Prediction method, quality evaluation method, prediction program, prediction system, and manufacturing system

Abstract

This prediction method comprises: acquiring an infrared spectrum of a resin composition containing a resin and a formulation; and predicting a physical property of the resin composition and / or a characteristic of the formulation from the infrared spectrum of the resin composition using a prediction model.

Description

Prediction method, quality evaluation method, prediction program, prediction system, and manufacturing system CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to Japanese Patent Application No. 2023-132694, filed on August 16, 2023, the entire disclosure of which is incorporated herein by reference.

[0002] The present disclosure relates to a prediction method, a quality evaluation method, a prediction program, a prediction system, and a manufacturing system.

[0003] As disclosed in Patent Document 1, a method for inspecting the quality of a resin using a near-infrared spectrometer is known.

[0004] Japanese Patent Application Laid-Open No. 2000-88553

[0005] When analyzing the physical properties of a resin composition that combines a resin and a blend by infrared spectroscopy, the presence of the blend affects the infrared spectrum (IR spectrum). Therefore, when quantitatively analyzing the physical properties or characteristics of a resin composition containing a blend using IR spectroscopy, it is difficult to simply apply the analytical method for a resin composition that does not contain the blend. Therefore, various analyses are performed to quantitatively analyze the physical properties or characteristics of a resin composition that contains a blend.

[0006] However, analyzing a resin composition requires a lot of time or cost. In other words, when a resin composition needs to be analyzed to quantitatively analyze the physical properties of the resin composition or the characteristics of the blend, the quantitative analysis of the physical properties of the resin composition or the characteristics of the blend cannot be easily performed. There is a need to easily perform quantitative analysis of the physical properties of the resin composition or the characteristics of the blend. In addition, there is a need to improve the quality of the resin composition based on the analysis results of the physical properties of the resin composition or the characteristics of the blend.

[0007] In view of the above circumstances, an object of the present disclosure is to provide a prediction method, a quality evaluation method, a prediction program, and a prediction system that can easily realize quantitative measurement of the physical properties of a resin composition or the characteristics of a blend, as well as a manufacturing system that can improve the quality of a resin composition.

[0008] (1) A prediction method according to one embodiment of the present disclosure includes obtaining an infrared spectrum of a resin composition including a resin and a blend, and predicting at least one of a physical property of the resin composition or a characteristic of the blend from the infrared spectrum of the resin composition using a prediction model.

[0009] (2) The prediction method described in (1) above may include predicting at least one of the properties of the blend as a transit property of the resin composition from the infrared spectrum of the resin composition using a first prediction model as the prediction model, and predicting at least one of the physical properties of the resin composition as a target property of the resin composition from the prediction result of the transit property of the resin composition using a second prediction model as the prediction model.

[0010] (3) In the prediction method described in (1) or (2) above, the infrared spectrum of the resin composition may include a wavenumber range in which the absorbance changes depending on changes in the properties of the blend.

[0011] (4) In the prediction method described in any one of (1) to (3) above, the characteristics of the blend may include at least one of information regarding the content of the blend or information regarding the size of the blend.

[0012] (5) In the prediction method described in any one of (1) to (4) above, the physical properties of the resin composition may include at least one of the mechanical properties of the resin composition, the fluidity of the resin composition, or the molecular weight of the resin composition.

[0013] (6) In the prediction method according to any one of (1) to (5) above, the prediction model may be a trained model generated by performing machine learning using training data that associates the infrared spectrum of the resin composition with at least one measured value or theoretical value of a physical property of the resin composition or a characteristic of the blend.

[0014] (7) The prediction method according to any one of (1) to (6) above may further include measuring the infrared spectrum of the resin composition by an ATR method.

[0015] (8) In the prediction method described in (7) above, a spectrum in the MIR band may be obtained as the infrared spectrum of the resin composition.

[0016] (9) The prediction method according to (7) or (8) above may further include pressing the solid pellets of the resin composition to measure the infrared spectrum of the solid pellets by an ATR method.

[0017] (10) The prediction method according to (7) or (8) above may further include measuring the infrared spectrum of the resin composition in a molten state by an ATR method.

[0018] (11) The prediction method according to any one of (1) to (6) above may further include measuring the infrared spectrum of the resin composition by a transmission method.

[0019] (12) In the prediction method described in (11) above, a spectrum in the NIR band may be obtained as the infrared spectrum of the resin composition.

[0020] (13) The prediction method according to (11) or (12) above may further include measuring the infrared spectrum of the resin composition having a thickness of 1 mm or less by a transmission method.

[0021] (14) In the prediction method according to any one of (1) to (13) above, the resin of the resin composition may contain polyamide.

[0022] (15) In the prediction method according to any one of (1) to (14) above, the resin composition formulation may contain a colorant.

[0023] (16) In the prediction method according to any one of (1) to (15) above, the resin composition formulation may contain an inorganic filler.

[0024] (17) The prediction method according to any one of (1) to (16) above may include obtaining a spectral dataset including a plurality of spectral data as an infrared spectrum of the resin composition, and predicting at least one of the physical properties of the resin composition or the characteristics of the blend from the spectral dataset using a model that inputs the spectral dataset as the prediction model.

[0025] (18) In the prediction method described in (17) above, when a single infrared spectrum is calculated as the infrared spectrum of the resin composition by integrating or averaging multiple pre-treatment spectra, the spectral data set may include the multiple pre-treatment spectra as the multiple spectral data.

[0026] (19) In the prediction method described in (17) above, when a set of measurements is performed to measure a plurality of pre-treatment spectra in order to calculate a single infrared spectrum by accumulating or averaging the plurality of pre-treatment spectra as the infrared spectrum of the resin composition, the spectral data set may include, as the plurality of spectral data, at least one of the plurality of pre-treatment spectra or the single infrared spectrum obtained in each measurement by performing the set of measurements multiple times.

[0027] (20) The prediction method according to any one of (1) to (19) above may include calculating the reliability of a prediction result of at least one of the physical properties of the resin composition or the characteristics of the blend.

[0028] (21) The prediction method according to any one of (17) to (19) above may include calculating the reliability of a prediction result of at least one of the physical properties of the resin composition or the characteristics of the blend based on the variability of the plurality of spectral data.

[0029] (22) The prediction method according to any one of (1) to (21) above may include acquiring measurement environment information of the infrared spectrum of the resin composition, and predicting at least one of the physical properties of the resin composition or the characteristics of the blend based on the infrared spectrum of the resin composition and the measurement environment information using a model that inputs the infrared spectrum of the resin composition and the measurement environment information as the prediction model.

[0030] (23) The prediction method described in (22) above may include calculating the reliability of a prediction result of at least one of the physical properties of the resin composition or the characteristics of the blend based on the measurement environment information.

[0031] (24) The prediction method according to any one of (1) to (23) above may include acquiring manufacturing conditions for the resin composition, and predicting at least one of the physical properties of the resin composition or the characteristics of the blend based on the infrared spectrum of the resin composition and the manufacturing conditions, using a model that inputs the infrared spectrum of the resin composition and the manufacturing conditions as the prediction model.

[0032] (25) The prediction method described in (24) above may include calculating the reliability of the prediction result of at least one of the physical properties of the resin composition or the characteristics of the blend based on the manufacturing conditions.

[0033] (26) A quality evaluation method for a resin composition according to one embodiment of the present disclosure includes predicting at least one of the physical properties of the resin composition or the characteristics of a compound contained in the resin composition by executing the prediction method described in any one of (1) to (25) above, and outputting a result of evaluating the quality of the resin composition based on at least one of the predicted results of the physical properties of the resin composition or the predicted results of the characteristics of the compound.

[0034] (27) A prediction program according to one embodiment of the present disclosure causes a processor to execute the prediction method described in any one of (1) to (25) above.

[0035] (28) A prediction system according to one embodiment of the present disclosure includes a measurement device that measures an infrared spectrum of a resin composition containing a resin and a blend, and a prediction device that predicts at least one of the physical properties of the resin composition or the characteristics of the blend from the infrared spectrum of the resin composition using a prediction model.

[0036] (29) In the prediction system described in (28) above, the measurement device may measure the infrared spectrum of the resin composition in a production line for the resin composition.

[0037] (30) A manufacturing system according to an embodiment of the present disclosure includes the prediction system according to (29) above and a manufacturing apparatus installed in a manufacturing line of the resin composition, wherein the manufacturing apparatus controls manufacturing conditions of the resin composition so that an absolute value of a difference between a predicted value and a target value of at least one item of a physical property of the resin composition or a characteristic of the blend is less than a difference threshold value.

[0038] (31) In the manufacturing system described in (30) above, the manufacturing apparatus may manufacture the resin composition by changing manufacturing conditions of the resin composition so that a predicted value of at least one item of the physical properties of the resin composition or the characteristics of the blend approaches a target value. The measuring apparatus may measure an infrared spectrum of the resin composition manufactured by changing the manufacturing conditions. The prediction apparatus may predict at least one item of the physical properties of the resin composition or the characteristics of the blend manufactured by changing the manufacturing conditions. The manufacturing apparatus may repeatedly change the manufacturing conditions until an absolute value of a difference between a predicted value and a target value of at least one item of the physical properties of the resin composition manufactured by changing the manufacturing conditions or the characteristics of the blend becomes less than a difference threshold.

[0039] (32) In the manufacturing system described in (30) above, the manufacturing apparatus may control the manufacturing conditions of the resin composition using a model trained using learning data that associates pre-change manufacturing conditions with predicted or measured values ​​of at least one item of the physical properties of the resin composition or the characteristics of the blend manufactured under the pre-change manufacturing conditions, and changed manufacturing conditions obtained by changing the pre-change manufacturing conditions with predicted or measured values ​​of at least one item of the physical properties of the resin composition or the characteristics of the blend manufactured under the changed manufacturing conditions.

[0040] (33) In the manufacturing system described in any one of (30) to (32) above, the manufacturing apparatus may control the manufacturing conditions of the resin composition based on the prediction results of multiple items of the physical properties of the resin composition or the characteristics of the blend.

[0041] (34) In the manufacturing system described in (33) above, the manufacturing apparatus may control the manufacturing conditions of the resin composition based on a predicted value of the viscosity of the resin composition and a predicted value of the mechanical properties of the resin composition.

[0042] The prediction method, quality evaluation method, prediction program, and prediction system according to the present disclosure enable quantitative measurement of the physical properties of a resin composition or the characteristics of a blend to be easily realized. Furthermore, the production system according to the present disclosure improves the quality of the resin composition.

[0043] 1 is a block diagram showing an example of a configuration of a prediction system according to the present disclosure. FIG. 1 is a block diagram showing an example of a prediction model that outputs target characteristics of a sample by inputting an IR spectrum of the sample. FIG. 2 is a block diagram showing an example of a prediction model that generates intermediate characteristics along the way to output target characteristics of the sample by inputting an IR spectrum of the sample. FIG. 3 is a schematic diagram explaining measurement of an IR spectrum by an ATR method. FIG. 4 is a schematic diagram explaining measurement of an IR spectrum by a transmission method. FIG. 5 is a flowchart showing an example of a procedure for generating a prediction model for physical properties. FIG. 6 is a flowchart showing an example of a procedure for measuring the IR spectrum of a sample. FIG. 7 is a flowchart showing an example of a procedure for predicting physical properties of a sample from the IR spectrum of the sample. FIG. 8 is a graph showing an example of an IR spectrum obtained by measuring a pressed sample by the ATR method. FIG. 9 is a graph showing an example of an IR spectrum obtained by measuring a sample in a molten state by the ATR method. FIG. 10 is a graph showing an example of an IR spectrum obtained by measuring a sample by a transmission method. FIG. 11 is a block diagram showing an example of a configuration of a manufacturing system according to the present disclosure. FIG. 12 is a flowchart showing an example of a procedure for a method for manufacturing a resin composition.

[0044] In the production of resin compositions in which other materials are blended with a resin, maintaining or improving the quality of the resin composition is required. To maintain or improve the quality of a resin composition, it is necessary to confirm the physical properties of the resin composition. While analyzing a resin composition to confirm its physical properties is conceivable, analytical methods applicable to resin compositions blended with other materials are limited. For example, infrared spectroscopy is known as a simple method for measuring the physical properties of a resin. However, when a resin contains other materials, such as glass fiber or carbon black, the blended materials can absorb or scatter infrared light, making it difficult to obtain a normal IR spectrum or reducing the reproducibility of the IR spectrum shape. Furthermore, when variations in the properties of the blend affect the variations in the physical properties of the resin composition, the variations in the properties of the blend must be detected using IR spectroscopy. When using an IR spectrum measurement method that detects variations due to the resin components in the resin composition, it is difficult to evaluate the physical properties of the resin composition or the properties of the blend. Furthermore, using other methods, such as destructive testing, to analyze a resin composition can be time-consuming or cost-intensive. There is a demand for a method for easily confirming the physical properties of resin compositions. Furthermore, there is a demand for improving the quality of resin compositions. For example, recycled raw materials from process offcuts or recycled raw materials derived from market-recovered products have large variations in the quality of the raw materials, which increases the frequency with which the quality of the final product deviates from specifications. Therefore, there is a demand for a method for easily performing quantitative analysis and for improving the quality of resin compositions.

[0045] Hereinafter, a prediction system 1 (see FIG. 1 ) that can easily predict at least one of the physical properties of a resin composition containing a resin and a blend or the properties of the blend contained in the resin composition will be described. The blend is another material blended into the resin. In the present disclosure, the resin may include polyamide (PA) or the like. The blend may include an inorganic filler such as glass fiber (GF). The blend may include a colorant such as carbon black (CB). The resin composition may be a mixture obtained by kneading the blend into the resin.

[0046] The physical properties of the sample predicted by the prediction system 1 may be, for example, mechanical properties such as tensile strength or flexural strength of the resin composition, fluidity, or molecular weight. The physical properties of the sample are expressed as numerical values. The fluidity of the resin composition represents the ease of flow of the resin composition when it is in a molten state, such as MFR (melt flow rate), MVR (melt volume rate), or shear viscosity, or the specific viscosity, relative viscosity, intrinsic viscosity, or reduced viscosity calculated from the solution viscosity of the resin composition. The ease of flow of the resin composition may be expressed as a numerical value. The properties of the compound predicted by the prediction system 1 may be, for example, the glass fiber content or fiber length if the compound is glass fiber.

[0047] The prediction system 1 can calculate the predicted results of the physical properties of a sample or the characteristics of a blend as predicted values. That is, the predicted values ​​of the physical properties of a sample or the characteristics of a blend may be numerical values ​​that represent the predicted results of the physical properties of the sample or the characteristics of the blend. The numerical values ​​may be continuously changing values ​​or discretely changing values. The prediction system 1 may generate the predicted results of the physical properties of a sample or the characteristics of a blend as categorical variables. The categorical variables may be nominal or ordinal scales.

[0048] The accuracy of the predicted values ​​of the sample's physical properties or blend's characteristics, and the accuracy of the prediction model 80 (see FIG. 2A or 2B ), are evaluated using known evaluation indices such as the coefficient of determination (R), mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), or mean absolute percentage error (MAPE) based on the measured and predicted values ​​of the sample's physical properties or blend's characteristics. Specifically, for the coefficient of determination, the closer the value is to 1, the higher the prediction accuracy of the sample's physical properties or blend's characteristics is evaluated. For the mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), the closer the value is to 0, the higher the prediction accuracy is evaluated. Furthermore, if cross-validation is performed when constructing the prediction model 80, the prediction accuracy may be evaluated using the average value of the evaluation indices obtained for each fold. The measured values ​​of the sample's physical properties or blend's characteristics are values ​​obtained by actually measuring the sample's physical properties or blend's characteristics. As the measured value of the physical property of the sample, a value obtained by actually measuring the mechanical property such as tensile strength or flexural strength of the resin composition, the flowability, the molecular weight, etc. may be obtained. As the measured value of the property of the compound contained in the sample, for example, if the compound is glass fiber, a value obtained by actually measuring the content or fiber length of the glass fiber, etc. may be obtained.

[0049] (Configuration Example of Prediction System 1) As shown in FIG. 1 , a prediction system 1 according to an embodiment of the present disclosure includes a prediction device 10 and a measurement device 20. The prediction system 1 predicts the physical properties of a sample or the characteristics of a blend using a resin composition containing a resin and a blend as the sample. In the prediction system 1, the measurement device 20 measures the IR spectrum of the sample. The IR spectrum is represented as multidimensional data that is a set of light intensities of wavelengths included in at least a portion of the wavelength band of infrared light. The prediction device 10 predicts at least one of the physical properties of the sample or the characteristics of the blend based on the measurement results of the IR spectrum of the sample. In other words, the prediction device 10 calculates a predicted value of at least one of the physical properties of the sample or the characteristics of the blend based on the multidimensional data representing the IR spectrum.

[0050] <Prediction Device 10 > The prediction device 10 includes a data acquisition unit 11 , a data processing unit 12 , a storage unit 15 , and an output unit 16 .

[0051] The data acquiring unit 11 acquires, from the measuring device 20, the IR spectrum of the sample measured by the measuring device 20. The data acquiring unit 11 may acquire information or data not only from the measuring device 20 but also from other external devices. The data acquiring unit 11 may acquire, for example, measured values ​​of the physical properties of the sample.

[0052] The data acquisition unit 11 may include a communication interface for connecting to the measurement device 20 or an external device, etc., via a wired or wireless connection. The communication interface may be configured to communicate with the measurement device 20 or an external device, etc., based on a communication standard such as RS-232C or RS-485. The communication interface may be configured to communicate with the measurement device 20 based on a communication standard such as Bluetooth (registered trademark). The communication interface may be configured to communicate with the measurement device 20 or an external device, etc., via a network such as a LAN (Local Area Network).

[0053] The data processing unit 12 predicts the physical properties of the sample based on the IR spectrum of the sample. The data processing unit 12 may include a spectrum processing unit 13 that processes the IR spectrum of the sample. The data processing unit 12 may include a prediction unit 14 that predicts the physical properties of the sample based on the IR spectrum of the sample.

[0054] The data processing unit 12 may be configured to include at least one processor, such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The data processing unit 12 may be configured with one processor or multiple processors. The processor constituting the data processing unit 12 may realize the functions of the data processing unit 12 by reading and executing a program stored in the storage unit 15 (described later). The data processing unit 12 may be configured to include one or more dedicated circuits. The dedicated circuits may include, for example, an FPGA (Field-Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit). The data processing unit 12 may be configured by combining a processor and a dedicated circuit.

[0055] The storage unit 15 stores various types of information or data to be processed by the prediction device 10. The storage unit 15 may store, for example, a program executed in the data processing unit 12, or data or processing results used in the processing executed in the data processing unit 12. The storage unit 15 may also function as a work memory for the data processing unit 12. The storage unit 15 may be configured to include, for example, a semiconductor memory, etc., but is not limited to this. For example, the storage unit 15 may be configured as an internal memory of a processor used as the data processing unit 12, or as a hard disk drive (HDD) accessible from the data processing unit 12. The storage unit 15 may be configured as a non-transitory readable medium. The storage unit 15 may be configured integrally with the data processing unit 12, or may be configured separately from the data processing unit 12.

[0056] The output unit 16 outputs the prediction results of the physical properties of the sample. The output unit 16 may include a display device. The display device may be configured to include, for example, an LCD (Liquid Crystal Display), an organic EL (Electro-Luminescence) display, an inorganic EL display, or a PDP (Plasma Display Panel). The display device is not limited to these displays and may be configured to include displays of various other types. The output unit 16 may be configured to output the prediction results of the physical properties of the sample to an external device.

[0057] As illustrated in FIG. 2A , the prediction device 10 may predict the physical properties of a sample using a prediction model 80. The prediction model 80 is realized by the prediction unit 14. The prediction unit 14 inputs the IR spectrum of the sample acquired by the data acquisition unit 11 into the prediction model 80. The prediction model 80 outputs a target characteristic of the sample based on the input IR spectrum of the sample. The target characteristic is a property of the sample that corresponds to a target physical property to be predicted. In other words, the prediction model 80 is configured to output the target characteristic of the sample when the IR spectrum of the sample is input.

[0058] As illustrated in FIG. 2B , the prediction device 10 may predict the physical properties of a sample using a prediction model 80 including a first prediction model 81 and a second prediction model 82 .

[0059] The prediction unit 14 inputs the IR spectrum acquired by the data acquisition unit 11 into the first prediction model 81. The first prediction model 81 outputs the transit characteristics of the sample based on the input IR spectrum. The transit characteristics are intermediately predicted characteristics for predicting the target characteristics. For example, the content or size of a compound blended in the resin composition may be predicted as the transit characteristics. In other words, the first prediction model 81 is configured to output the transit characteristics of the sample when the IR spectrum of the sample is input.

[0060] The prediction unit 14 inputs the route characteristics to the second prediction model 82. The second prediction model 82 outputs the target characteristics of the sample based on the input route characteristics. In other words, the second prediction model 82 is configured to output the target characteristics of the sample when the route characteristics of the sample are input.

[0061] The prediction model 80, which includes a first prediction model 81 and a second prediction model 82, is configured to output a target property of a sample when an IR spectrum of the sample is input.

[0062] The prediction model 80 may be a trained model generated by performing learning using the IR spectrum of a sample as training data. The prediction model 80 may be a trained model generated by performing machine learning using, as training data, data correlating the IR spectrum of a sample with measured values ​​or theoretical values ​​of the sample's target characteristics. In other words, the prediction model 80 may be a trained model generated by performing machine learning using, as training data, data correlating the IR spectrum of a sample resin composition with measured values ​​or theoretical values ​​of at least one of the resin composition's physical properties or blend characteristics, which are the target characteristics of the resin composition. The prediction model 80 may be a regression analysis model obtained by performing regression analysis using the IR spectrum of the sample as an explanatory variable and the sample's target characteristics as a response variable. The prediction model 80 may also be a calibration curve model generated using a chemometrics technique.

[0063] The first prediction model 81 may be a trained model generated by performing learning using data correlating the IR spectrum of a sample with the transit characteristics of the sample as training data. The first prediction model 81 may be a trained model generated by performing learning using the IR spectrum of a sample as training data. The first prediction model 81 may be a regression analysis model obtained by performing regression analysis using the IR spectrum of the sample as an explanatory variable and the transit characteristics of the sample as a response variable. The first prediction model 81 may be a calibration curve model generated using a chemometrics technique.

[0064] The second prediction model 82 may be a trained model generated by performing learning using data correlating the sample's transit characteristics with the sample's target characteristics as training data. The second prediction model 82 may be a trained model generated by performing learning using the sample's transit characteristics as training data. The second prediction model 82 may be a regression analysis model obtained by performing regression analysis using the sample's transit characteristics as explanatory variables and the sample's target characteristics as target variables. The second prediction model 82 may be a calibration curve model generated using a chemometrics technique.

[0065] The prediction model 80, the first prediction model 81, or the second prediction model 82 is not limited to the trained model, regression analysis model, or calibration curve model described above, and may be configured as a model of various types.

[0066] As described above, when predicting the physical properties of a sample using the prediction model 80, the more explanatory variable parameters that can be input into the prediction model 80, the more complex the fluctuations in the physical properties can be accommodated. In other words, even in cases where the spectral fluctuations are complex and cannot be predicted by regression analysis using one or a few explanatory variables, the prediction model 80 can be constructed based on a large number of explanatory variables. As a result of constructing the prediction model 80 for more complex spectral fluctuations, the prediction accuracy is improved. When the IR spectrum of a sample is input into the prediction model 80 as an explanatory variable, the parameters of the explanatory variable are the light intensity at each wavelength in the IR spectrum. The higher the resolution of the measured IR spectrum, the more explanatory variable parameters there are, and the more complex the spectral fluctuations can be constructed as the prediction model 80.

[0067] When a large number of explanatory variables are used in constructing the prediction model 80, the number of possible combinations of hyperparameters of the prediction model 80, parameters for spectrum preprocessing, and the like becomes enormous. Setting optimal parameters can be difficult. Furthermore, the possibility of overfitting the prediction model 80 increases, which can lead to a decrease in prediction accuracy for unknown data. In this case, by using a machine learning technique, even when a large number of explanatory variables are used in constructing the prediction model 80, it is possible to determine optimal parameters while preventing overfitting of the prediction model 80, thereby improving prediction accuracy.

[0068] The prediction device 10 may be realized using a cloud service or in an on-premise environment. Furthermore, the prediction device 10 may be realized in a form that combines a cloud service and an on-premise environment. When the prediction device 10 is realized using a cloud service, the output unit 16 may be configured as a display terminal used by a user, for example.

[0069] <Measurement Device 20> The measurement device 20 is assumed to be an infrared spectrometer capable of measuring an IR spectrum. The measurement device 20 may be configured to measure an IR spectrum in a wavelength band including at least one of near infrared (NIR), mid infrared (MIR), and far infrared (FIR), or in a wavelength band including all of these.

[0070] The measurement device 20 may be configured to measure an IR spectrum using the ATR (Attenuated Total Reflection) method. When measuring an IR spectrum using the ATR method, as shown in FIG. 3A , the measurement device 20 includes a light source 21, a detector 22, and a prism 23. The measurement device 20 is configured so that the prism 23 can be placed in close contact with the sample 30. The measurement device 20 emits light including infrared light from the light source 21 and causes the light to enter the prism 23. The light entering the prism 23 is also referred to as incident light. The light entering the prism 23 is totally reflected at the interface where the prism 23 and the sample 30 are in close contact with each other. The totally reflected light is also referred to as reflected light. The reflected light passes through the prism 23 and enters the detector 22. The detector 22 detects the reflected light. The measuring device 20 measures the difference between the spectrum of incident light emitted from the light source 21 and the spectrum of reflected light detected by the detector 22 as the reflected IR spectrum of the sample 30 .

[0071] When measuring the IR spectrum of sample 30 by the ATR method, sample 30 may be processed in a solid state so as to be in close contact with prism 23 and placed in measurement device 20. Sample 30 may be processed so that the compound is present in a range where light totally reflected at the surface in close contact with prism 23 seeps into sample 30 (seepage depth of evanescent waves). Sample 30 may be processed so that the compound is exposed at the surface in close contact with prism 23.

[0072] The sample 30 may be, for example, a solid pellet. When the sample 30 is a solid pellet, the solid pellet may be pressed before being measured by the ATR method. Pressing the solid pellet smoothes the pressed surface. Pressing the solid pellet also improves adhesion to the ATR prism 23. Pressing the solid pellet also makes it easier for the compound contained in the solid pellet to be exposed at the surface that is in close contact with the prism 23. Pressing the solid pellet also makes it easier for the compound contained in the solid pellet to be present at the penetration depth of the evanescent wave at the surface that is in close contact with the prism 23. The solid pellet may be pressed so that the compound is exposed at the surface that is in close contact with the prism 23 or is present at the penetration depth of the evanescent wave.

[0073] In the step of pressing solid pellets, the pressing conditions are not particularly limited. From the viewpoint of minimizing the influence of heat on the composition, the temperature during pressing may be set to 300°C or less, preferably 200°C or less, more preferably 100°C or less, and even more preferably 50°C or less. Furthermore, when pressing pellets that have cut surfaces during production, such as pellets pelletized by strand cutting or underwater cutting, the blend is appropriately exposed to the press surface by pressing so that the cut surfaces are parallel to the press surface of the press. By appropriately exposing the blend to the press surface, the accuracy of predicting physical properties is improved.

[0074] The sample 30 may be placed in the measurement device 20 in a molten state. When the sample 30 is in a molten state, the sample 30 in the molten state is likely to come into close contact with the prism 23. Furthermore, when the sample 30 is in a molten state, the compound is likely to be exposed at the surface of the sample 30 that comes into close contact with the prism 23. By configuring the measurement device 20 to be able to place the sample 30 in a molten state, the measurement device 20 can be used to measure the resin composition being produced as the sample 30 in-line during the resin composition production process.

[0075] The measurement device 20 may be configured to measure an IR spectrum by a transmission method. When measuring an IR spectrum by a transmission method, as shown in FIG. 3B , the measurement device 20 includes a light source 21 and a detector 22. The measurement device 20 emits light including infrared light from the light source 21 and causes the light to be incident on the sample 30. The light incident on the sample 30 is also referred to as incident light. The light incident on the sample 30 is transmitted through the sample 30. The light transmitted through the sample 30 is also referred to as transmitted light. The transmitted light is incident on the detector 22. The detector 22 detects the transmitted light. The measurement device 20 measures the difference between the spectrum of the incident light emitted from the light source 21 and the spectrum of the transmitted light detected by the detector 22 as the transmitted IR spectrum of the sample 30.

[0076] The shape of the infrared spectrum of the resin composition is not particularly limited, but preferably satisfies the following conditions when expressed in terms of absorbance: the maximum absorbance is 0.3 to 2.0, the minimum absorbance is 1.0 or less, and the difference between the maximum and minimum absorbance values ​​is 0.03 or more. The shape of the infrared spectrum of the resin composition may more preferably satisfies the following conditions: the maximum absorbance is 0.5 to 1.7, the minimum absorbance is 0.5 or less, and the difference between the maximum and minimum absorbance values ​​is 0.05 or more.

[0077] The maximum or minimum absorbance value of the infrared spectrum is preferably determined based on the raw data of the measured infrared spectrum.

[0078] If the shape of the spectrum satisfies the above-mentioned conditions, the influence of noise or measurement variations is reduced, and the prediction accuracy is improved by reducing the influence of noise or measurement variations.

[0079] When measuring the IR spectrum of the sample 30 by the transmission method, the sample 30 may be processed into a shape that allows easy light transmission, such as a sheet or thin plate. The sample 30 may be processed so that the thickness in the light transmission direction is, for example, 1 mm or less, preferably 0.5 mm or less, and more preferably 0.2 mm or less. By processing the sample 30 so that the thickness in the light transmission direction of the sample 30 is within the above-mentioned range, the shape of the spectrum becomes appropriate. By having the appropriate shape of the spectrum, prediction accuracy improves.

[0080] (Example of Operation of Prediction System 1) An example of operation of the prediction system 1 will now be described in detail.

[0081] <Model Generation> In the prediction system 1, the prediction device 10 may predict at least one of the physical properties of the sample or the properties of the blend by inputting the IR spectrum of the sample into the prediction model 80 and obtaining a predicted value of at least one of the physical properties of the sample or the properties of the blend output from the prediction model 80, as described above. The prediction device 10 may obtain a pre-prepared model from an external device and use it as the prediction model 80. The prediction device 10 may also generate and use the prediction model 80 itself. An example of the operation of the prediction device 10 when generating the prediction model 80 will be described below. The prediction system 1 predicts at least one of the physical properties of the sample or the properties of the blend contained in the sample. In the following description, when it is stated that the physical properties of a sample are to be predicted, it also includes predicting the properties of the blend contained in the sample.

[0082] The data processing unit 12 of the prediction device 10 may execute a model generation method including the steps of the flowchart illustrated in Fig. 4 in order to generate the prediction model 80. The model generation method may be realized as a model generation program executed by a processor constituting the data processing unit 12. The model generation program may be stored in a non-transitory computer-readable medium.

[0083] The data processing unit 12 acquires the IR spectrum of the sample (step S1). The IR spectrum of the sample is measured by the measurement device 20 and acquired from the measurement device 20 by the data acquisition unit 11.

[0084] The IR spectrum of the sample resin composition preferably includes a wavenumber range in which the absorbance changes depending on changes in the properties of the components contained in the resin composition. For example, the IR spectrum of the resin composition may include a wavenumber range in which the absorption peak of the components contained in the resin composition appears, or a wavenumber range in which the height or slope of the baseline absorbance changes depending on changes in the content or size of the components contained in the resin composition. The IR spectrum of the resin composition may also include a wavenumber range in which the absorption peak originating from the resin components contained in the resin composition changes depending on changes in the properties of the components contained in the resin composition. The IR spectrum of the resin composition may also include a wavenumber range in which the absorption peak originating from the resin components contained in the resin composition does not change depending on changes in the properties of the components contained in the resin composition. In this way, the accuracy of predicting the physical properties of the resin composition and the properties of the components is improved by using an IR spectrum that includes not only the wavenumber range in which the absorbance changes depending on changes in the properties of the components contained in the resin composition, but also the wavenumber range in which the absorbance originating from the resin components is included.

[0085] The data processing unit 12 acquires measured values ​​of the physical properties of the sample (step S2). The measured values ​​of the physical properties of the sample are measured by an apparatus for measuring the physical properties of the sample. The apparatus for measuring the physical properties of the sample may include, for example, an apparatus for measuring mechanical properties such as tensile strength or flexural strength of the sample, an apparatus for measuring the fluidity of the sample, or an apparatus for measuring the molecular weight of the sample.

[0086] The prediction system 1 may include a device that measures the physical properties of a sample as the measurement device 20. When the prediction system 1 includes a device that measures the physical properties of a sample, the data processing unit 12 can acquire the measured values ​​of the physical properties of the sample inside the prediction system 1.

[0087] The prediction system 1 does not need to include a device for measuring the physical properties of the sample. That is, the physical properties of the sample may be measured by an external device. When the physical properties of the sample are measured by an external device, the data acquisition unit 11 of the prediction device 10 may be configured to be able to communicate with the external device so as to acquire the physical properties of the sample from the external device.

[0088] The data processing unit 12 associates the IR spectrum of the sample with the measured values ​​of the sample's physical properties (step S3). The data processing unit 12 generates a prediction model 80 by performing learning using data in which the IR spectrum of the sample and the measured values ​​of the sample's physical properties are associated as training data (step S4). The data processing unit 12 may store the generated prediction model 80 in the storage unit 15. After performing the procedure of step S4, the data processing unit 12 ends the execution of the procedure of the flowchart in FIG. 4.

[0089] The method for generating the prediction model 80 from the training data is not particularly limited. However, in order to describe the relationship between the spectrum, which is multidimensional data, and physical properties and to achieve higher prediction accuracy, it is preferable to generate the prediction model 80 using a machine learning technique. The prediction model 80 generated using a machine learning technique can be a known model such as a linear regression model or a nonlinear regression model. Examples of linear regression models include multiple regression analysis, PLS (Partial Least Squares), PCR (Principal Component Regression), ridge regression, lasso regression, and elastic net. Furthermore, the prediction model 80 can also be a tree model such as a regression tree, random forest, or boosting tree; a kernel method such as support vector regression, kernel ridge regression, or Gaussian process regression; or a neural network such as a recurrent net or long short-term memory. Furthermore, the prediction model 80 can also be a linear discriminant analysis, logistic regression, k-nearest neighbor method, or the like.

[0090] When constructing the predictive model 80 using a machine learning technique, a pre-trained model can also be utilized using techniques such as transfer learning or fine tuning. The model used as the source of transfer learning or fine tuning is not particularly limited, and may be a trained model generated by performing machine learning using, as training data, data that associates the IR spectrum of the resin composition with at least one measured value or theoretical value of the physical properties of the resin composition or the characteristics of the blend. The training data used in this case may be data on a resin composition manufactured in a process different from the actual process in which the resin composition to be predicted is manufactured, for example, data on a resin composition manufactured using a testing machine.

[0091] Before being used for model generation, the acquired spectra may be processed by performing spectral processing such as smoothing, baseline correction, normalization, SNV (Standard Normal Variate) transformation, or differentiation. Furthermore, when generating a model using machine learning, parameters in the above-mentioned spectral processing may be determined so as to improve the predictive accuracy of the generated model.

[0092] When generating the prediction model 80 using a machine learning technique, the IR spectra used as training data preferably include a wavenumber range in which absorbance changes in response to changes in the properties of the compound contained in the resin composition, which is the sample from which the IR spectrum was measured. Furthermore, it is preferable that the degree of change in absorbance in response to changes in the compound properties falls within a specific range. Specifically, the IR spectra used as training data preferably include a wavenumber range that satisfies the condition that the difference between the absorbance of the IR spectrum corresponding to the resin composition in which the compound properties, such as the compound content or size, are at their minimum and the absorbance of the IR spectrum corresponding to the resin composition in which the compound properties are at their maximum is 0.05 or more and 2.0 or less. More preferably, the IR spectra used as training data may include a wavenumber range that satisfies the condition that the difference in absorbance between the IR spectra is 0.1 or more and 1.5 or less. By including the wavenumber range in which the change in absorbance satisfies the above-mentioned condition in the IR spectra used as training data, the effects of noise or measurement variation are reduced. As a result, prediction accuracy is improved.

[0093] The wavenumber range in which the absorbance changes according to changes in the properties of the blends contained in the resin composition described above may be identified based on raw IR spectrum data obtained by measuring the resin composition. Furthermore, when machine learning is performed using preprocessed IR spectra as training data to generate the prediction model 80, the wavenumber range in which the absorbance changes according to changes in the properties of the blends contained in the resin composition described above is preferably identified based on the preprocessed IR spectrum.

[0094] In particular, it is desirable that the wavenumber region in which the height or slope of the baseline absorbance changes depending on the change in the content or size of the compound is identified based on the spectrum before baseline correction in the preprocessing of the IR spectrum. Alternatively, the wavenumber region in which the height or slope of the baseline absorbance changes depending on the change in the content or size of the compound may be identified based on the IR spectrum that has been subjected to preprocessing other than baseline correction.

[0095] It is desirable that the wavenumber range of the IR spectrum used as learning data or teacher data for generating the prediction model 80 be the same as the wavenumber range of the IR spectrum input into the prediction model 80 when using the prediction model 80 to predict the physical properties of the resin composition or the characteristics of the blends contained in the resin composition.

[0096] The data processing unit 12 may not execute the step of acquiring the measured values ​​of the physical properties of the sample in step S2 and the step of associating the IR spectrum of the sample with the measured values ​​of the physical properties of the sample in step S3. In this case, the data processing unit 12 may generate the prediction model 80 by executing learning using the IR spectrum of the sample as learning data in the step S4.

[0097] The data processing unit 12 may further acquire measured values ​​of the sample's transit characteristics in step S2. In this case, the data processing unit 12 may associate the IR spectrum of the sample with the measured values ​​of the sample's transit characteristics in step S3, and perform learning using the data in which the IR spectrum of the sample and the measured values ​​of the sample's transit characteristics are associated as teacher data in step S4, thereby generating the first prediction model 81. Alternatively, the data processing unit 12 may associate the measured values ​​of the sample's transit characteristics with measured values ​​of the sample's physical properties in step S3, and perform learning using the data in which the measured values ​​of the sample's transit characteristics and the measured values ​​of the sample's physical properties are associated as teacher data in step S4, thereby generating the second prediction model 82.

[0098] The data processing unit 12 may include a model generation unit. When the data processing unit 12 includes a model generation unit, the model generation unit may execute a model generation method.

[0099] <Sample Measurement> In the prediction system 1, the prediction device 10 inputs the IR spectrum of a sample into the prediction model 80, as described above. The IR spectrum of the sample is measured by the measurement device 20. The prediction device 10 acquires the IR spectrum of the sample measured by the measurement device 20. An example of the procedure of a spectrum measurement method executed by a user of the prediction system 1 to measure the IR spectrum of a sample will be described below with reference to the flowchart in FIG. 5 .

[0100] The user processes the sample (step S11). Specifically, the user processes the sample according to the method for measuring the IR spectrum of the sample. When measuring the IR spectrum of the sample using the ATR method, the user may process the sample by pressing it so that the compound contained in the resin composition sample is present at the penetration depth of the evanescent wave. When measuring the IR spectrum of the sample using the ATR method, the user may process the resin composition sample into a molten state. When measuring the IR spectrum of the sample using a transmission method, the user may process the sample into a shape such as a sheet or thin plate so that the sample can easily transmit light. The user may process the sample so that the thickness of the sample in the direction of light transmission is, for example, 1 mm or less.

[0101] The user sets the processed sample in the measuring device 20 (step S2). The user measures the IR spectrum of the sample using the measuring device 20 (step S3). The measuring device 20 outputs the measurement results of the IR spectrum of the sample to the prediction device 10. The prediction device 10 acquires the IR spectrum of the sample using the data acquisition unit 11. The prediction device 10 may output the acquired IR spectrum of the sample to the data processing unit 12 or may store it in the storage unit 15. After executing the procedure of step S3, the user ends the execution of the procedure of the flowchart in FIG. 5. The user may measure the IR spectrum by executing the procedure from steps S11 to S13 for each of multiple samples.

[0102] The prediction system 1 may be configured to automatically perform at least some of the steps of the spectrum measurement method. For example, the prediction system 1 may include a means for processing a sample, or a means for setting the sample in the measurement device 20. The measurement device 20 may automatically start measurement when the sample is set.

[0103] <Prediction of Physical Properties of Sample> In the prediction system 1, as described above, the prediction device 10 may predict the physical properties of a sample by inputting the IR spectrum of the sample into the prediction model 80 and obtaining a predicted value of the physical properties of the sample output from the prediction model 80. An example of the operation of the prediction device 10 when predicting the physical properties of a sample using the prediction model 80 will be described below.

[0104] The data processing unit 12 of the prediction device 10 may execute a prediction method including the steps of the flowchart illustrated in Fig. 6 in order to predict the physical properties of a sample. The prediction method may be realized as a prediction program executed by a processor constituting the data processing unit 12. The prediction program may be stored in a non-transitory computer-readable medium.

[0105] The data processing unit 12 acquires the IR spectrum of the sample (step S21). The data processing unit 12 may acquire the IR spectrum of the sample from the measurement device 20 via the data acquisition unit 11, or may acquire the IR spectrum of the sample that has been previously acquired by the data acquisition unit 11 and stored in the storage unit 15.

[0106] The data processing unit 12 inputs the acquired IR spectrum of the sample to the prediction model 80 (step S22). The data processing unit 12 may process the IR spectrum of the sample using the spectrum processing unit 13 so that the IR spectrum of the sample can be input to the prediction model 80. The data processing unit 12 inputs the IR spectrum of the sample to the prediction unit 14, which realizes the prediction model 80. The data processing unit 12 may input the IR spectrum processed by the spectrum processing unit 13 to the prediction unit 14, or may input the IR spectrum as acquired without being processed by the spectrum processing unit 13 to the prediction unit 14.

[0107] The IR spectrum may be processed by performing spectral processing such as smoothing, baseline correction, normalization, SNV conversion, or differentiation before being input to the prediction unit 14 and used by the prediction model 80. The selection of the spectral processing method or the spectral processing conditions are preferably the same as the spectral processing performed when generating the prediction model 80.

[0108] The data processing unit 12 obtains predicted values ​​of the physical properties of the sample from the prediction model 80 (step S23). Specifically, the prediction model 80 outputs predicted values ​​of the physical properties corresponding to the input IR spectrum. The data processing unit 12 obtains the predicted values ​​of the physical properties of the sample output from the prediction model 80. The data processing unit 12 may output the predicted values ​​of the physical properties of the sample to the output unit 16. The output unit 16 may display the predicted values ​​of the physical properties of the sample so that the user can recognize the predicted values ​​of the physical properties of the sample. The data processing unit 12 may store the predicted values ​​of the physical properties of the sample in the memory unit 15. The output unit 16 may display the predicted values ​​of the physical properties of the sample stored in the memory unit 15 in response to a query operation from the user. After performing the procedure of step S23, the data processing unit 12 ends execution of the procedure of the flowchart of FIG. 6. The data processing unit 12 may perform the procedure of steps S21 to S23 for each of a plurality of samples to predict the physical properties of each sample.

[0109] <Evaluation of Sample Quality> The prediction device 10 may evaluate the quality of a resin composition based on the prediction results of the physical properties of the resin composition sample, and output the quality evaluation results. In other words, the prediction device 10 may execute a quality evaluation method that includes a procedure for evaluating the quality of a sample.

[0110] For example, the quality of a resin composition may be determined to be acceptable when a numerical value representing a physical property of the resin composition falls within a predetermined numerical range. The numerical range for determining the quality of a resin composition may be stored in the memory unit 15 of the prediction device 10. The data processing unit 12 of the prediction device 10 may acquire a numerical range for determining the quality of a sample resin composition from the memory unit 15, and may determine that the quality of the sample is acceptable when the predicted value of the physical property of the sample resin composition falls within the numerical range. The data processing unit 12 may output the evaluation result of the quality of the sample to the output unit 16. The output unit 16 may display the evaluation result of the quality of the sample to allow a user to recognize it.

[0111] As described above, the prediction system 1 according to the present disclosure can calculate predicted values ​​of the physical properties of a sample from the IR spectrum of the sample using the prediction model 80. In this way, the physical properties of the sample can be quantitatively predicted without analyzing the sample. As a result, quantitative measurement of the physical properties of the sample can be easily achieved.

[0112] (Example) A specific example in which the prediction system 1 according to the present disclosure predicts the physical properties of a sample will be described below.

[0113] <Prediction of Physical Properties Based on IR Spectrum in MIR Band Measured by ATR Method> In the prediction system 1, the measurement device 20 may measure the IR spectrum of a sample that is a pressed solid pellet by the ATR method. The prediction device 10 may predict at least one of the physical properties of the sample resin composition or the characteristics of a blend contained in the sample resin composition, based on the IR spectrum of the sample solid pellet measured by the ATR method.

[0114] As an example, the sample is a resin composition containing polyamide and glass fiber. Polyamide and glass fiber have absorption peaks in the mid-infrared (MIR) band. Therefore, the data processing unit 12 of the prediction device 10 acquires an IR spectrum of the sample in the mid-infrared (MIR) band, as illustrated in FIG. 7 . The horizontal axis of the graph in FIG. 7 represents wavenumber. The vertical axis of the graph in FIG. 7 represents absorbance at each wavenumber. The data processing unit 12 may extract the MIR band using the spectrum processing unit 13.

[0115] The IR spectrum illustrated in FIG. 7 was measured under the following conditions.

[0116] The samples for which IR spectra were measured were prepared as follows. First, a twin-screw extruder (TEM26SS, manufactured by Toshiba Machine) was set at a barrel temperature of 280°C, a screw rotation speed of 200-400 rpm, and a discharge rate of 20 kg / h. Polyamide 66 (1402S, manufactured by Asahi Kasei Corporation), glass fiber (ECS3T-275H, manufactured by Nippon Electric Glass Co., Ltd.), and carbon black were melt-kneaded, and pellets were obtained by strand cutting. The amount of carbon black added was 0.08 wt%. As shown in Table 1 below, 13 different combinations of glass fiber (GF) content and kneading screw rotation speed were used to prepare 13 types of pellet samples A-1 to A-13.

[0117]

[0118] The 13 types of pellets obtained as described above were pressed at room temperature using a "Quick Press" manufactured by Chromato Sciences Co., Ltd. During the pressing process, the pellets were pressed so that the strand cut surface was parallel to the pressing surface of the pressing machine, so that the glass fibers would be properly aligned on the pressing surface.

[0119] The IR spectra of Samples A-1 to A-13 obtained by pressing the pellets were measured by the ATR method using an infrared spectrophotometer. Three pellets were used for each sample, and each pellet was measured once by the ATR method. In other words, three IR spectra were obtained for each of Samples A-1 to A-13. The measurement conditions are as follows: Infrared spectrophotometer: Perkin-Elmer FT-IR Spectrum Two ATR crystal: Diamond Number of reflections: 1 Measurement wavenumber: 450 to 3800 cm －１ Wavenumber resolution: 2cm －１ Number of integrations: 16 times Detector: TGS detector

[0120] The IR spectrum graph illustrated in Fig. 7 includes 13 IR spectrum graphs obtained for 13 types of samples A-1 to A-13, respectively. Preprocessing was performed on each IR spectrum so that the maximum absorbance was 1 and the minimum absorbance was 0. The IR spectrum illustrated in Fig. 7 is the IR spectrum after preprocessing.

[0121] As described above, the IR spectrum graphs in Figure 7 are graphs of the IR spectra of 13 types of resin compositions obtained by varying the combination of the glass fiber content and the kneading screw rotation speed in 13 different ways. －１ In the wavenumber range around 900 to 1100 cm, an absorption peak due to the Si-O bond of the glass fiber appears. －１ In particular, the absorbance of each spectrum tends to increase as the glass fiber content increases. For the sample with a glass fiber content of 0 wt % and the sample with a glass fiber content of 60 wt %, the absorbance at 1000 cm is different. －１ In other words, the difference in the absorbance between the glass fiber content and the kneading screw rotation speed was －１The screw rotation speed of the kneading process correlates with the fiber length of the glass fiber. Therefore, the difference in the combination of the glass fiber content and size affects the absorbance of each spectrum in the wavenumber range around 900 to 1100 cm. －１ The IR spectrum shown in Fig. 7 includes a wavenumber range in which the absorbance changes depending on the content or size of the compound contained in the resin composition. The IR spectrum in the measured wavenumber range was used to construct the prediction model 80.

[0122] The data processing unit 12 inputs the IR spectrum of the MIR band of the resin composition as a sample into the prediction model 80, and obtains predicted values ​​of the physical properties of the resin composition or the properties of the blends contained in the resin composition output from the prediction model 80. The data processing unit 12 outputs the predicted values ​​of the physical properties of the resin composition or the properties of the blends contained in the resin composition to the output unit 16. The output unit 16 displays the predicted values ​​of the physical properties of the resin composition or the properties of the blends contained in the resin composition as prediction results of the physical properties of the resin composition or the properties of the blends contained in the resin composition, allowing the user to recognize the prediction results.

[0123] The prediction model 80 used in this example was created using code written in the programming language Python. Using data combining multiple samples for which actual measured values ​​of physical properties have been obtained and the IR spectra of those samples as training data, the prediction model 80 was created using PLS regression analysis, including a preprocessing step for the IR spectra. In this case, the preprocessing steps for the IR spectra included baseline correction, smoothing, normalization, and differentiation. The presence or absence of each preprocessing step, the various parameters for each preprocessing step, and the parameters for the PLS regression analysis performed following the preprocessing were set so as to select the optimum index value for prediction accuracy in cross-validation prediction verification.

[0124] In this embodiment, in order to evaluate the results predicted by the data processing unit 12 for the physical properties of the resin composition as a sample and the characteristics of the compounds contained in the resin composition, the values ​​actually measured for the physical properties of the resin composition and the characteristics of the compounds contained in the resin composition were compared with the predicted values ​​for the physical properties of the resin composition and the characteristics of the compounds contained in the resin composition.

[0125] As physical properties of the resin composition, tensile strength, flexural modulus, and Charpy impact strength were measured. In addition, as a property of the compound contained in the resin composition, the glass fiber (GF) content (ash content) was measured. Tensile strength was measured in accordance with ISO 527. Flexural modulus was measured in accordance with ISO 178. Charpy impact strength was measured in accordance with ISO 179. GF content (ash content) was measured in accordance with ISO 3451-1.

[0126] The prediction accuracy was calculated based on a comparison between the actual measured values ​​of the physical properties of the resin composition and the properties of the blends contained in the resin composition and the predicted values ​​of the physical properties of the resin composition and the properties of the blends contained in the resin composition. As an index of prediction accuracy, the average coefficient of determination (R2CV) for each fold was calculated in the cross-validation performed when creating the prediction model 80. The prediction accuracy of the physical properties of the resin composition and the properties of the blends contained in the resin composition using the above-mentioned prediction model 80 is shown in Table 2 below.

[0127]

[0128] According to the results shown in Table 2, the R2CV value for each item predicted in this example was 0.7 or higher. An R2CV value of 0.7 or higher is often considered to have good accuracy. Therefore, this example demonstrates that the physical properties of a resin composition and the characteristics of the ingredients contained in the resin composition can be predicted with high accuracy from the IR spectrum.

[0129] The prediction model 80 used in this embodiment may be a trained model generated by performing training using training data extracted from the IR spectrum of the MIR band.

[0130] The prediction model 80 may be divided into a first prediction model 81 and a second prediction model 82. In this embodiment, the first prediction model 81 may be configured to output a glass fiber content when an IR spectrum in the MIR band is input. The second prediction model 82 may be configured to output a predicted value of a physical property of the resin composition when the glass fiber content is input.

[0131] As described above, by pressing a sample to obtain an IR spectrum used to predict the physical properties of the sample, the accuracy of predictions of the physical properties of the sample from the IR spectrum alone is improved.

[0132] The accuracy of predicting the physical properties of the sample can be further improved by processing the sample so that the compound is exposed on the surface of the sample that comes into close contact with the prism 23 in the measuring device 20 or so that the compound is present at the penetration depth of the evanescent wave. In other words, the accuracy of predicting the physical properties of the sample can be further improved by appropriately selecting the processing conditions for pressing the sample.

[0133] Furthermore, by predicting the physical properties of a sample using the IR spectrum in the MIR band that includes the absorption peaks of polyamide and glass fiber, the accuracy of predicting the physical properties of a resin composition in which glass fiber is blended with polyamide can be improved.

[0134] The ability to predict the physical properties of a sample with high accuracy from the IR spectrum alone eliminates the need to analyze the sample using other techniques, and as a result, the physical properties of the sample can be measured easily.

[0135] <Prediction of Physical Properties Based on IR Spectrum of Molten Sample Measured by ATR Method> In the prediction system 1, the measurement device 20 may measure the IR spectrum of the molten sample by the ATR method. An example of an IR spectrum obtained by measuring the molten sample by the ATR method is shown in FIG. 8. The horizontal axis of the graph in FIG. 8 represents wavenumber. The vertical axis of the graph in FIG. 8 represents absorbance at each wavenumber.

[0136] The IR spectrum illustrated in FIG. 8 includes graphs of five IR spectra obtained by measuring five types of pellets, Samples A-1, A-3, A-6, A-9, and A-12, shown in Table 1 above. One IR spectrum was obtained for each type of sample. The measurement conditions are as follows: Infrared spectrophotometer: Bruker FT-IR Tensor27 ATR crystal: Diamond Number of reflections: 1 Measurement wavenumber: 500 to 4000 cm －１ Wavenumber resolution: 2cm －１ Number of integrations: 16 times Detector: MCT detector Accessory: Golgen Gate Measurement temperature: 280°C

[0137] The data processing unit 12 inputs the IR spectrum of each sample into the prediction model 80, and obtains predicted values ​​of the physical properties of the sample or the properties of the compound contained in the sample, which are output from the prediction model 80. The prediction model 80 used in this example is the same as the prediction model 80 used to predict the physical properties of the pressed sample described above.

[0138] In this example, the glass fiber (GF) content (ash content) contained in a molten sample was predicted. Meanwhile, the glass fiber (GF) content (ash content) contained in the sample was actually measured. Then, the prediction accuracy was calculated based on a comparison between the actually measured value of the glass fiber (GF) content (ash content) contained in the sample and the predicted value of the glass fiber (GF) content (ash content) contained in the sample. As an indicator of prediction accuracy, the average value of the coefficient of determination (R2CV) for each fold was calculated in the cross-validation performed when creating the prediction model 80. The prediction accuracy of the physical properties of the resin composition and the properties of the blends contained in the resin composition using the above-mentioned prediction model 80 is shown in Table 3 below.

[0139]

[0140] According to the results shown in Table 3, the R2CV value of the GF content (ash content) that was the target of prediction in this example was 0.98. On the other hand, in the results shown in Table 2 above, the R2CV value of the GF content (ash content) was 0.85. In other words, it was shown that measuring the sample in a molten state improves prediction accuracy compared to measuring a solid sample.

[0141] <IR spectrum measured by transmission method> In the prediction system 1, the measurement device 20 may measure the IR spectrum of a sample processed into a sheet by the transmission method. FIG. 9 shows an example of an IR spectrum obtained by measuring a polyamide resin containing glass fiber (GF) as a sample by the transmission method. The horizontal axis of the graph in FIG. 9 represents wavenumber. The vertical axis of the graph in FIG. 9 represents absorbance at each wavenumber. In addition, a sample with a high GF content and a sample with a low GF content were measured. In the graph in FIG. 9, the IR spectrum with a high absorbance overall is the IR spectrum obtained by measuring a sample with a high GF content. Conversely, the IR spectrum with a low absorbance overall is the IR spectrum obtained by measuring a sample with a low GF content.

[0142] The wavenumber of the IR spectrum illustrated in FIG. 9 is 5000 to 5600 cm －１ and a wavenumber range in which the wavenumber is 6000 to 7800 cm －１ No clear absorption peaks are observed in any of the wavenumber ranges. However, in these wavenumber ranges, the absorbance or slope of the baseline changes depending on the GF content. The reason why the absorbance or slope of the baseline changes depending on the GF content is that as the GF content increases, the proportion of infrared light scattered by the glass fibers increases, the proportion of light transmitted through the sample decreases overall, and the baseline absorbance increases.

[0143] Furthermore, not only when the GF content but also when the fiber length of the glass fiber varies, the absorbance or slope of the baseline changes depending on the fiber length of the glass fiber. This is because the rate at which infrared light is scattered by the glass fiber changes, causing the baseline absorbance to fluctuate. Furthermore, the degree to which infrared light is scattered at each wavenumber varies depending on the fiber length of the glass fiber. As a result, the slope of the baseline changes depending on the fiber length of the glass fiber. As a result of the investigation, it was found that in the near-infrared (NIR) band, differences in the fiber length of the glass fiber tend to be manifested as the absorbance or slope of the baseline. Therefore, it was discovered that the shape of the IR spectrum in the wavenumber range where no clear absorption peak is observed is important for predicting the physical properties of a sample or the characteristics of the compound contained in the sample.

[0144] (Manufacturing system 100) The prediction system 1 according to the present disclosure may be included in a resin composition manufacturing system 100, as shown in Fig. 10. The manufacturing system 100 includes a prediction system 1 and a manufacturing apparatus 40. The prediction system 1 includes a measurement apparatus 20 and a prediction apparatus 10. The manufacturing apparatus 40 and the measurement apparatus 20 are installed in a resin composition manufacturing line.

[0145] The manufacturing apparatus 40 may include a polymerization tank, an extruder, or the like. The manufacturing apparatus 40 is equipped with a control unit 42. The control unit 42 controls the polymerization tank, the extruder, or the like, and controls the production of the resin composition by the manufacturing apparatus 40. The control unit 42 may be configured to include at least one processor such as a CPU or a GPU, or may be configured to include at least one dedicated circuit. The control unit 42 may be configured similarly to the data processing unit 12.

[0146] The manufacturing system 100 may operate as follows to manufacture the resin composition.

[0147] The measuring device 20 measures the IR spectrum of the resin composition produced in-line by the production device 40. Specifically, the measuring device 20 measures the IR spectrum of the resin composition in the production line. The measuring device 20 may measure the IR spectrum of the resin composition within a predetermined time from the production of the resin composition. The measurement of the IR spectrum of the resin composition by the measuring device 20 may be performed automatically or manually by a manager or person in charge of the production line.

[0148] The prediction device 10 predicts at least one of the physical properties of the resin composition produced by the production device 40 or the characteristics of the blend contained in the resin composition, based on the IR spectrum measured in-line by the measurement device 20. The prediction device 10 may output the prediction result to the production device 40.

[0149] The prediction device 10 may notify the manager or person in charge of the production line of the prediction result via the output unit 16. By notifying the manager or person in charge of the production line of the prediction result, the manager or person in charge can determine whether to change the production conditions or stop the input of raw materials to the production line, etc.

[0150] The control unit 42 of the manufacturing apparatus 40 controls the manufacturing conditions of the resin composition based on the prediction results from the prediction system 1, and manufactures the resin composition. In other words, the control unit 42 feeds back the prediction results of the physical properties of the resin composition or the characteristics of the blends contained in the resin composition to the manufacturing conditions of the resin composition. Specifically, the control unit 42 may control the manufacturing conditions of the resin composition so that the absolute value of the difference between the predicted value and the target value of at least one of the physical properties of the resin composition manufactured by the manufacturing apparatus 40 or the characteristics of the blends contained in the resin composition is less than the difference threshold. The target value and the difference threshold may be set as appropriate for each item to be predicted based on the specifications or quality required of the resin composition.

[0151] The manufacturing conditions of the resin composition may include, but are not limited to, parameters such as temperature, pressure, or time in a polymerization tank or extruder.

[0152] The control unit 42 may repeatedly measure the IR spectrum and predict the properties of the resin composition or the properties of the blends contained in the resin composition, and change the manufacturing conditions of the resin composition so that the predicted values ​​of the properties of the resin composition or the properties of the blends contained in the resin composition approach the target values. The control unit 42 may change the manufacturing conditions so that the predicted values ​​gradually approach the target values, for example, by applying PID (Proportional Integral Differential) control as a feedback control method. In this way, the predicted values ​​of the properties of the resin composition or the properties of the blends contained in the resin composition are less likely to overshoot the target values. As a result, the quality of the resin composition is stabilized.

[0153] The control unit 42 may perform feedback control so that the absolute value of the difference between the predicted value and the target value becomes less than the difference threshold when the cycle of measuring the IR spectrum, predicting the physical properties or characteristics, and changing the manufacturing conditions is repeated a predetermined number of times. This allows the manufacturing conditions to be stabilized quickly. The rate of defective resin compositions on the manufacturing line is reduced, and the quality of the resin compositions is improved.

[0154] The control unit 42 may control the manufacturing conditions of the resin composition using a model trained using learning data that associates the pre-change manufacturing conditions with a predicted value or actual measurement value of at least one item of the physical properties of the resin composition or the characteristics of the compound manufactured under the pre-change manufacturing conditions, and the changed manufacturing conditions obtained by changing the pre-change manufacturing conditions with a predicted value or actual measurement value of at least one item of the physical properties of the resin composition or the characteristics of the compound manufactured under the changed manufacturing conditions.

[0155] The control unit 42 may change the production conditions for the resin composition based on the magnitude relationship between the predicted value and the target value of the physical properties of the resin composition produced by the production apparatus 40 or the properties of the blends contained in the resin composition. For example, when the predicted value is greater than the target value, the control unit 42 may change a parameter of the production conditions that is known to have the effect of decreasing the predicted value. Conversely, when the predicted value is smaller than the target value, the control unit 42 may change a parameter of the production conditions that is known to have the effect of increasing the predicted value. In other words, when the magnitude relationship between the predicted value and the target value and the relationship between an increase or decrease in the parameter of the production condition are known, the control unit 42 may change the parameter of the production condition by applying the magnitude relationship between the predicted value and the target value to the known relationship.

[0156] When the prediction device 10 predicts multiple items, the control unit 42 may change the parameters of the manufacturing conditions based on a combination of the magnitude relationships between the predicted values ​​and target values ​​of the multiple items. When the relationship between a pattern of combinations of the magnitude relationships between the predicted values ​​and target values ​​of the multiple items and an increase or decrease in one parameter of the manufacturing conditions is known, the control unit 42 may apply the combination of the magnitude relationships between the predicted values ​​and target values ​​of the multiple items to the known relationship and determine whether to increase, decrease, or leave the one parameter of the manufacturing conditions unchanged. When the relationship between a pattern of combinations of the magnitude relationships between the predicted values ​​and target values ​​of the multiple items and a combination of an increase or decrease in the multiple parameters of the manufacturing conditions is known, the control unit 42 may apply the combination of the magnitude relationships between the predicted values ​​and target values ​​of the multiple items to the known relationship and determine whether to increase, decrease, or leave each of the multiple parameters of the manufacturing conditions unchanged.

[0157] The relationship between the predicted and target values ​​of the physical properties of the resin composition or the properties of the ingredients contained in the resin composition, and the relationship between the increase and decrease of the parameters of the production conditions of the resin composition may be determined based on production experiments of the resin composition or on the track record of producing the resin composition on a production line.

[0158] When the prediction device 10 predicts the viscosity and mechanical properties of the resin composition as multiple items, the control unit 42 may control the manufacturing conditions of the resin composition based on the predicted viscosity and mechanical properties of the resin composition. The viscosity and mechanical properties of the resin composition are predicted with high accuracy from the IR spectrum of the resin composition. Therefore, by controlling the manufacturing conditions of the resin composition based on the predicted viscosity and mechanical properties of the resin composition, the viscosity and mechanical properties of the resin composition are controlled with high accuracy. As a result, the quality of the resin composition is improved.

[0159] The manufacturing system 100 may execute a manufacturing method for a resin product including the steps of the flowchart illustrated in Fig. 11 in order to manufacture a resin composition by controlling the manufacturing conditions of the resin composition. The manufacturing method for a resin composition may be realized as a manufacturing program for a resin composition executed by a processor constituting the control unit 42 of the manufacturing apparatus 40. The manufacturing program for a resin composition may be stored in a non-transitory computer-readable medium.

[0160] The control unit 42 of the manufacturing apparatus 40 controls the manufacturing conditions to manufacture a resin composition (step S31). The measurement device 20 measures the IR spectrum of the resin composition manufactured by the manufacturing apparatus 40 (step S32). The prediction device 10 predicts the physical properties of the resin composition or the characteristics of the ingredients contained in the resin composition based on the IR spectrum of the resin composition (step S33).

[0161] The control unit 42 determines whether the difference between the predicted value and the target value of the physical property of the resin composition or the characteristic of the blend is less than the difference threshold (step S34). When the prediction device 10 predicts multiple items of the physical property of the resin composition or the characteristic of the blend, the control unit 42 may determine that the difference between the predicted value and the target value is less than the difference threshold if the difference between the predicted value and the target value of all items is less than the difference threshold. The control unit 42 may determine that the difference between the predicted value and the target value is less than the difference threshold if the difference between the predicted value and the target value of at least one item selected or designated by a production line manager or person in charge among the multiple items is less than the difference threshold.

[0162] If the difference between the predicted value and the target value is not less than the difference threshold value (step S34: NO), the control unit 42 changes the manufacturing conditions of the resin composition (step S35). After executing the procedure of step S35, the control unit 42 returns to the procedure of step S31 and manufactures the resin composition under the changed manufacturing conditions.

[0163] If the difference between the predicted value and the target value is less than the difference threshold value (step S34: YES), the control unit 42 returns to the procedure of step S31 and continues producing the resin composition without changing the production conditions.

[0164] The manufacturing system 100 may perform the procedures from steps S32 to S35 in the flowchart of FIG. 11 on a sample extracted from the manufactured resin composition, or may perform the procedures on all the manufactured resin compositions. When extracting samples and measuring the IR spectrum, one sample may be extracted every time a predetermined number of resin compositions are manufactured. One sample may be extracted every predetermined time. One sample may be extracted irregularly or randomly. The sample may be extracted automatically or manually by a manager or person in charge of the manufacturing line.

[0165] As described above, the manufacturing system 100 can measure the IR spectrum of a resin composition in-line to predict the physical properties of the resin composition or the characteristics of the ingredients contained in the resin composition, and can manufacture the resin composition by feeding back the prediction results to the manufacturing conditions. In this way, the rate of defective resin compositions on the manufacturing line is reduced, and the quality of the resin composition is improved.

[0166] Other Embodiments Other embodiments will be described below.

[0167] <Input of Multiple IR Spectra into Prediction Model 80> In the embodiment described above, the measurement device 20 generated one IR spectrum as a measurement result of one resin composition. The prediction model 80 was configured to output a prediction result of the physical properties of the resin composition to be measured or the characteristics of a compound contained in the resin composition when one IR spectrum obtained as a measurement result of one resin composition was input.

[0168] The prediction model 80 may be configured to accept input of multiple sets of IR spectra as measurement results for one resin composition. In this case, when multiple sets of IR spectra are input as measurement results for one resin composition, the prediction model 80 outputs prediction results for the physical properties of the resin composition or the characteristics of the blends contained in the resin composition. The measurement device 20 may output multiple sets of IR spectra to be input to the prediction model 80 described above as measurement results for one resin composition. The set of multiple IR spectra to be input to the prediction model 80 is also referred to as a spectral dataset. Furthermore, the data elements constituting a spectral dataset are also referred to as spectral data. In other words, a spectral dataset includes at least one piece of spectral data.

[0169] For example, the measurement device 20 may perform IR spectrum detection multiple times using one resin composition as a sample and generate one IR spectrum as a measurement result for one resin composition by integrating or averaging the set of detected IR spectra. The measurement device 20 may output the set of IR spectra detected from one resin composition as a measurement result for one resin composition without integrating or averaging. The detected IR spectrum to be used for the integration or averaging process is also referred to as a pre-processing spectrum. The operation of measuring multiple pre-processing spectra by performing the operation of detecting one pre-processing spectrum multiple times is also referred to as one set of measurements. The position and orientation of the sample relative to the detector 22 are assumed to remain unchanged during the execution of one set of measurements. In other words, the position and orientation of the sample are assumed to be the same when detecting each of the multiple pre-processing spectra included in one set of measurements. The measurement device 20 may generate one IR spectrum for each set of measurements by integrating or averaging the multiple pre-processing spectra obtained by performing one set of measurements.

[0170] The measurement device 20 may output each of the pre-treatment spectra obtained in one set of measurements as a measurement result of the IR spectrum of one resin composition without performing an integration or averaging process on the pre-treatment spectra. In other words, the measurement device 20 may detect multiple pre-treatment spectra as the IR spectrum of one resin composition, and output a spectral data set including the multiple pre-treatment spectra as multiple spectral data.

[0171] The measuring device 20 may perform one set of measurements multiple times using one resin composition as a sample. The measuring device 20 may perform one set of measurements multiple times by changing the position or orientation of the sample relative to the detector 22 for each measurement in the set. The measuring device 20 may output a spectral dataset including, as multiple spectral data, multiple IR spectra, which are a collection of single IR spectra generated from the results of each of the multiple measurements in the set, by integrating or averaging multiple pre-processing spectra acquired in each measurement in the set to generate a single IR spectrum. In this case, the number of spectral data included in the spectral dataset is equal to the number of times the measurement in the set was performed. One set of measurements may be performed multiple times.

[0172] The measurement device 20 may output a spectral dataset including, as spectral data, a plurality of IR spectra that are a collection of a plurality of unprocessed spectra generated from the results of each of a plurality of measurements in a set, without performing an integration or averaging process on the plurality of unprocessed spectra acquired in each of the measurements in a set. In this case, the number of spectral data included in the spectral dataset is equal to the product of the number of times the set of measurements was performed and the number of unprocessed spectra acquired in each of the measurements in a set.

[0173] When the predictive model 80 is configured to allow input of a set of multiple IR spectra, i.e., a spectral data set, as the measurement results of one resin composition, the predictive model 80 may be generated by performing learning using, as training data, data that matches the spectral data set measured from one resin composition with measured values ​​of the physical properties of the resin composition being measured or the characteristics of the compounds contained in the resin composition.

[0174] The prediction model 80 may be configured to allow input of statistical values ​​such as the average value, standard deviation, maximum value, or minimum value of the light intensity at each wavelength of the plurality of spectral data included in the spectral data set as the measurement result of one resin composition. In this case, when the statistical values ​​of the light intensity at each wavelength of the plurality of spectral data are input as the measurement result of one resin composition, the prediction model 80 outputs a prediction result of the physical properties of the resin composition or the characteristics of the blend contained in the resin composition. The measurement device 20 may output the statistical values ​​of the light intensity at each wavelength of the plurality of spectral data input to the above-mentioned prediction model 80 as the measurement result of one resin composition.

[0175] When the predictive model 80 is configured to allow input of statistical values ​​of light intensity at each wavelength of multiple spectral data as measurement results for one resin composition, the predictive model 80 may be generated by performing learning using, as training data, data that matches the statistical values ​​of light intensity at each wavelength of multiple spectral data measured from one resin composition with measured values ​​of the physical properties of the resin composition being measured or the characteristics of the compounds contained in the resin composition.

[0176] As described above, the prediction model 80 may be configured to accept input of a spectral data set containing multiple spectral data as measurement results for a single resin composition, or information regarding IR spectral variability, such as statistical values ​​of the light intensity at each wavelength of the multiple spectral data. This increases the amount of information available for predicting the properties of the resin composition or the characteristics of the blends contained in the resin composition. As a result, the prediction accuracy may be improved.

[0177] For example, when the resin composition contains an anisotropic compound such as GF, the variation in the IR spectrum reflects the variation in the orientation or length of the compound, or the dispersion state of the compound. The variation in the IR spectrum reflects, for example, the variation in how anisotropy appears in the resin composition. By configuring the prediction model 80 to allow input of information related to the variation in the IR spectrum, information related to the length or anisotropy of the compound, or the anisotropy of the resin composition, is included in the input data to the prediction model 80. As a result, the prediction accuracy can be improved.

[0178] The prediction model 80 may be configured to accept either a single IR spectrum or a single spectral data set as the measurement result of a single resin composition, thereby enhancing the versatility of the prediction model 80.

[0179] <Input of Information Other Than IR Spectrum into Prediction Model 80> The prediction model 80 may be configured to allow input of measurement environment information when the measurement device 20 measures the resin composition, in addition to the IR spectrum of the resin composition. The measurement environment information may include the temperature, humidity, or air pressure of the space in which the sample 30 is placed in the measurement device 20 (see FIGS. 3A and 3B ). The measurement environment information may include the temperatures of the light source 21, the detector 22, or the prism 23 of the measurement device 20.

[0180] When the prediction model 80 is configured to allow measurement environment information to be input in addition to the IR spectrum of the resin composition, it may be generated by performing learning using data that corresponds the IR spectrum of the resin composition, the measurement environment information when the IR spectrum was measured, and measured values ​​of the physical properties of the resin composition to be measured or the characteristics of the compounds contained in the resin composition as training data.

[0181] The prediction model 80 may be configured to allow input of status information of the manufacturing apparatus 40 when the resin composition is manufactured, in addition to the IR spectrum of the resin composition. The status information may include the temperature, pressure, polymerization time, etc. of the polymerization tank of the manufacturing apparatus 40. The status information may include the temperature, pressure, extrusion speed, etc. of the extruder of the manufacturing apparatus 40.

[0182] When the predictive model 80 is configured to allow input of state information in addition to the IR spectrum of the resin composition, it may be generated by performing learning using as training data data that correlates the IR spectrum of the resin composition, state information of the manufacturing apparatus 40 when the resin composition was manufactured, and measured values ​​of the physical properties of the resin composition to be measured or the characteristics of the compounds contained in the resin composition.

[0183] The prediction model 80 may be configured to allow both measurement environment information and state information to be input in addition to the IR spectrum of the resin composition. When the prediction model 80 is configured to allow both measurement environment information and state information to be input in addition to the IR spectrum of the resin composition, the prediction model 80 may be generated by performing learning using, as training data, data that associates the IR spectrum of the resin composition, measurement environment information when the IR spectrum was measured, state information of the manufacturing apparatus 40 when the resin composition was manufactured, and measured values ​​of the physical properties of the resin composition to be measured or the characteristics of the blends contained in the resin composition.

[0184] As described above, the prediction accuracy can be improved by inputting at least one of the measured environmental information and the state information into the prediction model 80.

[0185] <Output from Prediction Model 80> The prediction device 10 may output the reliability of the prediction result in addition to the prediction result of the physical properties of the resin composition or the characteristics of the blends contained in the resin composition. In this case, the prediction model 80 may be configured to output the reliability of the prediction result in addition to the prediction result.

[0186] The prediction device 10 may output the reliability of the prediction result using a model that outputs the reliability of the prediction result as a separate model from the prediction model 80 that outputs the prediction result of the physical properties of the resin composition or the characteristics of the compound contained in the resin composition.

[0187] In this disclosure, the term reliability should be understood broadly and may be expressed as, for example, uncertainty. The reliability may include information regarding the reliability of the prediction model 80 itself or information regarding the reliability of the prediction results, such as predicted values, output by the prediction model 80.

[0188] When the reliability represents information about the reliability of the prediction model 80 itself, known evaluation indices such as the coefficient of determination (R2), mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), or mean absolute percentage error (MAPE), which are calculated based on predicted values ​​and actual measured values ​​of the physical properties of the resin composition that is the sample 30 or the properties of the blend contained in the resin composition, may be used as the reliability. Furthermore, when cross-validation is performed when constructing the prediction model 80, the average value of the values ​​of the above evaluation indices obtained for each fold may be used as the reliability. For example, R2CV may be used as the average value of R2.

[0189] When the reliability represents information about the reliability of a predicted value, an index such as the variance or standard deviation of the predicted value, or a confidence interval may be used as the reliability. Furthermore, a model such as Gaussian Process Regression (GPR), a confusion matrix, or the like may be used to calculate the reliability.

[0190] The reliability may be expressed as a rank corresponding to the difference between the predicted value and the measured value of the physical property of the resin composition or the property of the compound contained in the resin composition. For example, the reliability may be expressed as rank A when the absolute value of the difference between the predicted value and the measured value is less than a first threshold, as rank B when the absolute value of the difference between the predicted value and the measured value is equal to or greater than the first threshold and less than a second threshold, or as rank C when the absolute value of the difference between the predicted value and the measured value is equal to or greater than the second threshold.

[0191] The reliability may be associated with the probability that the predicted value approximately matches the measured value. For example, the reliability may be expressed as being higher as the probability that the predicted value approximately matches the measured value increases.

[0192] The manner in which the reliability is expressed is not limited to the above-mentioned examples, and may include various other manners.

[0193] The calculated reliability can be used for various purposes, such as judging or ranking the quality of a resin composition, detecting abnormalities in a resin composition production line, or determining whether maintenance or updating of the prediction model 80 is necessary. The reliability may be used as a criterion for determining whether to execute a proposed change in manufacturing conditions based on the predicted results of the physical properties of the resin composition or the properties of the blends contained in the resin composition. For example, the manufacturing system 100 may be configured to automatically execute a change in manufacturing conditions when the numerical value representing the reliability is equal to or greater than a threshold, and not execute a change in manufacturing conditions and display an alert to notify the operator when the numerical value representing the reliability is equal to or less than the threshold. As described above, calculating the reliability can improve the quality of a resin composition.

[0194] When multiple IR spectra are input as the IR spectrum of a resin composition, or when statistical values ​​of the light intensity at each wavelength of the IR spectrum of a resin composition are input, i.e., when information regarding the variability of the IR spectrum of a resin composition is input, the prediction model 80 outputs a prediction result of the physical properties of the resin composition or the characteristics of the blends contained in the resin composition, and may also output the reliability of the prediction result based on the variability of the IR spectrum of the resin composition. For example, the reliability may be output so that the greater the variability of the IR spectrum of the resin composition, the lower the reliability. The prediction model 80 may output a value based on statistical values ​​such as the variance or standard deviation of multiple IR spectra as the reliability.

[0195] When measurement environment information is input in addition to the IR spectrum of the resin composition, the prediction model 80 may output the reliability of the prediction result of the physical properties of the resin composition or the characteristics of the blends contained in the resin composition based on the measurement environment information. For example, the reliability may be output so that it is low when the temperature of the measurement environment is outside the recommended temperature range defined in the specifications of the measurement device 20.

[0196] When state information of the manufacturing apparatus 40 is input in addition to the IR spectrum of the resin composition, the prediction model 80 may output the reliability of the prediction result of the physical properties of the resin composition or the characteristics of the blends contained in the resin composition based on the state information. For example, the reliability may be output so that it is low when the value of a parameter representing the state of the manufacturing apparatus 40 deviates from the value of a parameter set as a manufacturing condition.

[0197] That is, the reliability may indicate that it decreases due to disturbances. For example, the prediction model 80 may output the reliability such that the reliability decreases as the disturbance increases. Disturbances may include, but are not limited to, temperature, humidity, vibrations applied to the device, light source intensity, fluctuations or changes in the atmosphere inside the device, changes in pressure applied to the measurement unit, or changes in the person performing the measurement.

[0198] When the prediction model 80 is configured to output the reliability of the predicted results of the physical properties of the resin composition or the characteristics of the compound contained in the resin composition, the prediction model 80 may be generated by performing learning using as training data data that corresponds the actual difference between the predicted results of the physical properties of the resin composition or the characteristics of the compound contained in the resin composition by the prediction model 80 and the actual measured value, and the variation in the IR spectrum of the resin composition when the actual result was obtained, measurement environment information, or status information of the manufacturing equipment 40.

[0199] Although the embodiments of the present disclosure have been described based on the drawings and examples, it should be noted that those skilled in the art could make various modifications or alterations based on the present disclosure. Therefore, it should be noted that these modifications and alterations are included within the scope of the present disclosure. For example, the functions included in each component or step can be rearranged so as not to cause logical inconsistencies, and multiple components or steps can be combined or divided into one. The embodiments of the present disclosure can also be realized as a program executed by a processor included in an apparatus or a storage medium on which a program is recorded. It should be understood that these are also included within the scope of the present disclosure.

[0200] 1 Prediction system 10 Prediction device (11: data acquisition unit, 12: data processing unit, 13: spectrum analysis unit, 14: prediction unit, 15: memory unit, 16: output unit) 20 Measuring device (21: light source, 22: detection unit, 23: prism) 30 Sample 40 Manufacturing device (42: control unit) 80 Prediction model (81: first prediction model, 82: second prediction model) 100 Manufacturing system

Claims

1. A method for predicting a resin composition comprising: acquiring an infrared spectrum of the resin composition comprising a resin and a blend; and predicting at least one of a physical property of the resin composition or a characteristic of the blend from the infrared spectrum of the resin composition using a predictive model.

2. The prediction method according to claim 1, comprising: predicting at least one of the properties of the blend as a intermediate property of the resin composition from the infrared spectrum of the resin composition using a first prediction model as the prediction model; and predicting at least one of the physical properties of the resin composition as a target property of the resin composition from the prediction result of the intermediate property of the resin composition using a second prediction model as the prediction model.

3. The prediction method according to claim 1 or 2, wherein the infrared spectrum of the resin composition includes a wavenumber range in which the absorbance changes in response to changes in the properties of the blend.

4. The prediction method of claim 3, wherein the characteristics of the blend include at least one of information regarding the content of the blend or the size of the blend.

5. The prediction method according to claim 1 or 2, wherein the physical properties of the resin composition include at least one of the mechanical properties of the resin composition, the fluidity of the resin composition, or the molecular weight of the resin composition.

6. The prediction method according to claim 1 or 2, wherein the prediction model is a trained model generated by performing machine learning using training data that matches the infrared spectrum of the resin composition with at least one actual measured value or theoretical value of a physical property of the resin composition or a characteristic of the blend.

7. The prediction method according to claim 1 or 2, further comprising measuring the infrared spectrum of the resin composition by an ATR method.

8. The prediction method according to claim 7, wherein a spectrum in the MIR band is obtained as the infrared spectrum of the resin composition.

9. The prediction method according to claim 7, further comprising pressing the solid pellets of the resin composition to measure the infrared spectrum of the solid pellets by an ATR method.

10. The prediction method according to claim 7, further comprising measuring an infrared spectrum of the resin composition in a molten state by an ATR method.

11. The prediction method according to claim 1 or 2, further comprising measuring the infrared spectrum of the resin composition by a transmission method.

12. The prediction method according to claim 11, wherein a spectrum in the NIR band is obtained as the infrared spectrum of the resin composition.

13. The prediction method according to claim 11, wherein the infrared spectrum of the resin composition, when expressed in terms of absorbance, has a maximum absorbance of 0.3 or more and 2.0 or less, a minimum absorbance of 1.0 or less, and a difference between the maximum and minimum absorbance values ​​of 0.05 or more.

14. The prediction method according to claim 1 or 2, wherein the resin of the resin composition comprises polyamide.

15. The prediction method according to claim 1 or 2, wherein the resin composition formulation includes a colorant.

16. The prediction method according to claim 1 or 2, wherein the resin composition formulation includes an inorganic filler.

17. The prediction method according to claim 1, comprising: obtaining a spectral dataset including a plurality of spectral data as an infrared spectrum of the resin composition; and predicting at least one of the physical properties of the resin composition or the characteristics of the blend from the spectral dataset using a model that inputs the spectral dataset as the prediction model.

18. A prediction method as described in claim 17, wherein, in the case where a single infrared spectrum is calculated by integrating or averaging multiple pre-treatment spectra as the infrared spectrum of the resin composition, the spectral data set includes the multiple pre-treatment spectra as the multiple spectral data.

19. The prediction method described in claim 17, wherein, when a set of measurements is performed to measure a plurality of pre-treatment spectra in order to calculate a single infrared spectrum by integrating or averaging the plurality of pre-treatment spectra as the infrared spectrum of the resin composition, the spectral data set includes, as the plurality of spectral data, at least one of the plurality of pre-treatment spectra or the one infrared spectrum obtained in each measurement by performing the set of measurements a plurality of times.

20. The prediction method according to claim 1 or 2, further comprising calculating a reliability of the predicted result of at least one of the physical properties of the resin composition or the characteristics of the blend.

21. A prediction method according to any one of claims 17 to 19, comprising calculating a reliability of a prediction result of at least one of the physical properties of the resin composition or the characteristics of the blend based on the variability of the plurality of spectral data.

22. A prediction method according to claim 1 or 2, comprising: acquiring measurement environment information of the infrared spectrum of the resin composition; and predicting at least one of the physical properties of the resin composition or the characteristics of the blend based on the infrared spectrum of the resin composition and the measurement environment information, using a model into which the infrared spectrum of the resin composition and the measurement environment information are input as the prediction model.

23. The prediction method according to claim 22, further comprising calculating a reliability of a prediction result of at least one of the physical properties of the resin composition or the characteristics of the blend based on the measurement environment information.

24. A prediction method according to claim 1 or 2, comprising: acquiring manufacturing conditions of the resin composition; and predicting at least one of the physical properties of the resin composition or the characteristics of the blend based on the infrared spectrum of the resin composition and the manufacturing conditions, using a model in which the infrared spectrum of the resin composition and the manufacturing conditions are input as the prediction model.

25. The prediction method according to claim 24, further comprising calculating a reliability of a prediction result of at least one of the physical properties of the resin composition or the characteristics of the blend based on the manufacturing conditions.

26. A method for evaluating the quality of a resin composition, comprising: predicting at least one of the physical properties of the resin composition or the characteristics of a blend contained in the resin composition by executing the prediction method described in claim 1 or 2; and outputting a result of evaluating the quality of the resin composition based on at least one of the predicted results of the physical properties of the resin composition or the predicted results of the characteristics of the blend.

27. A prediction program for causing a processor to execute the prediction method according to claim 1 or 2.

28. A prediction system comprising: a measurement device that measures an infrared spectrum of a resin composition containing a resin and a blend; and a prediction device that predicts at least one physical property of the resin composition or a characteristic of the blend from the infrared spectrum of the resin composition using a prediction model.

29. The prediction system according to claim 28, wherein the measuring device measures the infrared spectrum of the resin composition within a production line for the resin composition.

30. A manufacturing system comprising the prediction system according to claim 29 and a manufacturing device installed in a production line for the resin composition, wherein the manufacturing device controls the manufacturing conditions for the resin composition so that the absolute value of the difference between a predicted value and a target value for at least one item of a physical property of the resin composition or a characteristic of the blend is less than a difference threshold value.

31. The manufacturing system described in claim 30, wherein the manufacturing apparatus manufactures the resin composition by changing manufacturing conditions of the resin composition so as to bring a predicted value of at least one item of the physical properties of the resin composition or the characteristics of the blend closer to a target value, the measuring apparatus measures the infrared spectrum of the resin composition manufactured by changing the manufacturing conditions, the predicting apparatus predicts at least one item of the physical properties of the resin composition or the characteristics of the blend manufactured by changing the manufacturing conditions, and the manufacturing apparatus repeats the change in manufacturing conditions until an absolute value of the difference between the predicted value and the target value of at least one item of the physical properties of the resin composition manufactured by changing the manufacturing conditions or the characteristics of the blend becomes less than a difference threshold value.

32. A manufacturing system as described in claim 30, wherein the manufacturing apparatus controls the manufacturing conditions of the resin composition using a model trained using learning data that matches pre-change manufacturing conditions, a predicted value or actual measurement value of at least one item of the physical properties of the resin composition or the characteristics of the blend produced under the pre-change manufacturing conditions, and changed manufacturing conditions obtained by changing the pre-change manufacturing conditions, and a predicted value or actual measurement value of at least one item of the physical properties of the resin composition or the characteristics of the blend produced under the changed manufacturing conditions.

33. A manufacturing system according to any one of claims 30 to 32, wherein the manufacturing device controls the manufacturing conditions of the resin composition based on the prediction results of a plurality of items of physical properties of the resin composition or characteristics of the blend.

34. The manufacturing system according to claim 33, wherein the manufacturing device controls the manufacturing conditions of the resin composition based on a predicted value of the viscosity of the resin composition and a predicted value of a mechanical property of the resin composition.

Images