WO2022264885A1 - Dispositif de prédiction, appareil de production et procédé de prédiction - Google Patents

Dispositif de prédiction, appareil de production et procédé de prédiction Download PDF

Info

Publication number
WO2022264885A1
WO2022264885A1 PCT/JP2022/022970 JP2022022970W WO2022264885A1 WO 2022264885 A1 WO2022264885 A1 WO 2022264885A1 JP 2022022970 W JP2022022970 W JP 2022022970W WO 2022264885 A1 WO2022264885 A1 WO 2022264885A1
Authority
WO
WIPO (PCT)
Prior art keywords
fluororesin
value
prediction
polymerization
polymerization process
Prior art date
Application number
PCT/JP2022/022970
Other languages
English (en)
Japanese (ja)
Inventor
淳平 伊與田
Original Assignee
ダイキン工業株式会社
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 ダイキン工業株式会社 filed Critical ダイキン工業株式会社
Priority to CN202280042560.6A priority Critical patent/CN117500845A/zh
Publication of WO2022264885A1 publication Critical patent/WO2022264885A1/fr

Links

Images

Classifications

    • C—CHEMISTRY; METALLURGY
    • C08—ORGANIC MACROMOLECULAR COMPOUNDS; THEIR PREPARATION OR CHEMICAL WORKING-UP; COMPOSITIONS BASED THEREON
    • C08F—MACROMOLECULAR COMPOUNDS OBTAINED BY REACTIONS ONLY INVOLVING CARBON-TO-CARBON UNSATURATED BONDS
    • C08F14/00—Homopolymers and copolymers of compounds having one or more unsaturated aliphatic radicals, each having only one carbon-to-carbon double bond, and at least one being terminated by a halogen
    • C08F14/18—Monomers containing fluorine
    • C—CHEMISTRY; METALLURGY
    • C08—ORGANIC MACROMOLECULAR COMPOUNDS; THEIR PREPARATION OR CHEMICAL WORKING-UP; COMPOSITIONS BASED THEREON
    • C08F—MACROMOLECULAR COMPOUNDS OBTAINED BY REACTIONS ONLY INVOLVING CARBON-TO-CARBON UNSATURATED BONDS
    • C08F2/00—Processes of polymerisation
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring

Definitions

  • the present disclosure relates to a prediction device for predicting the MFR value in the polymerization process of fluororesin, a prediction method, and a manufacturing apparatus for manufacturing fluororesin.
  • Patent Document 1 describes a method for producing a thermoplastic resin composition. Moreover, Patent Document 1 describes a preferable range of the melt flow rate (MFR) of the fluororesin and a method for measuring the same. In this way, the performance is evaluated by measuring the MFR during the production of the fluororesin or the like.
  • MFR melt flow rate
  • the present disclosure provides a prediction device, a prediction method, and a manufacturing device capable of predicting the MFR value of a mixture containing a fluororesin without requiring measurement of the MFR in the polymerization process of the fluororesin.
  • a prediction device of the present disclosure is a prediction device for predicting the MFR value of a mixture containing the fluororesin in a polymerization tank at a timing after charging raw materials and before finishing production in a fluororesin polymerization process, wherein the fluorine an acquisition unit that acquires information including a pressure value in the polymerization tank measured in the current polymerization process of the resin and a stirring current value of a stirrer that stirs the inside of the polymerization tank as a prediction sensor value;
  • the relationship between the learning sensor value measured in the past polymerization process of the same type of fluororesin as the resin and the MFR value measured for the fluororesin at a predetermined timing in the past polymerization process has been learned by machine learning.
  • a prediction unit that predicts the MFR value at the predetermined timing of the mixture containing the fluororesin in the current polymerization process using the prediction sensor value acquired by the acquisition unit from a model.
  • the prediction device further includes a timer that measures the elapsed time from the start of the current polymerization process to the present time as a polymerization time, and the prediction unit measures the polymerization timed by the timer together with the prediction sensor value. Time can be used to predict the MFR value.
  • the prediction device may further include an output processing section that outputs information about the current completion time of the polymerization process obtained using the MFR value predicted by the prediction section.
  • the prediction device it is possible to add additional material to the polymerization tank in the polymerization process of the fluororesin, and the acquisition unit is added to the polymerization tank together with the pressure value and the stirring current value.
  • the prediction unit acquires the integrated amount of the material as a prediction sensor value, and the prediction unit calculates the prediction sensor value by a trained model that has been trained with the learning sensor value including the integrated amount obtained in the past polymerization step. can be used to predict MFR values.
  • the prediction unit can predict the MFR value using the pressure value measured during a part of the fluororesin polymerization process.
  • the prediction unit can predict the MFR value using at least one of the average value, maximum value, minimum value, or variance value of the pressure values within the partial period.
  • the prediction unit can predict the MFR value using the stirring current value measured during a part of the fluororesin polymerization process.
  • the prediction unit can predict the MFR value using the average value, maximum value, minimum value, or variance value of the stirring current values within the partial period.
  • the partial period can be selected from a plurality of periods obtained by dividing the time from the start to the end of the polymerization process of the fluororesin.
  • the fluororesin polymerization step is performed at least twice, the acquisition unit acquires the MFR value of the fluororesin obtained in the previous polymerization step, and the prediction unit acquires the MFR value obtained by the acquisition unit.
  • the MFR value of the fluororesin can be predicted using the prediction sensor value including the MFR value of the fluororesin obtained in the previous polymerization step.
  • the production apparatus of the present disclosure is a production apparatus for producing a fluororesin, comprising a polymerization vessel into which the material of the fluororesin is charged, a stirrer for stirring the inside of the polymerization vessel, and a current polymerization process of the fluororesin.
  • An acquisition unit that acquires the pressure value in the polymerization tank measured by and the stirring current value of the stirrer as prediction sensor values, and the fluororesin measured in the past polymerization process of the same type of fluororesin.
  • the relationship between the learning sensor value, the time required for the past polymerization step, and the MFR value measured for the fluororesin obtained in the past polymerization step is learned by a learned model by machine learning.
  • a prediction unit that predicts the MFR value of the mixture containing the fluororesin in the current polymerization process using the prediction sensor value acquired by the acquisition unit, and the MFR value predicted by the prediction unit. and an output processing unit for outputting information about the completion time of the polymerization process of the development.
  • the method for producing a fluororesin of the present disclosure includes the steps of: charging a fluororesin material into a polymerization tank; stirring the inside of the polymerization tank; A step of acquiring a pressure value and a stirring current value of the stirrer as prediction sensor values, a learning sensor value measured in a past polymerization process of the same type of fluororesin as the fluororesin, and the past polymerization
  • the relationship between the time required for the process and the MFR value measured for the fluororesin obtained in the past polymerization process is learned by a trained model that has been learned by machine learning, using the prediction sensor value obtained. , predicting the MFR value of the mixture containing the fluororesin in the current polymerization process, and using information about the completion time of the current polymerization process obtained using the predicted MFR value, and terminating.
  • the prediction method of the present disclosure is a prediction method for predicting the MFR value of a mixture containing the fluororesin in a polymerization tank in the polymerization process of the fluororesin, and the polymerization measured in the current polymerization process of the fluororesin
  • the relationship between the learning sensor value, the time required for the past polymerization step, and the MFR value measured for the fluororesin obtained in the past polymerization step is learned by a learned model by machine learning. and a prediction step of predicting the MFR value of the mixture containing the fluororesin in the current polymerization process, using the prediction sensor value acquired by the acquisition unit.
  • the generation method is a method of generating a trained model for predicting the MFR value of the fluororesin in a polymerization tank in the polymerization process of the fluororesin, and is measured in the polymerization process of the same type of fluororesin as the fluororesin.
  • teacher data including the sensor value and the polymerization time required for the polymerization process; and data including the MFR value measured for the fluororesin obtained in the polymerization process, which is the correct data for the teacher data.
  • the prediction device, manufacturing device, and prediction method of the present disclosure it is possible to predict the MFR value of a mixture containing a fluororesin in the polymerization process during the production of a fluororesin.
  • FIG. 4 is a conceptual diagram showing generation of a trained model
  • FIG. 4 is a conceptual diagram showing use of a trained model
  • FIG. 4 is a conceptual diagram illustrating learning data used for generating a trained model
  • It is a figure of an example of the display screen which outputs the prediction result obtained by the prediction apparatus.
  • FIG. 4 is a conceptual diagram for explaining learning data; It is a figure explaining an example of the data item of learning data.
  • 4 is a conceptual diagram illustrating an example of acquired data
  • FIG. 4 is a flowchart for explaining processing in the manufacturing apparatus;
  • the prediction device, manufacturing device, and prediction method according to the embodiment will be described below with reference to the drawings.
  • the prediction apparatus of the present disclosure makes predictions using various measured sensor values without actually measuring the MFR value of the mixture containing the fluororesin in the polymerization process.
  • the predictor is therefore a so-called "soft sensor”.
  • the manufacturing apparatus of the present disclosure predicts the MFR value of the fluororesin.
  • the prediction device will be described as being included in a manufacturing device that manufactures fluororesin.
  • the same reference numerals are given to the same configurations, and the description thereof is omitted.
  • the "polymerization process” is assumed to be the polymerization process of the fluororesin. Specifically, the “polymerization step” is described as a process from materials for producing a fluororesin to obtaining a fluororesin by a polymerization reaction.
  • fluororesin is a resin containing a fluorine atom, and is a resin obtained by polymerizing a monomer containing at least one fluorine-containing monomer.
  • the above Rf a is preferably a linear or branched perfluoroalkyl group having 1 to 3 carbon atoms, more preferably a linear or branched perfluoroalkyl group having 1 or 2 carbon atoms. In one aspect, the perfluoroalkyl group is linear. In another aspect, the perfluoroalkyl group is branched.
  • the above Rf b is preferably a linear or branched perfluoroalkylene group having 1 to 3 carbon atoms, more preferably a linear or branched perfluoroalkylene group having 1 or 2 carbon atoms. In one aspect, the perfluoroalkylene group is linear. In another aspect, the perfluoroalkylene group is branched.
  • R is a linear or branched alkyl group having 1 to 6 carbon atoms.
  • R above is preferably a linear or branched alkyl group having 1 to 3 carbon atoms, more preferably a linear or branched alkyl group having 1 or 2 carbon atoms.
  • the alkyl group is straight chain. In another aspect, the alkyl group is branched.
  • the fluororesin may be either a homopolymer of the fluoromonomer or a copolymer of two or more monomers containing at least one of the fluoromonomers.
  • fluororesin examples include polytetrafluoroethylene (PTFE), polychlorotrifluoroethylene (PCTFE), tetrafluoroethylene (TFE)/hexafluoropropylene (HFP) copolymer, and TFE/HFP/perfluoro(alkyl vinyl ether).
  • PAVE copolymer TFE/PAVE copolymer, (tetrafluoroethylene-perfluoroalkyl vinyl ether copolymer (PFA) and tetrafluoroethylene-perfluoromethyl vinyl ether copolymer (MFA)), ethylene (Et) /TFE copolymer, Et/TFE/HFP copolymer, chlorotrifluoroethylene (CTFE)/TFE copolymer, Et/CTFE copolymer, TFE/vinylidene fluoride (VDF) copolymer, VDF/HFP /TFE copolymer, VDF/HFP copolymer and the like.
  • PFA tetrafluoroethylene-perfluoroalkyl vinyl ether copolymer
  • MFA tetrafluoroethylene-perfluoromethyl vinyl ether copolymer
  • Et ethylene
  • Et/TFE/HFP copolymer Et/TFE/HFP copolymer
  • the fluororesin is a fluororesin containing tetrafluoroethylene units.
  • fluororesins containing tetrafluoroethylene units include TFE homopolymers, TFE/HFP copolymers, TFE/PAVE, TFE/ethylene, TFE/vinyl ether, TFE/vinyl ester, TFE/vinyl ester/vinyl ether , or a TFE/vinyl ether/allyl ether copolymer.
  • the fluororesin can be TFE/ethylene.
  • the fluororesin is a fluororesin containing a tetrafluoroethylene unit
  • the content of the tetrafluoroethylene unit is preferably 20 mol% to 80 mol%, more preferably 30 mol% to 70 mol%, still more preferably 35 It can be from mol % to 60 mol %.
  • the fluororesin is an ethylene/tetrafluoroethylene copolymer
  • the molar ratio of ethylene/tetrafluoroethylene is preferably 20/80 to 80/20, more preferably 25/75 to 70/30, and even more preferably can be from 30/70 to 60/40.
  • the melting point of the fluororesin is preferably 200°C or higher, more preferably 230°C or higher.
  • the melt flow rate of the fluororesin is preferably 0.1-60, more preferably 1-50.
  • the melt flow rate is a value obtained by heating a resin to be measured in a cylindrical cylinder, extruding it with a certain weight, and measuring the amount of resin extruded in 10 minutes. Under the same temperature and load conditions, the higher the MFR value, the better the fluidity.
  • the above values are an example of values in the case of a Hastelloy cylinder, a temperature condition of 300° C., and a load condition of 5.00 kg.
  • Machine learning is a method of learning features contained in input data and generating a "model” that estimates results corresponding to newly input data. Specifically, machine learning is learned by a “learner”. A model generated in this manner is referred to as a "learned model”.
  • FIG. 2A Multiple sets of learning data are input to the "learning device" as shown in FIG. 2A.
  • the learning device learns the relationship of the learning data and generates a trained model representing the relationship of the learning data with parameters. Also, by using the generated trained model, it is possible to obtain desired output for a new input, as shown in FIG. 2B.
  • Machine learning includes supervised learning, unsupervised learning, reinforcement learning, etc. In the following embodiments, an example using supervised learning will be described. Therefore, FIG. 2A also shows an example of supervised learning in which each "explanatory variable" is associated with the "objective variable” which is the correct data.
  • the learner for example, as a machine learning method, neural network (NN), support vector machine (SVM), decision tree, gradient boosting, random forest, XGBoost, linear regression, Ridge regression, Lasso regression, ElasticNet, partial minimum 2 Square regression, Gaussian process regression, principal component analysis (PCA), etc. can be used. Also, the learning device may use a combination of these methods.
  • NN neural network
  • SVM support vector machine
  • XGBoost linear regression
  • Ridge regression Lasso regression
  • ElasticNet partial minimum 2 Square regression
  • Gaussian process regression Gaussian process regression
  • PCA principal component analysis
  • learning data is the data used when creating a learning model in machine learning.
  • an "explanatory variable” and a “objective variable” which is the correct answer corresponding to the explanatory variable are used as a set of learning data, and are input to a learning device.
  • a "teacher dataset ”. After creating a learning model using a part of the prepared data set as a "learning data set”, another part of the data set is used as a "validation data set" to verify and improve the prediction accuracy of the created learning model.
  • the "explanatory variable” includes the pressure value in the polymerization tank, the stirring current value of the stirrer, and the like.
  • the "objective variable” is the MFR value of the mixture containing the fluororesin.
  • the manufacturing apparatus 1 includes a prediction device 10, a polymerization tank 2, a stirrer 3, and a control device 4.
  • Polymerization tank 2 is charged with fluororesin material. For example, materials are introduced at the start of the polymerization process. Also, some of the materials may be additionally introduced in the polymerization process.
  • the polymerization vessel 2 is hermetically sealed so as to be isolated from the outside air, pressure and temperature.
  • a pressure sensor 21 for measuring the internal pressure value and a temperature sensor 22 for measuring the internal temperature value are connected to the polymerization tank 2 . Sensor values connected by the sensors 21 and 22 are acquired by the prediction device 10 .
  • the stirrer 3 stirs the mixture containing the materials introduced into the polymerization tank 2 from the start to the end of the polymerization process during the production of the fluororesin.
  • the stirrer 3 has a motor 31 , a rotating shaft 32 connected to the motor 31 , and stirring blades 33 supported by the rotating shaft 32 .
  • the stirrer 3 can stir the mixture containing the materials in the polymerization tank 2 by rotating the stirring blades 33 positioned in the polymerization tank 2 using the motor 31 .
  • a polymerization reaction proceeds by stirring with the stirrer 3 , and a fluororesin is produced from the mixture in the polymerization tank 2 .
  • the power and rotation speed of the stirrer 3 depend on the stirring current, which is the current given to the motor 31 .
  • the control device 4 controls the stirrer 3.
  • the controller 4 controls the stirring current applied to the motor 31, for example, so that the rotation speed of the stirrer 3 is constant in the polymerization process.
  • the control device 4 may perform feedback control of the stirring current value measured by the motor 31 and the stirring current given to the motor 31 according to the rotational speed of the stirrer 3 .
  • the stirring current may be feedforward controlled using data from the prediction device 10, which will be described later.
  • the controller 4 controls the environment inside the polymerization tank 2 .
  • the control device 4 controls the compressor so that the pressure value measured by the pressure sensor 21 becomes a target value.
  • the control device 4 controls the temperature controller so that the temperature value measured by the temperature sensor 22 becomes a target value.
  • the control device 4 may feedback-control the pressure and temperature in the polymerization tank 2 according to the sensor values measured by the pressure sensor 21 and the temperature sensor 22 .
  • the pressure and temperature may be feedforward controlled using data from the prediction device 10, which will be described later.
  • the control device 4 does not necessarily require measurement and control of both pressure and temperature.
  • the control device 4 may, for example, measure and control only pressure.
  • the prediction device 10 is an information processing device that includes a control unit 11 , a storage unit 12 and a communication unit 13 .
  • the control unit 11 is a controller that controls the prediction device 10 as a whole.
  • the control unit 11 reads out and executes the prediction program 123 stored in the storage unit 12 to realize processing as the clock unit 111 , the acquisition unit 112 , the prediction unit 113 and the output processing unit 114 .
  • the control unit 11 is not limited to one that realizes a predetermined function by cooperation of hardware and software, and may be a hardware circuit designed exclusively for realizing a predetermined function.
  • the control unit 11 can be realized by various processors such as CPU, MPU, GPU, FPGA, DSP, and ASIC.
  • the storage unit 12 is a recording medium for recording various information.
  • the storage unit 12 is realized by, for example, RAM, ROM, flash memory, SSD (Solid State Device), hard disk, other storage devices, or an appropriate combination thereof.
  • the storage unit 12 stores the prediction program 123 executed by the control unit 11 as well as various data used for learning and prediction.
  • the storage unit 12 stores acquired data 121 , a trained model 122 , and a prediction program 123 .
  • the communication unit 13 is an interface circuit (module) for enabling data communication with an external device (not shown).
  • the prediction device 10 can also include an input unit 14 and an output unit 15 .
  • the input unit 14 is input means such as operation buttons, a mouse, and a keyboard used for inputting operation signals and data.
  • the output unit 15 is output means such as a display used for outputting processing results and data.
  • the prediction device 10 may be realized by one computer, or may be realized by a combination of multiple computers connected via a network. Also, although illustration is omitted, for example, all or part of the data stored in the storage unit 12 is stored in an external recording medium connected via a network, and the prediction device 10 stores data in the external recording medium. It may be configured to use stored data.
  • the timer unit 111 measures the elapsed time from the start of the fluororesin polymerization process in the manufacturing apparatus 1 as the polymerization time.
  • the fluororesin polymerization step can be performed multiple times.
  • the timer 111 measures the polymerization time for each polymerization step. Therefore, the clock unit 111 clocks the polymerization time of the current polymerization process.
  • the acquisition unit 112 acquires the pressure value in the polymerization tank measured in the polymerization process of the fluororesin and the stirring current value of the stirrer that stirs the material in the polymerization tank or the mixture containing the fluororesin as prediction sensor values. do.
  • the acquisition unit 112 associates the acquired prediction sensor value with the time information and stores it in the storage unit 12 as the acquired data 121 .
  • the time information is the superimposed time measured by the timer 111 .
  • the time information may be the time, but it is desirable that the time information be associated with the polymerization time at the time of data acquisition.
  • the timing at which the acquisition unit 112 acquires each sensor value can be determined according to the mode of use.
  • the timing for acquiring the sensor value is a regular timing such as every second, minute, or ten minutes.
  • the prediction unit 113 predicts the MFR value of the fluororesin in the current polymerization process at a predetermined timing, using the prediction sensor values acquired by the acquisition unit 112, based on a learned model that has been learned by machine learning.
  • the predetermined timing is timing such as every 10 hours.
  • each timing need not be evenly spaced as long as it is a predetermined timing.
  • the prediction unit 113 can predict the MFR value using the prediction sensor value and the polymerization time clocked by the clock unit 111 .
  • This trained model consists of the learning sensor values measured in the past polymerization process of the same type of fluororesin as the fluororesin manufactured in the current polymerization process, the time required for the past polymerization process, and the time required for the past polymerization process. It is generated by learning the relationship with the MFR value measured for the obtained fluororesin. The generation of this trained model will be detailed later.
  • the prediction unit 113 can predict the MFR value using pressure values measured during a part of the fluororesin polymerization process.
  • the prediction unit 113 can also predict the MFR value using the stirring current value measured during a part of the fluororesin polymerization process.
  • This partial period is selected from a plurality of periods obtained by dividing the time from the start to the end of the fluororesin polymerization process.
  • the prediction unit 113 can predict the MFR value using the average value of pressure values measured during a partial period.
  • the prediction unit 113 can predict the MFR value using the average value of the stirring current values measured during a partial period.
  • the partial period used for averaging the pressure values and the partial period used for averaging the stirring current values may be the same or different.
  • the prediction in the prediction unit 113 will also be detailed later. In this way, the prediction device 10 can efficiently improve the prediction accuracy of the MFR value by using the sensor values of the partial period that have a high degree of influence on the MFR value.
  • the MFR value is likely to vary depending on the operator, making highly accurate measurement difficult.
  • by predicting the MFR value by the prediction unit 113 using sensor values a highly reliable MFR value can be obtained.
  • the prediction unit 113 uses the predicted MFR value to obtain information about the current polymerization end time.
  • the 'information about end time' is the 'estimated end time' at which the current polymerization is expected to end.
  • the 'information about the end time' may be the 'remaining time' required from the present until the polymerization process is expected to end.
  • the 'information about end time' may be the timing at which a specific operation for the end of manufacturing is performed. Specific operations for the end of production include, for example, stopping the charging of raw materials, stopping stirring, lowering the tank pressure, and starting cooling.
  • the "information on end time" can be obtained according to sensor values obtained in the production of the polymer.
  • the prediction unit 113 can obtain information about the end time by using the obtained MFR value and a predetermined calculation formula or correspondence formula. In this way, by obtaining the information about the end time by the prediction unit 113, in the polymerization process of the fluororesin, the information about the end time can be determined without actually taking out the polymer from the polymerization tank and measuring its performance. becomes possible.
  • the output processing unit 114 outputs information regarding the current polymerization end time obtained by the prediction unit 113 .
  • the accuracy of the fluororesin obtained by the production apparatus 1 can be improved by using the information on the predicted completion time in this way to judge the completion of the polymerization process.
  • the output processing unit 114 may also output the expected MFR value of the current mixture itself.
  • the output processing unit 114 displays, for example, a display unit w1 that displays a graph representing changes in the predicted MFR value from the start as shown in FIG. 4, and a display unit w2 that displays the "remaining time" required until the The screen W is output to the output unit 15 .
  • the display portion w1 may include an index t1 indicating the current time and an index t2 indicating the end point predicted by the prediction portion 113.
  • the prediction unit 113 uses a learning device to generate a trained model.
  • a learning device is contained in the prediction part 113, it is not limited to this.
  • the learner may exist outside the prediction device 10 .
  • the prediction unit 113 uses a trained model generated by an external learning device and stored in the storage unit 12 .
  • the “explanatory variables” used by the prediction unit 113 for machine learning are, as described above with reference to FIG.
  • the data includes the "stirring current value” of the stirrer 3 and the "polymerization time” which is the time required for the past polymerization process.
  • the correct “objective variable” is the "MFR value” measured for the fluororesin polymerized in the corresponding past polymerization process.
  • the learning device uses multiple sets of data including the explanatory variable and the objective variable. Therefore, the trained model used by the prediction unit 113 indicates the relationship between a plurality of "pressure value”, “stirring current value”, “polymerization time” and “MFR value” obtained in the past polymerization process with parameters. is.
  • the MFR value measured for the fluororesin polymerized in the past polymerization process includes the MFR value measured for the fluororesin finally obtained as a result, as well as the sampling during the polymerization process. It may also include the MFR value of the mixture withdrawn from the run.
  • the prediction unit 113 predicts the MFR value using the pressure value and the stirring current value from the learned model generated by machine learning in this way.
  • the learning device may use learning data obtained by preprocessing instead of using the plurality of pressure values and current values as learning data as learning data.
  • the learning device can use learning data including, as explanatory variables, an average value of a plurality of pressure values and an average value of a plurality of stirring current values in a specific period. For example, as shown in FIG.
  • groups are set by equally dividing the total time for each different polymerization process, and the average values of the pressure value and the stirring current value are obtained for each group. is used as the training data with the explanatory variables.
  • the polymerization steps A to C are assumed to be polymerization steps of the same type of fluororesin performed at different timings.
  • the example of FIG. 5 is an example in which the polymerization time for each polymerization step A to C is divided into a plurality of groups of 10 hours each, and the values used in the learning data are obtained for each group.
  • the pressure values and stirring current values measured from the start to 10 hours are the data for Group 1, and the pressure values and stirring current values measured from 10 hours to 20 hours after stirring.
  • Group 2 data is the current value.
  • the average value of the pressure values acquired from the start to 10 hours after the start (pressure average value A1) and the average value of the stirring current values acquired at the same time (stirring current average The value a1) is learning data for group 1.
  • the pressure values and stirring current values measured from 10 hours after the start to 20 hours after the start serve as data for Group 2. Therefore, the average value of the pressure values acquired from 10 hours after the start to 20 hours after the start (pressure average value A2) and the average value of the stirring current values acquired at the same time (stirring current average value b2) , become the training data for group 2. Similarly, the pressure average value and stirring current average value are obtained for groups 3 to 20, and these serve as learning data for each group in the polymerization process A. Similarly, for the polymerization steps B and C, each group is determined according to the polymerization time, and the pressure average value and the stirring current average value are obtained for each group and used as learning data. The learner can use the learning data thus preprocessed. In the example shown in FIG. 5, each group has been described as having 10 hours, but this time is not limited to this and can be arbitrarily set according to the situation.
  • the value used for learning differs depending on which data is used to determine the MFR value at which timing.
  • the learning device does not need to use the average pressure value and the average stirring current value of all groups as learning data, and may use the average pressure value and average stirring current value of a specific group as learning data.
  • FIG. 6 is an example of data items of learning sensor values used in the learning device.
  • the average pressure values of groups 20 and 19, which correspond to the final stage of the polymerization process, and the average pressure value of group 1, which corresponds to the initial stage of the polymerization process are used as the learning sensor values.
  • the average stirring current values of groups 20 and 18 corresponding to the initial stage of the polymerization process and the average stirring current value of group 8 corresponding to the middle period of the polymerization process are used as learning sensor values.
  • the learning device uses sensor values obtained during a selected partial period of the polymerization process as learning sensor values.
  • the sensor values used here are determined by selecting valid sensor values from each group. In this case, it is not necessary to obtain the average pressure value and the average stirring current value for all the groups, and it is sufficient to obtain the average pressure value and the average stirring current value for the selected group.
  • the learning device includes the "average pressure value” of each group, which is the “explanatory variable”, the “average stirring current value” of each group, and the “polymerization time”, and the "MFR value” which is the “objective variable”. is obtained as a learned model that predicts the MFR value from inputs such as the pressure value and the stirring current value.
  • the prediction unit 113 uses the learned model generated as described above and the sensor values acquired by the acquisition unit 112 to predict the MFR value of the mixture in the polymerization vessel in the current polymerization process.
  • the learning device learned using learning data including the average pressure value and the average stirring current value for each of a plurality of groups set for a partial period of the polymerization process. Therefore, the prediction unit 113 also preprocesses the acquired data using the average values obtained for the same number of groups as the learning data. Then, using the preprocessed acquired data, the MFR value is predicted by the trained model.
  • the prediction unit 113 divides the acquired data 121 into groups at predetermined timings, and averages the pressure value and the stirring current value for each group. and predict the MFR value using the trained model.
  • the trained model trained using only the data of group 1 is used.
  • a trained model trained using the data of Groups 1 to 13 may be used.
  • the timing at which the prediction unit 113 predicts the MFR value is the timing at which all prediction sensor values of a predetermined group required for prediction are obtained. Therefore, the prediction unit 113 does not necessarily have to predict the MFR value each time a group ends. For example, in the example shown in FIG. 7, after 10 hours, after 20 hours, after 30 hours, etc., are considered to be sufficiently shorter than the typical end time of the polymerization process. Therefore, in such a case, without predicting the MFR value, for example, the MFR at the end of each group after the elapse of a predetermined time, determined from the shortest time of the past polymerization steps, in which there is no possibility of the polymerization step ending, value can be predicted.
  • the control device 4 under the control of the control device 4, the control of the environment in the polymerization tank 2 and the stirrer 3 is started.
  • the timer 111 starts measuring the polymerization time (S2).
  • a polymerization reaction of the fluororesin is started in the polymerization tank 2 under the control of the controller.
  • the acquisition unit 112 starts acquiring input sensor values (S3).
  • the obtaining unit 112 obtains, for example, the pressure value measured by the pressure sensor 21 and the stirring current value of the current applied to the motor 31 .
  • the prediction unit 113 predicts the MFR value using the trained model 122 (S4).
  • the predictive sensor values used in the trained model 122 may be preprocessed.
  • the prediction unit 113 predicts the end time using the MFR value predicted in step S4 (S5).
  • the end time here is the timing of a specific operation for the end of manufacturing, or the like, and the end may be determined in a later process according to this timing.
  • the output processing unit 114 outputs the end time predicted in step S5 (S6).
  • the output processing unit 114 can output the end time by displaying the display screen W as described above with reference to FIG. 4 on the output unit 15, for example.
  • the prediction device 10 returns to step S3 and repeats the processes of steps S3 to S7.
  • the end of the polymerization process is determined according to the end time predicted in step S5. For example, when the prediction unit 113 predicts the time required for completion, if the time predicted in step S5 becomes 0, the polymerization process is terminated.
  • the prediction device 10 can predict the MFR value using sensor values without actually measuring the MFR value of the mixture. Moreover, the prediction device 10 can improve the prediction accuracy of the MFR value by using the polymerization time. As a result, the manufacturing apparatus 1 can use the MFR value predicted by the prediction apparatus 10 to predict the completion timing of the fluororesin polymerization in the fluororesin polymerization step. As a result, by using the MFR value predicted by the prediction device 10, there is no need to actually measure the MFR value by taking out a part of the mixture from the polymerization tank during the polymerization process for measuring the MFR value. can do.
  • the prediction device 10 is described as generating the learned model 122 using the average pressure value and the average stirring current value, which are sensor values, as the learning data.
  • the maximum value, minimum value, or variance value of the sensor values may be used.
  • the average value, the maximum value, the minimum value and the variance value may be used in combination.
  • the prediction unit 113 predicts the MFR value using a combination of at least one of the average value, maximum value, minimum value and variance value.
  • the prediction device 10 can effectively use various values obtained from the measured values to improve the prediction accuracy of the MFR value.
  • the acquisition unit 112 acquires the pressure value and the stirring current value as well as the integrated amount of the material introduced into the polymerization tank 2 as the prediction sensor value.
  • the prediction unit 113 uses the learned model 122 that has been learned with learning data including the integrated amounts of materials used in the past polymerization as well as the pressure value and the stirring current value. The prediction unit 113 also predicts the MFR value using the prediction sensor value including the integrated amount of the material acquired by the acquisition unit 112 .
  • the prediction device 10 can improve the prediction accuracy by predicting the MFR value using the integrated amount of material. This is because the amount of material input affects the MFR value of the fluororesin in the polymerization process. In this manner, the prediction device 10 can improve the prediction accuracy of the MFR value by using the integrated amount of materials to be input.
  • the acquisition unit 112 may acquire the MFR value of the fluororesin obtained in the previous polymerization step. In other words, the obtaining unit 112 may obtain the MFR value of the fluororesin obtained in the previous batch.
  • the prediction unit 113 uses the learned model 122 trained with learning data including the MFR value of the fluororesin obtained in the previous polymerization step, along with each sensor value. Further, the prediction unit 113 predicts the MFR value using the MFR value of the fluororesin obtained in the previous polymerization process obtained by the obtaining unit 112 together with the prediction sensor value.
  • the prediction accuracy can be further improved.
  • various data used in the polymerization process or various data acquired in the polymerization process can be used as learning data.
  • the same kind of data is used as the data used in the trained model.
  • temperature values measured by the temperature sensor 22 may be used as the learning sensor value and the prediction sensor value.
  • the number of trained models 122 used by the prediction unit 113 is not mentioned, but the number of trained models used by the prediction unit 113 is not limited.
  • the prediction unit 113 may use two or more trained models 122 in combination.
  • the prediction device, manufacturing device, and prediction method described in all claims of this disclosure are realized by cooperation with hardware resources, such as processors, memories, and programs.
  • the prediction device, manufacturing device, and prediction method of the present disclosure are useful, for example, in specifying the timing of completion of polymerization in the polymerization process of fluororesin.

Landscapes

  • Chemical & Material Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Medicinal Chemistry (AREA)
  • Polymers & Plastics (AREA)
  • Organic Chemistry (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Automation & Control Theory (AREA)
  • Addition Polymer Or Copolymer, Post-Treatments, Or Chemical Modifications (AREA)
  • Testing And Monitoring For Control Systems (AREA)
  • Polymerisation Methods In General (AREA)

Abstract

La présente invention prédit, lors de la production d'une résine fluorée, une valeur de MFR d'un mélange contenant la résine fluorée lors d'une étape de polymérisation. Un dispositif de prédiction (10) est destiné à prédire une valeur de MFR d'une résine fluorée dans une cuve de polymérisation à un moment prescrit après l'entrée de matière première mais avant la fin de la production lors d'une étape de polymérisation pour obtenir la résine fluorée. Le dispositif de prédiction comprend : une unité d'acquisition (112) qui acquiert, en tant que valeurs de capteur de prédiction, une valeur de pression dans la cuve de polymérisation mesurée au cours de la présente étape de polymérisation pour obtenir la résine fluorée et une valeur de courant électrique d'agitation d'un agitateur chargé d'effectuer une agitation dans la cuve de polymérisation ; et une unité de prédiction (113) qui prédit une valeur de MFR au moment prescrit d'un mélange contenant la résine fluorée au cours de la présente étape de polymérisation à partir des valeurs de capteur de prédiction acquises par l'unité d'acquisition (112), à l'aide d'un modèle formé formé par apprentissage machine réalisé sur des relations entre des valeurs de capteur de formation mesurées au cours d'étapes de polymérisation passées pour obtenir des résines fluorées du même type que ladite résine fluorée et des valeurs de MFR mesurées pour les résines fluorées au moment prescrit pendant les étapes de polymérisation passées.
PCT/JP2022/022970 2021-06-17 2022-06-07 Dispositif de prédiction, appareil de production et procédé de prédiction WO2022264885A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202280042560.6A CN117500845A (zh) 2021-06-17 2022-06-07 预测装置、制造装置和预测方法

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2021101048A JP2023000306A (ja) 2021-06-17 2021-06-17 予測装置、製造装置及び予測方法
JP2021-101048 2021-06-17

Publications (1)

Publication Number Publication Date
WO2022264885A1 true WO2022264885A1 (fr) 2022-12-22

Family

ID=84527455

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2022/022970 WO2022264885A1 (fr) 2021-06-17 2022-06-07 Dispositif de prédiction, appareil de production et procédé de prédiction

Country Status (3)

Country Link
JP (1) JP2023000306A (fr)
CN (1) CN117500845A (fr)
WO (1) WO2022264885A1 (fr)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH05140229A (ja) * 1991-11-22 1993-06-08 Mitsubishi Kasei Corp ポリオレフインの製造方法
JPH05140230A (ja) * 1991-11-22 1993-06-08 Mitsubishi Kasei Corp ポリオレフインを製造するための重合反応運転支援装置
JP2003076934A (ja) * 2001-09-03 2003-03-14 Tosoh Corp ポリマーの物性予測方法及びそれを用いたプラントの運転制御方法
JP2007191657A (ja) * 2006-01-23 2007-08-02 Hitachi Ltd ポリマー重合装置
JP2019151833A (ja) * 2018-02-28 2019-09-12 ダイキン工業株式会社 熱可塑性樹脂組成物およびその製造方法
WO2020054183A1 (fr) * 2018-09-10 2020-03-19 富士フイルム株式会社 Dispositif et procédé d'aide à la réaction d'écoulement, équipement et procédé de réaction d'écoulement
WO2022004880A1 (fr) * 2020-07-03 2022-01-06 ダイキン工業株式会社 Dispositif de prédiction, dispositif de calcul, dispositif de production et procédé de production

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH05140229A (ja) * 1991-11-22 1993-06-08 Mitsubishi Kasei Corp ポリオレフインの製造方法
JPH05140230A (ja) * 1991-11-22 1993-06-08 Mitsubishi Kasei Corp ポリオレフインを製造するための重合反応運転支援装置
JP2003076934A (ja) * 2001-09-03 2003-03-14 Tosoh Corp ポリマーの物性予測方法及びそれを用いたプラントの運転制御方法
JP2007191657A (ja) * 2006-01-23 2007-08-02 Hitachi Ltd ポリマー重合装置
JP2019151833A (ja) * 2018-02-28 2019-09-12 ダイキン工業株式会社 熱可塑性樹脂組成物およびその製造方法
WO2020054183A1 (fr) * 2018-09-10 2020-03-19 富士フイルム株式会社 Dispositif et procédé d'aide à la réaction d'écoulement, équipement et procédé de réaction d'écoulement
WO2022004880A1 (fr) * 2020-07-03 2022-01-06 ダイキン工業株式会社 Dispositif de prédiction, dispositif de calcul, dispositif de production et procédé de production

Also Published As

Publication number Publication date
CN117500845A (zh) 2024-02-02
JP2023000306A (ja) 2023-01-04

Similar Documents

Publication Publication Date Title
CN110377447B (zh) 一种异常数据检测方法、装置及服务器
US11275345B2 (en) Machine learning Method and machine learning device for learning fault conditions, and fault prediction device and fault prediction system including the machine learning device
US20230119336A1 (en) Prediction device, calculation device, manufacturing device, and manufacturing method
WO2020188328A1 (fr) Procédé de réalisation d'un processus et optimisation de signaux de commande utilisés dans le procédé
JP2017120649A (ja) 故障条件を学習する機械学習方法及び機械学習装置、並びに該機械学習装置を備えた故障予知装置及び故障予知システム
JP7274330B2 (ja) 製造システム、情報処理方法、および製造方法
US8250006B2 (en) Inferential sensors developed using three-dimensional pareto-front genetic programming
JP2017083985A (ja) 時系列データ処理方法、時系列データ処理プログラム、および、時系列データ処理装置
WO2005019948A1 (fr) Modélisation et fabrication de produits assistées par ordinateur
JP6711323B2 (ja) プロセスの異常状態診断方法および異常状態診断装置
US20230376855A1 (en) Anomaly detection and tuning recommendation system
US20200293011A1 (en) Production system, information processing method, and production method
WO2020136859A1 (fr) Dispositif d'estimation, procédé d'estimation et support d'enregistrement lisible par ordinateur
WO2022264885A1 (fr) Dispositif de prédiction, appareil de production et procédé de prédiction
JP5216643B2 (ja) 混練判定システム
CN108959787B (zh) 考虑实际工况的宏宏双驱动***的热变形预测方法及***
CN112966399B (zh) 一种基于机器学习的脉管制冷机工况预测方法及***
Derks et al. Parallel processing of chemical information in a local area network—II. A parallel cross-validation procedure for artificial neural networks
TWI825490B (zh) 用於通過近紅外光譜學和機器學習技術在製造過程中進行端點檢測之設備和方法，及相關電腦可讀取媒體
JP7243082B2 (ja) 混練異常度学習装置、混練異常度推定装置、混練異常度学習方法、混練異常度推定方法及びプログラム
JP2008165412A (ja) 性能演算装置
KR101837464B1 (ko) 폴리프로필렌 중합반응 시 중합체의 시뮬레이션 방법
JP2011145905A (ja) 予測関数生成装置、方法、及び、プログラム
Wei et al. An adaptive soft sensor based on multi-state partial least squares regression
Othman et al. Polymer composition control in emulsion co-and terpolymerizations

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: 22824877

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 202280042560.6

Country of ref document: CN

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 13/03/2024)