Learning device, prediction device, learning method, and prediction method

Abstract

Provided is a learning device that generates a machine learning model for predicting impact strength of a polymer composite material, the learning device comprising: a first learning unit that causes a first machine learning model to learn correspondence between material information and transition temperatures of brittle fracture and ductile fracture of a polymer composite material to be learned; a first prediction unit that outputs a predicted transition temperature of the polymer composite material to be learned by inputting the material information of the polymer composite material to be learned into the trained first machine learning model; and a second learning unit that causes a second machine learning model for predicting impact strength of a polymer composite material to be predicted to learn correspondence between the material information of the polymer composite material to be learned and the predicted transition temperature, and the impact strength.

Description

Learning device, prediction device, learning method, and prediction method

[0001] The present disclosure relates to a learning device, a prediction device, a learning method, and a prediction method.

[0002] Since polymer composite materials are lightweight and have excellent mechanical properties, they are widely used in various fields such as aerospace, automobiles, electronic devices, or construction materials. In particular, in applications where excellent physical properties such as heat resistance, impact resistance, abrasion resistance, and chemical stability are required, their performance is extremely important.

[0003] The physical properties of polymer composite materials greatly depend on the temperature change of the use environment. As a result, the temperature change of the use environment may affect the reliability and durability of polymer composite materials. Impact strength is known as a physical property with high temperature dependence.

[0004] Non-Patent Document 1 discloses that brittle fracture is likely to occur at low temperatures, and at high temperatures, the motion mode changes and ductile fracture becomes dominant, resulting in a significant change in physical properties. In addition, Patent Document 1 discloses a method for predicting the physical properties of polymer composite materials from parameters such as the composition and structure of polymer composite materials.

[0005] Japanese Patent No. 7397949

[0006] Shigeki Higa, "Effect of Temperature on Impact Properties of PP / Elastomer / Filler Composites," Journal of the Adhesion Society of Japan Vol. 48 No. 10 (2012) p. 348-353

[0007] However, in the method of predicting the physical properties of polymer composite materials from parameters such as the composition and structure of polymer composite materials, there is a problem that the prediction accuracy decreases near the transition temperature where the motion mode changes.

[0008] An object of the present disclosure is to provide a learning device and a learning method for generating a machine learning model that predicts the impact strength of polymer composite materials with higher accuracy, and a prediction device and a prediction method for predicting the impact strength of polymer composite materials with higher accuracy.

[0009] The present disclosure has the following configuration.

[0010] [1] A learning device for generating a machine learning model for predicting the impact strength of a polymer composite material, comprising: a first learning unit that causes a first machine learning model to learn the correspondence between the transition temperatures of brittle fracture and ductile fracture of a polymer composite material to be learned and material information; a first prediction unit that outputs a predicted transition temperature value of the polymer composite material to be learned by inputting the material information of the polymer composite material to be learned into the learned first machine learning model; and a second learning unit that causes a second machine learning model for predicting the impact strength of a polymer composite material to be predicted to learn the correspondence between the material information of the polymer composite material to be learned and the predicted transition temperature value and impact strength.

[0011] [2] The learning apparatus according to [1], further comprising: an extraction unit for extracting the transition temperature of the polymer composite material to be learned from a plurality of measurement data obtained by measuring the impact strength of the polymer composite material to be learned at a plurality of temperatures.

[0012] [3] The learning device according to [1] or [2], wherein the material information includes a first information which is an index representing the fluidity of the thermoplastic resin contained in the polymer composite material; a second information which is the weight fraction of the ethylene-propylene copolymer in a polypropylene resin containing a propylene polymer and an ethylene-propylene copolymer; a third information which is the weight fraction of the propylene component of the ethylene-propylene copolymer in the polypropylene resin; a fourth information which is the intrinsic viscosity of the polypropylene resin; a fifth information which is the weight fraction of the sum of the second information derived from the polypropylene resin in the compound and the amount of elastomer; a sixth information which is the amount of inorganic filler added; and a seventh information which is the weight fraction of the sum of the amount of polypropylene polymer and the amount of the propylene component of the ethylene-propylene copolymer in the compound.

[0013] [4] A prediction device for predicting the impact strength of a polymer composite material, comprising: an acquisition unit that acquires material information of a polymer composite material to be predicted; a second prediction unit that outputs a predicted transition temperature value of the polymer composite material to be predicted by inputting the material information of the polymer composite material to be predicted into a first machine learning model that has been trained to determine the correspondence between the transition temperatures of brittle fracture and ductile fracture of a polymer composite material to be learned and the material information; and a third prediction unit that outputs a predicted impact strength value of the polymer composite material to be predicted by inputting the material information of the polymer composite material to be predicted and the predicted transition temperature value of the polymer composite material to be predicted into a second machine learning model that has been trained to determine the correspondence between the material information of the polymer composite material to be learned, the predicted transition temperature value of the polymer composite material to be learned predicted by the trained first machine learning model, and the impact strength.

[0014] [5] The prediction device according to [4], wherein the material information includes a first information which is an index representing the fluidity of the thermoplastic resin contained in the polymer composite material; a second information which is the weight fraction of the ethylene-propylene copolymer in a polypropylene resin containing a propylene polymer and an ethylene-propylene copolymer; a third information which is the weight fraction of the propylene component of the ethylene-propylene copolymer in the polypropylene resin; a fourth information which is the intrinsic viscosity of the polypropylene resin; a fifth information which is the weight fraction of the sum of the second information derived from the polypropylene resin in the compound and the amount of elastomer; a sixth information which is the amount of inorganic filler; and a seventh information which is the weight fraction of the sum of the amount of polypropylene polymer and the amount of the propylene component of the ethylene-propylene copolymer in the compound.

[0015] [6] A learning method for generating a machine learning model in which a learning device predicts the impact strength of a polymer composite material, comprising: a first learning step of causing a first machine learning model to learn the correspondence between the transition temperatures of brittle fracture and ductile fracture of a polymer composite material to be learned and material information; a first prediction step of inputting material information of the polymer composite material to be learned into the learned first machine learning model to output a predicted transition temperature value of the polymer composite material to be learned; and a second learning step of causing a second machine learning model for predicting the impact strength of a polymer composite material to be predicted to learn the correspondence between the material information of the polymer composite material to be learned and the predicted transition temperature value and impact strength.

[0016] [7] A prediction method for predicting the impact strength of a polymer composite material using a prediction device, comprising: an acquisition step of acquiring material information of a polymer composite material to be predicted; a second prediction step of outputting a predicted transition temperature of the polymer composite material to be predicted by inputting the material information of the polymer composite material to be predicted into a first machine learning model that has been trained to determine the correspondence between the transition temperatures of brittle fracture and ductile fracture of a polymer composite material to be learned and the material information; and a third prediction step of outputting a predicted impact strength of the polymer composite material to be predicted by inputting the material information of the polymer composite material to be predicted and the predicted transition temperature of the polymer composite material to be predicted into a second machine learning model that has been trained to determine the correspondence between the material information of the polymer composite material to be learned, the predicted transition temperature of the polymer composite material to be learned predicted by the trained first machine learning model, and the impact strength.

[0017] According to this disclosure, we can provide a learning device and learning method for generating a machine learning model that predicts the impact strength of polymer composite materials with higher accuracy, and a prediction device and prediction method for predicting the impact strength of polymer composite materials with higher accuracy.

[0018] This is a diagram showing the configuration of an example of the physical property prediction system 1 according to this embodiment. This is a diagram showing the hardware configuration of an example of the computer according to this embodiment. This is a diagram showing the functional configuration of an example of the physical property prediction system 1 according to this embodiment. This is a flowchart showing an example of the processing of the learning device 10 according to this embodiment. This is an image diagram showing an example of the measurement of Charpy impact strength. This is a graph showing an example of the change in measurement data of the impact strength of a polymer composite material to be learned with respect to temperature. This is an explanatory diagram showing a part of the dataset from which the transition temperature has been extracted. This is a graph showing an example of the relationship between the measurement temperature "Temp" and the impact strength measurement data "Charpy" in the dataset of Figure 7. This is an explanatory diagram showing a part of the dataset used to train the first machine learning model. This is a graph showing an example of the prediction accuracy of the trained first machine learning model. This is an explanatory diagram showing a part of the dataset used to train the second machine learning model. This is a graph showing an example of the prediction accuracy of the trained second machine learning model. This is a flowchart showing an example of the processing of the prediction device 12 according to this embodiment. This is an explanatory diagram showing an example of the processing of the prediction device 12 according to this embodiment.

[0019] Next, embodiments of the present invention will be described in detail. However, the present invention is not limited to the following embodiments.

[0020] [First Embodiment] <System Configuration> Figure 1 is a configuration diagram of an example of a physical property prediction system 1 according to this embodiment. The physical property prediction system 1 in Figure 1 has a learning device 10, a prediction device 12, and a worker terminal 14. The learning device 10 and the prediction device 12 may be implemented as a physical property prediction device 16. The learning device 10, the prediction device 12, and the worker terminal 14 of the physical property prediction system 1 shown in Figure 1 are connected to enable data communication via a communication network 18 such as a local area network (LAN) or the Internet.

[0021] The worker terminal 14 is an information processing device such as a PC, tablet, or smartphone operated by the worker. The worker terminal 14 displays a screen and accepts operations from the worker. The worker terminal 14 transmits the content of the operations received from the worker to the learning device 10 or prediction device 12. The worker terminal 14 displays the results of the processing by the learning device 10 or prediction device 12 on the screen for the worker to confirm.

[0022] The learning device 10 is an information processing device such as a PC or workstation that executes a process to generate a machine learning model for predicting the impact strength of polymer composite materials based on the content of the operation received from the worker. The learning device 10 may display the processing results on a display device such as the display of the worker terminal 14 for the worker to confirm, or it may display them on the display device of the learning device 10 for the worker to confirm. The learning device 10 may also display a screen on the display device and accept operations from the worker.

[0023] The prediction device 12 is an information processing device such as a PC or workstation that performs a process to predict the impact strength of a polymer composite material based on the content of the operation received from the operator. The prediction device 12 uses a machine learning model trained by the learning device 10 to predict the impact strength of the polymer composite material. The prediction device 12 may display the processing results on a display device such as the display of the operator terminal 14 for the operator to confirm, or it may display them on the display device of the prediction device 12 for the operator to confirm. The prediction device 12 may also display a screen on the display device and accept operations from the operator.

[0024] In the material property prediction system 1 according to this embodiment, a first machine learning model and a second machine learning model are constructed. The first machine learning model predicts the transition temperatures for brittle fracture and ductile fracture of a polymer composite material from the material information of the polymer composite material. The second machine learning model predicts the impact strength of the polymer composite material from the transition temperature predicted by the first machine learning model (hereinafter referred to as the predicted transition temperature value) and the material information of the polymer composite material.

[0025] Thus, the property prediction system 1 according to this embodiment provides a prediction method that predicts the impact strength of a polymer composite material from its material information with higher accuracy by using a first machine learning model and a second machine learning model. Furthermore, since the property prediction system 1 according to this embodiment can predict the impact strength of a polymer composite material from its material information, the amount of experimental work can be reduced.

[0026] It should be noted that the physical property prediction system 1 shown in Figure 1 is merely an example, and there are various system configurations depending on the application and purpose. For example, the learning device 10 and the prediction device 12 may be implemented using multiple computers, or they may be implemented as a cloud computing service. Alternatively, the learning device 10 and the prediction device 12 may be implemented as a single computer as the physical property prediction device 16. Furthermore, the physical property prediction system 1 shown in Figure 1 may be implemented using a standalone computer.

[0027] <Hardware Configuration> The learning device 10, prediction device 12, and worker terminal 14 in Figure 1 are implemented by a computer 500 with the hardware configuration shown in Figure 2, for example.

[0028] Figure 2 is a hardware configuration diagram of an example of a computer according to this embodiment. The computer 500 shown in Figure 2 includes an input device 501, a display device 502, an external interface 503, RAM 504, ROM 505, CPU 506, a communication interface 507, and an HDD 508, and each of these is interconnected via bus B. The input device 501 and the display device 502 may also be used by connecting them to the computer 500 via the external interface 503.

[0029] The input device 501 includes a touch panel, operation keys and buttons, a keyboard and mouse, etc., used by the operator to input various signals. The display device 502 consists of a display such as a liquid crystal or organic EL that displays the screen, and a speaker that outputs sound data such as voice and sound.

[0030] The communication interface 507 is an interface for the computer 500 to perform data communication. The HDD 508 is an example of a non-volatile storage device that stores programs and data. The programs and data stored include the OS, which is the basic software that controls the entire computer 500, and applications that provide various functions on the OS. The computer 500 may also use a drive device that uses flash memory as a storage medium (for example, a solid-state drive: SSD, etc.) instead of the HDD 508.

[0031] The external I / F 503 is an interface to an external device. An external device may be a recording medium 503a, etc. This allows the computer 500 to read from and / or write to the recording medium 503a via the external I / F 503. The recording medium 503a may be a flexible disk, CD, DVD, SD memory card, or USB memory, etc.

[0032] ROM 505 is an example of a non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. ROM 505 stores programs and data such as the BIOS, OS settings, and network settings that are executed when the computer 500 starts up. RAM 504 is an example of a volatile semiconductor memory (storage device) that temporarily holds programs and data.

[0033] The CPU 506 is an arithmetic unit that controls and implements the functions of the entire computer 500 by reading programs and data from storage devices such as ROM 505 and HDD 508 onto RAM 504 and executing processing. In this embodiment, the computer 500 can implement various functions of the learning device 10, prediction device 12, and worker terminal 14 described later by executing a program. The CPU 506 may also read and execute a program from the recording medium 503a where the program is stored via an external I / F 503.

[0034] <Functional Configuration> The configuration of the physical property prediction system 1 according to this embodiment will now be described. Figure 3 is a functional configuration diagram of an example of the physical property prediction system 1 according to this embodiment. Note that parts of the configuration diagram in Figure 3 that are not necessary for the explanation of this embodiment have been appropriately omitted.

[0035] The learning device 10 has a configuration comprising a request receiving unit 20, a response transmission unit 22, a first learning unit 24, a first prediction unit 26, a second learning unit 28, an extraction unit 30, a dataset storage unit 32, and a machine learning model storage unit 34. The prediction device 12 has a configuration comprising a request receiving unit 40, a response transmission unit 42, an acquisition unit 44, a second prediction unit 46, a third prediction unit 48, and a trained machine learning model storage unit 50. The operator terminal 14 has a configuration comprising an information display unit 60, an operation reception unit 62, a request transmission unit 64, and a response receiving unit 66.

[0036] The information display unit 60 of the worker terminal 14 displays a screen for the worker to view on the display device 502. The operation reception unit 62 receives various operations from the worker. The request transmission unit 64 transmits a processing request to the learning device 10 or prediction device 12 based on the various operations received from the worker. The response reception unit 66 receives a response from the learning device 10 or prediction device 12 to the processing request transmitted by the request transmission unit 64.

[0037] The request receiving unit 20 of the learning device 10 receives a processing request from the worker terminal 14 based on the worker's operation. The response transmission unit 22 transmits a response to the processing request received by the request receiving unit 20 to the worker terminal 14.

[0038] The first learning unit 24 trains the first machine learning model on the correspondence between the transition temperatures of brittle fracture and ductile fracture of the polymer composite material to be studied and the material information. The first learning unit 24 trains the first machine learning model by using a training dataset in which the material information of the polymer composite material to be studied, which is the explanatory variable, and the transition temperatures of brittle fracture and ductile fracture of the polymer composite material to be studied, which is the objective variable.

[0039] The first prediction unit 26 inputs the material information of the polymer composite material to be learned into the first machine learning model learned by the first learning unit 24, and outputs a predicted value of the transition temperature of the polymer composite material to be learned.

[0040] The second learning unit 28 causes the second machine learning model that predicts the impact strength of the polymer composite material to be predicted to learn the correspondence between the material information of the polymer composite material to be learned, the predicted value of the transition temperature, and the impact strength. The second learning unit 28 uses a learning dataset in which the material information and the predicted value of the transition temperature of the polymer composite material to be learned, which are explanatory variables, and the impact strength of the polymer composite material to be learned, which is the target variable, are associated with each other to learn the second machine learning model.

[0041] The extraction unit 30 extracts the transition temperature of the polymer composite material to be learned from a plurality of measurement data obtained by measuring the impact strength of the polymer composite material to be learned at a plurality of temperatures. The transition temperature of the polymer composite material to be learned extracted by the extraction unit 30 is used as the target variable of the learning dataset of the first machine learning model.

[0042] The dataset storage unit 32 stores the datasets used by the first learning unit 24 or the second learning unit 28 for machine learning. The machine learning model storage unit 34 stores the machine learning models learned by the first learning unit 24 or the second learning unit 28. The machine learning model storage unit 34 may store the machine learning models learned by the first learning unit 24 or the second learning unit 28.

[0043] In this way, the learning device 10 can generate the first machine learning model that predicts the transition temperature of the polymer composite material from the material information of the polymer composite material. Further, the learning device 10 can generate the second machine learning model that predicts the impact strength of the polymer composite material from the material information and the predicted value of the transition temperature of the polymer composite material.

[0044] The request receiving unit 40 of the prediction device 12 receives a processing request based on the operation of the operator from the operator terminal 14. The response transmitting unit 42 transmits a response to the processing request received by the request receiving unit 40 to the operator terminal 14.

[0045] The acquisition unit 44 acquires the material information of the polymer composite material to be predicted, for example, from the operator terminal 14.

[0046] The second prediction unit 46 inputs the material information of the polymer composite material to be predicted into the first machine learning model that has been learned by the learning device 10 for the correspondence between the transition temperature of brittle fracture and ductile fracture of the polymer composite material to be learned and the material information, so as to output the predicted value of the transition temperature of the polymer composite material to be predicted.

[0047] The third prediction unit 48 uses the second machine learning model that has been learned for the correspondence between the material information of the polymer composite material to be learned, the predicted value of the transition temperature of the polymer composite material to be learned predicted by the learned first machine learning model, and the impact strength. The third prediction unit 48 inputs the material information of the polymer composite material to be predicted and the predicted value of the transition temperature of the polymer composite material to be predicted output by the second prediction unit 46 into the learned second machine learning model, so as to output the predicted value of the impact strength of the polymer composite material to be predicted.

[0048] The learned machine learning model storage unit 50 stores the machine learning model that has been learned by the learning device 10. A regression model can be applied to the machine learning model. For example, an algorithm of Gaussian process regression can be applied to the machine learning model. Note that linear regression, least squares regression, multiple regression, logistic regression, random forest regression and other algorithms may be used for the machine learning model.

[0049] In this way, the prediction device 12 can predict the transition temperature of the polymer composite material to be predicted from the material information of the polymer composite material by using the first machine learning model that predicts the transition temperature of the polymer composite material from the material information. Further, the prediction device 12 can predict the impact strength of the polymer composite material to be predicted by using the second machine learning model that predicts the impact strength of the polymer composite material from the material information and the predicted value of the transition temperature.

[0050] <Process> <<Learning method>> FIG. 4 is a flowchart of an example of the processing of the learning device 10 according to the present embodiment. [[ID=十六]] [[ID=十七]]

[0051] In step S10, the extraction unit 30 of the learning device 10 acquires measurement data of the impact strength of the polymer composite material to be learned. The measurement data of the impact strength of the polymer composite material to be learned acquired in step S10 consists of multiple measurement data of the impact strength of the polymer composite material to be learned measured at multiple temperatures. Each item in the dataset will be explained using polypropylene resin as an example of a polymer composite material.

[0052] The impact strength obtained in step S10 can be, for example, the Charpy impact strength. The Charpy impact strength is measured, for example, as follows:

[0053] (1) In accordance with JIS K6921-2, a multipurpose test piece (Type A1) as specified in JIS K7139 is injection molded from polypropylene resin using an injection molding machine (FANUC ROBOSHOT S2000i manufactured by FANUC Corporation) under the following conditions: molten resin temperature of 200°C, mold temperature of 40°C, average injection speed of 200 mm / second, holding pressure time of 40 seconds, and total cycle time of 60 seconds.

[0054] (2) In accordance with JIS K7111-1, a notching tool A-4 manufactured by Toyo Seiki Seisakusho Co., Ltd. is used to process the material to a width of 10 mm, a thickness of 4 mm, and a length of 80 mm, and then a 2 mm notch is made in the width direction to obtain a test piece for measurement.

[0055] (3) For the test specimens used for measurement, a fully automatic impact testing machine with a low-temperature chamber (No. 258-ZA) manufactured by Yasuda Seiki Seisakusho Co., Ltd. is used to measure the Charpy impact strength (edgewise impact, 1eA method) while increasing the temperature in 5°C increments from -30°C. Near temperatures where the impact strength value changes significantly, the temperature interval is shortened as needed for measurement.

[0056] Figure 5 is an illustrative diagram of an example of Charpy impact strength measurement. Charpy impact strength is measured by applying a high-speed impact to a notched, prismatic test specimen to break it, and then evaluating the energy required for fracture and the resulting toughness.

[0057] Charpy impact strength is measured using the following formula (1) based on the impact energy absorbed by the test specimen and the cross-sectional area of ​​the test specimen.

[0058]

[0059] Figure 6 is a graph showing an example of how impact strength measurement data for a polymer composite material being studied changes with temperature. In the example in Figure 6, the physical properties change between 20 and 40°C. In the example in Figure 6, around 30°C is the transition temperature at which the state changes from one dominated by brittle fracture to one dominated by ductile fracture.

[0060] In step S12, the extraction unit 30 extracts the transition temperature of the polymer composite material to be learned from the measurement data of the impact strength of the polymer composite material to be learned acquired in step S10, using the following equation (2).

[0061]

[0062] In equation (2), "a" is the rate of increase, "b" is the inflection point, "c" is the lower asymptote, and "d" is the upper asymptote. The extraction unit 30 fits equation (2) to the temperature dependence data of the Charpy impact strength of the polymer composite material to be learned, "Charpy(T)", acquired in step S10, and extracts "b". In this embodiment, the inflection point "b" in equation (2) is defined as the transition temperature. Fitting is the process of extracting "a", "b", "c", and "d" so that the curve of equation (2) matches the plot of the measurement data. The extraction unit 30 extracts the extracted "b" as the transition temperature of the polymer composite material to be learned, acquired in step S10.

[0063] Figure 7 is an explanatory diagram showing a portion of the dataset from which transition temperatures were extracted. Figure 8 is an example graph visualizing the relationship between the measured temperature "Temp" and the impact strength measurement data "Charpy" from the dataset in Figure 7.

[0064] "Charpy" represents the measured impact strength data (actual values).

[0065] "MFR" is one of the indicators that represent the flowability (fluidity) of thermoplastic resins (plastics). "MFR" is measured according to JIS K6921-2 under conditions of a temperature of 230°C and a load of 2.16 kg. "MFR" is an example of the first type of information that represents the fluidity of thermoplastic resins contained in polymer composite materials.

[0066] "BIPO" is the weight fraction of ethylene-propylene copolymer in a polypropylene resin (heterophasic copolymer) containing propylene polymer and ethylene-propylene copolymer. However, in the case of compounds with added elastomers or talc, it is the weight fraction of ethylene-propylene copolymer in the compound. "BIPO" is an example of the second piece of information, which is the weight fraction of ethylene-propylene copolymer in a polypropylene resin containing propylene polymer and ethylene-propylene copolymer.

[0067] "Pc" is the weight fraction of propylene in the ethylene-propylene copolymer component of the polypropylene resin. "Pc" is an example of the third piece of information, which is the weight fraction of propylene in the ethylene-propylene copolymer component of the polypropylene resin.

[0068] "XSIV" is the intrinsic viscosity measured at 135°C after collecting the xylene-soluble component of a polypropylene resin, dissolving it in tetrahydronaphthalene, and observing the result. "XSIV" is an example of the fourth type of information regarding the intrinsic viscosity of polypropylene resins.

[0069] "Tot_Rubber" represents the weight fraction of the sum of the amount of "BIPO" derived from the polypropylene resin in the compound and the amount of elastomer. "Tot_Rubber" is an example of a fifth piece of information, which is the weight fraction of the second piece of information derived from the polypropylene resin in the compound and the sum of the amounts of elastomer.

[0070] "Talc" is the amount of talc used as an inorganic filler. "Talc" is an example of the sixth piece of information, which represents the amount of inorganic filler.

[0071] "Tot_PP" is the weight fraction of the sum of the amount of polypropylene polymer in the compound and the amount of propylene component in the ethylene-propylene copolymer. "Tot_PP" is an example of the seventh piece of information, which is the weight fraction of the sum of the amount of polypropylene polymer in the compound and the amount of propylene component in the ethylene-propylene copolymer.

[0072] "Temp" is the measured temperature. "MFR", "BIPO", "Pc", "XSIV", "Tot_Rubber", "Talc", and "Tot_PP" shown in Figure 7 are examples of material information.

[0073] In step S14, the first learning unit 24 trains the first machine learning model using the dataset shown in Figure 9, for example, which associates the transition temperature and material information of the polymer composite material to be learned, extracted by the extraction unit 30 in step S12.

[0074] Figure 9 is an explanatory diagram showing a portion of the dataset used to train the first machine learning model. The dataset in Figure 9 associates the transition temperature of the polymer composite material being trained with material information.

[0075] The first learning unit 24 uses the material information of the polymer composite material to be learned as an explanatory variable and the transition temperature of the polymer composite material to be learned as an objective variable, and trains the first machine learning model to learn the correspondence between the material information of the polymer composite material to be learned and the transition temperature. The algorithm of the first machine learning model is, for example, Gaussian process regression. The learning method is, for example, five-fold cross-validation.

[0076] For example, the prediction accuracy of a first machine learning model trained on a dataset partially shown in Figure 9 is as shown in Figure 10. Figure 10 is an example graph showing the prediction accuracy of a trained first machine learning model. Prediction accuracy can be expressed, for example, by the test data loss value (MSE) or (R²). For example, in the example in Figure 10, the test data loss value (MSE) is "168.6" and the test data loss value (R²) is "0.485".

[0077] In step S16, the first prediction unit 26 outputs a predicted transition temperature value for the polymer composite material to be learned by inputting material information of the polymer composite material to be learned into the first machine learning model that has been trained by the first learning unit 24.

[0078] In step S18, the second learning unit 28 trains a second machine learning model that predicts the impact strength of a polymer composite material by learning the material information of the polymer composite material to be learned, the predicted transition temperature, and the correspondence between these values ​​and the impact strength.

[0079] Figure 11 is an explanatory diagram showing a portion of the dataset used to train the second machine learning model. The dataset in Figure 11 associates material information and predicted transition temperatures of the polymer composite material to be trained with the impact strength of the polymer composite material to be trained.

[0080] The second learning unit 28 uses the material information and predicted transition temperature of the polymer composite material to be learned as explanatory variables, and the impact strength of the polymer composite material to be learned as the objective variable, to train the second machine learning model on the correspondence between the material information and predicted transition temperature of the polymer composite material to be learned and the impact strength of the polymer composite material to be learned. The algorithm of the second machine learning model is, for example, Gaussian process regression. The learning method is, for example, five-fold cross-validation.

[0081] Here, we will explain using a second machine learning model for comparison, which was trained on a dataset in which the predicted transition temperature of the polymer composite material being trained was removed from the explanatory variables, in order to compare it with the prediction accuracy of the second machine learning model, which was trained on a dataset partially shown in Figure 11.

[0082] For example, the prediction accuracy of a second machine learning model trained on the dataset partially shown in Figure 11 is as shown in Figure 12A. Furthermore, the prediction accuracy of a comparative second machine learning model trained by removing the predicted transition temperature of the polymer composite material from the explanatory variables of the dataset partially shown in Figure 11 is as shown in Figure 12B. Figures 12A and 12B are example graphs showing the prediction accuracy of the trained second machine learning model. Prediction accuracy can be expressed, for example, by the loss value (MSE) or (R²) of the test data.

[0083] For example, in the case of Figure 12A, the test data loss (MSE) of the second machine learning model trained on the dataset partially shown in Figure 11 is "21.8" and the test data loss (R2) is "0.961". Also, in the case of Figure 12B, the test data loss (MSE) of the second comparative machine learning model trained on the dataset partially shown in Figure 11, excluding the predicted transition temperature of the polymer composite material, is "28.6" and the test data loss (R2) is "0.949".

[0084] A second machine learning model, trained with material information and predicted transition temperatures of the polymer composite material as explanatory variables and the impact strength of the polymer composite material as the objective variable, shows improved prediction accuracy compared to a comparative second machine learning model trained by excluding the predicted transition temperature from the explanatory variables.

[0085] In particular, the second machine learning model, which was trained using the material information and predicted transition temperature of the polymer composite material as explanatory variables and the impact strength of the polymer composite material as the objective variable, shows improved prediction accuracy around the transition temperature (e.g., around 40-60°C) compared to the second comparative machine learning model, which was trained by removing the predicted transition temperature from the explanatory variables.

[0086] In this way, the learning device 10 can generate a first machine learning model and a second machine learning model for accurately predicting the impact strength of a polymer composite material from the material information of the polymer composite material.

[0087] <<Prediction Processing>> Figure 13 is a flowchart of an example of the processing of the prediction device 12 according to this embodiment. Figure 14 is an explanatory diagram of an example of the processing of the prediction device 12 according to this embodiment.

[0088] In step S30, the acquisition unit 44 of the prediction device 12 acquires material information of the polymer composite material to be predicted from, for example, the operator terminal 14.

[0089] In step S32, the second prediction unit 46 outputs a predicted transition temperature value for the polymer composite material by inputting the material information of the polymer composite material to be predicted into the first machine learning model 100, which has already been learned by the learning device 10 to determine the correspondence between the transition temperatures of brittle fracture and ductile fracture of the polymer composite material to be learned and the material information.

[0090] In step S34, the third prediction unit 48 uses the material information of the polymer composite material to be learned, and the second machine learning model 102 which has been trained to determine the correspondence between the predicted transition temperature of the polymer composite material to be learned, as predicted by the trained first machine learning model 100, and the impact strength. The third prediction unit 48 inputs the material information of the polymer composite material to be predicted and the predicted transition temperature of the polymer composite material output by the trained first machine learning model 100 to the trained second machine learning model 102, and outputs a predicted impact strength value for the polymer composite material to be predicted.

[0091] In step S36, for example, the response transmission unit 42 of the prediction device 12 transmits the predicted impact strength value of the polymer composite material to be predicted to the operator terminal 14 as a result of the processing for predicting the polymer composite material to be predicted, and displays the predicted impact strength value of the polymer composite material to be predicted on the operator terminal 14. The prediction device 12 may also display the predicted impact strength value of the polymer composite material to be predicted on the display device of the prediction device 12.

[0092] Thus, the prediction device 12 can accurately predict the impact strength of a polymer composite material from its material information using the first and second machine learning models generated by the learning device 10. Furthermore, since the prediction device 12 can predict the impact strength from its material information of the polymer composite material without relying on measured impact strength data (actual values), it can reduce the amount of experimental work required.

[0093] [Other Embodiments] In this embodiment, an example of predicting impact strength from material information of a polymer composite material to be predicted has been described. However, the method is not limited to impact strength, and other physical properties may be predicted using the procedure described above.

[0094] The material information of the new polymer composite material developed by the material property prediction system 1 according to this embodiment may be used, for example, as input information to a manufacturing apparatus that produces the polymer composite material according to the material information. The manufacturing apparatus can produce the polymer composite material by blending multiple materials according to the material information.

[0095] Although this embodiment has been described above, it will be understood that various modifications to the form and details are possible without departing from the spirit and scope of the claims. Although the present invention has been described above based on examples, the present invention is not limited to the above examples, and various modifications are possible within the scope described in the claims. This application claims priority to Basic Application No. 2024-232247 filed with the Japan Patent Office on December 27, 2024, the entire contents of which are incorporated herein by reference.

[0096] 1. Material property prediction system 10. Learning device 12. Prediction device 14. Operator terminal 18. Communication network 24. First learning unit 26. First prediction unit 28. Second learning unit 30. Extraction unit 44. Acquisition unit 46. Second prediction unit 48. Third prediction unit 100. First trained machine learning model 102. Second trained machine learning model

Claims

1. A learning device for generating a machine learning model for predicting the impact strength of a polymer composite material, comprising: a first learning unit that causes a first machine learning model to learn the correspondence between the transition temperatures of brittle fracture and ductile fracture of a polymer composite material to be learned and material information; a first prediction unit that outputs a predicted transition temperature value of the polymer composite material to be learned by inputting the material information of the polymer composite material to be learned into the learned first machine learning model; and a second learning unit that causes a second machine learning model for predicting the impact strength of a polymer composite material to be predicted to learn the correspondence between the material information of the polymer composite material to be learned, the predicted transition temperature value, and impact strength.

2. The learning apparatus according to claim 1, further comprising: an extraction unit for extracting the transition temperature of the polymer composite material to be learned from a plurality of measurement data obtained by measuring the impact strength of the polymer composite material to be learned at a plurality of temperatures.

3. The learning device according to claim 1 or 2, wherein the material information includes: first information which is an index representing the fluidity of the thermoplastic resin contained in the polymer composite material; second information which is the weight fraction of the ethylene-propylene copolymer in a polypropylene resin containing a propylene polymer and an ethylene-propylene copolymer; third information which is the weight fraction of the propylene component of the ethylene-propylene copolymer in the polypropylene resin; fourth information which is the intrinsic viscosity of the polypropylene resin; fifth information which is the weight fraction of the sum of the second information derived from the polypropylene resin in the compound and the amount of elastomer; sixth information which is the amount of inorganic filler added; and seventh information which is the weight fraction of the sum of the amount of polypropylene polymer and the amount of propylene component of the ethylene-propylene copolymer in the compound.

4. A prediction device for predicting the impact strength of a polymer composite material, comprising: an acquisition unit that acquires material information of a polymer composite material to be predicted; a second prediction unit that outputs a predicted transition temperature value of the polymer composite material to be predicted by inputting the material information of the polymer composite material to be predicted into a first machine learning model that has been trained to determine the correspondence between the transition temperatures of brittle fracture and ductile fracture of a polymer composite material to be learned and the material information; and a third prediction unit that outputs a predicted impact strength value of the polymer composite material to be predicted by inputting the material information of the polymer composite material to be predicted and the predicted transition temperature value of the polymer composite material to be predicted into a second machine learning model that has been trained to determine the correspondence between the material information of the polymer composite material to be learned, the predicted transition temperature value of the polymer composite material to be learned predicted by the trained first machine learning model, and the impact strength.

5. The prediction device according to claim 4, wherein the material information includes: first information which is an index representing the fluidity of the thermoplastic resin contained in the polymer composite material; second information which is the weight fraction of the ethylene-propylene copolymer in a polypropylene resin containing a propylene polymer and an ethylene-propylene copolymer; third information which is the weight fraction of the propylene component of the ethylene-propylene copolymer in the polypropylene resin; fourth information which is the intrinsic viscosity of the polypropylene resin; fifth information which is the weight fraction of the sum of the second information derived from the polypropylene resin in the compound and the amount of elastomer; sixth information which is the amount of inorganic filler added; and seventh information which is the weight fraction of the sum of the amount of polypropylene polymer and the amount of propylene component of the ethylene-propylene copolymer in the compound.

6. A learning method for generating a machine learning model in which a learning device predicts the impact strength of a polymer composite material, comprising: a first learning step of causing a first machine learning model to learn the correspondence between the transition temperatures of brittle fracture and ductile fracture of a polymer composite material to be learned and material information; a first prediction step of inputting material information of the polymer composite material to be learned into the learned first machine learning model to output a predicted transition temperature value of the polymer composite material to be learned; and a second learning step of causing a second machine learning model to learn the correspondence between the material information of the polymer composite material to be learned, the predicted transition temperature value, and impact strength.

7. A prediction method for predicting the impact strength of a polymer composite material using a prediction device, comprising: an acquisition step of acquiring material information of a polymer composite material to be predicted; a second prediction step of inputting the material information of the polymer composite material to be predicted into a first machine learning model that has been trained to determine the correspondence between the transition temperatures of brittle fracture and ductile fracture of a polymer composite material to be learned and the material information, thereby outputting a predicted transition temperature value for the polymer composite material to be predicted; and a third prediction step of inputting the material information of the polymer composite material to be predicted and the predicted transition temperature value for the polymer composite material to be predicted into a second machine learning model that has been trained to determine the correspondence between the material information of the polymer composite material to be learned, the predicted transition temperature value for the polymer composite material to be learned predicted by the trained first machine learning model, and the impact strength, thereby outputting a predicted impact strength value for the polymer composite material to be predicted.

Images