Prediction method, prediction device

Abstract

The present invention predicts, in a process for producing resin powder, the physical properties of water-absorbing resin powder from a near-infrared absorption spectrum. A prediction device (100) comprises: a measurement data acquisition unit (11) that acquires near-infrared measurement data; and a prediction unit (13) that inputs at least one of the near-infrared measurement data and one or more pieces of processed data which have been prepared on the basis of the near-infrared measurement data into a prediction model, and that outputs prediction information pertaining to the physical properties of the resin powder.

Description

[Technical Field]

[0001] This disclosure relates to a prediction method and apparatus for predicting the physical properties of water-absorbent resin powder. [Background technology]

[0002] Superabsorbent polymers (SAP) are resins that are both water-swellable and water-insoluble. SAP is often in powder (or granular) form. Known properties of superabsorbent polymers include water absorption ratio (CRC), water absorption ratio under load (AAP), water absorption rate, and SFC (saline flow induction). The required physical properties and their ranges differ depending on the application, specifically the type and composition of the sanitary material used. Therefore, a wide variety of SAPs exhibiting diverse physical properties are required depending on the form of the final product.

[0003] To confirm the physical properties of SAP powder, it is necessary to apply different measurement methods for each property measurement item, and each measurement requires a predetermined amount of time. In SAP manufacturing, it is difficult to grasp the physical property values ​​of SAP at each process in real time, which may lead to the production of products that do not meet specifications. In other words, it may lead to a decrease in yield during SAP manufacturing.

[0004] Patent Document 1 discloses a method for predicting the physical properties of a water-absorbent resin using a specific Raman spectrum. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] International Publication No. 2020 / 109601 [Overview of the project] [Problems that the invention aims to solve]

[0006] In Raman spectroscopy, a specific single wavelength is irradiated onto the sample, and scattered light within a specific wavenumber range is measured. Due to the characteristics of this irradiated light, Raman spectra are less affected by the particle size of the object being measured. For this reason, Raman spectra are unsuitable for accurately measuring the particle size of an object.

[0007] In contrast, the method using near-infrared absorption spectroscopy involves irradiating the target sample with multiple wavelengths or a continuous spectrum of near-infrared light (generally wavelengths of 750 nm to 2500 nm) and measuring the transmitted, absorbed, refracted, reflected, and diffused light. This allows for measurements that include information about the particle size of the target sample. Furthermore, compared to Raman spectroscopy, using near-infrared light allows for measurements over a wider wavelength range, thus providing more accurate average information about the target sample.

[0008] One aspect of this disclosure aims to enable the prediction of the physical properties of a water-absorbent resin powder from its near-infrared absorption spectrum during the manufacturing process of the water-absorbent resin powder. [Means for solving the problem]

[0009] To solve the above problems, a prediction method according to one aspect of the present disclosure is a method for predicting the physical properties of a resin powder, (Note: The resin powder refers to either a water-absorbing resin powder or an intermediate product generated in a manufacturing process for producing the water-absorbing resin powder), and includes: a measurement data acquisition step of acquiring near-infrared measurement data showing the near-infrared absorption spectrum of the resin powder; and a prediction step of inputting at least one of the near-infrared measurement data and one or more processing data generated based on the near-infrared measurement data into a prediction model to output prediction information related to the physical properties of the resin powder.

[0010] Furthermore, in order to solve the above problems, a prediction device according to one aspect of the present disclosure is a prediction device for predicting the physical properties of a resin powder, (Note: The resin powder refers to either a water-absorbing resin powder or an intermediate product generated in a manufacturing process for producing the water-absorbing resin powder), a measurement data acquisition unit (which acquires measurement data showing the near-infrared absorption spectrum measured for the resin powder), and a prediction unit (which inputs at least one of the near-infrared measurement data and one or more processing data generated based on the near-infrared measurement data into a prediction model and outputs prediction information related to the physical properties of the resin powder). [Effects of the Invention]

[0011] According to one aspect of this disclosure, the physical properties of a water-absorbing resin powder can be predicted from the near-infrared absorption spectrum. [Brief explanation of the drawing]

[0012] [Figure 1] This is a block diagram showing an example of the configuration of a prediction system equipped with a prediction device according to Embodiment 1 of this disclosure. [Figure 2] This is a functional block diagram showing an example of the main components of a prediction device. [Figure 3] This is a flowchart showing the processing flow performed by the prediction device. [Figure 4] This is a functional block diagram showing an example of the main components of a prediction device that generates a prediction model. [Figure 5] This diagram shows the data structure of near-infrared measurement data. [Figure 6] This is a diagram showing the data structure of physical property information. [Figure 7] This flowchart shows the processing flow performed by a predictive machine that executes machine learning. [Figure 8] This is a block diagram showing an example of a prediction system in Embodiment 2 of this disclosure. [Figure 9] This table shows the correspondence between the MAC address acquired by the prediction device according to Embodiment 2 of this disclosure and the near-infrared spectrophotometer. [Figure 10] It is a graph showing the correlation between the measured value and the predicted value of gel D50. [Figure 11] It is a graph showing the correlation between the measured value and the predicted value of CRC. [Figure 12] It is a graph showing the correlation between the measured value and the predicted value of AAP. [Figure 13] It is a graph showing the correlation between the measured value and the predicted value of SFC. [Figure 14] It is a graph showing the correlation between the measured value and the predicted value of D50. [Figure 15] It is a graph showing the correlation between the measured value and the predicted value of the water content (solid content).

Mode for Carrying Out the Invention

[0013] 〔Embodiment 1〕 Hereinafter, embodiments of the present disclosure will be described in detail.

[0014] (Configuration of Prediction System 1000) First, the configuration of a prediction system 1000 including a prediction device 100 according to an embodiment of the present disclosure will be described with reference to FIG. 1. FIG. 1 is a block diagram showing an example of the configuration of the prediction system 1000.

[0015] The prediction system 1000 includes a prediction device 100, a near-infrared spectrophotometer 3, and an external device 4.

[0016] The prediction device 100 includes a CPU 1 and a memory 2. As shown in FIG. 1, the prediction device 100 may be communicably connected to the near-infrared spectrophotometer 3 and the external device 4. The communication between the prediction device 100 and the near-infrared spectrophotometer 3 may be any of short-range wireless communication, wired connection, and communication via a network such as the Internet. Alternatively, the communication between the prediction device 100 and the near-infrared spectrophotometer 3 may be configured to be directly connected by a connector such as a USB terminal. The communication between the prediction device 100 and the external device 4 is the same as the communication between the prediction device 100 and the near-infrared spectrophotometer 3.

[0017] Figure 1 shows a case where there is one near-infrared spectrophotometer 3 and one external device 4, both of which are communicatively connected to the prediction device 100, but the system is not limited to this. There may be one or more near-infrared spectrophotometers 3 and external devices 4, each of which are communicatively connected to the prediction device 100.

[0018] The prediction device 100 inputs at least one of the following into a prediction model: near-infrared measurement data showing a near-infrared absorption spectrum acquired from the near-infrared spectrophotometer 3, and one or more processing data generated based on the near-infrared measurement data, and outputs prediction information related to the physical properties of the resin powder. In this specification, "measurement data showing a near-infrared absorption spectrum" may be simply referred to as near-infrared absorption spectrum (near-infrared measurement data).

[0019] The near-infrared spectrophotometer 3 is an instrument that measures the reflected light and transmitted light from a resin powder when it is irradiated with near-infrared light, and calculates the near-infrared absorption spectrum that represents the near-infrared absorption characteristics of the resin powder. Here, near-infrared light is light having a wavelength range of 750 to 2500 nm. The near-infrared absorption spectrum will be explained later.

[0020] External device 4 may be any device that receives the prediction results output from prediction device 100. For example, external device 4 may be any display device, or it may be a computer used by an administrator who manages the resin powder manufacturing process. Alternatively, external device 4 may be any manufacturing device that performs the processing of the resin powder manufacturing process.

[0021] (Configuration of prediction device 100) In the following section, the configuration of the prediction device 100, which predicts the physical properties of water-absorbent resin powder (hereinafter sometimes simply referred to as "resin powder") using the prediction model 22, will be explained with reference to Figure 2. Figure 2 is a functional block diagram showing an example of the main components of the prediction device 100.

[0022] Here, the predictive model 22 may be generated by machine learning processing using at least one of the following (1) and (2) as training data.

[0023] (1) A combination of near-infrared measurement data, including near-infrared absorption spectra, of several previously manufactured resin powders with known physical properties, and physical property information of the final product associated with said near-infrared measurement data.

[0024] (2) A combination of near-infrared measurement data, including near-infrared absorption spectra of multiple pre-produced intermediate products with known physical properties, generated in the manufacturing process for producing each pre-produced resin powder, and physical property information of the intermediate products associated with the near-infrared measurement data.

[0025] In one example, a pre-trained prediction model 22 may be pre-installed in the prediction device 100. Alternatively, the prediction device 100 may further include a function to perform machine learning processing using at least one of the above (1) and (2) as training data.

[0026] The prediction device 100 can accurately predict the physical properties of the resin powder from which the near-infrared absorption spectrum was measured, by using the prediction model 22 generated by such machine learning. The method for generating the prediction model 22 will be described later.

[0027] As shown in Figure 2, the prediction device 100 includes a control unit 10 that comprehensively controls each part of the prediction device 100, a storage unit 20 that stores various data used by the control unit 10, and a communication unit 50 for outputting prediction results to an external device 4. The control unit 10 corresponds to the CPU 1 in Figure 1, and the storage unit 20 corresponds to the memory 2 in Figure 1.

[0028] The communication unit 50 is for data communication with the external device 4. Communication between the prediction device 100 and the external device 4 may be via short-range wireless communication, wired connection, or communication via a network such as the Internet. Alternatively, the prediction device 100 and the external device 4 may be directly connected by a connector such as a USB terminal.

[0029] The control unit 10 includes a measurement data acquisition unit 11 and a prediction unit 13.

[0030] The measurement data acquisition unit 11 acquires a near-infrared absorption spectrum from the near-infrared spectrophotometer 3. The measurement data acquisition unit 11 may store the acquired near-infrared absorption spectrum in the storage unit 20 as near-infrared measurement data (not shown). The measurement data acquisition unit 11 may also read a previously stored near-infrared absorption spectrum from the near-infrared measurement data and use it for subsequent predictions.

[0031] The prediction unit 13 inputs the near-infrared absorption spectrum into the prediction model 22, which will be described later, and outputs prediction information related to the physical properties of the resin powder.

[0032] Furthermore, the prediction unit 13 may generate one or more processed data based on the near-infrared absorption spectrum. Here, processed data is data that differs from the raw near-infrared absorption spectrum data, and is obtained by applying one or more predetermined preprocessing steps to the near-infrared absorption spectrum. The prediction unit 13 may perform preprocessing defined in the prediction model 22 on the near-infrared absorption spectrum acquired by the measurement data acquisition unit 11 (preprocessing step). The preprocessing includes at least one of the following steps. • Outlier removal process Outlier removal is performed on multiple resin powders. Article This process involves comparing near-infrared absorption spectra measured at a given location, detecting near-infrared absorption spectra that are significantly different from other near-infrared absorption spectra, and removing those near-infrared absorption spectra. Article"Measured at a specific location" means that the measurement was performed by irradiating multiple different regions of a sample consisting of resin powder and having a predetermined area with measurement light. Specific outlier detection methods include One Class Support Vector Machine (One Class SVM) processing, detection using Mahalanobis distance, LOF (Local Outlier Factor) method, Tukey method, and nearest neighbor method. • Averaging process The averaging process involves multiple resin powders. Article This process calculates a single average spectral data from multiple near-infrared absorption spectra measured at the same location. • Wavelength selection process The wavelength selection process is the process of selecting the wavelength range of spectral data to be input to the prediction model 22, which will be described later. In the wavelength range processing, for example, a wavelength range in which a characteristic absorption pattern appears for each resin powder for which the near-infrared absorption spectrum has been measured may be selected. Differential processing Differential processing generates differential data by differentiating the spectral data with respect to wavelength. The differential data may include data obtained by first-order differentiation and second-order differentiation with respect to wavelength. • Baseline correction processing Baseline correction processing involves multiple resin powders. Article This process aligns the baselines of multiple near-infrared absorption spectra measured at the same location.

[0033] The preprocessing described above is merely an example, and the preprocessing performed by the prediction unit 13 is not limited to these. For example, the prediction unit 13 may perform the following processing on the near-infrared absorption spectrum. • Smoothing process (weighted moving average processing, smoothing spline processing, etc.) • Difference spectral processing • Standardization process (SNV (Standard Normal Variate) process) • Multiple scattering correction processing (MSC (Multiplicative Scatter Correction) processing) • Dimensionality reduction using Principal Component Analysis (PCA) Other methods such as classification and clustering may also be performed.

[0034] In the prediction method performed by the prediction unit 13, for example, near-infrared measurement data showing the near-infrared absorption spectrum of the resin powder is used for multiple resin powders. Article The process may include an averaging step in which multiple near-infrared absorption spectra are acquired at a location, and an averaging process is performed on the acquired spectra to calculate average spectral data. In the prediction step, the average spectral data may be input to the prediction model 22 as processed data. Alternatively, the averaging step may be a process defined by the prediction model 22.

[0035] The prediction method performed by the prediction unit 13 may further include, for example, a wavelength range selection step for selecting the wavelength range of average spectral data to be input to the prediction model 22. Furthermore, in the prediction step, the average spectral data within that wavelength range may be input to the prediction model 22 as processed data. The wavelength range selection step may also be a process defined by the prediction model 22.

[0036] The prediction method performed by the prediction unit 13 may further include, for example, a differential data generation step that generates differential data obtained by differentiating the average spectral data in the wavelength range described above with respect to wavelength. In addition, in the prediction step, the differential data may be input to the prediction model 22 as processed data. Furthermore, the differential data generation step may be a process defined by the prediction model 22.

[0037] (Processing performed by the prediction device 100) The processes performed by the prediction device 100 will be explained below using Figure 3. Figure 3 is a flowchart showing the flow of processes performed by the prediction device 100.

[0038] First, the measurement data acquisition unit 11 acquires near-infrared measurement data, which is the near-infrared absorption spectrum measured by the near-infrared spectrophotometer 3 (Step S1: Near-infrared measurement data acquisition step).

[0039] Next, the prediction unit 13 reads the prediction model 22 from the memory unit 20 (step S2).

[0040] The prediction unit 13 inputs the near-infrared measurement data acquired in step S1 into the prediction model 22 (step S3). At this time, the prediction unit 13 may perform the above-described preprocessing on the acquired near-infrared absorption spectrum based on the prediction model 22. The prediction unit 13 may perform one or more of the above-described preprocessing steps. The preprocessing performed by the prediction unit 13 will be explained later with specific examples.

[0041] Next, the prediction unit 13 predicts the physical properties of the target object based on the prediction model 22, using either pre-processed or unprocessed near-infrared measurement data (step S4: prediction step).

[0042] The communication unit 50 outputs prediction information, which shows the prediction result output from the prediction unit 13, to the external device 4 (step S4).

[0043] <Example of pretreatment> Here, as an example, we will explain the specific pretreatment performed by the prediction unit 13 when predicting gel D50 from the near-infrared absorption spectrum of a hydrated gel, which is an intermediate product in the manufacturing process of resin powder.

[0044] The prediction unit 13 performs outlier detection processing (e.g., One Class SVM) on multiple near-infrared absorption spectra acquired by the measurement data acquisition unit 11, and removes near-infrared absorption spectra that are significantly different from the other near-infrared absorption spectra.

[0045] Next, the prediction unit 13 performs an averaging process on the remaining multiple near-infrared absorption spectra to generate a single average spectral data.

[0046] The preprocessing performed by the prediction unit 13 may vary depending on which stage in the manufacturing process of the resin powder the near-infrared absorption spectrum measurement target is and what the physical properties of the target are. That is, the prediction unit 13 may perform a wavelength selection process to select the wavelength range of the spectral data. Alternatively, the prediction unit 13 may perform a differentiation process to generate differential data by differentiating the spectral data with respect to wavelength. Furthermore, these processes may be performed in combination.

[0047] Thus, by performing appropriate pretreatment depending on which stage in the manufacturing process of the resin powder the near-infrared absorption spectrum measurement target is and what the physical properties to be predicted are, the prediction accuracy of the prediction device 100 can be improved.

[0048] (Configuration of prediction device 100) Next, the configuration of the prediction device 100, which performs machine learning to generate the prediction model 22, will be explained using Figure 4. Figure 4 is a functional block diagram showing an example of the main components of the prediction device 100 that generates the prediction model 22. For the sake of explanation, components having the same function as those described in Figure 1 are denoted by the same reference numerals, and their descriptions are not repeated. The prediction device 100 may generate the prediction model 22 by performing any known supervised machine learning.

[0049] The control unit 10 includes a measurement data acquisition unit 11, a prediction unit 13, and a prediction model generation unit 18.

[0050] The measurement data acquisition unit 11 acquires multiple near-infrared absorption spectra (also referred to as a group of near-infrared absorption spectra) included in the near-infrared measurement data 21, as specified by the prediction model generation unit 18, and outputs the group of near-infrared absorption spectra to the prediction unit 13.

[0051] The prediction unit 13 reads the prediction model candidate (described later) generated by the prediction model generation unit 18 from the prediction model generation unit 18. Furthermore, it inputs the near-infrared absorption spectrum group included in the near-infrared measurement data 21, as specified by the prediction model generation unit 18, into the prediction model candidate, and outputs the prediction result, which predicts the group of physical properties corresponding to the input near-infrared absorption spectrum group, to the prediction model generation unit 18.

[0052] The prediction model generation unit 18 generates a candidate prediction model to be trained and validated using machine learning. A candidate prediction model is a prediction model for which prior machine learning has not been completed. Once predetermined machine learning is completed and the prediction accuracy meets the criteria, the candidate prediction model is stored in the storage unit 20 as a prediction model 22. The prediction model generation unit 18 also selects a group of data to be used for machine learning from the near-infrared measurement data 21 and physical property information 23 stored in the storage unit 20.

[0053] The prediction model generation unit 18 may calculate a model evaluation index by comparing (1) and (2) below.

[0054] (1) Prediction results predicted by the prediction model candidate output from the prediction unit 13 (2) A group of physical properties included in the physical property information 23 read from the memory unit 20, which is associated with the group of near-infrared absorption spectra input to the candidate prediction model. Here, the model evaluation index is, for example, an index for evaluating the error between the prediction result in (1) and the group of physical properties included in the physical property information 23 in (2). The model evaluation index may be any index that can evaluate the accuracy of the prediction result, and may be the mean squared error or the coefficient of determination (R 2 ) is also acceptable.

[0055] The prediction model generation unit 18 determines, based on the model evaluation index, whether the prediction model candidate satisfies predetermined evaluation criteria. The predetermined evaluation criteria are arbitrary criteria set in advance to evaluate the prediction accuracy of the prediction model candidate.

[0056] If a candidate prediction model meets predetermined evaluation criteria, the prediction model generation unit 18 stores the candidate prediction model in the prediction model 22 as the optimal prediction model. On the other hand, if the generated candidate prediction model does not meet the predetermined evaluation criteria, the prediction model generation unit 18 updates the candidate prediction model.

[0057] "Updating the candidate prediction model" may include updating the candidate prediction model by updating its weights, hyperparameters, etc., so that the error between the prediction result and the physical properties included in the physical property information 23 is minimized, and generating a new candidate prediction model. The backpropagation method or the like may be applied to updating the candidate prediction model.

[0058] The physical property information 23 includes physical property information of the final product, which is associated with near-infrared measurement data, including near-infrared absorption spectra of multiple previously manufactured resin powders with known physical properties. Furthermore, the physical property information 23 includes physical property information of intermediate products, which is associated with near-infrared measurement data 21, showing the near-infrared absorption spectra of multiple previously generated intermediate products with known physical properties, produced during the manufacturing process for each manufactured resin powder. The physical property information may be information related to the physical properties of the water-absorbing resin powder, as will be explained later.

[0059] The physical property information 23 may include measured values ​​of the resin powder being measured or measured values ​​of the intermediate product as physical properties. Each physical property may also be assigned a measurement ID. The physical property information 23 may also include a set of physical properties used in machine learning to generate the prediction model 22 from the candidate prediction model.

[0060] The near-infrared measurement data 21 includes data files of the near-infrared absorption spectra of the resin powder or intermediate product that was measured. These data files may be, for example, CSV files and text files. Each near-infrared absorption spectrum data file may also be assigned a measurement ID. The near-infrared measurement data 21 may also include a group of near-infrared absorption spectra used in machine learning to generate a prediction model 22 from candidate prediction models.

[0061] Here, the correspondence between the near-infrared absorption spectrum contained in the near-infrared measurement data 21 and the physical properties contained in the physical property information 23 will be explained using Figures 5 and 6.

[0062] Figure 5 shows the data structure of the near-infrared measurement data 21, and Figure 6 shows the data structure of the physical property information 23. In Figure 5, the near-infrared measurement data 21 has multiple near-infrared absorption spectrum data files, and each near-infrared absorption spectrum data file is assigned a measurement ID.

[0063] Furthermore, in Figure 6, the physical property information 23 has data files for multiple physical properties (measured values), and each physical property is assigned a measurement ID. As shown in Figure 6, this measurement ID may be the same as the measurement ID assigned to the near-infrared absorption spectrum data file mentioned above, and the near-infrared absorption spectrum and physical property information with the same ID may be the result of measuring the same product. For example, the near-infrared absorption spectrum assigned measurement ID "001" in Figure 5 and the physical property assigned measurement ID "001" in Figure 6 may be data measured from the same resin powder or intermediate product.

[0064] The prediction unit 13 uses a candidate prediction model specified by the prediction model generation unit 18. The prediction unit 13 may obtain the measurement ID of the near-infrared absorption spectrum specified by the prediction model generation unit 18 and compare it with a group of physical properties corresponding to the group of near-infrared absorption spectra read from the storage unit 20 that has the same measurement ID as the prediction result of the prediction unit 13, as information about the same product. Alternatively, the prediction result output from the prediction unit 13 may be assigned the same measurement ID as the measurement ID of the near-infrared absorption spectrum input to the candidate prediction model. The prediction model generation unit 18 may compare the prediction result output from the prediction unit 13 with a group of physical properties corresponding to the group of near-infrared absorption spectra read from the storage unit 20 that has the same measurement ID as the prediction result, as information about the same product.

[0065] Note that the measurement ID assigned to the near-infrared absorption spectrum and the measurement ID assigned to the physical properties may be different. If the respective measurement IDs are different, it is sufficient that the measurement ID of the physical properties corresponds to the measurement ID of the near-infrared absorption spectrum.

[0066] The prediction model 22 may be generated using machine learning methods that either linear regression or nonlinear regression. Examples of machine learning methods for generating the prediction model 22 include, for linear regression, PLS (partial least squares regression), PCR (principal component regression), simple regression, multiple regression, ridge regression, Lasso regression, and Bayesian linear regression. Examples of nonlinear regression include (convolutional) neural networks, support vector regression, k-nearest neighbors, and regression trees. Ensemble learning combining the methods listed above may also be used. Furthermore, the prediction model 22 may be a model that makes numerical predictions of various physical properties, or a model that makes judgment predictions on whether the physical properties are pass or fail.

[0067] In this disclosure, it is preferable that machine learning of PLS ​​and PCR is used to generate the predictive model 22.

[0068] The prediction model 22 may specify the necessary pretreatment to be performed on the near-infrared absorption spectrum.

[0069] (Process to generate predictive model 22) Next, the processing performed by the prediction device 100 will be explained using Figure 7. Figure 7 is a flowchart showing the flow of processing performed by the prediction device 100, which executes machine learning. Here, we will explain using the example where the prediction device 100 generates a prediction model 22 using a combination of near-infrared measurement data 21 and physical property information 23 corresponding to the near-infrared measurement data 21 as training data.

[0070] First, the measurement data acquisition unit 11 reads from the storage unit 20 a group of near-infrared absorption spectra included in the near-infrared measurement data 21, which the prediction model generation unit 18 has designated for use as a candidate prediction model. The measurement data acquisition unit 11 also reads the group of physical properties included in the physical property information 23 that is associated with the group of near-infrared absorption spectra (step S11).

[0071] Next, the prediction model generation unit 18 generates a candidate prediction model and outputs the candidate prediction model to the prediction unit 13 (step S12).

[0072] The prediction unit 13 inputs the group of near-infrared absorption spectra acquired by the measurement data acquisition unit 11 into the candidate prediction model (step S13).

[0073] The prediction unit 13 outputs prediction results that predict the group of physical properties corresponding to the group of near-infrared absorption spectra input to the candidate prediction model (step S14).

[0074] The prediction model generation unit 18 compares the group of physical properties associated with the input near-infrared absorption spectrum group with the prediction results output from the prediction unit 13 to calculate a model evaluation index (step S15).

[0075] The prediction model generation unit 18 determines whether the prediction model candidate satisfies predetermined evaluation criteria based on the model evaluation index (step S16). If the prediction model candidate satisfies predetermined evaluation criteria (YES in step S16), the prediction model generation unit 18 stores the prediction model candidate as the optimal prediction model candidate in the prediction model 22 (step S 17 ).

[0076] On the other hand, if the prediction model candidate does not meet the evaluation criteria (NO in step S16), the prediction model generation unit 18 updates the prediction model candidate (step S12). In step S12, the prediction model generation unit 18 may update the weights, hyperparameters, etc., of the prediction model candidate that did not meet the evaluation criteria, or it may generate a new prediction model candidate.

[0077] Repeat steps S12 to S16 until step S16 is YES.

[0078] Figure 7 illustrates an example in which the prediction model generation unit 18 generates one prediction model candidate and then generates a prediction model 22 from that candidate using machine learning. However, the process is not limited to this. For example, the prediction model generation unit 18 may generate multiple prediction model candidates. In this case, after performing the machine learning shown in Figure 7 on each of the prediction model candidates, the prediction model candidate with the highest prediction accuracy may be stored as the optimal prediction model candidate in the prediction model 22. Alternatively, machine learning may be performed on the prediction model including preprocessing.

[0079] (Measurement method for obtaining near-infrared absorption spectra) The method for measuring near-infrared absorption spectra described herein is a method for measuring the near-infrared absorption spectrum of a resin powder for use in the prediction method of the prediction device 100 described above. The measurement method includes the steps of irradiating the resin powder with near-infrared light and calculating the near-infrared absorption spectrum of the resin powder from measured values ​​of at least one of the reflected light and transmitted light from the resin powder. The resin powder is either a water-absorbing resin powder or an intermediate product produced in the manufacturing process for producing the water-absorbing resin powder. Near-infrared absorption spectra are described below.

[0080] (Near-infrared absorption spectrum) This section describes the near-infrared absorption spectrum used by the prediction device 100 for predicting the physical properties of the resin powder.

[0081] <Measuring equipment> Near-infrared absorption spectra are measured using near-infrared spectroscopy, which involves irradiating a sample with near-infrared light in a specific wavelength range and detecting the transmitted or reflected light. Near-infrared absorption spectra can be measured, for example, using a near-infrared spectrophotometer. While not particularly limited, examples of near-infrared spectrophotometers that can be used include the FT-NIR NIRFlex (trademark) N-500 series and NIRMaster series (manufactured by BUCHI), the IRMA51 series and IRMD51 series (manufactured by Chino Corporation), the IR Tracer100 NIR system (manufactured by Shimadzu Corporation), Spectrum3 NIR (manufactured by PerkinElmer), and the MATRIX series FT-NIR spectrometer (manufactured by BRUKER). The obtained spectral data can be analyzed using commercially available software. Note that near-infrared spectrophotometers may be referred to by different names depending on the manufacturer, such as near-infrared multi-component analyzer or near-infrared analyzer.

[0082] <Wavelengths of near-infrared light> Near-infrared light is light having wavelengths in the 750-2500 nm wavelength range. The near-infrared spectrum is measured by irradiating with light containing near-infrared light having wavelengths in the aforementioned wavelength range. The wavelength of the irradiated light may be all wavelengths in the near-infrared wavelength range, or it may be one or more selected specific wavelengths. In the near-infrared absorption spectrum measurement of the present disclosure, the above-mentioned irradiated light is irradiated onto the water-absorbing resin to be measured, and the transmitted, absorbed, refracted, reflected, and diffused light is measured, so not only chemical information but also physical information can be collected. In other words, it is affected by the temperature of the sample to be measured, the atmosphere in the measurement optical path (presence or absence of vapor, presence or absence of nitrogen displacement, atmospheric pressure, etc.), surface roughness, sample thickness, sample filling state, and time until measurement. Therefore, from the viewpoint of prediction accuracy, when obtaining a near-infrared absorption spectrum, it is preferable to measure under conditions where the above-mentioned physical conditions are as uniform as possible. If necessary, the above-mentioned physical conditions (e.g., the sample temperature) may be measured separately, and the corresponding near-infrared absorption spectrum may be corrected based on these measured values.

[0083] In this disclosure, the near-infrared absorption spectrum is measured at at least one of the following points in time: before the polymerization step, between the polymerization step and the drying step, or after the drying step, and the prediction information output in the prediction step may be used to control one or more manufacturing apparatus used in the resin powder manufacturing process.

[0084] The advantages of using near-infrared absorption spectroscopy in this disclosure include (1) rapid acquisition of analytical results, (2) non-contact and non-destructive analysis, (3) simultaneous quantitative analysis of multiple components, (4) measurement of physical quantities (such as particle size), and (5) ease of operation.

[0085] (Physical properties of water-absorbent polymer powder) The prediction method of this disclosure outputs predictive information relating to the physical properties of at least one of the water-absorbent resin powder and the intermediate products generated in the manufacturing process for producing the water-absorbent resin powder.

[0086] The physical properties that the prediction device 100 can predict may include at least one of the following (1) to (16).

[0087] (1) Gel D50 (2) CRC (3) AAP (4) SFC (5) T20, U20, K20 (6) Vortex (7) D50 (8) Water content of the water-containing gel (9) Solid content (10) Residual Monomers (11) FSR (12) FSC (13) Flow Rate (14) Density (15) Ext (16) Gel Ext Preferably, the physical properties that the prediction device 100 can predict are at least one of the following: (1) gel D50, (2) CRC, (3) AAP, (4) SFC, (6) Vortex, (7) D50, (8) water content of the water-containing gel, and (9) solid content.

[0088] <Water absorbent resin> In this disclosure, "water-absorbent polymer" means a water-swellable, water-insoluble crosslinked polymer, which is generally in particulate form. Furthermore, "water-swellable" means that the unpressurized absorption ratio (CRC) as defined in NWSP 241.0.R2(15) is 5 g / g or more, and "water-insoluble" means that the soluble content (Ext) as defined in NWSP 270.0.R2(15) is 50% by mass or less.

[0089] The water-absorbing resin can be designed appropriately according to its application and is not particularly limited, but it is preferably a hydrophilic crosslinked polymer obtained by crosslinking an unsaturated monomer having a carboxyl group. Furthermore, it is not limited to a form in which the entire amount (100% by weight) is polymer, and may also be a surface-crosslinked material or a composition containing additives, etc., within the range that maintains the above performance.

[0090] For example, "absorbent polymer" is "poly(meth)acrylic acid (salt)," and may contain (meth)acrylic acid and / or its salt as repeating units as the main component.

[0091] "NWSP" stands for "Non-Woven Standard Procedures-Edition 2015," which was jointly issued by EDANA (European Disposales And Nonwovens Association) and INDA (Association of the Nonwoven Fabrics Industry) to unify evaluation methods for nonwoven fabrics and their products in the United States and Europe, and represents a standard measurement method for superabsorbent polymers. Unless otherwise specified, this disclosure measures the physical properties of superabsorbent polymers in accordance with "NWSP."

[0092] <Gel D50> Gel D50 is the mass-average particle size converted to the solid content of the intermediate product, the hydrated gel. Gel D50 is measured in accordance with the method described in WO2016 / 204302. The Gel D50 in this disclosure is the value corresponding to SolidD50 described in WO2016 / 204302.

[0093] In this disclosure, gel D50 can be measured from after the polymerization step until before the drying step. When a water-absorbent resin is produced by aqueous solution polymerization, it is measured after the gel grinding step described later, or before the drying step.

[0094] <crc>(NWSP241.0.R2(15)) "CRC" is an abbreviation for Centrifuge Retention Capacity, which refers to the water absorption ratio of a water-absorbent polymer under no pressure (sometimes called "water absorption ratio").

[0095] Specifically, this refers to the water absorption ratio (unit: g / g) after 0.2g of superabsorbent resin is placed in a nonwoven fabric bag, immersed in a large excess of 0.9 wt% sodium chloride aqueous solution for 30 minutes to allow the superabsorbent resin to swell freely, and then the water is drained using a centrifuge (250G).

[0096] <aap>(NWSP242.0.R2(15)) "AAP" is an abbreviation for Absorption Against Pressure, and refers to the water absorption ratio of superabsorbent polymers under pressure.

[0097] Specifically, AAP refers to the water absorption ratio (unit: g / g) after swelling 0.9g of superabsorbent polymer in a large excess of 0.9 wt% sodium chloride aqueous solution under a load of 2.06 kPa (21 g / cm², 0.3 psi) for 1 hour. 2 In some cases, the measurement may be taken using a different pressure setting (0.7 psi).

[0098] <sfc> "SFC" is an abbreviation for Saline Flow Conductivity, which refers to the permeability of a 0.69 wt% sodium chloride aqueous solution through an absorbent polymer under a load of 2.07 kPa (unit: ×10 -7 ·cm 3 ·s·g -1 ) refers to the SFC. "SFC" is measured in accordance with the SFC test method disclosed in U.S. Patent No. 5,669,894.

[0099] <t20> "T20" refers to the water absorption time, which is the time (in seconds) required for 1 g of resin powder to absorb 20 g of a 0.9 wt% sodium chloride aqueous solution, and is measured in accordance with the measurement method disclosed in U.S. Patent Publication US2012 / 0318046.

[0100] <u20> "U20" represents the absorption over 20 minutes (in units of g / g) and is measured in accordance with the measurement method disclosed in U.S. Patent Publication US2012 / 0318046.

[0101] <k20> "K20" represents the effective transmittance (unit: m) over 20 minutes. 2 ) and is measured in accordance with the measurement method disclosed in U.S. Patent Publication US2012 / 0318046.

[0102] <vortex> The vortex (water absorption time) is measured according to the following procedure. First, 0.02 parts by mass of food additive Brilliant Blue No. 1 is added to 1000 parts by mass of pre-prepared physiological saline (0.9% by mass sodium chloride aqueous solution), and then the liquid temperature is adjusted to 30°C.

[0103] Next, measure 50 ml of the above physiological saline solution into a 100 ml beaker, and add 2.0 g of superabsorbent resin while stirring at 600 rpm using a stirrer tip with a length of 40 mm and a diameter of 8 mm. Starting from the moment the superabsorbent resin is added, measure the time it takes for the superabsorbent resin to absorb the physiological saline solution and cover the stirrer tip as Vortex (absorption time) (unit: seconds).

[0104] <d50> In this disclosure, "D50" is the mass-average particle diameter of the resin powder produced in the drying process described later. The mass-average particle diameter (D50) is measured in the same manner as described in "(3) Mass-Average Particle Diameter (D50) and Logarithmic Standard Deviation of Particle Diameter Distribution" in U.S. Patent No. 7,638,570.

[0105] <Water content and solids content of water-containing gel> (NWSP230.0.R2) The water content and solids content of a hydrated gel refer to the water content and resin solids content of the hydrated gel before drying. The water content and solids content of a hydrated gel can be measured from after the polymerization process until before the drying process. That is, it may be the water content and solids content of the hydrated gel before pulverization, or it may be the water content and solids content of the pulverized particulate hydrated gel.

[0106] The water content of the hydrated gel is measured in accordance with NWSP. For the measurement, the sample mass is changed to 2.0 g, the drying temperature to 180°C, and the drying time to 24 hours. Specifically, 2.0 g of hydrated gel is placed in an aluminum cup with a base diameter of 50 mm, and the total mass W1 (g) of the sample (hydrated gel and aluminum cup) is accurately weighed. Next, the sample is placed in an oven set to an ambient temperature of 180°C. After 24 hours, the sample is removed from the oven, and the total mass W2 (g) is accurately weighed. When the mass of the hydrated gel used in this measurement is M (g), the water content (100-α) (mass%) of the hydrated gel is calculated according to the following formula (Equation 1). α is the solid content (mass%) of the hydrated gel.

[0107] (100-α)(mass%)={(W1-W2) / M}×100...(Equation 1).

[0108] <Gel CRC> "Gel CRC" refers to the CRC of the hydrated gel before drying. Gel CRC can be measured from after the polymerization process until before the drying process. That is, it may be the CRC of the hydrated gel before pulverization, or it may be the CRC of the particulate hydrated gel after pulverization.

[0109] Specifically, "Gel CRC" refers to the water absorption ratio (unit: g / g) obtained after 0.6g of water-containing gel is placed in a nonwoven fabric bag, immersed in a large excess of 0.9 wt% sodium chloride aqueous solution for 24 hours to allow the water-absorbing resin to swell freely, and then the water is drained using a centrifuge (250G).

[0110] <ext>(NWSP270.0.R2(15)) "Ext" is an abbreviation for Extractables, and refers to the water-soluble content (amount of water-soluble components). Specifically, it is the amount of dissolved polymer (in wt%) after adding 1.0 g of superabsorbent polymer to 200 mL of 0.9 wt% sodium chloride aqueous solution and stirring for 16 hours. The amount of dissolved polymer is measured using pH titration.

[0111] <Gel Ext> "Gel Ext" refers to the Ext of the hydrated gel before drying. Gel Ext can be measured from after the polymerization process until before the drying process. That is, it may be the Ext of the hydrated gel before pulverization, or it may be the Ext of the particulate hydrated gel after pulverization.

[0112] Specifically, based on the "Ext" measurement method described above, the sample is changed to 2.0g and measured, and the result is calculated as the mass percentage of water-soluble content per solid.

[0113] <Residual Monomers> (NWSP210.0.R2(19)) "Residual Monomers" refers to the amount of monomers remaining in the water-absorbent resin and is measured in accordance with NWSP210.0.R2(19).

[0114] <fsr> "FSR" is the water absorption rate (unit: g / g / s) and is measured in accordance with the measurement method disclosed in International Publication No. 2009 / 016055.

[0115] <fsc>(NWSP240.0.R2(15)) "FSC" is an abbreviation for Free Swell Capacity, and refers to the water absorption ratio of a water-absorbent polymer when suspended under no pressure. FSC is measured in accordance with NWSP240.0.R2(15).

[0116] <Flow Rate> (NWSP251.0.R2(15)) "Flow Rate" refers to the flow velocity of the water-absorbing resin. The flow rate is measured in accordance with NWSP251.0.R2(15).

[0117] <density> "Density" refers to the bulk density of the water-absorbent polymer. Density is measured in accordance with NWSP251.0.R2(15).

[0118] The physical properties described above may be measured at any point in the resin powder manufacturing process.

[0119] (Method for manufacturing resin powder) Below The method for producing resin powder is described below.

[0120] In this disclosure, known methods or combinations thereof can be used as the method for producing the water-absorbent polymer powder. The method for producing the water-absorbent polymer powder may, for example, include a polymerization step and a drying step, and preferably includes a gel grinding step, a post-crosslinking step, and a sizing step. Each step is described below.

[0121] <Polymerization process> The polymerization process, as an example, involves polymerizing a monomer mainly composed of acrylic acid (salt) and an aqueous monomer aqueous solution containing at least one polymerizable internal crosslinking agent to obtain a hydrated gel-like crosslinked polymer (hereinafter referred to as "hydrated gel").

[0122] [Polymerization initiator] The polymerization initiator used in this disclosure is not particularly limited, as it is appropriately selected depending on the polymerization form, etc., but examples include pyrolysis-type polymerization initiators, photodegradation-type polymerization initiators, or redox-type polymerization initiators used in combination with reducing agents that promote the decomposition of these polymerization initiators. Specifically, one or more of the polymerization initiators disclosed in U.S. Patent No. 7,265,190 are used. Furthermore, from the viewpoint of the handling of the polymerization initiator and the physical properties of the particulate water absorbent or water absorbent resin, peroxides or azo compounds are preferably used, more preferably peroxides, and even more preferably persulfates.

[0123] Alternatively, the polymerization reaction may be carried out by irradiating with active energy rays such as radiation, electron beams, or ultraviolet rays instead of the polymerization initiator, or these active energy rays may be used in combination with the polymerization initiator.

[0124] [Polymerization form] The polymerization methods applicable to this disclosure are not particularly limited, but from the viewpoint of the water absorption characteristics of the water-containing gel and the ease of polymerization control, spray droplet polymerization, aqueous solution polymerization, inverted-phase suspension polymerization, more preferably aqueous solution polymerization, inverted-phase suspension polymerization, and even more preferably aqueous solution polymerization are preferred. Among these, continuous aqueous solution polymerization is particularly preferred, and either continuous belt polymerization or continuous kneader polymerization can be applied.

[0125] <Gel grinding process> The gel crushing step is a process of crushing the water-containing gel obtained in the polymerization step to obtain particulate water-containing gel. When producing superabsorbent resins by spray droplet polymerization or reverse-phase suspension polymerization, particulate water-containing gel can be obtained, so the gel crushing step does not need to be performed. Alternatively, the gel crushing step may be performed simultaneously with the polymerization step, as in continuous kneader polymerization. In particular, from the viewpoint of obtaining SAP with a high water absorption rate, it is preferable to granulate the water-containing gel in this gel crushing step to produce particulate water-containing gel in which the gel D50 is within the desired range.

[0126] [Gel grinder] The gel grinding equipment used in this process during or after polymerization is not particularly limited and includes gel grinders equipped with multiple rotating stirring blades, such as batch-type or continuous-type double-arm kneaders, single-screw extruders, twin-screw extruders, meat choppers, screw-type extruders, and double-screw kneaders equipped with crushing means.

[0127] Among these, a screw-type extruder in which a perforated plate is installed at one end of the casing is preferred. Specifically, examples include the screw-type extruders disclosed in Japanese Patent Publication No. 2000-63527 and WO2011 / 126079.

[0128] [Gel grinding area] In this disclosure, the gel grinding is performed during and / or after the polymerization process, and more preferably on the hydrated gel polymer after the polymerization process. In the case of a method in which gel grinding is performed during polymerization, such as kneader polymerization, the monomer aqueous solution continuously changes into a hydrated gel polymer as the polymerization time progresses. Therefore, it is sufficient to gel grind the hydrated gel polymer after the point in time when the polymerization temperature is at its maximum, or the hydrated gel polymer with a monomer polymerization rate of 90 mol% or more. Here, the maximum polymerization temperature is also called the polymerization peak temperature. The monomer polymerization rate is sometimes called the conversion rate. The monomer polymerization rate is calculated from the amount of polymer calculated from the pH titration of the hydrated gel polymer and the amount of residual monomer.

[0129] Furthermore, if the polymerization process is belt polymerization, the hydrated gel polymer during and / or after the polymerization process, preferably the hydrated gel polymer after the polymerization process, can be cut or roughly crushed to a size of several tens of centimeters before gel pulverization. This operation makes it easier to fill the gel pulverizer with the hydrated gel polymer, allowing the gel pulverization process to be carried out more smoothly. The means for cutting or roughly crushing the hydrated gel polymer should preferably be such that it does not knead the polymer, for example, a guillotine cutter. The size and shape of the hydrated gel polymer obtained by cutting or roughly crushing are not particularly limited, as long as they can be filled into the gel pulverizer.

[0130] <Drying process> The drying process involves drying the particulate water-containing gel to a desired solid content to obtain granular dried material. While the drying method is not particularly limited, examples include heating drying, hot air drying, reduced pressure drying, fluidized bed drying, infrared drying, microwave drying, drum dryer drying, azeotropic dehydration with hydrophobic organic solvents, and high-humidity drying using high-temperature steam. 、 A post-crosslinking agent, as described later, may be used in this drying process to obtain a water-absorbing resin powder that has been post-crosslinked (also called surface crosslinked) in the drying process.

[0131] [Drying equipment] The drying apparatus used in the drying process is not particularly limited, and one or more types such as heat transfer conduction type dryers, radiant heat transfer type dryers, hot air heat transfer type dryers, and dielectric heating type dryers can be appropriately selected. The drying apparatus may be batch type or continuous type. The drying apparatus may also be direct heating type or indirect heating type. Furthermore, the drying apparatus may be any of the following types: material stationary type, material agitation type, material transfer type, and hot air conveying type. Examples of heat transfer type dryers include vented band type, vented circuit type, vented vertical type, parallel flow band type, vented tunnel type, vented agitation type, vented rotary type, rotary type with heating tube, fluidized bed type, and airflow type.

[0132] [Drying temperature] The drying temperature in the drying process is 80°C or higher, preferably 100°C or higher, more preferably 120°C or higher, and particularly preferably 150°C or higher. Furthermore, the drying temperature is 250°C or lower, preferably 230°C or lower, and more preferably 220°C or lower. Any combination of the upper and lower limits of the drying temperature is acceptable. A drying temperature below 80°C is undesirable because it prolongs the drying time required to obtain a suitable resin solid content (moisture content). Also, undried material may be generated, potentially causing clogging during the subsequent grinding process. A drying temperature exceeding 250°C is undesirable due to safety concerns and the generation of discolored foreign matter. Note that the drying temperature refers to the temperature of the heat transfer medium used for drying in the case of direct heating, the temperature of the hot air used for drying in the case of hot air drying, and the temperature of the heat transfer surface used for drying in the case of indirect heating.

[0133] [Drying time] The drying time in the drying process refers to the time until the solid content reaches 80% by weight or more, and is preferably 60 minutes or less, followed by 40 minutes or less, 30 minutes or less, and 25 minutes or less in that order. The lower limit of the drying time is about 1 minute, taking drying efficiency into consideration. Furthermore, the total drying time is preferably 120 minutes or less, followed by 100 minutes or less, 80 minutes or less, and 60 minutes or less in that order. If the drying time is too short, undried material will be generated, which may cause clogging during the subsequent grinding process.

[0134] [Resin solids] The particulate water-containing gel obtained in the gel pulverization step described above is dried in the drying step described above to obtain a dried polymer. The resin solids content, determined from the loss on drying of the dried polymer (measured by heating 1 g of powder or particles at 180°C for 3 hours), is preferably 80% by weight or more, more preferably 85-99% by weight, and even more preferably 86-98% by weight.

[0135] <Post-crosslinking process> This process involves the polymerization-containing gel, and / or This process involves adding a post-crosslinking agent to the dried material, which reacts with the functional groups (especially carboxyl groups) of the water-absorbent resin, to cause a crosslinking reaction. Since crosslinking mainly occurs from the surface of the water-absorbent resin particles, this is also called surface crosslinking or secondary crosslinking. As an example, in this process, a post-crosslinking agent is added to granular water-containing gel and / or granular dried material and reacted. This process comprises a post-crosslinking agent addition step and a heat treatment step, and may optionally include a cooling step after the heat treatment step.

[0136] <Sizing process> This process involves adjusting the particle size of granular dried material or post-crosslinked granular dried material. This sizing process yields a water-absorbing resin powder in which the particle size or particle size distribution is more actively controlled.

[0137] Preferably, the sizing step includes a crushing step and / or a classification step. The crushing step is a step in which loosely aggregated granular dried material, obtained through a drying step or heat treatment step, is broken up with a crusher to adjust the particle size. The classification step is a step in which coarse particles and fine powder are removed from the granular dried material, post-crosslinked granular dried material, or their crushed material using a classifier. Ideally, the sizing step should be such that a water-absorbing resin powder with controlled particle size and particle size distribution is obtained by the crushing step alone. Since the water absorption performance, handling, and feel when applied to sanitary materials such as diapers and sanitary products vary depending on the particle size and particle size distribution of the water-absorbing resin powder, it is preferable to obtain a water-absorbing resin powder with a desired particle size and particle size distribution through the above-mentioned sizing step.

[0138] <Other processes> In addition to the steps described above, the method for producing water-absorbent polymer powder may also include a cooling step, a monomer aqueous solution preparation step, a step of adding various additives, a fine powder removal step, and a fine powder recycling step. Furthermore, it may include other known steps.

[0139] [Embodiment 2] Other embodiments of this disclosure are described below. For the sake of clarity, components having the same function as those described in the above embodiments are denoted by the same reference numerals, and their descriptions are omitted.

[0140] (Configuration of prediction system 1000a) For example, the prediction device 100 may measure the near-infrared absorption spectrum of the intermediate product at at least one of the following stages in the manufacturing process of the water-absorbent resin powder: before the polymerization step, between the polymerization step and the drying step, and after the drying step, and output prediction information related to the physical properties of the intermediate product (or the finished resin powder) at any of the above steps.

[0141] Furthermore, in the method for producing resin powder, the manufacturing conditions in one or more manufacturing steps of the water-absorbing resin powder may be controlled based on the prediction information output by the prediction device 100.

[0142] Furthermore, prediction information output by the prediction device 100 may be used to control the method for manufacturing the resin powder.

[0143] Furthermore, any manufacturing apparatus (corresponding to external apparatus 4 in Figure 1) that performs any processing step included in the resin powder manufacturing process may be controlled based on the prediction information output from the prediction apparatus 100. A prediction system 1000a having such a configuration will be described in Figure 8. Figure 8 is a block diagram showing an example of the configuration of a prediction system 1000a according to another embodiment of the present disclosure.

[0144] In Figure 8, the prediction system 1000a comprises a prediction device 100, near-infrared spectrophotometers 3a-3f, and external devices 4a-4e. The prediction device 100 is connected to the near-infrared spectrophotometers 3a-3f and the external devices 4a-4e. The external devices 4a-4e are, for example, control devices for performing each process (polymerization process, grinding process, etc.).

[0145] Here, as an example, we will explain the case where the gel D50 of a water-absorbent resin powder is predicted. Gel D50 is a physical property of the water-absorbent resin powder measured, for example, after the gel grinding process. In Figure 8, we assume that the external device 4b is a device that controls the gel grinding process, and the near-infrared spectrophotometer 3c is a near-infrared spectrophotometer that measures the near-infrared absorption spectrum after the gel grinding process.

[0146] First, the near-infrared spectrophotometer 3c performed the measurement. Particulate water-containing gel The near-infrared absorption spectrum is output to the prediction device 100. The prediction device 100 preprocesses the acquired near-infrared absorption spectrum based on the prediction model. Examples of preprocessing include outlier removal and averaging. Based on the prediction model, the prediction device 100 predicts gel D50 from the preprocessed near-infrared absorption spectrum. The prediction device 100 outputs the prediction result to an external device 4c, for example. The external device 4c may be, for example, a device that controls the drying process.

[0147] For example, if the prediction device 100 predicts that the size of gel D50 is larger than a predetermined value, the external device 4c that controls the drying process may change the conditions in the process, such as controlling the heating of the resin powder at a higher temperature than predetermined.

[0148] Furthermore, if the prediction device 100 predicts that the size of gel D50 is larger than a predetermined value, the external device 4b that controls the gel grinding process may increase the gel grinding load beyond the predetermined value. Specifically, the external device 4b may change the conditions in the process, such as controlling the rotation speed to increase the speed so that a higher shear force is applied to the gel.

[0149] Other examples include, for instance, if the prediction device 100 predicts that the CRC of the final product will be higher than the product standard, then controlling the amount of crosslinking agent in the polymerization process to be reduced. Also, for example, if the AAP of the final product is lower than the product standard, then controlling the composition of the treatment agent in the post-crosslinking process to be changed.

[0150] In this embodiment, the prediction device 100 may be able to identify which of the multiple near-infrared spectrophotometers the near-infrared measurement data was acquired from. For example, the prediction device 100 may pre-acquire the MAC address of each near-infrared spectrophotometer in the prediction system 1000a and the installation location of each near-infrared spectrophotometer. As an example, Figure 9 shows a correspondence table between MAC addresses and near-infrared spectrophotometers. In this way, by acquiring the MAC address of the near-infrared spectrophotometer along with the near-infrared measurement data, the prediction device 100 can identify which process in the prediction system 1000a the near-infrared measurement data is from.

[0151] A prediction system 1000a with this configuration can accurately predict in a short time the physical properties of intermediate products in each process of resin powder manufacturing (e.g., pulverized gel properties such as gel particle size) or the physical properties of the final resin powder product (e.g., the aforementioned CRC and AAP). By using the prediction information to control the manufacturing equipment (external devices 4a to 4e) that perform each process in the resin powder manufacturing process, the physical property manipulating factors in each manufacturing process can be adjusted in real time, effectively suppressing the occurrence of products that do not meet specifications.

[0152] Furthermore, near-infrared spectrophotometers 3a-3f are generally inexpensive (at least cheaper than Raman spectrometers). Therefore, it is possible to keep the cost of deploying multiple near-infrared spectrophotometers 3a-3f in the resin powder manufacturing process low.

[0153] [Examples of implementation using software] Control block of prediction device 100 described in Embodiment 1 and Embodiment 2 (especially control unit 10 )teeth This may be implemented by logic circuits (hardware) formed on an integrated circuit (IC chip), or by software.

[0154] In the latter case, the prediction device 100 includes a computer that executes instructions for a program, which is software that realizes each function. This computer includes, for example, one or more processors and a computer-readable recording medium that stores the program. The object of this disclosure is achieved when the processor reads the program from the recording medium and executes it in the computer. For example, a CPU (Central Processing Unit) can be used as the processor. As the recording medium, a "tangible medium that is not temporary," such as ROM (Read Only Memory), can be used, as well as tape, disk, card, semiconductor memory, programmable logic circuit, etc. It may also further include RAM (Random Access Memory) for deploying the program. Furthermore, the program may be supplied to the computer via any transmission medium capable of transmitting the program (such as a communication network or broadcast wave). One aspect of this disclosure can also be realized in the form of a data signal embedded in a carrier wave, in which the program is embodied by electronic transmission.

[0155] This disclosure is not limited to the embodiments described above, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of this disclosure. [Examples]

[0156] An embodiment of this disclosure is described below. In order to adjust the physical properties of the sample to be used for near-infrared absorption spectrum measurement, the conditions of the polymerization process, gel grinding process, drying process, post-crosslinking process, sizing process, and other processes described above were appropriately changed to obtain the sample to be measured. In the polymerization process described later, for example, the amount of polyethylene glycol diacrylate, which is the internal crosslinking agent, was changed. In the gel grinding process described later, for example, the pore size of the perforated plate was changed. In the drying process described later, for example, the drying time was changed. Furthermore, in the post-crosslinking process described later, for example, the type and amount of post-crosslinking agent used, as well as the temperature and time during heat treatment were changed.

[0157] <Preparation of superabsorbent polymer (SAP)> [Polymerization process] Solution (A) was prepared by mixing 23.2 g of acrylic acid, 0.135 g (0.080 mol%) of polyethylene glycol diacrylate (weight-average molecular weight (Mw) 523Da), 0.071 g of 2.0 wt% diethylenetriaminepentaacetic acid trisodium aqueous solution, 22.2 g of ion-exchanged water, and 9.6 g of 48.5 wt% sodium hydroxide aqueous solution in a polypropylene container with an inner diameter of 50 mm and a capacity of 120 mL.

[0158] While stirring with a magnetic stirrer, 9.8 g of a 48.5 wt% sodium hydroxide aqueous solution was added to the above solution (A), which had been heated to 45°C, in an open system over approximately 5 seconds and mixed to prepare monomer aqueous solution (1). During the mixing process, the heat of neutralization and heat of dissolution caused the temperature of monomer aqueous solution (1) to rise to approximately 80°C.

[0159] Subsequently, when the temperature of the obtained monomer aqueous solution (1) reached 78°C, 1.01 g of a 4.5 wt% sodium persulfate aqueous solution was added and stirred for approximately 3 seconds. Then, the obtained reaction solution (1) was poured into a stainless steel petri dish in an open system.

[0160] The stainless steel petri dish described above had an inner diameter of 88 mm and a height of 20 mm. The surface temperature of the stainless steel petri dish was preheated to 50°C using a hot plate (NEO HOTPLATE H1-1000, manufactured by Inouchi Seieido Co., Ltd.).

[0161] Immediately after supplying the reaction solution (1) described above, the stainless steel petri dish was covered with a glass container having an exhaust port, and a vacuum pump was used to draw air out the contents so that the pressure inside the case was 85 kPa. The pressure outside the case was 101.3 kPa (atmospheric pressure).

[0162] After the reaction solution (1) was poured into the stainless steel petri dish, polymerization began after a short time. The polymerization proceeded with expansion and foaming upwards in all directions, generating water vapor, and then contracted to a size slightly larger than the bottom surface. This expansion and contraction was completed within approximately 1 minute. After being held in the polymerization container (i.e., a stainless steel petri dish covered with a glass container) for 3 minutes, the hydrated gel-like crosslinked polymer (hereinafter referred to as "hydrated gel") (1) was removed.

[0163] [Gel grinding process] The obtained hydrated gel (1) was pulverized using a screw extruder (meat chopper) having the following specifications. The screw extruder was equipped with a perforated plate at its tip, with a diameter of 82 mm, a pore size of 8.0 mm, 33 pores, and a thickness of 9.5 mm. As for the gel pulverization conditions, the amount of hydrated gel (1) added was approximately 360 g / min, and the gel pulverization was carried out while adding 90°C deionized water at a rate of 50 g / min in parallel with the gel addition. This pulverized particulate hydrated gel (1) was used to evaluate gel D50, which will be described later.

[0164] [Drying process] The pulverized particulate water-containing gel (1) was spread on a stainless steel wire mesh with a mesh size of 850 μm and dried with hot air at 190°C for 30 minutes. Subsequently, the dried polymer (1) obtained in this drying operation was pulverized using a roll mill (WML type roll mill manufactured by Inoguchi Giken Co., Ltd.), and then classified using JIS standard sieves with mesh sizes of 710 μm and 175 μm to obtain water-absorbing resin powder (1).

[0165] [Post-crosslinking process] 100 g of the above-mentioned superabsorbent polymer powder (1) was sprayed with a surface crosslinking agent solution consisting of 0.025 g of ethylene glycol diglycidyl ether, 0.3 g of ethylene carbonate, 0.5 g of propylene glycol, and 2.0 g of deionized water, and mixed. This mixture was heat-treated at 200°C for 35 minutes to obtain surface-crosslinked superabsorbent polymer powder (2).

[0166] Through the above series of operations, irregularly shaped crushed superabsorbent polymer powders (1) and (2) were obtained. These superabsorbent polymer powders were used for the evaluation of CRC, AAP, SFC, D50, and water content (solids) described later.

[0167] <Measurement of water-absorbing resins> The equipment and conditions used for measuring the near-infrared absorption spectrum are as follows:

[0168] (i) Equipment: FT-NIR NIRFlex (trademark registered) N-500 (manufactured by BUCHI) Measurement wavelength: 800~2500nm Measurement method: Diffuse reflectance measurement (ii) Equipment: IRMA5184S (manufactured by Chino Corporation) Measurement wavelength (8 wavelengths): 1320, 1460, 1600, 1720, 1800, 1960, 2100, 2310nm Measurement method: Near-infrared absorption type.

[0169] <Performance evaluation of prediction devices for various physical properties of water-absorbent resin powders> The wavelength data of the obtained near-infrared absorption spectra were used as features, and the physical property information of the measured samples was used as the objective variable. The relationship between these was determined by PCR or PLS (Partial least square) regression analysis. The performance of the prediction device was evaluated below for each of the physical properties of the superabsorbent polymer powder: (1) Gel D50, (2) CRC, (3) AAP, (4) SFC, (5) D50, and (6) Solids content.

[0170] The dataset used for evaluation includes multiple combinations of near-infrared measurement data and the physical properties associated with that near-infrared measurement data. The dataset is divided into training data and validation data for evaluation. Here, the training data includes near-infrared absorption spectra and measured values ​​of the physical properties associated with those near-infrared absorption spectra, and is used for prior machine learning. The validation data is data not included in the training data. In this example, the dataset used for evaluation was randomly divided, and a predictive model was created on the training data using PLS or PCR.

[0171] First, the near-infrared absorption spectra and physical properties of several water-absorbing resin powders (1) and (2) were measured, and a dataset was prepared containing N combinations of near-infrared absorption spectra and the physical properties associated with those near-infrared absorption spectra. The dataset was then divided into training data and validation data, with 80% of the combinations used for training and the remaining 20% ​​used for validation.

[0172] The following graphs show the plots for each physical property. In each graph, the data labeled "training" is the training data, and the data labeled "test" is the validation data. The dotted line in the graph represents the true regression line obtained when the measured value of the physical property and the predicted value of the physical property perfectly match. The predicted values ​​of the physical properties are plotted against the measured values ​​of the physical properties for both the training and validation data. In each graph, the closer the points plotting the measured and predicted values ​​are to the regression line, the higher the performance of the prediction device can be judged.

[0173] (1) Gel D50 The dataset has 36 combinations of near-infrared measurement absorption spectra and physical properties associated with the near-infrared absorption spectra. The dataset was randomly divided into training data and validation data, and 80% was used for training and 20% for validation. A prediction model by PCR was created for the training data. Figure 10 is a graph plotting the predicted values against the measured values of gel D50 in the range of 80 to 190 μm.

[0174] (2) CRC The dataset has 79 combinations of near-infrared measurement absorption spectra and physical properties associated with the near-infrared absorption spectra. The dataset was randomly divided into training data and validation data, and 80% was used for training and 20% for validation. A prediction model by PLS was created for the training data. Figure 11 is a graph plotting the predicted values against the measured values of CRC in the range of 24 to 31 g / g.

[0175] (3) AAP The dataset has 69 combinations of near-infrared measurement absorption spectra and physical properties associated with the near-infrared absorption spectra. The dataset was randomly divided into training data and validation data, and 80% was used for training and 20% for validation. A prediction model by PLS was created for the training data. Figure 12 is a graph plotting the predicted values against the measured values of AAP in the range of 24.5 to 27 g / g.

[0176] (4) SFC The dataset has 64 combinations of near-infrared measurement absorption spectra and physical properties associated with the near-infrared absorption spectra. The dataset was randomly divided into training data and validation data, and 80% was used for training and 20% for validation. A prediction model by PLS was created for the training data. Figure 13 is a graph plotting the predicted values against the measured values of SFC in the range of 20 to 110 (×10 -7 ·cm 3 ·s·g -1 )

[0177] (5) D50 The dataset contains 90 combinations of near-infrared absorption spectra and the corresponding physical properties. The dataset was randomly split into training and validation data, with 80% used for training and 20% for validation. A PLS predictive model was created for the training data. Figure 14 is a graph plotting the predicted values ​​of D50 against the measured values ​​in the range of 250 to 450 μm.

[0178] (6) Moisture content (solid content) The dataset contains 29 combinations of near-infrared absorption spectra and the corresponding physical properties. The dataset was randomly split into training and validation data, with 80% used for training and 20% for validation. A PLS predictive model was created on the training data. Figure 15 is a graph plotting the predicted values ​​against the measured water content in the range of 96.5 to 98.5 wt%. Since the solid content is calculated as 100 - water content (weight %), this graph can also be said to show the predicted values ​​against the measured values ​​in solid content.

[0179] <Evaluation Results> For all physical properties, the predicted values ​​showed good correlation with the measured values. Furthermore, even for validation data not included in the training data, each physical property could be predicted with the same accuracy as the training data, demonstrating that the prediction device 100 has good performance. [Explanation of symbols]

[0180] 100 Prediction Devices 11 Measurement data acquisition unit 13 Prediction Section 22 Predictive Models 23. Physical Properties Information< / density> < / fsc> < / fsr> < / ext> < / vortex> < / sfc> < / aap> < / crc>

Claims

1. A method for predicting the physical properties of resin powder, The resin powder is either a water-absorbent resin powder or an intermediate product generated in the manufacturing process for producing the water-absorbent resin powder. A near-infrared measurement data acquisition step is to acquire near-infrared measurement data showing the near-infrared absorption spectrum of the resin powder, The prediction step includes inputting at least one of the near-infrared measurement data and one or more processing data generated based on the near-infrared measurement data into a prediction model to output prediction information related to the physical properties of the resin powder, The aforementioned prediction information is (1) The mass-average particle size (gel D50) of the water-containing gel which is the intermediate product, (2) The absorption ratio (CRC) of the resin powder under no pressure, (3) The absorption ratio (AAP) of the resin powder under pressure, (4) The saline flow induction property (SFC) of the resin powder, (5) Water absorption time (Vortex) of the resin powder, (6) The mass-average particle size (D50) of the resin powder, (7) The amount of monomers remaining in the resin powder, (8) Water absorption rate (FSR) of the resin powder, (9) Water absorption ratio (FSC) of the resin powder when suspended under no pressure, (10) Flow rate of the resin powder, (11) The bulk density of the resin powder, (12) Amount of water-soluble component of the resin powder (Ext), (13) The absorption ratio of the water-containing gel under no pressure (gel CRC), and (14) Amount of water-soluble components in the water-containing gel of the resin powder before drying (gel Ext), Includes at least one of the following: Prediction method.

2. The aforementioned predictive model is generated by machine learning using at least one of the following as training data: (1) a combination of near-infrared measurement data including near-infrared absorption spectra of multiple previously manufactured resin powders with known physical properties and physical property information of the final product associated with said near-infrared measurement data; and (2) a combination of near-infrared measurement data including near-infrared absorption spectra of multiple previously produced intermediate products with known physical properties generated in the manufacturing process for each manufactured resin powder and physical property information of the intermediate products associated with said near-infrared measurement data. The prediction method according to claim 1.

3. The aforementioned prediction model is generated using either linear regression or nonlinear regression. The prediction method according to claim 2.

4. The aforementioned prediction model was generated using either principal component regression or partial least squares regression. The prediction method according to claim 2 or 3.

5. This includes a preprocessing step for generating the aforementioned processing data, In the aforementioned preprocessing step, one or more of the following are performed: outlier removal, averaging, wavelength range selection, and differentiation. The prediction method according to any one of claims 1 to 4.

6. The manufacturing process for the resin powder includes a polymerization step and a drying step. The near-infrared absorption spectrum is measured before the polymerization step, between the polymerization step and the drying step, and after the drying step. Based on the prediction information output in the prediction step, one or more manufacturing devices used in the resin powder manufacturing process are controlled. The prediction method according to any one of claims 1 to 5.

7. A predictive device for predicting the physical properties of resin powder, The resin powder is either a water-absorbent resin powder or an intermediate product generated in the manufacturing process for producing the water-absorbent resin powder. The system comprises: a measurement data acquisition unit that acquires near-infrared measurement data showing the near-infrared absorption spectrum measured for the resin powder; and a prediction unit that inputs at least one of the near-infrared measurement data and one or more processing data generated based on the near-infrared measurement data into a prediction model and outputs prediction information related to the physical properties of the resin powder. The aforementioned prediction information is (1) The mass-average particle size (gel D50) of the water-containing gel which is the intermediate product, (2) The absorption ratio (CRC) of the resin powder under no pressure, (3) The absorption ratio (AAP) of the resin powder under pressure, (4) The saline flow induction property (SFC) of the resin powder, (5) Water absorption time (Vortex) of the resin powder, (6) The mass-average particle size (D50) of the resin powder, (7) The amount of monomers remaining in the resin powder, (8) Water absorption rate (FSR) of the resin powder, (9) Water absorption ratio (FSC) of the resin powder when suspended under no pressure, (10) Flow rate of the resin powder, (11) The bulk density of the resin powder, (12) Amount of water-soluble component of the resin powder (Ext), (13) The absorption ratio of the water-containing gel under no pressure (gel CRC), and (14) Amount of water-soluble components in the water-containing gel of the resin powder before drying (gel Ext), Includes at least one of the following: Prediction device.

8. A method for producing resin powder, comprising a polymerization step and a drying step, A method for producing resin powder, wherein the manufacturing conditions of one or more of the manufacturing steps of the resin powder are controlled based on prediction information obtained by the prediction method described in any one of claims 1 to 6.

9. A measurement method for measuring the near-infrared absorption spectrum of a resin powder used in the prediction method described in any one of claims 1 to 5, The steps include irradiating the resin powder with near-infrared light, The step includes calculating the near-infrared absorption spectrum of the resin powder from a measurement value obtained by measuring at least one of the reflected light and transmitted light from the resin powder, The resin powder is either a water-absorbing resin powder or an intermediate product generated in the manufacturing process for producing the water-absorbing resin powder. Measurement method.

Images