executes it, and controls each part of the computer 900 via the bus 910.
  • the program 930 may be stored in the memory 914 instead of the storage device 920.
  • the program 930 may be recorded in the medium 970 in an installable file format or an executable file format, and provided to the computer 900 via the media input/output unit 928.
  • the program 930 may be provided to the computer 900 by downloading it via the network 940 via the communication I/F unit 922.
  • the computer 900 may realize various functions realized by the processor 912 executing the program 930 using hardware such as an FPGA or ASIC.
  • the computer 900 is, for example, a stationary computer or a portable computer, and is an electronic device of any type.
  • the computer 900 may be a client-type computer, a server-type computer, or a cloud-type computer.
  • the computer 900 may be applied to devices other than the control unit 23, the machine learning device 5, and the blow condition adjustment device 6.
  • the blow condition adjustment device is a device that determines the target pre-blow conditions, which are the target values of multiple pre-blow conditions that are set in the pre-blow process during the blow molding process for blow molding the preform 3.
  • the pre-blow process is a process of introducing pre-blow air into the preform 3.
  • output data consisting of target pre-blow condition information is obtained based on input data consisting of pre-blow condition information and characteristic distribution information.
  • the pre-blow condition information is information on a plurality of pre-blow conditions.
  • the characteristic distribution information is information on the mass distribution of the blow molded container molded by applying the plurality of pre-blow conditions.
  • the target pre-blow condition information is information on the target pre-blow conditions.
  • the pre-blow conditions include at least one of the pre-blow air pressure, the pressure maintenance period during which the pre-blow air pressure is maintained, the pre-blow air flow rate, the stretch amount Lx of the preform 3, and the stretch speed of the preform 3.
  • the stretch amount is the amount by which the preform 3 is stretched by inserting the stretch rod 222 into the preform 3.
  • the stretch speed is the speed at which the preform 3 is stretched.
  • the mass distribution of a blow molded container is highly dependent on the pre-blow air pressure, the pressure maintenance period, the pre-blow air flow rate, the stretch amount Lx of the preform 3, and the stretch speed of the preform 3. Therefore, by setting at least one of these parameters as a pre-blow condition and adjusting at least one parameter, the characteristic distribution of the blow molded container can be made more uniform.
  • the characteristic distribution also includes at least one of the mass distribution of the blow molded container and the wall thickness distribution of the blow molded container.
  • the relationship between the pre-blow air pressure and the stretch amount Lx of the preform 3 mainly affects the variation in the wall thickness of the blow molded container in the central axial direction of the preform 3. Therefore, by setting at least one of the mass distribution of the blow molded container and the wall thickness distribution of the blow molded container as the characteristic distribution and using the evaluation result of at least one of these as input data for the machine learning device 5, more accurate learning results can be obtained.
  • design pre-blow condition information is information about pre-blow conditions created based on design information for a blow-molded container.
  • design characteristic distribution information is information about the characteristic distribution of a blow molding device molded by applying pre-blow conditions created based on the design information.
  • the blow condition adjustment device 6 determines target pre-blow conditions using a learning model 12 in which the correlation between the input data and the output data is learned by machine learning, and outputs information related to the determined target pre-blow conditions as output data.
  • the machine learning device 5 also generates a learning model 12 for inferring target pre-blow conditions.
  • the machine learning device 5 includes a learning data storage unit 52, a machine learning unit 501, and a learned model storage unit 53.
  • the learning data storage unit 52 stores multiple sets of learning data 13 composed of input data consisting of pre-blow condition information and characteristic distribution information, and output data corresponding to the input data and consisting of target pre-blow condition information.
  • the machine learning unit 501 receives multiple sets of learning data 13 and causes the learning model 12 to learn the correlation between the input data and the output data.
  • the learned model storage unit stores the learning model learned by the machine learning unit 501.
  • the inference device includes a memory 914 and a processor 912, and infers target pre-blow conditions.
  • the processor 912 executes information acquisition processing, inference processing, and output processing.
  • the information acquisition processing is a processing for acquiring input data consisting of pre-blow condition information and characteristic distribution information.
  • the inference processing is a processing for inferring target pre-blow conditions using a learning model 12 based on machine learning stored in the memory 914 upon acquiring the input data.
  • the output processing is a processing for outputting output data including the inferred target pre-blow conditions.
  • the information processing method is also a method for determining target pre-blow conditions.
  • the information processing method determines output data consisting of target pre-blow condition information based on input data consisting of pre-blow condition information and characteristic distribution information.
  • the machine learning method is a method of generating a learning model 12 for inferring target pre-blow conditions.
  • the machine learning method includes a learning data storage process, a machine learning process, and a learned model storage process.
  • the learning data storage process is a process of storing multiple sets of learning data composed of input data consisting of pre-blow condition information and characteristic distribution information, and output data corresponding to the input data and consisting of target pre-blow condition information.
  • the machine learning process is a process of inputting multiple sets of learning data 13, thereby making the learning model 12 learn the correlation between the input data and the output data.
  • the learned model storage process is a process of storing the learning model learned in the machine learning process in the learned model storage unit 53.
  • the heating conditions of the preform 3 include the heating temperature, the heating time, and a combination of these.
  • a learning model may be prepared for each heating condition of the preform 3.
  • the learning model 12 outputs the target pre-blow condition information as the inference result, but it may also output the difference between the previously inferred target pre-blow condition information and the currently inferred target pre-blow condition information as the inference result.
  • the machine learning unit 501 learns the learning model 12
  • data on the difference between the previous pre-blow condition information and the current pre-blow condition information is prepared as the target pre-blow condition information (correct answer label) included in the learning data 13.
  • the characteristic distribution information included in the learning data 13 was at least one of the evaluation results of the mass distribution and the evaluation results of the wall thickness distribution, but it may be at least one of the measurement results of the mass distribution itself and the measurement results of the wall thickness distribution itself.
  • the pre-blow conditions included the amount of stretching, but instead of the amount of stretching, the amount of advancement of the stretch rod 222 may be included.
  • a neural network is used as the learning model 12 for realizing machine learning by the machine learning unit 501, but other machine learning models may be used.
  • machine learning models include tree types such as decision trees and regression trees, ensemble learning such as bagging and boosting, neural network types (including deep learning) such as recurrent neural networks, convolutional neural networks, and LSTM (Long Short Term Memory), clustering types such as hierarchical clustering, non-hierarchical clustering, k-nearest neighbors, and k-means, multivariate analyses such as principal component analysis, factor analysis, and logistic regression, and support vector machines.
  • a mathematical optimization model may be used instead of the learning model based on machine learning. That is, input data may be input to the mathematical optimization model, and output data may be output from the mathematical optimization model.
  • a mathematical optimization model may be stored in the memory 914 instead of the learning model 12, and the processor 912 may infer the target pre-blow conditions using the mathematical optimization model stored in the memory 914.
  • Bayesian optimization or a genetic algorithm may be used as the mathematical optimization model.
  • the present invention can also be provided in the form of a program (machine learning program) that causes the computer 900 to function as each component of the machine learning device 5, or a program (machine learning program) that causes the computer 900 to execute each step of the machine learning method.
  • the present invention can also be provided in the form of a program (inference program) that causes the computer 900 to function as each unit of the blow condition adjustment device 6.
  • the blow condition adjustment device includes a memory 914 and a processor 912, of which the processor 912 can execute a series of processes.
  • This series of processes includes an information acquisition process (information acquisition step) for acquiring pre-blow condition information, which is information relating to one or more pre-blow conditions, and characteristic distribution information, which is information relating to a specific distribution of a blow-molded container molded by applying the one or more pre-blow conditions, and an inference process (inference step) for inferring target pre-blow conditions using a learning model stored in memory 914 once the pre-blow condition information and characteristic distribution information have been acquired by the information acquisition process.
  • information acquisition process information acquisition step
  • characteristic distribution information which is information relating to a specific distribution of a blow-molded container molded by applying the one or more pre-blow conditions
  • inference process inference step
  • blow condition adjustment device inference method or inference program
  • the blow condition adjustment device inference method or inference program
  • it may apply the inference method implemented by the inference unit 601 using the machine learning device 5 and the trained learning model 12 generated by the machine learning method according to the above embodiment.
  • a blow condition adjusting device for determining a target pre-blow condition which is a target value of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, among blow molding steps for blow molding a preform, comprising: A blow condition adjusting device that calculates output data consisting of target pre-blow condition information that is information on the one or more pre-blow conditions, based on input data consisting of pre-blow condition information that is information on the one or more pre-blow conditions, and characteristic distribution information that is information on the characteristic distribution of a product molded by applying the one or more pre-blow conditions.
  • the one or more pre-blow conditions include at least one of a pressure of the pre-blow fluid, a pressure maintenance period during which the pressure is maintained, a flow rate of the pre-blow fluid, a stretch amount by which the preform is stretched by inserting a stretch rod into the preform, and a stretch speed by which the preform is stretched.
  • Appendix 3 The blow condition adjusting device according to claim 1 or 2, wherein the characteristic distribution includes at least one of a mass distribution of the product and a wall thickness distribution of the product.
  • a machine learning device that generates a learning model for inferring a target pre-blow condition, which is a target value of one or more pre-blow conditions set in a pre-blow process, which is a process of introducing a pre-blow fluid into a preform, among blow molding processes for blow molding the preform, comprising: a learning data storage unit that stores a plurality of sets of learning data each composed of input data including pre-blow condition information, which is information related to the one or more pre-blow conditions, and characteristic distribution information, which is information related to a characteristic distribution of a product molded by applying the one or more pre-blow conditions, and output data, which is associated with the input data and includes target pre-blow condition information, which is information related to the target pre-blow conditions; a machine learning unit that causes a learning model to learn a correlation between the input data and the output data by inputting a plurality of sets of the learning data; and a trained model storage unit that stores the learning model trained by the machine
  • An inference device comprising a memory and at least one processor, the inference device infers a target pre-blow condition which is a target value of one or more pre-blow conditions set in a pre-blow step which is a step of introducing a pre-blow fluid into a preform in a blow molding step for blow molding the preform, the inference device comprising: The at least one processor an information acquisition process for acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions; an inference process for inferring the target pre-blow condition by using a learning model based on machine learning stored in the memory when the input data is acquired; and an output process for outputting output data including the inferred target pre-blow condition.
  • pre-blow condition information which is information regarding the one or more pre-blow conditions
  • characteristic distribution information which is information regarding the characteristic distribution of a
  • An inference device comprising a memory and at least one processor, the inference device infers a target pre-blow condition which is a target value of one or more pre-blow conditions set in a pre-blow step which is a step of introducing a pre-blow fluid into a preform in a blow molding step for blow molding the preform, the inference device comprising: The at least one processor an information acquisition process for acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions; an inference process for inferring the target pre-blow condition using a mathematical optimization model stored in the memory when the input data is acquired; and an output process for outputting output data including the inferred target pre-blow condition.
  • pre-blow condition information which is information regarding the one or more pre-blow conditions
  • characteristic distribution information which is information regarding the characteristic distribution of a product molded by applying
  • An information processing method for determining a target pre-blow condition which is a target value of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, in a blow molding step for blow molding a preform, comprising: An information processing method for determining output data consisting of target pre-blow condition information, which is information on the target pre-blow condition, based on input data consisting of pre-blow condition information, which is information on the one or more pre-blow conditions, and characteristic distribution information, which is information on the characteristic distribution of a product molded by applying the one or more pre-blow conditions.
  • a machine learning method for generating a learning model for inferring and determining a target pre-blow condition, which is a target value of one or more pre-blow conditions set in a pre-blow process, which is a process of introducing a pre-blow fluid into a preform, among blow molding processes for blow molding the preform comprising: a learning data storage step of storing a plurality of sets of learning data each including input data including pre-blow condition information, which is information on the one or more pre-blow conditions, and characteristic distribution information, which is information on the characteristic distribution of a product molded by applying the one or more pre-blow conditions, and output data, which is associated with the input data and includes target pre-blow condition information, which is information on the target pre-blow conditions; a machine learning process in which a correlation between the input data and the output data is learned by a learning model by inputting a plurality of sets of the learning data; and a trained model storage step of storing the learned model trained in the
  • An inference method for inferring target pre-blow conditions which are target values of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, in a blow molding step for blow molding a preform
  • the inference method comprising: The at least one processor an information acquisition step of acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions; an inference step of inferring the target pre-blow condition by using a learning model based on machine learning stored in the memory when the input data is acquired; and an output step of outputting output data including the inferred target pre-blow condition.
  • An inference method for inferring target pre-blow conditions which are target values of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, in a blow molding step for blow molding a preform
  • the inference method comprising: The at least one processor an information acquisition step of acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions; an inference step of inferring the target pre-blow condition using a mathematical optimization model stored in the memory when the input data is obtained; and an output step of outputting output data including the inferred target pre-blow condition.

Landscapes

  • Engineering & Computer Science (AREA)
  • Manufacturing & Machinery (AREA)
  • Mechanical Engineering (AREA)
  • Blow-Moulding Or Thermoforming Of Plastics Or The Like (AREA)

Abstract

[Problem] To obtain a blow conditions adjustment device, a machine learning device, an inference device, an information processing method, a machine learning method and an inference method which make it possible to easily determine a pre-blow conditions target value. [Solution] In order to adjust the blow conditions of a pre-blow step, this pre-blow conditions adjustment device obtains output data comprising target pre-blow conditions, on the basis of input data comprising a plurality of pre-blow conditions and the mass distribution of a blow-molded container which is molded by applying said plurality of pre-blow conditions. The pre-blow step involves stretching a pre-form 3 by inserting a stretching rod 222 into the heated pre-form 3, and introducing air into the pre-form 3.

Description

ブロー条件調整装置、機械学習装置、推論装置、情報処理方法、機械学習方法、及び、推論方法BLOW CONDITION ADJUSTING DEVICE, MACHINE LEARNING DEVICE, INFERENCE DEVICE, INFORMATION PROCESSING METHOD, MACHINE LEARNING METHOD, AND INFERENCE METHOD
　本発明は、ブロー条件調整装置、機械学習装置、推論装置、情報処理方法、機械学習方法、及び、推論方法に関する。 The present invention relates to a blow condition adjustment device, a machine learning device, an inference device, an information processing method, a machine learning method, and an inference method.
　従来、プリフォームをブロー成形することで中空容器を製造するブロー成形装置では、ブロー成形工程を制御する少なくとも１つのパラメータの量が、シミュレーションモデルに基づいて設定されていた。このようなシミュレーションモデルでは、ブローガス供給領域における流動横断面積、流動抵抗、発生する圧力、ブローガスの体積流、及び、体積変化等が考慮されている（例えば、特許文献１）。 　In conventional blow molding devices that manufacture hollow containers by blow molding preforms, the amount of at least one parameter that controls the blow molding process is set based on a simulation model. Such simulation models take into account the cross-sectional area of flow in the blow gas supply area, the flow resistance, the generated pressure, the volumetric flow of the blow gas, and the volume change (for example, Patent Document 1).
特表２０１２－５０８６５８号公報JP 2012-508658 A
しかし、従来のブロー成形装置では、シミュレーションモデルを実際のブロー成形環境に整合させることが難しく、目標とする特性分布を実現するための制御条件、特に、適切な質量分布や肉厚分布を実現するために重要となるプレブロー工程における制御条件であるプレブロー条件を得るまで、多くの試作を繰り返す必要があった。 However, with conventional blow molding equipment, it is difficult to match the simulation model to the actual blow molding environment, and it is necessary to repeat many prototypes until the control conditions for achieving the target property distribution are obtained, in particular the pre-blow conditions, which are the control conditions in the pre-blow process that are important for achieving appropriate mass distribution and wall thickness distribution.
　本発明は、上記のような課題を解決するために為されたものであり、プレブロー条件の目標値を容易に決定することができるブロー条件調整装置、機械学習装置、推論装置、情報処理方法、機械学習方法、及び、推論方法を得ることを目的とする。 The present invention has been made to solve the above problems, and aims to provide a blow condition adjustment device, machine learning device, inference device, information processing method, machine learning method, and inference method that can easily determine target values for pre-blow conditions.
　本発明に係るブロー条件調整装置は、プリフォームをブロー成形するためのブロー成形工程のうち、プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を決定するブロー条件調整装置であって、１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データに基づいて、目標プレブロー条件に関する情報である目標プレブロー条件情報からなる出力データを求める。 The blow condition adjustment device according to the present invention is a blow condition adjustment device that determines target pre-blow conditions, which are target values of one or more pre-blow conditions that are set in a pre-blow process, which is a process of introducing a pre-blow fluid into a preform, among the blow molding processes for blow molding a preform, and determines output data consisting of target pre-blow condition information, which is information regarding the target pre-blow conditions, based on input data consisting of pre-blow condition information, which is information regarding one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions.
　本発明に係るブロー条件調整装置、機械学習装置、推論装置、情報処理方法、機械学習方法、及び、推論方法によれば、プレブロー条件の目標値を容易に決定することができる。 The blow condition adjustment device, machine learning device, inference device, information processing method, machine learning method, and inference method of the present invention make it easy to determine the target value of the pre-blow condition.
第１の実施形態に係るブロー成形装置の一例を示すブロック図である。1 is a block diagram showing an example of a blow molding device according to a first embodiment. FIG. 第１の実施形態に係るブロー成形容器の成形ユニットの一例を示す概略構成図である。FIG. 2 is a schematic configuration diagram showing an example of a molding unit for the blow-molded container according to the first embodiment. ブロー成形工程の流れを示すフローチャートである。1 is a flowchart showing a flow of a blow molding process. 延伸量を説明するための図である。FIG. 13 is a diagram for explaining the amount of stretching. 第１の実施形態に係る機械学習装置の一例を示すブロック図である。FIG. 1 is a block diagram showing an example of a machine learning device according to a first embodiment. 第１の実施形態に係る学習モデル及び学習用データの一例を示す図である。FIG. 2 is a diagram illustrating an example of a learning model and learning data according to the first embodiment. 機械学習装置により実行される機械学習ルーチンの一例を示すフローチャートである。1 is a flowchart illustrating an example of a machine learning routine executed by the machine learning device. 第１の実施形態に係るブロー条件調整装置の一例を示すブロック図である。1 is a block diagram showing an example of a blow condition adjustment device according to a first embodiment. FIG. 図８のブロー条件調整装置の機能の一例を示す機能説明図である。FIG. 9 is a functional explanatory diagram showing an example of a function of the blow condition adjustment device of FIG. 8 . コンピュータの一例を示すハードウェア構成図である。FIG. 2 is a hardware configuration diagram illustrating an example of a computer.
　以下、図面を参照して本発明を実施するための実施形態について説明する。以下では、本発明の目的を達成するための説明に必要な範囲を模式的に示し、本発明の該当部分の説明に必要な範囲を主に説明することとし、説明を省略する箇所については公知技術によるものとする。 Below, an embodiment for carrying out the present invention will be described with reference to the drawings. Below, the scope necessary for the explanation to achieve the object of the present invention will be shown diagrammatically, and the scope necessary for the explanation of the relevant parts of the present invention will be mainly explained, and the parts where explanation is omitted will be based on publicly known technology.
（第１の実施形態）
　図１は、第１の実施形態に係るブロー成形装置の一例を示すブロック図である。ブロー成形装置２は、その主要な構成要素として、ロータリユニット２１、複数の成形ユニット２２、及び、制御ユニット２３を備えている。 First Embodiment
1 is a block diagram showing an example of a blow molding apparatus according to a first embodiment. The blow molding apparatus 2 includes, as its main components, a rotary unit 21, a plurality of molding units 22, and a control unit 23.
　ロータリユニット２１は、ロータリ支持部及び回転機構部を有している。ロータリ支持部は、円板状に形成されており、ロータリ支持部の周方向へ等間隔に配置された複数の成形ユニット２２を支持している。回転機構部は、ロータリ支持部を既定の回転速度で回転させる。 The rotary unit 21 has a rotary support and a rotation mechanism. The rotary support is formed in a disk shape and supports a number of molding units 22 that are arranged at equal intervals around the circumference of the rotary support. The rotation mechanism rotates the rotary support at a predetermined rotation speed.
　制御ユニット２３は、成形ユニット２２が備えるモジュール群及びセンサ群と電気的に接続されている。図１には、モジュール群の一部としてのプレブローバルブ２２４２、メインブローバルブ２２４３及び排気バルブ２２４４と、センサ群の一部としての圧力センサ２２４５及び流量センサ２２４６とが示されている。図１には、その他のモジュール群及びセンサ群の図示は省略されている。 The control unit 23 is electrically connected to the module group and the sensor group of the molding unit 22. FIG. 1 shows a pre-blow valve 2242, a main blow valve 2243, and an exhaust valve 2244 as part of the module group, and a pressure sensor 2245 and a flow rate sensor 2246 as part of the sensor group. Other module groups and sensor groups are omitted from FIG. 1.
　制御ユニット２３は、例えば、汎用又は専用のコンピュータで構成されている。制御ユニット２３は、その主要な構成要素として、制御部２３０、通信部２３１、入力部２３２、出力部２３３及び記憶部２３４を有している。 The control unit 23 is composed of, for example, a general-purpose or dedicated computer. The control unit 23 has, as its main components, a control unit 230, a communication unit 231, an input unit 232, an output unit 233, and a memory unit 234.
　制御部２３０は、例えば、演算処理装置又はシーケンサにより構成されている。制御部２３０は、例えば、記憶部２３４に記憶されているブロー成形プログラム２３４０を実行することにより、ブロー成形制御部２３００、圧力監視部２３０１、及び、流量監視部２３０２として機能する。 The control unit 230 is configured, for example, by an arithmetic processing device or a sequencer. The control unit 230 functions as a blow molding control unit 2300, a pressure monitoring unit 2301, and a flow rate monitoring unit 2302, for example, by executing a blow molding program 2340 stored in the memory unit 234.
　通信部２３１は、通信ネットワークに接続され、例えば、ブロー成形装置２のユーザが使用する端末装置との間において各種のデータを送受信する通信インタフェースとして機能する。入力部２３２は、ブロー成形装置２のユーザによる各種の入力操作を受け付ける。出力部２３３は、画面の表示、シグナルタワーの点灯、及び、ブザーの鳴動を介して各種の情報をユーザに出力することにより、ユーザインタフェースとして機能する。 The communication unit 231 is connected to a communication network and functions as a communication interface for transmitting and receiving various types of data between, for example, a terminal device used by a user of the blow molding device 2. The input unit 232 accepts various input operations by the user of the blow molding device 2. The output unit 233 functions as a user interface by outputting various types of information to the user through the display of a screen, the lighting of a signal tower, and the sounding of a buzzer.
　記憶部２３４は、ブロー成形装置２の動作において使用される各種のプログラム及びデータを記憶する。プログラムには、オペレーティングシステム及びブロー成形プログラム２３４０が含まれる。データには、装置設定情報２３４１が含まれる。装置設定情報２３４１は、ブロー成形装置２がブロー成形処理を実行するときの各種の動作条件を登録可能な情報であり、例えば、表示画面を介してユーザにより編集可能に構成されている。 The storage unit 234 stores various programs and data used in the operation of the blow molding device 2. The programs include an operating system and a blow molding program 2340. The data includes device setting information 2341. The device setting information 2341 is information that can register various operating conditions when the blow molding device 2 executes the blow molding process, and is configured to be editable by the user via a display screen, for example.
　ブロー成形制御部２３００は、成形ユニット２２が備えるモジュール群を動作させる。 The blow molding control unit 2300 operates the modules included in the molding unit 22.
　図２は、第１の実施形態に係るブロー成形容器の成形ユニット２２の一例を示す概略構成図である。成形ユニット２２は、金型支持機構部２２０、シール支持部２２１、ストレッチロッド２２２、ストレッチロッド支持機構部２２３、ブロー流体給排部２２４、及び、温調機構部２２５を有している。 FIG. 2 is a schematic diagram showing an example of a molding unit 22 for a blow molded container according to the first embodiment. The molding unit 22 has a mold support mechanism 220, a seal support mechanism 221, a stretch rod 222, a stretch rod support mechanism 223, a blow fluid supply/discharge mechanism 224, and a temperature control mechanism 225.
　金型支持機構部２２０は、金型２０を開閉可能に支持している。シール支持部２２１は、プリフォーム３を支持している。金型２０は、金型支持機構部２２０によって閉じられたとき、プリフォーム３を挟み込む。これにより、プリフォーム３は、金型２０内で密閉状態になる。ストレッチロッド２２２は、プリフォーム３の開口部からプリフォーム３の内部に挿入可能に配置されている。ストレッチロッド支持機構部２２３は、ストレッチロッド２２２を進退移動可能に支持している。 The mold support mechanism 220 supports the mold 20 so that it can be opened and closed. The seal support mechanism 221 supports the preform 3. When the mold 20 is closed by the mold support mechanism 220, it sandwiches the preform 3. This causes the preform 3 to be sealed within the mold 20. The stretch rod 222 is arranged so that it can be inserted into the interior of the preform 3 from the opening of the preform 3. The stretch rod support mechanism 223 supports the stretch rod 222 so that it can move back and forth.
　ブロー流体給排部２２４は、プリフォーム３に対するブロー流体の供給又は排出を行う。本実施の形態では、ブロー流体として空気（エア）が用いられている。温調機構部２２５は、金型２０の温度を調節する。なお、本実施の形態では、ブロー流体は、空気であるものとして説明するが、空気以外の任意の気体であってもよいし、液体であってもよい。 The blow fluid supply/discharge unit 224 supplies or discharges blow fluid to the preform 3. In this embodiment, air is used as the blow fluid. The temperature adjustment mechanism unit 225 adjusts the temperature of the mold 20. Note that in this embodiment, the blow fluid is described as being air, but it may be any gas other than air, or it may be a liquid.
　ブロー流体給排部２２４は、主配管２２４０、３つの分岐配管２２４１Ａ、２２４１Ｂ及び２２４１Ｃ、プレブローバルブ２２４２、メインブローバルブ２２４３、排気バルブ２２４４、圧力センサ２２４５、及び、流量センサ２２４６を有している。 The blow fluid supply/discharge section 224 has a main pipe 2240, three branch pipes 2241A, 2241B, and 2241C, a pre-blow valve 2242, a main blow valve 2243, an exhaust valve 2244, a pressure sensor 2245, and a flow rate sensor 2246.
　主配管２２４０は、シール支持部２２１を介してストレッチロッド２２２に接続されている。３つの分岐配管２２４１Ａ、２２４１Ｂ及び２２４１Ｃは、主配管２２４０から分岐されている。分岐配管２２４１Ａは、図示しないプレブローエア供給源に接続されている。プレブローエア供給源は、プレブロー流体としてのプレブローエアの供給源である。分岐配管２２４１Ｂは、図示しないメインブローエア供給源に接続されている。メインブローエア供給源は、高圧のメインブローエアの供給源である。メインブローエア供給源の圧力は、プレブローエア供給源の圧力よりも高い。分岐配管２２４１Ｃは、図示しない排気系統に接続されている。 The main pipe 2240 is connected to the stretch rod 222 via the seal support 221. Three branch pipes 2241A, 2241B, and 2241C are branched off from the main pipe 2240. The branch pipe 2241A is connected to a pre-blow air supply source (not shown). The pre-blow air supply source is a supply source of pre-blow air as a pre-blow fluid. The branch pipe 2241B is connected to a main blow air supply source (not shown). The main blow air supply source is a supply source of high-pressure main blow air. The pressure of the main blow air supply source is higher than the pressure of the pre-blow air supply source. The branch pipe 2241C is connected to an exhaust system (not shown).
　プレブローバルブ２２４２は、分岐配管２２４１Ａに設けられている。メインブローバルブ２２４３は、分岐配管２２４１Ｂに設けられている。排気バルブ２２４４は、分岐配管２２４１Ｃに設けられている。 The pre-blow valve 2242 is provided in the branch pipe 2241A. The main blow valve 2243 is provided in the branch pipe 2241B. The exhaust valve 2244 is provided in the branch pipe 2241C.
　圧力センサ２２４５は、主配管２２４０に設けられている。圧力センサ２２４５は、プリフォーム３内に供給されるエアの圧力を所定の時間間隔で測定し、その結果を圧力監視部２３０１へ出力する。流量センサ２２４６は、主配管２２４０に設けられている。流量センサ２２４６は、プリフォーム３内に供給されるエアの流量を所定の時間間隔で測定し、その結果を流量監視部２３０２へ出力する。 The pressure sensor 2245 is provided in the main pipe 2240. The pressure sensor 2245 measures the pressure of the air supplied into the preform 3 at predetermined time intervals and outputs the result to the pressure monitoring unit 2301. The flow rate sensor 2246 is provided in the main pipe 2240. The flow rate sensor 2246 measures the flow rate of the air supplied into the preform 3 at predetermined time intervals and outputs the result to the flow rate monitoring unit 2302.
　なお、図２において、金型支持機構部２２０、シール支持部２２１及びストレッチロッド支持機構部２２３の具体的な構成は省略されている。これらの機構は、例えば、サーボモータ、シリンダ等の駆動力発生用のモジュールと、リニアガイド、ボールスクリュ、ギヤ、カム、ベルト、カップリング、軸受等の駆動力伝達機構と、リニアセンサ、エンコーダセンサ、リミットセンサ等のセンサとが適宜組み合わされて構成されている。 In addition, in Figure 2, the specific configurations of the mold support mechanism 220, the seal support mechanism 221, and the stretch rod support mechanism 223 are omitted. These mechanisms are configured by appropriately combining, for example, modules for generating driving force such as servo motors and cylinders, driving force transmission mechanisms such as linear guides, ball screws, gears, cams, belts, couplings, and bearings, and sensors such as linear sensors, encoder sensors, and limit sensors.
　また、図２において、温調機構部２２５の具体的な構成は省略されている。温調機構部２２５は、例えば、電熱ヒータ等の温度調節用のモジュールと、温度センサ等のセンサとが適宜組み合わされて構成されている。また、圧力センサ２２４５及び流量センサ２２４６は、主配管２２４０ではなく、シール支持部２２１に設けられていてもよい。 In addition, in FIG. 2, the specific configuration of the temperature adjustment mechanism 225 is omitted. The temperature adjustment mechanism 225 is configured, for example, by appropriately combining a temperature adjustment module such as an electric heater and a sensor such as a temperature sensor. In addition, the pressure sensor 2245 and the flow rate sensor 2246 may be provided in the seal support part 221 instead of in the main pipe 2240.
　図３は、ブロー成形工程の流れを示すフローチャートである。ブロー成形工程は、ブロー成形処理を実行する工程であり、金型２０に配置されたプリフォーム３をブロー成形し、ブロー成形容器の成形体を得るための工程である。ブロー成形工程は、プリフォーム３のセット（ステップＳ０）、金型２０の閉動作（ステップＳ１）、ストレッチ（ステップＳ２）、プレブローエア供給開始（ステップＳ３）、プレブロー圧力維持（ステップＳ４）、メインブローエア供給開始（ステップＳ５）、メインブロー圧力維持（ステップＳ６）、ブローエア排気開始（ステップＳ７）、金型２０の開動作（ステップＳ８）、成形体の取出し（ステップＳ９）の順に実行される。 FIG. 3 is a flow chart showing the flow of the blow molding process. The blow molding process is a process for performing a blow molding process, and for blow molding the preform 3 placed in the mold 20 to obtain a molded body of a blow molded container. The blow molding process is carried out in the following order: setting the preform 3 (step S0), closing the mold 20 (step S1), stretching (step S2), starting the supply of pre-blow air (step S3), maintaining the pre-blow pressure (step S4), starting the supply of main blow air (step S5), maintaining the main blow pressure (step S6), starting the exhaust of blow air (step S7), opening the mold 20 (step S8), and removing the molded body (step S9).
　ステップＳ０において、ブロー成形制御部２３００は、予め加熱されたプリフォーム３をシール支持部２２１に支持させる。ステップＳ１において、ブロー成形制御部２３００は、金型支持機構部２２０に金型２０を閉じさせる。ステップＳ２において、ブロー成形制御部２３００は、ストレッチロッド２２２をプリフォーム３の中心軸に沿って進出させることにより、プリフォーム３を延伸させる。 In step S0, the blow molding control unit 2300 causes the seal support unit 221 to support the preheated preform 3. In step S1, the blow molding control unit 2300 causes the mold support mechanism unit 220 to close the mold 20. In step S2, the blow molding control unit 2300 causes the stretch rod 222 to advance along the central axis of the preform 3, thereby stretching the preform 3.
　ステップＳ３において、ブロー成形制御部２３００は、プレブローバルブ２２４２を開く。ステップＳ４において、ブロー成形制御部２３００は、ストレッチロッド２２２を進出させながら、プレブローエアの供給を継続させることによりプレブロー圧力を維持させる。 In step S3, the blow molding control unit 2300 opens the pre-blow valve 2242. In step S4, the blow molding control unit 2300 advances the stretch rod 222 while continuing to supply pre-blow air to maintain the pre-blow pressure.
　ステップＳ５において、ブロー成形制御部２３００は、プレブローバルブ２２４２を閉じて、プレブローエアの供給を停止するとともに、メインブローバルブ２２４３を開いてブロー圧力を上昇させる。ステップＳ６において、ブロー成形制御部２３００は、メインブローエアの供給を継続させることにより、ブロー圧力を目標圧力に維持させる。ステップＳ７において、ブロー成形制御部２３００は、メインブローバルブ２２４３を閉じるとともに、排気バルブ２２４４を開いてブロー流体を主配管２２４０から外部へ排出させる。 In step S5, the blow molding control unit 2300 closes the pre-blow valve 2242 to stop the supply of pre-blow air, and opens the main blow valve 2243 to increase the blow pressure. In step S6, the blow molding control unit 2300 maintains the blow pressure at the target pressure by continuing the supply of main blow air. In step S7, the blow molding control unit 2300 closes the main blow valve 2243 and opens the exhaust valve 2244 to discharge the blow fluid from the main pipe 2240 to the outside.
　ステップＳ８において、ブロー成形制御部２３００は、ストレッチロッド２２２をプリフォーム３内部から退出させながら、金型支持機構部２２０により金型２０を開かせる。ステップＳ９において、ブロー成形制御部２３００は、シール支持部２２１により成形体が固定されている状態を解除する。これにより、成形体は、シール支持部２２１から取出し可能にされる。 In step S8, the blow molding control unit 2300 causes the mold support mechanism 220 to open the mold 20 while retracting the stretch rod 222 from inside the preform 3. In step S9, the blow molding control unit 2300 releases the state in which the molded body is fixed by the seal support unit 221. This makes the molded body removable from the seal support unit 221.
　ブロー成形工程のうち、ステップＳ０からステップＳ４までの工程は、プレブローエアを用いた工程であり、プレブロー工程と呼ばれる。また、ブロー成形工程のうち、ステップＳ５からステップＳ９までの工程は、メインブローエアを用いた工程であり、メインブロー工程と呼ばれる。 In the blow molding process, steps S0 to S4 use pre-blow air and are called the pre-blow process. In addition, in the blow molding process, steps S5 to S9 use main blow air and are called the main blow process.
　即ち、プレブロー工程は、ブロー成形工程のうち、加熱されたプリフォーム３にストレッチロッド２２２をプリフォーム３の中心軸に沿って挿入することによりプリフォーム３を中心軸方向に延伸させるとともに、プリフォーム３内にエアを導入する工程である。 In other words, the pre-blow process is a blow molding process in which a stretch rod 222 is inserted into the heated preform 3 along the central axis of the preform 3 to stretch the preform 3 in the central axis direction and introduce air into the preform 3.
　プレブロー工程におけるプレブロー条件としては、ブロー流体としてのエアの圧力、圧力維持期間、エアの圧力が維持されているときのエアの流量、プリフォーム３の延伸量、及び、プリフォーム３の延伸速度が挙げられる。圧力維持期間は、エアの圧力を維持している期間であり、図３のステップＳ４の期間に相当する。プリフォーム３の延伸量は、プリフォーム３が中心軸方向に延伸する量である。プリフォーム３の延伸速度は、プリフォーム３が中心軸方向に延伸する速度である。 The pre-blow conditions in the pre-blow process include the air pressure as the blow fluid, the pressure maintenance period, the air flow rate when the air pressure is maintained, the stretch amount of the preform 3, and the stretch speed of the preform 3. The pressure maintenance period is the period during which the air pressure is maintained, and corresponds to the period of step S4 in FIG. 3. The stretch amount of the preform 3 is the amount by which the preform 3 stretches in the central axis direction. The stretch speed of the preform 3 is the speed at which the preform 3 stretches in the central axis direction.
　図４は、延伸量を説明するための図である。図４の左側の図は、図３のステップＳ１におけるプリフォーム３とストレッチロッド２２２の相対位置関係を示している。図４の中央の図は、図３のステップＳ２において、進出中のストレッチロッド２２２がプリフォーム３に突き当たった状態を示している。図４の右側の図は、図３のステップＳ４において、ストレッチロッド２２２が停止したときの状態を示している。このように、延伸量Ｌｘは、プリフォーム３にストレッチロッド２２２を挿入することによりプリフォーム３が延伸する量である。 Figure 4 is a diagram for explaining the stretch amount. The diagram on the left side of Figure 4 shows the relative positional relationship between the preform 3 and the stretch rod 222 in step S1 of Figure 3. The diagram in the center of Figure 4 shows the state in step S2 of Figure 3 where the advancing stretch rod 222 hits the preform 3. The diagram on the right side of Figure 4 shows the state in step S4 of Figure 3 when the stretch rod 222 stops. In this way, the stretch amount Lx is the amount by which the preform 3 is stretched by inserting the stretch rod 222 into the preform 3.
　例えば、延伸量が比較的小さい段階において、エアの圧力を増加させると、成形されたブロー成形容器の肩部分における肉厚は、ブロー成形容器の底部分における肉厚よりも厚くなる傾向がある。つまり、ブロー成形容器の中心軸方向における肉厚分布は、底部分よりも肩部分が厚くなる傾向がある。即ち、この場合、ブロー成形容器の中心軸方向における質量分布としては、底部分よりも肩部分が重くなる傾向がある。 For example, when the amount of stretching is relatively small, if the air pressure is increased, the thickness of the shoulder portion of the blow molded container tends to be thicker than the bottom portion of the blow molded container. In other words, the thickness distribution in the central axis direction of the blow molded container tends to be thicker in the shoulder portion than in the bottom portion. In other words, in this case, the mass distribution in the central axis direction of the blow molded container tends to be heavier in the shoulder portion than in the bottom portion.
　一方、延伸量が比較的大きい段階において、エアの圧力を増加させると、成形されたブロー成形容器の肩部分における肉厚は、ブロー成形容器の底部分における肉厚よりも薄くなる傾向がある。つまり、ブロー成形容器の中心軸方向における肉厚分布は、肩部分よりも底部分が厚くなる傾向がある。即ち、この場合、ブロー成形容器の中心軸方向における質量分布としては、肩部分よりも底部分が重くなる傾向がある。このように、プレブロー条件の設定と、製造後のブロー成形容器の中心軸方向の特性分布との関連性は強いことから、プレブロー条件の調整が、ブロー成形容器の中心軸方向の特性分布に大きな影響を与えるものであることがわかる。 On the other hand, when the air pressure is increased at a stage where the amount of stretching is relatively large, the thickness of the shoulder portion of the resulting blow-molded container tends to be thinner than the thickness of the bottom portion of the blow-molded container. In other words, the thickness distribution in the central axis direction of the blow-molded container tends to be thicker at the bottom portion than at the shoulder portion. In other words, in this case, the mass distribution in the central axis direction of the blow-molded container tends to be heavier at the bottom portion than at the shoulder portion. In this way, there is a strong correlation between the setting of the pre-blow conditions and the distribution of properties in the central axis direction of the blow-molded container after manufacture, and it can be seen that adjusting the pre-blow conditions has a significant impact on the distribution of properties in the central axis direction of the blow-molded container.
　図５は、第１の実施形態に係る機械学習装置の一例を示すブロック図である。機械学習装置５は、制御部５０、通信部５１、学習用データ記憶部５２、及び、学習済みモデル記憶部５３を備えている。 FIG. 5 is a block diagram showing an example of a machine learning device according to the first embodiment. The machine learning device 5 includes a control unit 50, a communication unit 51, a learning data storage unit 52, and a trained model storage unit 53.
　制御部５０は、学習用データ取得部５００及び機械学習部５０１として機能する。通信部５１は、ネットワーク７を介して外部装置と接続され、各種のデータを送受信する通信インタフェースとして機能する。外部装置は、例えば、作業者端末装置１、ブロー成形装置２、及び、特性分布測定装置８である。特性分布測定装置８は、ブロー成形容器の肉厚分布及び質量分布の少なくともいずれか１つを測定する装置である。 The control unit 50 functions as a learning data acquisition unit 500 and a machine learning unit 501. The communication unit 51 is connected to an external device via the network 7 and functions as a communication interface for sending and receiving various data. The external device is, for example, the operator terminal device 1, the blow molding device 2, and the characteristic distribution measuring device 8. The characteristic distribution measuring device 8 is a device that measures at least one of the wall thickness distribution and the mass distribution of the blow molded container.
　学習用データ取得部５００は、通信部５１及びネットワーク７を介して、作業者端末装置１、ブロー成形装置２、及び、特性分布測定装置８と接続されており、作業者端末装置１、ブロー成形装置２、及び、特性分布測定装置８の少なくともいずれか１つから学習用データ１３を取得する。学習用データ１３は、入力データと出力データとにより構成されている。 The learning data acquisition unit 500 is connected to the worker terminal device 1, the blow molding device 2, and the characteristic distribution measurement device 8 via the communication unit 51 and the network 7, and acquires learning data 13 from at least one of the worker terminal device 1, the blow molding device 2, and the characteristic distribution measurement device 8. The learning data 13 is composed of input data and output data.
　入力データは、プレブロー条件情報及び当該プレブロー条件情報に対応する特性分布情報とから構成されている。出力データは、目標プレブロー条件情報により構成されている。目標プレブロー条件情報は、ブロー成形容器の特性分布が目標の特性分布となるときのプレブロー条件の目標値に関する情報である。ブロー成形容器の特性分布は、ブロー成形容器の中心軸方向における質量分布である。なお、ブロー成形容器の特性分布は、ブロー成形容器の中心軸方向における肉厚分布であってもよい。 The input data consists of pre-blow condition information and characteristic distribution information corresponding to the pre-blow condition information. The output data consists of target pre-blow condition information. The target pre-blow condition information is information on the target value of the pre-blow condition when the characteristic distribution of the blow molded container becomes the target characteristic distribution. The characteristic distribution of the blow molded container is the mass distribution in the central axial direction of the blow molded container. The characteristic distribution of the blow molded container may also be the wall thickness distribution in the central axial direction of the blow molded container.
ここで、発明者の鋭意研究の成果から、ブロー成形容器、即ち、製品である中空容器において、中心軸方向の特性分布が強度向上や搬送性の向上等の性能向上に大きな影響を与えることが明らかとなった。 Here, the inventors' intensive research has revealed that the distribution of properties in the central axis direction of blow molded containers, i.e., the finished hollow container, has a significant impact on improving performance such as strength and transportability.
具体的には、ブロー成形容器の中心軸方向の特性分布を設定された特性分布に近づけることで、当該ブロー成形容器の各部に必要な強度を付与することができる。また、ブロー成形容器の中心軸方向の特性分布を、設定された特性分布に近づけることで、当該ブロー成形容器の全体の重量配分及びボトルの賦形が設計通りとなり、搬送中などに発生する当該ブロー成形容器の転倒を低減させることができ、重量配分が偏ることによるメインブロー時のボトルのバースト発生を低減させることができる。 Specifically, by bringing the characteristic distribution in the central axis direction of the blow molded container closer to a set characteristic distribution, it is possible to impart the necessary strength to each part of the blow molded container. In addition, by bringing the characteristic distribution in the central axis direction of the blow molded container closer to a set characteristic distribution, the overall weight distribution of the blow molded container and the shape of the bottle will be as designed, which can reduce the tipping of the blow molded container during transportation, etc., and reduce the occurrence of bottle bursting during the main blow due to uneven weight distribution.
一方で、図４に示された延伸量に基づいて前述したように、プレブロー条件の調整は、ブロー成形容器の中心軸方向の特性分布に大きな影響を与えるものである。これらのことから、ブロー成形容器の性能向上においては、ブロー成形容器の中心軸方向の特性分布を好適に付与する必要があり、そのためには、中心軸方向の特性分布に大きな影響を及ぼすプレブロー条件を適切に設定する必要があることが明らかとなった。 On the other hand, as described above based on the stretching amounts shown in Figure 4, adjusting the pre-blow conditions has a significant effect on the distribution of properties in the central axis direction of the blow-molded container. From these findings, it has become clear that in order to improve the performance of blow-molded containers, it is necessary to give the blow-molded container an appropriate distribution of properties in the central axis direction, and to do so, it is necessary to appropriately set the pre-blow conditions, which have a significant effect on the distribution of properties in the central axis direction.
　従って、学習用データ１３の入力データとして、ブロー成形容器の性能に大きな影響を及ぼす中心軸方向の特性分布情報と、特性分布に大きな影響を及ぼすプレブロー条件情報とから構成することとしている。 Therefore, the input data for the learning data 13 is composed of information on the characteristic distribution in the central axis direction, which has a significant effect on the performance of the blow-molded container, and information on the pre-blow conditions, which has a significant effect on the characteristic distribution.
　学習用データ１３は、教師あり学習における教師データ（トレーニングデータ）、検証データ及びテストデータとして用いられるデータである。また、目標プレブロー条件情報は、教師あり学習における正解ラベルとして用いられるデータである。 The learning data 13 is data used as teacher data (training data), verification data, and test data in supervised learning. In addition, the target pre-blow condition information is data used as a correct answer label in supervised learning.
　学習用データ記憶部５２は、学習用データ取得部５００において取得された複数組の学習用データ１３のセットを記憶するデータベースである。なお、学習用データ記憶部５２を構成しているデータベースの具体的な構成は、適宜設計される。 The learning data storage unit 52 is a database that stores multiple sets of learning data 13 acquired by the learning data acquisition unit 500. The specific configuration of the database that constitutes the learning data storage unit 52 is designed as appropriate.
　機械学習部５０１は、学習用データ記憶部５２に記憶されている複数組の学習用データ１３のセットを用いて機械学習を実施する。すなわち、機械学習部５０１は、学習モデル１２に複数組の学習用データ１３のセットを入力し、学習用データ１３に含まれる入力データ、即ち、プレブロー条件情報及び特性分布情報と、出力データ、即ち、目標プレブロー条件情報との相関関係を学習モデル１２に学習させることにより、学習済みの学習モデル１２を生成する。 The machine learning unit 501 performs machine learning using multiple sets of learning data 13 stored in the learning data storage unit 52. That is, the machine learning unit 501 inputs multiple sets of learning data 13 to the learning model 12, and generates a learned learning model 12 by having the learning model 12 learn the correlation between the input data included in the learning data 13, i.e., the pre-blow condition information and characteristic distribution information, and the output data, i.e., the target pre-blow condition information.
　学習済みモデル記憶部５３は、機械学習部５０１により生成された学習済みの学習モデル１２を記憶するデータベースである。学習済みの学習モデル１２は、具体的には、調整済みの重みパラメータ群である。学習済みモデル記憶部５３に記憶されている学習済みの学習モデル１２は、ネットワーク７又は記録媒体を介して実システム、例えば、ブロー成形装置２に提供される。なお、図５において、学習用データ記憶部５２及び学習済みモデル記憶部５３は別々の記憶部として示されているが、これらは、単一の記憶部により構成されていてもよい。 The trained model storage unit 53 is a database that stores the trained learning model 12 generated by the machine learning unit 501. Specifically, the trained learning model 12 is a set of adjusted weight parameters. The trained learning model 12 stored in the trained model storage unit 53 is provided to the actual system, for example, the blow molding device 2, via the network 7 or a recording medium. Note that, although the training data storage unit 52 and the trained model storage unit 53 are shown as separate storage units in FIG. 5, they may be configured as a single storage unit.
　図６は、第１の実施形態に係る学習モデル１２及び学習用データ１３の一例を示す図である。学習モデル１２の機械学習に用いられる学習用データ１３は、プレブロー条件情報と特性分布情報とにより構成されている。 FIG. 6 is a diagram showing an example of the learning model 12 and learning data 13 according to the first embodiment. The learning data 13 used for machine learning of the learning model 12 is composed of pre-blow condition information and characteristic distribution information.
　プレブロー条件情報は、成形されたブロー成形容器のプレブロー条件を含んでいる。本実施形態では、プレブロー条件情報に含まれるプレブロー条件は、プレブロー工程において維持されるエアの圧力、プレブロー工程における圧力維持期間、プレブロー工程においてエアの圧力が維持されているときのエアの流量、プリフォーム３の延伸量、及び、プリフォーム３の延伸速度である。 The pre-blow condition information includes the pre-blow conditions of the blow molded container that is produced. In this embodiment, the pre-blow conditions included in the pre-blow condition information are the air pressure maintained in the pre-blow process, the pressure maintenance period in the pre-blow process, the air flow rate when the air pressure is maintained in the pre-blow process, the stretch amount of the preform 3, and the stretch speed of the preform 3.
　特性分布情報は、上記プレブロー条件を適用して成形されたブロー成形容器の中心軸方向における質量分布の評価結果を含んでいる。質量分布は、実際に成形されたブロー成形容器を測定することにより求められる。質量分布の評価結果は、例えば、中心軸方向に沿った異なる位置ごとに複数測定された質量の平均・分散・標準偏差・最大・最小等の統計値に基づいて求めることができる。 The characteristic distribution information includes the evaluation results of the mass distribution in the central axis direction of the blow molded container molded by applying the above pre-blow conditions. The mass distribution is obtained by measuring the actual blow molded container. The evaluation results of the mass distribution can be obtained, for example, based on statistical values such as the average, variance, standard deviation, maximum, and minimum of mass measured multiple times at different positions along the central axis direction.
　なお、特性分布情報は、上記プレブロー条件を適用して成形されたブロー成形容器の中心軸方向における肉厚分布を含んでいてもよい。肉厚分布は、実際に成形されたブロー成形容器を測定することにより求められる。肉厚分布の評価結果は、例えば、中心軸方向に沿った異なる位置ごとに複数測定された肉厚の平均・分散・標準偏差・最大・最小等の統計値に基づいて求めることができる。また、特性分布情報は、質量分布の評価結果と肉厚分布の評価結果とを含んでいてもよい。質量分布及び肉厚分布は、自動測定装置を用いてセンシングすることも可能である。 The characteristic distribution information may include the thickness distribution in the central axis direction of the blow molded container molded by applying the above pre-blow conditions. The thickness distribution is obtained by measuring the blow molded container that has actually been molded. The evaluation result of the thickness distribution can be obtained, for example, based on statistical values such as the average, variance, standard deviation, maximum, and minimum of the thickness measured multiple times at different positions along the central axis. The characteristic distribution information may also include the evaluation result of the mass distribution and the evaluation result of the thickness distribution. The mass distribution and the thickness distribution can also be sensed using an automatic measuring device.
　学習用データ取得部５００は、試作されたブロー成形容器の質量分布が測定された後、試作されたブロー成形容器のプレブロー条件情報を、ブロー成形装置２又は作業者端末装置１から受信するとともに、特性分布情報を作業者端末装置１から受信する。これにより、学習用データ取得部５００は、学習用データ１３を取得する。また、学習用データ取得部５００は、正解ラベルとしての目標プレブロー条件情報を、ブロー成形装置２又は作業者端末装置１から受信する。目標プレブロー条件は、例えば、ブロー成形工程における熟練作業者によるブロー条件調整結果に基づいている。 After the mass distribution of the prototype blow-molded container is measured, the learning data acquisition unit 500 receives pre-blow condition information for the prototype blow-molded container from the blow molding device 2 or the worker terminal device 1, and also receives characteristic distribution information from the worker terminal device 1. In this way, the learning data acquisition unit 500 acquires learning data 13. In addition, the learning data acquisition unit 500 receives target pre-blow condition information as a correct label from the blow molding device 2 or the worker terminal device 1. The target pre-blow conditions are based on, for example, the results of blow condition adjustments made by a skilled worker in the blow molding process.
　学習モデル１２には、例えば、ニューラルネットワークの構造が採用されている。学習モデル１２は、入力層１２０、中間層１２１、及び、出力層１２２を備えている。入力層１２０、中間層１２１、及び、出力層１２２の間には、複数のニューロンをそれぞれ接続する複数のシナプスが張られており、各シナプスには、重みがそれぞれ対応付けられている。各シナプスの重みからなる重みパラメータ群が、機械学習により調整される。 The learning model 12 employs, for example, a neural network structure. The learning model 12 includes an input layer 120, an intermediate layer 121, and an output layer 122. A plurality of synapses are provided between the input layer 120, the intermediate layer 121, and the output layer 122, connecting a plurality of neurons, and each synapse is associated with a weight. A group of weight parameters consisting of the weights of each synapse is adjusted by machine learning.
　入力層１２０は、入力データとしてのプレブロー条件情報及び特性分布情報に対応する数のニューロンを有している。入力層１２０には、プレブロー条件情報及び特性分布情報の各値が、各ニューロンにそれぞれ入力される。出力層１２２は、出力データとしての目標プレブロー条件情報に対応する数のニューロンを有している。出力層１２２からは、プレブロー条件情報に対する目標プレブロー条件情報の予測結果、即ち、推論結果が出力データとして出力される。 The input layer 120 has a number of neurons corresponding to the pre-blow condition information and characteristic distribution information as input data. In the input layer 120, each value of the pre-blow condition information and characteristic distribution information is input to each neuron. The output layer 122 has a number of neurons corresponding to the target pre-blow condition information as output data. The output layer 122 outputs the prediction result of the target pre-blow condition information for the pre-blow condition information, i.e., the inference result, as output data.
　図７は、機械学習装置５により実行される機械学習ルーチンの一例を示すフローチャートである。図７のルーチンが開始されると、ステップＳ１００において、学習用データ取得部５００は、機械学習を開始するための事前準備として、複数の学習用データ１３を取得し、取得された学習用データ１３を学習用データ記憶部５２に記憶する。ここで、準備のために取得される学習用データの数については、最終的に得られる学習モデル１２に求められる推論精度を考慮して設定されればよい。 FIG. 7 is a flowchart showing an example of a machine learning routine executed by the machine learning device 5. When the routine of FIG. 7 is started, in step S100, the learning data acquisition unit 500 acquires multiple pieces of learning data 13 as preparation for starting machine learning, and stores the acquired learning data 13 in the learning data storage unit 52. Here, the number of pieces of learning data acquired for preparation may be set taking into consideration the inference accuracy required for the ultimately obtained learning model 12.
　次いで、ステップＳ１１０において、機械学習部５０１は、機械学習を開始するために、学習前の学習モデル１２を準備する。ここで準備する学習前の学習モデル１２は、図６に例示したニューラルネットワークモデルにより構成されている。この時点において、各シナプスの重みは、初期値に設定されている。 Next, in step S110, the machine learning unit 501 prepares a pre-learning learning model 12 in order to start machine learning. The pre-learning learning model 12 prepared here is configured with the neural network model exemplified in FIG. 6. At this point, the weights of each synapse are set to their initial values.
　次いで、ステップＳ１２０において、機械学習部５０１は、学習用データ記憶部５２に記憶されている複数組の学習用データ１３から、例えば、ランダムに１組の学習用データ１３を取得する。 Next, in step S120, the machine learning unit 501 acquires, for example, one set of training data 13 randomly from the multiple sets of training data 13 stored in the training data storage unit 52.
　次いで、ステップＳ１３０において、機械学習部５０１は、１組の学習用データ１３に含まれるプレブロー条件情報及び特性分布情報（入力データ）を、準備された学習前又は学習中の学習モデル１２の入力層１２０に入力する。その結果、学習モデル１２の出力層１２２から推論結果として目標プレブロー条件情報（出力データ）が出力される。この出力データは、学習前又は学習中の学習モデル１２によって生成されたデータであるため、推論結果として出力された出力データは、学習用データ１３に含まれる目標プレブロー条件情報（正解ラベル）とは異なる情報を示す。 Next, in step S130, the machine learning unit 501 inputs the pre-blow condition information and characteristic distribution information (input data) contained in the set of learning data 13 to the input layer 120 of the prepared learning model 12 before or during learning. As a result, target pre-blow condition information (output data) is output as an inference result from the output layer 122 of the learning model 12. Because this output data is data generated by the learning model 12 before or during learning, the output data output as an inference result indicates information different from the target pre-blow condition information (correct label) contained in the learning data 13.
　次いで、ステップＳ１４０において、機械学習部５０１は、ステップＳ１２０において取得された１組の学習用データ１３に含まれる目標プレブロー条件情報（正解ラベル）と、ステップＳ１３０において出力層１２３から推論結果として出力された目標プレブロー条件情報（出力データ）とを比較する。機械学習部５０１は、正解ラベルと出力データとの比較結果に基づいて、各シナプスの重みを調整する処理、即ち、バックプロパゲーションを実施することにより、機械学習を行う。これにより、機械学習部５０１は、プレブロー条件情報及び特性分布情報と、目標プレブロー条件情報との相関関係を学習モデル１２に学習させる。 Next, in step S140, the machine learning unit 501 compares the target pre-blow condition information (correct label) included in the set of learning data 13 acquired in step S120 with the target pre-blow condition information (output data) output from the output layer 123 as an inference result in step S130. The machine learning unit 501 performs machine learning by implementing a process of adjusting the weight of each synapse, i.e., backpropagation, based on the comparison result between the correct label and the output data. In this way, the machine learning unit 501 causes the learning model 12 to learn the correlation between the pre-blow condition information and characteristic distribution information, and the target pre-blow condition information.
　次いで、ステップＳ１５０において、機械学習部５０１は、学習終了条件が成立したか否かを判定する。例えば、機械学習部５０１は、正解ラベルと出力データとに基づく誤差関数の評価値、及び、学習用データ記憶部５２内に記憶されている未学習の学習用データ１３の残数の少なくともいずれか一方に基づいて、学習終了条件が成立したか否かを判定する。 Next, in step S150, the machine learning unit 501 determines whether or not the learning end condition is met. For example, the machine learning unit 501 determines whether or not the learning end condition is met based on at least one of the evaluation value of the error function based on the correct label and the output data, and the remaining number of unlearned learning data 13 stored in the learning data storage unit 52.
　ステップＳ１５０において、学習終了条件が成立していない場合、機械学習部５０１は、学習中の学習モデル１２に対し、未学習の学習用データ１３を用いて、ステップＳ１２０～Ｓ１４０の処理を複数回実施する。一方、ステップＳ１５０において、学習終了条件が成立している場合、機械学習部５０１は、ステップＳ１６０において、生成された学習済みの学習モデル１２（調整済みの重みパラメータ群）を学習済みモデル記憶部５３に記憶させ、本ルーチンを一旦終了する。 If the learning end condition is not met in step S150, the machine learning unit 501 performs the processes of steps S120 to S140 multiple times on the learning model 12 being trained, using untrained learning data 13. On the other hand, if the learning end condition is met in step S150, the machine learning unit 501 stores the generated trained learning model 12 (adjusted weight parameter group) in the trained model storage unit 53 in step S160, and temporarily ends this routine.
　機械学習方法において、ステップＳ１００が学習用データ記憶工程、ステップＳ１１０～ステップＳ１５０が機械学習工程、ステップＳ１６０が学習済みモデル記憶工程に相当する。 In the machine learning method, step S100 corresponds to the learning data storage process, steps S110 to S150 correspond to the machine learning process, and step S160 corresponds to the trained model storage process.
　図８は、第１の実施形態に係るブロー条件調整装置の一例を示すブロック図である。ブロー条件調整装置６は、制御部６０、通信部６１、及び、学習済みモデル記憶部６２を備えている。 FIG. 8 is a block diagram showing an example of a blow condition adjustment device according to the first embodiment. The blow condition adjustment device 6 includes a control unit 60, a communication unit 61, and a trained model storage unit 62.
　制御部６０は、情報取得部６００、推論部６０１及び出力処理部６０２として機能する。通信部６１は、ネットワーク７を介して外部装置と接続されており、各種のデータを送受信する通信インタフェースとして機能する。外部装置は、例えば、作業者端末装置１、ブロー成形装置２、及び、特性分布測定装置８である。 The control unit 60 functions as an information acquisition unit 600, an inference unit 601, and an output processing unit 602. The communication unit 61 is connected to external devices via the network 7, and functions as a communication interface for sending and receiving various data. The external devices are, for example, the operator terminal device 1, the blow molding device 2, and the characteristic distribution measuring device 8.
　情報取得部６００は、通信部６１及びネットワーク７を介して外部装置と接続されており、外部装置から入力データを取得する情報取得処理を実行する。具体的には、情報取得部６００は、作業者端末装置１又はブロー成形装置２から、入力データとして、プレブロー条件情報を取得する。また、情報取得部６００は、作業者端末装置１又は特性分布測定装置８から、入力データとして、プレブロー条件情報に対応する特性分布情報を取得する。 The information acquisition unit 600 is connected to an external device via the communication unit 61 and the network 7, and executes an information acquisition process to acquire input data from the external device. Specifically, the information acquisition unit 600 acquires pre-blow condition information as input data from the worker terminal device 1 or the blow molding device 2. The information acquisition unit 600 also acquires characteristic distribution information corresponding to the pre-blow condition information as input data from the worker terminal device 1 or the characteristic distribution measuring device 8.
　推論部６０１は、情報取得部６００により取得されたプレブロー条件情報及び当該プレブロー条件情報に対応する特性分布情報を、入力データとして学習モデル１２に入力する。推論部６０１は、学習済みモデル記憶部６２に記憶されている学習モデル１２を用いて推論処理を実行する。 The inference unit 601 inputs the pre-blow condition information acquired by the information acquisition unit 600 and the characteristic distribution information corresponding to the pre-blow condition information as input data to the learning model 12. The inference unit 601 executes the inference process using the learning model 12 stored in the trained model storage unit 62.
　学習済みモデル記憶部６２は、推論部６０１において用いられる学習済みの学習モデル１２を記憶するデータベースである。なお、学習済みモデル記憶部６２に記憶される学習モデル１２の数は１つに限定されず、例えば、機械学習の手法、ブロー成形容器の形状、プリフォーム３の材料のように、条件が異なる複数の学習済みモデルが記憶され、選択的に利用されてもよい。 The trained model storage unit 62 is a database that stores trained learning models 12 used in the inference unit 601. The number of training models 12 stored in the trained model storage unit 62 is not limited to one, and multiple trained models with different conditions, such as the machine learning method, the shape of the blow molded container, and the material of the preform 3, may be stored and selectively used.
　また、学習済みモデル記憶部６２は、外部コンピュータの記憶部によって代用されてもよく、その場合には、推論部６０１は、当該外部コンピュータにアクセスすればよい。外部コンピュータとしては、例えば、サーバ型コンピュータ及びクラウド型コンピュータが挙げられる。 The trained model storage unit 62 may be replaced by a storage unit of an external computer, in which case the inference unit 601 only needs to access the external computer. Examples of external computers include server-type computers and cloud-type computers.
　出力処理部６０２は、推論部６０１において推論された目標プレブロー条件を含む出力データを外部装置に出力する出力処理を実行する。例えば、出力処理部６０２は、生成された目標プレブロー条件情報を作業者端末装置１に送信してもよいし、ブロー成形装置２に送信してもよい。 The output processing unit 602 executes an output process to output output data including the target pre-blow conditions inferred by the inference unit 601 to an external device. For example, the output processing unit 602 may transmit the generated target pre-blow condition information to the worker terminal device 1 or to the blow molding device 2.
　このように、制御部６０は、プレブロー条件情報と特性分布情報とからなる入力データに基づいて、目標プレブロー条件情報からなる出力データを求める。 In this way, the control unit 60 determines output data consisting of target pre-blow condition information based on input data consisting of pre-blow condition information and characteristic distribution information.
　図９は、図８のブロー条件調整装置の機能の一例を示す機能説明図である。ブロー条件調整装置６において、情報取得部６００は、入力データとしてプレブロー条件情報及び当該プレブロー条件情報に対応する特性分布情報を取得し、推論部６０１に入力する。推論部６０１は、プレブロー条件情報及び当該プレブロー条件情報に対応する特性分布情報を学習モデル１２に入力し、学習モデル１２による推論結果として目標プレブロー条件情報を生成する。 FIG. 9 is a functional explanatory diagram showing an example of the function of the blow condition adjustment device of FIG. 8. In the blow condition adjustment device 6, the information acquisition unit 600 acquires pre-blow condition information and characteristic distribution information corresponding to the pre-blow condition information as input data, and inputs them to the inference unit 601. The inference unit 601 inputs the pre-blow condition information and characteristic distribution information corresponding to the pre-blow condition information to the learning model 12, and generates target pre-blow condition information as an inference result by the learning model 12.
　入力データとしては、例えば、設計プレブロー条件情報と、設計特性分布情報とが情報取得部６００に与えられる。設計プレブロー条件情報は、ブロー成形容器の設計情報に基づいて作成されたプレブロー条件に関する情報である。設計特性分布情報は、設計情報に基づいて作成されたプレブロー条件を適用して成形されたブロー成形容器の特性分布に関する情報である。 As input data, for example, design pre-blow condition information and design characteristic distribution information are provided to the information acquisition unit 600. The design pre-blow condition information is information about pre-blow conditions created based on the design information of the blow molded container. The design characteristic distribution information is information about the characteristic distribution of the blow molded container molded by applying the pre-blow conditions created based on the design information.
　図１０は、コンピュータの一例を示すハードウェア構成図である。制御ユニット２３、機械学習装置５、及び、ブロー条件調整装置６は、汎用又は専用のコンピュータ９００により構成される。 FIG. 10 is a hardware configuration diagram showing an example of a computer. The control unit 23, the machine learning device 5, and the blow condition adjustment device 6 are configured by a general-purpose or dedicated computer 900.
　コンピュータ９００は、図１０に示すように、その主要な構成要素として、バス９１０、プロセッサ９１２、メモリ９１４、入力デバイス９１６、出力デバイス９１７、表示デバイス９１８、ストレージ装置９２０、通信Ｉ／Ｆ（インタフェース）部９２２、外部機器Ｉ／Ｆ部９２４、Ｉ／Ｏ（入出力）デバイスＩ／Ｆ部９２６、及び、メディア入出力部９２８を備えている。なお、上記の構成要素は、コンピュータ９００が使用される用途に応じて適宜省略されてもよい。 As shown in FIG. 10, the computer 900 has as its main components a bus 910, a processor 912, a memory 914, an input device 916, an output device 917, a display device 918, a storage device 920, a communication I/F (interface) unit 922, an external device I/F unit 924, an I/O (input/output) device I/F unit 926, and a media input/output unit 928. Note that the above components may be omitted as appropriate depending on the application for which the computer 900 is used.
　プロセッサ９１２は、１つ又は複数の演算処理装置（ＣＰＵ（Central Processing Unit）、ＭＰＵ（Micro Processing Unit）、ＤＳＰ（Digital Signal Processor）、ＧＰＵ（Graphics Processing Unit）等）で構成されており、コンピュータ９００全体を統括する制御部として動作する。メモリ９１４は、各種のデータ及びプログラム９３０を記憶しており、例えば、メインメモリとして機能する揮発性メモリ（ＤＲＡＭ、ＳＲＡＭ等）と、不揮発性メモリ（ＲＯＭ）、フラッシュメモリ等とにより構成される。 The processor 912 is composed of one or more arithmetic processing devices (CPU (Central Processing Unit), MPU (Micro Processing Unit), DSP (Digital Signal Processor), GPU (Graphics Processing Unit), etc.) and operates as a control unit that controls the entire computer 900. The memory 914 stores various data and programs 930, and is composed of, for example, volatile memory (DRAM, SRAM, etc.) that functions as main memory, non-volatile memory (ROM), flash memory, etc.
　入力デバイス９１６は、例えば、キーボード、マウス、テンキー、電子ペン等により構成され、入力部として機能する。出力デバイス９１７は、例えば、音（音声）出力装置、バイブレーション装置等により構成され、出力部として機能する。表示デバイス９１８は、例えば、液晶ディスプレイ、有機ＥＬディスプレイ、電子ペーパー、プロジェクタ等により構成され、出力部として機能する。入力デバイス９１６及び表示デバイス９１８は、タッチパネルディスプレイのように、一体的に構成されていてもよい。ストレージ装置９２０は、例えば、ＨＤＤ、ＳＳＤ（Solid State Drive）等により構成され、記憶部として機能する。ストレージ装置９２０は、オペレーティングシステムやプログラム９３０の実行に必要な各種のデータを記憶している。 The input device 916 is, for example, a keyboard, a mouse, a numeric keypad, an electronic pen, etc., and functions as an input unit. The output device 917 is, for example, a sound (audio) output device, a vibration device, etc., and functions as an output unit. The display device 918 is, for example, a liquid crystal display, an organic EL display, electronic paper, a projector, etc., and functions as an output unit. The input device 916 and the display device 918 may be integrated, such as a touch panel display. The storage device 920 is, for example, a HDD, an SSD (Solid State Drive), etc., and functions as a memory unit. The storage device 920 stores various data necessary for the execution of the operating system and the program 930.
　通信Ｉ／Ｆ部９２２は、インターネット、イントラネット等のネットワーク９４０（図５のネットワーク７と同じであってもよい）に有線又は無線により接続され、所定の通信規格に従って他のコンピュータとの間でデータの送受信を行う通信部として機能する。外部機器Ｉ／Ｆ部９２４は、カメラ、プリンタ、スキャナ、リーダライタ等の外部機器９５０に有線又は無線により接続されており、所定の通信規格に従って外部機器９５０との間においてデータの送受信を行う通信部として機能する。Ｉ／ＯデバイスＩ／Ｆ部９２６は、各種のセンサ、アクチュエータ等のＩ／Ｏデバイス９６０に接続されており、Ｉ／Ｏデバイス９６０との間において、例えば、センサによる検出信号やアクチュエータへの制御信号等の各種の信号やデータの送受信を行う通信部として機能する。メディア入出力部９２８は、例えば、ＤＶＤドライブ、ＣＤドライブ等のドライブ装置により構成され、ＤＶＤ、ＣＤ等のメディア（非一時的な記憶媒体）９７０に対してデータの読み書きを行う。 The communication I/F unit 922 is connected to a network 940 (which may be the same as the network 7 in FIG. 5) such as the Internet or an intranet by wire or wirelessly, and functions as a communication unit that transmits and receives data to and from other computers according to a predetermined communication standard. The external device I/F unit 924 is connected to an external device 950 such as a camera, printer, scanner, or reader/writer by wire or wirelessly, and functions as a communication unit that transmits and receives data to and from the external device 950 according to a predetermined communication standard. The I/O device I/F unit 926 is connected to an I/O device 960 such as various sensors and actuators, and functions as a communication unit that transmits and receives various signals and data, such as detection signals from sensors and control signals to actuators, between the I/O device 960. The media input/output unit 928 is composed of a drive device such as a DVD drive or a CD drive, and reads and writes data to and from media (non-temporary storage media) 970 such as DVDs and CDs.
　上記構成を有するコンピュータ９００において、プロセッサ９１２は、ストレージ装置９２０に記憶されたプログラム９３０をメモリ９１４に呼び出して実行し、バス９１０を介してコンピュータ９００の各部を制御する。なお、プログラム９３０は、ストレージ装置９２０に代えて、メモリ９１４に記憶されていてもよい。プログラム９３０は、インストール可能なファイル形式又は実行可能なファイル形式によりメディア９７０に記録され、メディア入出力部９２８を介してコンピュータ９００に提供されてもよい。プログラム９３０は、通信Ｉ／Ｆ部９２２を介してネットワーク９４０経由によりダウンロードすることによりコンピュータ９００に提供されてもよい。また、コンピュータ９００は、プロセッサ９１２がプログラム９３０を実行することにより実現する各種の機能を、例えば、ＦＰＧＡ、ＡＳＩＣ等のハードウェアにより実現するものでもよい。 In the computer 900 having the above configuration, the processor 912 calls up the program 930 stored in the storage device 920 into the memory 914, executes it, and controls each part of the computer 900 via the bus 910. The program 930 may be stored in the memory 914 instead of the storage device 920. The program 930 may be recorded in the medium 970 in an installable file format or an executable file format, and provided to the computer 900 via the media input/output unit 928. The program 930 may be provided to the computer 900 by downloading it via the network 940 via the communication I/F unit 922. In addition, the computer 900 may realize various functions realized by the processor 912 executing the program 930 using hardware such as an FPGA or ASIC.
　コンピュータ９００は、例えば、据置型コンピュータや携帯型コンピュータにより構成されており、任意の形態の電子機器である。コンピュータ９００は、クライアント型コンピュータでもよいし、サーバ型コンピュータやクラウド型コンピュータでもよい。コンピュータ９００は、制御ユニット２３、機械学習装置５、及び、ブロー条件調整装置６以外の装置に適用されてもよい。 The computer 900 is, for example, a stationary computer or a portable computer, and is an electronic device of any type. The computer 900 may be a client-type computer, a server-type computer, or a cloud-type computer. The computer 900 may be applied to devices other than the control unit 23, the machine learning device 5, and the blow condition adjustment device 6.
　このように、第１の実施形態に係るブロー条件調整装置は、プリフォーム３をブロー成形するためのブロー成形工程のうち、プレブロー工程において設定される複数のプレブロー条件の目標値である目標プレブロー条件を決定する装置である。 In this way, the blow condition adjustment device according to the first embodiment is a device that determines the target pre-blow conditions, which are the target values of multiple pre-blow conditions that are set in the pre-blow process during the blow molding process for blow molding the preform 3.
　プレブロー工程は、プリフォーム３内にプレブローエアを導入する工程である。このブロー条件調整方法では、プレブロー条件情報と特性分布情報とからなる入力データに基づいて、目標プレブロー条件情報からなる出力データが求められる。プレブロー条件情報は、複数のプレブロー条件に関する情報である。特性分布情報は、当該複数のプレブロー条件を適用して成形されたブロー成形容器の質量分布に関する情報である。目標プレブロー条件情報は、目標プレブロー条件に関する情報である。 The pre-blow process is a process of introducing pre-blow air into the preform 3. In this blow condition adjustment method, output data consisting of target pre-blow condition information is obtained based on input data consisting of pre-blow condition information and characteristic distribution information. The pre-blow condition information is information on a plurality of pre-blow conditions. The characteristic distribution information is information on the mass distribution of the blow molded container molded by applying the plurality of pre-blow conditions. The target pre-blow condition information is information on the target pre-blow conditions.
　これによれば、目標プレブロー条件を決定するまでのブロー成形容器の試作回数を低減させることができることから、目標プレブロー条件を容易に決定することができる。また、例えば、金型が交換された場合及びプリフォームの材料が変更された場合のように、成形条件が変更された場合であっても、熟練作業者の技能に頼ることなく、目標プレブロー条件を決定することができる。 This makes it possible to reduce the number of prototypes of blow molded containers required to determine the target pre-blow conditions, making it easier to determine the target pre-blow conditions. Furthermore, even if molding conditions change, such as when the mold is replaced or the preform material is changed, the target pre-blow conditions can be determined without relying on the skills of a skilled worker.
　また、プレブロー条件には、プレブローエアの圧力、プレブローエアの圧力を維持している期間である圧力維持期間、プレブローエアの流量、プリフォーム３の延伸量Ｌｘ、及び、プリフォーム３の延伸速度のうちの少なくともいずれか１つである。延伸量は、プリフォーム３にストレッチロッド２２２を挿入することによりプリフォーム３が延伸する量である。延伸速度は、プリフォーム３が延伸する速度である。 The pre-blow conditions include at least one of the pre-blow air pressure, the pressure maintenance period during which the pre-blow air pressure is maintained, the pre-blow air flow rate, the stretch amount Lx of the preform 3, and the stretch speed of the preform 3. The stretch amount is the amount by which the preform 3 is stretched by inserting the stretch rod 222 into the preform 3. The stretch speed is the speed at which the preform 3 is stretched.
　ブロー成形容器の質量分布は、プレブローエアの圧力、圧力維持期間、プレブローエアの流量、プリフォーム３の延伸量Ｌｘ、及び、プリフォーム３の延伸速度に大きく依存している。従って、これらのパラメータの少なくとも１つをプレブロー条件として設定し、少なくとも１つのパラメータを調整することにより、ブロー成形容器の特性分布をより均一にすることができる。 The mass distribution of a blow molded container is highly dependent on the pre-blow air pressure, the pressure maintenance period, the pre-blow air flow rate, the stretch amount Lx of the preform 3, and the stretch speed of the preform 3. Therefore, by setting at least one of these parameters as a pre-blow condition and adjusting at least one parameter, the characteristic distribution of the blow molded container can be made more uniform.
　また、特性分布には、ブロー成形容器の質量分布、及び、ブロー成形容器の肉厚分布の少なくともいずれか１つが含まれている。 The characteristic distribution also includes at least one of the mass distribution of the blow molded container and the wall thickness distribution of the blow molded container.
　プレブローエアの圧力とプリフォーム３の延伸量Ｌｘとの関係は、主に、プリフォーム３の中心軸方向におけるブロー成形容器の肉厚のばらつきに影響を与える。従って、特性分布として、ブロー成形容器の質量分布、及び、ブロー成形容器の肉厚分布の少なくともいずれか１つを設定し、これらの少なくともいずれか１つの評価結果を機械学習装置５への入力データとすることにより、より正確な学習結果を得ることができる。 The relationship between the pre-blow air pressure and the stretch amount Lx of the preform 3 mainly affects the variation in the wall thickness of the blow molded container in the central axial direction of the preform 3. Therefore, by setting at least one of the mass distribution of the blow molded container and the wall thickness distribution of the blow molded container as the characteristic distribution and using the evaluation result of at least one of these as input data for the machine learning device 5, more accurate learning results can be obtained.
　また、入力データとして、設計プレブロー条件情報と、設計特性分布情報とが与えられる。設計プレブロー条件情報は、ブロー成形容器の設計情報に基づいて作成されたプレブロー条件に関する情報である。設計特性分布情報は、設計情報に基づいて作成されたプレブロー条件を適用して成形されたブロー成形装置の特性分布に関する情報である。 In addition, design pre-blow condition information and design characteristic distribution information are given as input data. The design pre-blow condition information is information about pre-blow conditions created based on design information for a blow-molded container. The design characteristic distribution information is information about the characteristic distribution of a blow molding device molded by applying pre-blow conditions created based on the design information.
　これによれば、設計プレブロー条件から目標プレブロー条件を得ることができる。従って、設計プレブロー条件情報及び設計特性情報をブロー条件調整装置６に入力し、目標プレブロー条件を求め、求められた目標プレブロー条件を新たな設計プレブロー条件として、ブロー条件調整装置６に入力することを繰り返し行うことにより、より適切な目標プレブロー条件をより早期に求めることができる。その結果、ブロー成形容器の設計から、ブロー成形容器を完成させるまでの開発期間を短縮することができる。 This makes it possible to obtain target pre-blow conditions from the design pre-blow conditions. Therefore, by repeatedly inputting design pre-blow condition information and design characteristic information into the blow condition adjustment device 6, determining target pre-blow conditions, and inputting the determined target pre-blow conditions into the blow condition adjustment device 6 as new design pre-blow conditions, it is possible to determine more appropriate target pre-blow conditions at an earlier stage. As a result, it is possible to shorten the development period from designing a blow-molded container to completing the blow-molded container.
　なお、入力データとして、任意のプレブロー条件及び当該プレブロー条件に対応する特性分布情報を用いても、求められた目標プレブロー条件を新たなプレブロー条件として、ブロー条件調整装置６に入力することを繰り返し行うことにより、より適切な目標プレブロー条件を求めることも可能である。 In addition, even if any pre-blow conditions and characteristic distribution information corresponding to those pre-blow conditions are used as input data, it is possible to obtain more appropriate target pre-blow conditions by repeatedly inputting the obtained target pre-blow conditions as new pre-blow conditions into the blow condition adjustment device 6.
　また、本実施形態に係るブロー条件調整装置６は、入力データが入力されると、入力データと出力データとの相関関係が機械学習により学習された学習モデル１２を用いて、目標プレブロー条件を決定し、決定された目標プレブロー条件に関する情報を出力データとして出力する。 In addition, when input data is input, the blow condition adjustment device 6 according to this embodiment determines target pre-blow conditions using a learning model 12 in which the correlation between the input data and the output data is learned by machine learning, and outputs information related to the determined target pre-blow conditions as output data.
　また、機械学習装置５は、目標プレブロー条件を推論するための学習モデル１２を生成する。機械学習装置５は、学習用データ記憶部５２と、機械学習部５０１と、学習済みモデル記憶部５３とを備えている。学習用データ記憶部５２は、プレブロー条件情報と特性分布情報とからなる入力データと、入力データに対応付けられ、目標プレブロー条件情報からなる出力データとにより構成される学習用データ１３のセットを複数組記憶する。機械学習部５０１は、学習用データ１３のセットが複数組入力されることにより、入力データと出力データとの相関関係を学習モデル１２に学習させる。学習済みモデル記憶部は、機械学習部５０１により学習された学習モデルを記憶する。 The machine learning device 5 also generates a learning model 12 for inferring target pre-blow conditions. The machine learning device 5 includes a learning data storage unit 52, a machine learning unit 501, and a learned model storage unit 53. The learning data storage unit 52 stores multiple sets of learning data 13 composed of input data consisting of pre-blow condition information and characteristic distribution information, and output data corresponding to the input data and consisting of target pre-blow condition information. The machine learning unit 501 receives multiple sets of learning data 13 and causes the learning model 12 to learn the correlation between the input data and the output data. The learned model storage unit stores the learning model learned by the machine learning unit 501.
　推論装置は、メモリ９１４と、プロセッサ９１２とを備え、目標プレブロー条件を推論する。プロセッサ９１２は、情報取得処理と、推論処理と、出力処理とを実行する。情報取得処理は、プレブロー条件情報と、特性分布情報とからなる入力データを取得する処理である。推論処理は、入力データを取得すると、メモリ９１４に格納されている機械学習による学習モデル１２を用いて、目標プレブロー条件を推論する処理である。出力処理は、推論された目標プレブロー条件を含む出力データを出力する処理である。 The inference device includes a memory 914 and a processor 912, and infers target pre-blow conditions. The processor 912 executes information acquisition processing, inference processing, and output processing. The information acquisition processing is a processing for acquiring input data consisting of pre-blow condition information and characteristic distribution information. The inference processing is a processing for inferring target pre-blow conditions using a learning model 12 based on machine learning stored in the memory 914 upon acquiring the input data. The output processing is a processing for outputting output data including the inferred target pre-blow conditions.
　また、情報処理方法は、目標プレブロー条件を決定するための方法である。情報処理方法は、プレブロー条件情報と、特性分布情報とからなる入力データに基づいて、目標プレブロー条件情報からなる出力データを求める。 The information processing method is also a method for determining target pre-blow conditions. The information processing method determines output data consisting of target pre-blow condition information based on input data consisting of pre-blow condition information and characteristic distribution information.
　機械学習方法は、目標プレブロー条件を推論するための学習モデル１２を生成する方法である。機械学習方法は、学習用データ記憶工程と、機械学習工程と、学習済みモデル記憶工程とを備えている。学習用データ記憶工程は、プレブロー条件情報と、特性分布情報とからなる入力データと、入力データに対応付けられ、目標プレブロー条件情報からなる出力データとにより構成される学習用データのセットを複数組記憶する工程である。機械学習工程は、学習用データ１３のセットが複数組入力されることにより、入力データと出力データとの相関関係を学習モデル１２に学習させる工程である。学習済みモデル記憶工程は、機械学習工程において学習された学習モデルを学習済みモデル記憶部５３に記憶させる工程である。 The machine learning method is a method of generating a learning model 12 for inferring target pre-blow conditions. The machine learning method includes a learning data storage process, a machine learning process, and a learned model storage process. The learning data storage process is a process of storing multiple sets of learning data composed of input data consisting of pre-blow condition information and characteristic distribution information, and output data corresponding to the input data and consisting of target pre-blow condition information. The machine learning process is a process of inputting multiple sets of learning data 13, thereby making the learning model 12 learn the correlation between the input data and the output data. The learned model storage process is a process of storing the learning model learned in the machine learning process in the learned model storage unit 53.
（他の実施形態）
　本発明は、上述した実施形態に制約されるものではなく、本発明の主旨を逸脱しない範囲内において種々変更して実施することが可能である。そして、それらはすべて、本発明の技術思想に含まれる。 Other Embodiments
The present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit and scope of the present invention. All such modifications are included in the technical concept of the present invention.
　例えば、プレブロー条件情報に、プリフォーム３の加熱条件に関する情報が追加されてもよい。プリフォーム３の加熱条件は、加熱温度、加熱時間、及び、これらの組合せを含んでいる。また、プリフォーム３の加熱条件ごとに学習モデルが用意されてもよい。 For example, information regarding the heating conditions of the preform 3 may be added to the pre-blow condition information. The heating conditions of the preform 3 include the heating temperature, the heating time, and a combination of these. In addition, a learning model may be prepared for each heating condition of the preform 3.
　また、上記実施形態では、学習モデル１２は、推論結果として目標プレブロー条件情報を出力していたが、推論結果として前回推論した目標プレブロー条件情報と今回推論した目標プレブロー条件情報との差分を出力してもよい。この場合に、機械学習部５０１によって学習モデル１２を学習する際にも、学習用データ１３に含まれる目標プレブロー条件情報（正解ラベル）として前回のプレブロー条件情報と今回のプレブロー条件情報との差分のデータを用意することは、当業者にとって明らかである。 In addition, in the above embodiment, the learning model 12 outputs the target pre-blow condition information as the inference result, but it may also output the difference between the previously inferred target pre-blow condition information and the currently inferred target pre-blow condition information as the inference result. In this case, it will be clear to those skilled in the art that when the machine learning unit 501 learns the learning model 12, data on the difference between the previous pre-blow condition information and the current pre-blow condition information is prepared as the target pre-blow condition information (correct answer label) included in the learning data 13.
　また、上記実施形態では、学習用データ１３に含まれる特性分布情報は、質量分布の評価結果及び肉厚分布の評価結果の少なくともいずれか１つであったが、質量分布の測定結果そのもの、及び、肉厚分布の測定結果そのものの、少なくともいずれか１つであってもよい。 In addition, in the above embodiment, the characteristic distribution information included in the learning data 13 was at least one of the evaluation results of the mass distribution and the evaluation results of the wall thickness distribution, but it may be at least one of the measurement results of the mass distribution itself and the measurement results of the wall thickness distribution itself.
　また、上記実施形態では、プレブロー条件に、延伸量が含まれていたが、延伸量に代えて、ストレッチロッド２２２が進出する量が含まれていてもよい。 In addition, in the above embodiment, the pre-blow conditions included the amount of stretching, but instead of the amount of stretching, the amount of advancement of the stretch rod 222 may be included.
　また、上記実施形態では、機械学習部５０１による機械学習を実現する学習モデル１２として、ニューラルネットワークを採用した場合について説明したが、他の機械学習のモデルが採用されてもよい。他の機械学習のモデルとしては、例えば、決定木、回帰木等のツリー型、バギング、ブースティング等のアンサンブル学習、再帰型ニューラルネットワーク、畳み込みニューラルネットワーク、ＬＳＴＭ（Long Short Term Memory）等のニューラルネット型（ディープラーニングを含む）、階層型クラスタリング、非階層型クラスタリング、ｋ近傍法、ｋ平均法等のクラスタリング型、主成分分析、因子分析、ロジスティク回帰等の多変量解析、サポートベクターマシン等が挙げられる。 In the above embodiment, a neural network is used as the learning model 12 for realizing machine learning by the machine learning unit 501, but other machine learning models may be used. Examples of other machine learning models include tree types such as decision trees and regression trees, ensemble learning such as bagging and boosting, neural network types (including deep learning) such as recurrent neural networks, convolutional neural networks, and LSTM (Long Short Term Memory), clustering types such as hierarchical clustering, non-hierarchical clustering, k-nearest neighbors, and k-means, multivariate analyses such as principal component analysis, factor analysis, and logistic regression, and support vector machines.
　また、機械学習による学習モデルに代えて、数理最適化モデルが用いられてもよい。即ち、入力データは、数理最適化モデルに入力され、出力データは、数理最適化モデルから出力されてもよい。言い換えると、ブロー条件調整装置６において、メモリ９１４に、学習モデル１２に代えて、数理最適化モデルが格納され、プロセッサ９１２がメモリ９１４に格納されている数理最適化モデルを用いて目標プレブロー条件を推論してもよい。数理最適化モデルには、例えば、ベイズ最適化や遺伝的アルゴリズムが用いられる。 Furthermore, a mathematical optimization model may be used instead of the learning model based on machine learning. That is, input data may be input to the mathematical optimization model, and output data may be output from the mathematical optimization model. In other words, in the blow condition adjustment device 6, a mathematical optimization model may be stored in the memory 914 instead of the learning model 12, and the processor 912 may infer the target pre-blow conditions using the mathematical optimization model stored in the memory 914. For example, Bayesian optimization or a genetic algorithm may be used as the mathematical optimization model.
　また、本発明は、機械学習装置５が備える各部としてコンピュータ９００を機能させるプログラム（機械学習プログラム）や、機械学習方法が備える各工程をコンピュータ９００に実行させるためのプログラム（機械学習プログラム）の態様で提供することもできる。 The present invention can also be provided in the form of a program (machine learning program) that causes the computer 900 to function as each component of the machine learning device 5, or a program (machine learning program) that causes the computer 900 to execute each step of the machine learning method.
　また、本発明は、ブロー条件調整装置６が備える各部としてコンピュータ９００を機能させるプログラム（推論プログラム）の態様で提供することもできる。その場合、ブロー条件調整装置（推論プログラム）としては、メモリ９１４と、プロセッサ９１２とを含み、このうちのプロセッサ９１２が、一連の処理を実行するものとすることができる。 The present invention can also be provided in the form of a program (inference program) that causes the computer 900 to function as each unit of the blow condition adjustment device 6. In this case, the blow condition adjustment device (inference program) includes a memory 914 and a processor 912, of which the processor 912 can execute a series of processes.
　当該一連の処理とは、１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形されたブロー成形容器の特定分布に関する情報である特性分布情報とを取得する情報取得処理（情報取得工程）と、情報取得処理によりプレブロー条件情報と特性分布情報とを取得すると、メモリ９１４に格納されている学習モデルを用いて、目標プレブロー条件を推論する推論処理（推論工程）とを含んでいる。 This series of processes includes an information acquisition process (information acquisition step) for acquiring pre-blow condition information, which is information relating to one or more pre-blow conditions, and characteristic distribution information, which is information relating to a specific distribution of a blow-molded container molded by applying the one or more pre-blow conditions, and an inference process (inference step) for inferring target pre-blow conditions using a learning model stored in memory 914 once the pre-blow condition information and characteristic distribution information have been acquired by the information acquisition process.
　ブロー条件調整装置（推論方法又は推論プログラム）の態様で提供することにより、容易に種々の装置への適用が可能となる。ブロー条件調整装置（推論方法又は推論プログラム）が目標プレブロー条件情報を推論する際、上記実施形態に係る機械学習装置５及び機械学習方法により生成された学習済みの学習モデル１２を用いて、推論部６０１が実施する推論手法を適用してもよいことは、当業者にとって当然に理解され得るものである。 By providing it in the form of a blow condition adjustment device (inference method or inference program), it can be easily applied to various devices. It will be obvious to those skilled in the art that when the blow condition adjustment device (inference method or inference program) infers the target pre-blow condition information, it may apply the inference method implemented by the inference unit 601 using the machine learning device 5 and the trained learning model 12 generated by the machine learning method according to the above embodiment.
　以上、好ましい実施の形態等について詳説したが、上述した実施の形態等に制限されることはなく、特許請求の範囲に記載された範囲を逸脱することなく、上述した実施の形態等に種々の変形及び置換を加えることができる。 The above describes preferred embodiments in detail, but the present invention is not limited to the above-described embodiments, and various modifications and substitutions can be made to the above-described embodiments without departing from the scope of the claims.
　以下、本発明の諸態様を付記としてまとめて記載する。 The various aspects of the present invention are summarized below as appendices.
　（付記１）
　プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を決定するブロー条件調整装置であって、
　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データに基づいて、前記目標プレブロー条件に関する情報である目標プレブロー条件情報からなる出力データを求める
　ブロー条件調整装置。
　（付記２）
　前記１つ以上のプレブロー条件には、前記プレブロー流体の圧力、前記圧力を維持している期間である圧力維持期間、前記プレブロー流体の流量、前記プリフォームにストレッチロッドを挿入することにより前記プリフォームが延伸する量である延伸量、及び、前記プリフォームが延伸する速度である延伸速度のうちの少なくともいずれか１つが含まれている
　付記１に記載のブロー条件調整装置。
　（付記３）
　前記特性分布には、前記製品の質量分布、及び、前記製品の肉厚分布の少なくともいずれか１つが含まれている
　付記１又は付記２に記載のブロー条件調整装置。
　（付記４）
　前記入力データとして、前記製品の設計情報に基づいて作成されたプレブロー条件に関する情報である設計プレブロー条件情報と、前記設計情報に基づいて作成されたプレブロー条件を適用して成形された前記製品の特性分布に関する情報である設計特性分布情報とが与えられる
　付記１から付記３までのいずれか１項に記載のブロー条件調整装置。
　（付記５）
　前記入力データが入力されると、前記入力データと前記出力データとの相関関係が機械学習により学習された学習モデルを用いて、前記目標プレブロー条件を決定し、決定された前記目標プレブロー条件に関する情報を前記出力データとして出力する
　付記１から付記４までのいずれか１項に記載のブロー条件調整装置。
　（付記６）
　前記入力データが入力されると、前記入力データと前記出力データとの相関関係を定義した数理最適化モデルを用いて、前記目標プレブロー条件を決定し、決定された前記目標プレブロー条件に関する情報を前記出力データとして出力する
　付記１から付記４までのいずれか１項に記載のブロー条件調整装置。
　（付記７）
　プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論するための学習モデルを生成する機械学習装置であって、
　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データと、前記入力データに対応付けられ、前記目標プレブロー条件に関する情報である目標プレブロー条件情報からなる出力データとにより構成される学習用データのセットを複数組記憶する学習用データ記憶部と、
　前記学習用データのセットが複数組入力されることにより、前記入力データと前記出力データとの相関関係を学習モデルに学習させる機械学習部と、
　前記機械学習部により学習された前記学習モデルを記憶する学習済みモデル記憶部と
　を備えている機械学習装置。
　（付記８）
　メモリと、少なくとも１つのプロセッサとを備え、プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論する推論装置であって、
　前記少なくとも１つのプロセッサは、
　　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データを取得する情報取得処理と、
　　前記入力データを取得すると、前記メモリに格納されている機械学習による学習モデルを用いて、前記目標プレブロー条件を推論する推論処理と、
　　推論された前記目標プレブロー条件を含む出力データを出力する出力処理と
　を実行する推論装置。
　（付記９）
　メモリと、少なくとも１つのプロセッサとを備え、プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論する推論装置であって、
　前記少なくとも１つのプロセッサは、
　　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データを取得する情報取得処理と、
　　前記入力データを取得すると、前記メモリに格納されている数理最適化モデルを用いて、前記目標プレブロー条件を推論する推論処理と、
　　推論された前記目標プレブロー条件を含む出力データを出力する出力処理と
　を実行する推論装置。
　（付記１０）
　プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を決定するための情報処理方法であって、
　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データに基づいて、前記目標プレブロー条件に関する情報である目標プレブロー条件情報からなる出力データを求める
　情報処理方法。
　（付記１１）
　プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論する決定するための学習モデルを生成する機械学習方法であって、
　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データと、前記入力データに対応付けられ、前記目標プレブロー条件に関する情報である目標プレブロー条件情報からなる出力データとにより構成される学習用データのセットを複数組記憶する学習用データ記憶工程と、
　前記学習用データのセットが複数組入力されることにより、前記入力データと前記出力データとの相関関係を学習モデルに学習させる機械学習工程と、
　前記機械学習工程において学習された前記学習モデルを学習済みモデル記憶部に記憶させる学習済みモデル記憶工程と
　を実行する機械学習方法。
　（付記１２）
　メモリと、少なくとも１つのプロセッサとを備える推論装置により実行されて、プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論する推論方法であって、
　前記少なくとも１つのプロセッサは、
　　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データを取得する情報取得工程と、
　　前記入力データを取得すると、前記メモリに格納されている機械学習による学習モデルを用いて、前記目標プレブロー条件を推論する推論工程と、
　　推論された前記目標プレブロー条件を含む出力データを出力する出力工程と
　を実行する推論方法。
　（付記１３）
　メモリと、少なくとも１つのプロセッサとを備える推論装置により実行されて、プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論する推論方法であって、
　前記少なくとも１つのプロセッサは、
　　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データを取得する情報取得工程と、
　　前記入力データを取得すると、前記メモリに格納されている数理最適化モデルを用いて、前記目標プレブロー条件を推論する推論工程と、
　　推論された前記目標プレブロー条件を含む出力データを出力する出力工程と
　を実行する推論方法。 (Appendix 1)
A blow condition adjusting device for determining a target pre-blow condition, which is a target value of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, among blow molding steps for blow molding a preform, comprising:
A blow condition adjusting device that calculates output data consisting of target pre-blow condition information that is information on the one or more pre-blow conditions, based on input data consisting of pre-blow condition information that is information on the one or more pre-blow conditions, and characteristic distribution information that is information on the characteristic distribution of a product molded by applying the one or more pre-blow conditions.
(Appendix 2)
The one or more pre-blow conditions include at least one of a pressure of the pre-blow fluid, a pressure maintenance period during which the pressure is maintained, a flow rate of the pre-blow fluid, a stretch amount by which the preform is stretched by inserting a stretch rod into the preform, and a stretch speed by which the preform is stretched.
(Appendix 3)
The blow condition adjusting device according to claim 1 or 2, wherein the characteristic distribution includes at least one of a mass distribution of the product and a wall thickness distribution of the product.
(Appendix 4)
The blow condition adjustment device according to any one of Supplementary Note 1 to Supplementary Note 3, wherein the input data includes design pre-blow condition information, which is information on pre-blow conditions created based on design information of the product, and design characteristic distribution information, which is information on a characteristic distribution of the product molded by applying the pre-blow conditions created based on the design information.
(Appendix 5)
When the input data is input, the target pre-blow conditions are determined using a learning model in which a correlation between the input data and the output data is learned by machine learning, and information regarding the determined target pre-blow conditions is output as the output data. The blow condition adjustment device described in any one of Supplementary Note 1 to Supplementary Note 4.
(Appendix 6)
When the input data is input, the target pre-blow conditions are determined using a mathematical optimization model that defines a correlation between the input data and the output data, and information on the determined target pre-blow conditions is output as the output data. The blow condition adjustment device described in any one of Supplementary Note 1 to Supplementary Note 4.
(Appendix 7)
A machine learning device that generates a learning model for inferring a target pre-blow condition, which is a target value of one or more pre-blow conditions set in a pre-blow process, which is a process of introducing a pre-blow fluid into a preform, among blow molding processes for blow molding the preform, comprising:
a learning data storage unit that stores a plurality of sets of learning data each composed of input data including pre-blow condition information, which is information related to the one or more pre-blow conditions, and characteristic distribution information, which is information related to a characteristic distribution of a product molded by applying the one or more pre-blow conditions, and output data, which is associated with the input data and includes target pre-blow condition information, which is information related to the target pre-blow conditions;
a machine learning unit that causes a learning model to learn a correlation between the input data and the output data by inputting a plurality of sets of the learning data;
and a trained model storage unit that stores the learning model trained by the machine learning unit.
(Appendix 8)
An inference device comprising a memory and at least one processor, the inference device infers a target pre-blow condition which is a target value of one or more pre-blow conditions set in a pre-blow step which is a step of introducing a pre-blow fluid into a preform in a blow molding step for blow molding the preform, the inference device comprising:
The at least one processor
an information acquisition process for acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions;
an inference process for inferring the target pre-blow condition by using a learning model based on machine learning stored in the memory when the input data is acquired;
and an output process for outputting output data including the inferred target pre-blow condition.
(Appendix 9)
An inference device comprising a memory and at least one processor, the inference device infers a target pre-blow condition which is a target value of one or more pre-blow conditions set in a pre-blow step which is a step of introducing a pre-blow fluid into a preform in a blow molding step for blow molding the preform, the inference device comprising:
The at least one processor
an information acquisition process for acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions;
an inference process for inferring the target pre-blow condition using a mathematical optimization model stored in the memory when the input data is acquired;
and an output process for outputting output data including the inferred target pre-blow condition.
(Appendix 10)
1. An information processing method for determining a target pre-blow condition, which is a target value of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, in a blow molding step for blow molding a preform, comprising:
An information processing method for determining output data consisting of target pre-blow condition information, which is information on the target pre-blow condition, based on input data consisting of pre-blow condition information, which is information on the one or more pre-blow conditions, and characteristic distribution information, which is information on the characteristic distribution of a product molded by applying the one or more pre-blow conditions.
(Appendix 11)
A machine learning method for generating a learning model for inferring and determining a target pre-blow condition, which is a target value of one or more pre-blow conditions set in a pre-blow process, which is a process of introducing a pre-blow fluid into a preform, among blow molding processes for blow molding the preform, comprising:
a learning data storage step of storing a plurality of sets of learning data each including input data including pre-blow condition information, which is information on the one or more pre-blow conditions, and characteristic distribution information, which is information on the characteristic distribution of a product molded by applying the one or more pre-blow conditions, and output data, which is associated with the input data and includes target pre-blow condition information, which is information on the target pre-blow conditions;
a machine learning process in which a correlation between the input data and the output data is learned by a learning model by inputting a plurality of sets of the learning data;
and a trained model storage step of storing the learned model trained in the machine learning step in a trained model storage unit.
(Appendix 12)
An inference method for inferring target pre-blow conditions, which are target values of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, in a blow molding step for blow molding a preform, the inference method comprising:
The at least one processor
an information acquisition step of acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions;
an inference step of inferring the target pre-blow condition by using a learning model based on machine learning stored in the memory when the input data is acquired;
and an output step of outputting output data including the inferred target pre-blow condition.
(Appendix 13)
An inference method for inferring target pre-blow conditions, which are target values of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, in a blow molding step for blow molding a preform, the inference method comprising:
The at least one processor
an information acquisition step of acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions;
an inference step of inferring the target pre-blow condition using a mathematical optimization model stored in the memory when the input data is obtained;
and an output step of outputting output data including the inferred target pre-blow condition.
　１２　学習モデル、１３　学習用データ、２　ブロー成形装置、２２２　ストレッチロッド、３　プリフォーム、５　機械学習装置、５０１　機械学習部、５２　学習用データ記憶部、５３　学習済みモデル記憶部、６　ブロー条件調整装置、９１２　プロセッサ、９１４　メモリ、Ｌｘ　延伸量。
  12 Learning model, 13 Learning data, 2 Blow molding device, 222 Stretch rod, 3 Preform, 5 Machine learning device, 501 Machine learning unit, 52 Learning data storage unit, 53 Learned model storage unit, 6 Blow condition adjustment device, 912 Processor, 914 Memory, Lx Stretch amount.

Claims (13)

  1. 　プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を決定するブロー条件調整装置であって、
    　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データに基づいて、前記目標プレブロー条件に関する情報である目標プレブロー条件情報からなる出力データを求める
    　ブロー条件調整装置。     A blow condition adjusting device for determining a target pre-blow condition, which is a target value of one or more pre-blow conditions set in a pre-blow process, which is a process of introducing a pre-blow fluid into a preform, among blow molding processes for blow molding the preform, comprising:
    A blow condition adjusting device that calculates output data consisting of target pre-blow condition information that is information on the one or more pre-blow conditions, based on input data consisting of pre-blow condition information that is information on the one or more pre-blow conditions, and characteristic distribution information that is information on the characteristic distribution of a product molded by applying the one or more pre-blow conditions.
  2. 　前記１つ以上のプレブロー条件には、前記プレブロー流体の圧力、前記圧力を維持している期間である圧力維持期間、前記プレブロー流体の流量、前記プリフォームにストレッチロッドを挿入することにより前記プリフォームが延伸する量である延伸量、及び、前記プリフォームが延伸する速度である延伸速度のうちの少なくともいずれか１つが含まれている
    　請求項１に記載のブロー条件調整装置。     2. The blow condition adjusting device according to claim 1, wherein the one or more pre-blow conditions include at least one of a pressure of the pre-blow fluid, a pressure maintenance period during which the pressure is maintained, a flow rate of the pre-blow fluid, a stretch amount by which the preform is stretched by inserting a stretch rod into the preform, and a stretch speed by which the preform is stretched.
  3. 　前記特性分布には、前記製品の質量分布、及び、前記製品の肉厚分布の少なくともいずれか１つが含まれている
    　請求項１又は請求項２に記載のブロー条件調整装置。     The blow condition adjusting device according to claim 1 or 2, wherein the characteristic distribution includes at least one of a mass distribution of the product and a wall thickness distribution of the product.
  4. 　前記入力データとして、前記製品の設計情報に基づいて作成されたプレブロー条件に関する情報である設計プレブロー条件情報と、前記設計情報に基づいて作成されたプレブロー条件を適用して成形された前記製品の特性分布に関する情報である設計特性分布情報とが与えられる
    　請求項１又は請求項２に記載のブロー条件調整装置。     The blow condition adjustment device according to claim 1 or claim 2, wherein the input data includes design pre-blow condition information, which is information about pre-blow conditions created based on design information of the product, and design characteristic distribution information, which is information about a characteristic distribution of the product molded by applying the pre-blow conditions created based on the design information.
  5. 　前記入力データが入力されると、前記入力データと前記出力データとの相関関係が機械学習により学習された学習モデルを用いて、前記目標プレブロー条件を決定し、決定された前記目標プレブロー条件に関する情報を前記出力データとして出力する
    　請求項１又は請求項２に記載のブロー条件調整装置。     3. The blow condition adjustment device according to claim 1 or claim 2, wherein when the input data is input, the target pre-blow conditions are determined using a learning model in which a correlation between the input data and the output data is learned by machine learning, and information regarding the determined target pre-blow conditions is output as the output data.
  6. 　前記入力データが入力されると、前記入力データと前記出力データとの相関関係を定義した数理最適化モデルを用いて、前記目標プレブロー条件を決定し、決定された前記目標プレブロー条件に関する情報を前記出力データとして出力する
    　請求項１又は請求項２に記載のブロー条件調整装置。     3. The blow condition adjustment device according to claim 1 or 2, wherein when the input data is input, the target pre-blow conditions are determined using a mathematical optimization model that defines a correlation between the input data and the output data, and information on the determined target pre-blow conditions is output as the output data.
  7. 　プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論するための学習モデルを生成する機械学習装置であって、
    　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データと、前記入力データに対応付けられ、前記目標プレブロー条件に関する情報である目標プレブロー条件情報からなる出力データとにより構成される学習用データのセットを複数組記憶する学習用データ記憶部と、
    　前記学習用データのセットが複数組入力されることにより、前記入力データと前記出力データとの相関関係を学習モデルに学習させる機械学習部と、
    　前記機械学習部により学習された前記学習モデルを記憶する学習済みモデル記憶部と
    　を備えている機械学習装置。     A machine learning device that generates a learning model for inferring a target pre-blow condition, which is a target value of one or more pre-blow conditions set in a pre-blow process, which is a process of introducing a pre-blow fluid into a preform, among blow molding processes for blow molding the preform, comprising:
    a learning data storage unit that stores a plurality of sets of learning data each composed of input data including pre-blow condition information, which is information related to the one or more pre-blow conditions, and characteristic distribution information, which is information related to a characteristic distribution of a product molded by applying the one or more pre-blow conditions, and output data, which is associated with the input data and includes target pre-blow condition information, which is information related to the target pre-blow conditions;
    a machine learning unit that causes a learning model to learn a correlation between the input data and the output data by inputting a plurality of sets of the learning data;
    and a trained model storage unit that stores the learning model trained by the machine learning unit.
  8. 　メモリと、少なくとも１つのプロセッサとを備え、プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論する推論装置であって、
    　前記少なくとも１つのプロセッサは、
    　　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データを取得する情報取得処理と、
    　　前記入力データを取得すると、前記メモリに格納されている機械学習による学習モデルを用いて、前記目標プレブロー条件を推論する推論処理と、
    　　推論された前記目標プレブロー条件を含む出力データを出力する出力処理と
    　を実行する推論装置。     An inference device comprising a memory and at least one processor, the inference device infers a target pre-blow condition which is a target value of one or more pre-blow conditions set in a pre-blow step which is a step of introducing a pre-blow fluid into a preform in a blow molding step for blow molding the preform, the inference device comprising:
    The at least one processor
    an information acquisition process for acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions;
    an inference process for inferring the target pre-blow condition by using a learning model based on machine learning stored in the memory when the input data is acquired;
    and an output process for outputting output data including the inferred target pre-blow condition.
  9. 　メモリと、少なくとも１つのプロセッサとを備え、プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論する推論装置であって、
    　前記少なくとも１つのプロセッサは、
    　　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データを取得する情報取得処理と、
    　　前記入力データを取得すると、前記メモリに格納されている数理最適化モデルを用いて、前記目標プレブロー条件を推論する推論処理と、
    　　推論された前記目標プレブロー条件を含む出力データを出力する出力処理と
    　を実行する推論装置。     An inference device comprising a memory and at least one processor, the inference device infers a target pre-blow condition which is a target value of one or more pre-blow conditions set in a pre-blow step which is a step of introducing a pre-blow fluid into a preform in a blow molding step for blow molding the preform, the inference device comprising:
    The at least one processor
    an information acquisition process for acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions;
    an inference process for inferring the target pre-blow condition using a mathematical optimization model stored in the memory when the input data is acquired;
    and an output process for outputting output data including the inferred target pre-blow condition.
  10. 　プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を決定するための情報処理方法であって、
    　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データに基づいて、前記目標プレブロー条件に関する情報である目標プレブロー条件情報からなる出力データを求める
    　情報処理方法。     1. An information processing method for determining a target pre-blow condition, which is a target value of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, in a blow molding step for blow molding a preform, comprising:
    An information processing method for determining output data consisting of target pre-blow condition information, which is information regarding the one or more pre-blow conditions, based on input data consisting of pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions.
  11. 　プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論するための学習モデルを生成する機械学習方法であって、
    　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データと、前記入力データに対応付けられ、前記目標プレブロー条件に関する情報である目標プレブロー条件情報からなる出力データとにより構成される学習用データのセットを複数組記憶する学習用データ記憶工程と、
    　前記学習用データのセットが複数組入力されることにより、前記入力データと前記出力データとの相関関係を学習モデルに学習させる機械学習工程と、
    　前記機械学習工程において学習された前記学習モデルを学習済みモデル記憶部に記憶させる学習済みモデル記憶工程と
    　を実行する機械学習方法。     A machine learning method for generating a learning model for inferring a target pre-blow condition, which is a target value of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, among blow molding steps for blow molding the preform, comprising:
    a learning data storage step of storing a plurality of sets of learning data each including input data including pre-blow condition information, which is information on the one or more pre-blow conditions, and characteristic distribution information, which is information on the characteristic distribution of a product molded by applying the one or more pre-blow conditions, and output data, which is associated with the input data and includes target pre-blow condition information, which is information on the target pre-blow conditions;
    a machine learning process in which a correlation between the input data and the output data is learned by a learning model by inputting a plurality of sets of the learning data;
    and a trained model storage step of storing the learned model trained in the machine learning step in a trained model storage unit.
  12. 　メモリと、少なくとも１つのプロセッサとを備える推論装置により実行されて、プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論する推論方法であって、
    　前記少なくとも１つのプロセッサは、
    　　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データを取得する情報取得工程と、
    　　前記入力データを取得すると、前記メモリに格納されている機械学習による学習モデルを用いて、前記目標プレブロー条件を推論する推論工程と、
    　　推論された前記目標プレブロー条件を含む出力データを出力する出力工程と
    　を実行する推論方法。     An inference method for inferring target pre-blow conditions, which are target values of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, in a blow molding step for blow molding a preform, the inference method comprising:
    The at least one processor
    an information acquisition step of acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions;
    an inference step of inferring the target pre-blow condition by using a learning model based on machine learning stored in the memory when the input data is acquired;
    and an output step of outputting output data including the inferred target pre-blow condition.
  13. 　メモリと、少なくとも１つのプロセッサとを備える推論装置により実行されて、プリフォームをブロー成形するためのブロー成形工程のうち、前記プリフォーム内にプレブロー流体を導入する工程であるプレブロー工程において設定される１つ以上のプレブロー条件の目標値である目標プレブロー条件を推論する推論方法であって、
    　前記少なくとも１つのプロセッサは、
    　　前記１つ以上のプレブロー条件に関する情報であるプレブロー条件情報と、当該１つ以上のプレブロー条件を適用して成形された製品の特性分布に関する情報である特性分布情報とからなる入力データを取得する情報取得工程と、
    　　前記入力データを取得すると、前記メモリに格納されている数理最適化モデルを用いて、前記目標プレブロー条件を推論する推論工程と、
    　　推論された前記目標プレブロー条件を含む出力データを出力する出力工程と
    　を実行する
    　推論方法。
     
          An inference method for inferring target pre-blow conditions, which are target values of one or more pre-blow conditions set in a pre-blow step, which is a step of introducing a pre-blow fluid into a preform, in a blow molding step for blow molding a preform, the inference method comprising:
    The at least one processor
    an information acquisition step of acquiring input data including pre-blow condition information, which is information regarding the one or more pre-blow conditions, and characteristic distribution information, which is information regarding the characteristic distribution of a product molded by applying the one or more pre-blow conditions;
    an inference step of inferring the target pre-blow condition using a mathematical optimization model stored in the memory when the input data is obtained;
    and an output step of outputting output data including the inferred target pre-blow condition.
PCT/JP2023/025646 2022-10-07 2023-07-11 Blow conditions adjustment device, machine learning device, inference device, information processing method, machine learning method and inference method WO2024075358A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2022162816A JP2024055693A (en) 2022-10-07 2022-10-07 BLOW CONDITION ADJUSTING DEVICE, MACHINE LEARNING DEVICE, INFERENCE DEVICE, INFORMATION PROCESSING METHOD, MACHINE LEARNING METHOD, AND INFERENCE METHOD
JP2022-162816 2022-10-07

Publications (1)

Publication Number Publication Date
WO2024075358A1 true WO2024075358A1 (en) 2024-04-11

Family

ID=90608088

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2023/025646 WO2024075358A1 (en) 2022-10-07 2023-07-11 Blow conditions adjustment device, machine learning device, inference device, information processing method, machine learning method and inference method

Country Status (2)

Country Link
JP (1) JP2024055693A (en)
WO (1) WO2024075358A1 (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2002052560A (en) * 2000-08-10 2002-02-19 Toray Ind Inc System for supporting determination of injection-molded article production parameter
JP2010511541A (en) * 2006-12-05 2010-04-15 シデル パーティシペイションズ A method of forming a hollow container from a parison by adjusting the state with feedback when the pressure inside the parison rises due to blowing.
JP2014083813A (en) * 2012-10-25 2014-05-12 Toyo Seikan Kaisha Ltd Blow molding apparatus, and blow molding method
JP2019089235A (en) * 2017-11-14 2019-06-13 株式会社Fts Molding method for blow molding article
JP2019524485A (en) * 2016-07-01 2019-09-05 ブリュックナー・マシーネンバウ・ゲーエムベーハー・ウント・コー・カーゲー Control device for forming and / or processing resin foil and resin foil molding control method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2002052560A (en) * 2000-08-10 2002-02-19 Toray Ind Inc System for supporting determination of injection-molded article production parameter
JP2010511541A (en) * 2006-12-05 2010-04-15 シデル パーティシペイションズ A method of forming a hollow container from a parison by adjusting the state with feedback when the pressure inside the parison rises due to blowing.
JP2014083813A (en) * 2012-10-25 2014-05-12 Toyo Seikan Kaisha Ltd Blow molding apparatus, and blow molding method
JP2019524485A (en) * 2016-07-01 2019-09-05 ブリュックナー・マシーネンバウ・ゲーエムベーハー・ウント・コー・カーゲー Control device for forming and / or processing resin foil and resin foil molding control method
JP2019089235A (en) * 2017-11-14 2019-06-13 株式会社Fts Molding method for blow molding article

Also Published As

Publication number Publication date
JP2024055693A (en) 2024-04-18

Similar Documents

Publication Publication Date Title
Xie et al. Data-Driven Modeling Based on Two-Stream ${\rm {\lambda}} $ Gated Recurrent Unit Network With Soft Sensor Application
Yu et al. Optimization of extrusion blow molding processes using soft computing and Taguchi’s method
Yuan et al. Stacked isomorphic autoencoder based soft analyzer and its application to sulfur recovery unit
JP7274330B2 (en) Manufacturing system, information processing method, and manufacturing method
US11669056B2 (en) Generation of a control system for a target system
Angelov et al. Adaptive inferential sensors based on evolving fuzzy models
WO2022207223A1 (en) Method and system for optimizing a digital twin simulation model using machine learning
JP7151547B2 (en) Predictive control development device, predictive control development method, and predictive control development program
CN115335830A (en) Neural architecture search with weight sharing
WO2024075358A1 (en) Blow conditions adjustment device, machine learning device, inference device, information processing method, machine learning method and inference method
Yarlagadda et al. Development of a hybrid neural network system for prediction of process parameters in injection moulding
Esche et al. Architectures for neural networks as surrogates for dynamic systems in chemical engineering
Cubeddu et al. Hybrid artificial neural network for prediction and control of process variables in food extrusion
CN114239397A (en) Soft measurement modeling method based on dynamic feature extraction and local weighted deep learning
Qiu et al. Soft sensor based on localized semi-supervised relevance vector machine for penicillin fermentation process with asymmetric data
CN111491773B (en) Molding parameter processing method for injection press
WO2024122194A1 (en) Molding condition derivation device, machine learning device, inference device, information processing method, machine learning method, and inference method
JP2024083180A (en) Molding condition derivation device, machine learning device, inference device, information processing method, machine learning method, and inference method
JP7488638B2 (en) Operation amount determination device, molding device system, molding machine, computer program, operation amount determination method, and status display device
WO2024075852A1 (en) Machine learning device, information processing device, reasoning device, machine learning method, control conditions prediction method, and reasoning method
Ma et al. Intelligent recommendation system of the injection molding process parameters based on CAE simulation, process window, and machine learning
WO2022054463A1 (en) Machine learning method, computer program, machine learning device, and molding machine
WO2021256215A1 (en) Control device, high-level controller, control method, and control program
He et al. Hybrid robust model based on an improved functional link neural network integrating with partial least square (IFLNN-PLS) and its application to predicting key process variables
US20230122304A1 (en) Manufacturing equipment control via predictive sequence to sequence models

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23874492

Country of ref document: EP

Kind code of ref document: A1