JP7362337B2 - Information processing device, control method for information processing device, and program - Google Patents

Description

The present invention relates to an information processing device, a method of controlling the information processing device, and a program.

Various additives are added to resins to prepare the resins with desired mechanical, physical, and chemical properties. Therefore, knowing the concentration of additives added to resin is important in quality control and material design.

Methods for analyzing additives include separating the additives added to the resin and analyzing them using analyzers such as ultraviolet detectors, infrared detectors, and mass spectrometers, and directly analyzing the resin itself with the additives attached. Examples include methods of analysis.

Methods for separating additives added to the resin include methods of extracting the additives using a solvent and separating the additives using liquid chromatography, and methods of separating low molecular weight components including the additives using size exclusion chromatography (SEC). Examples include methods of separating. However, since such a method uses organic solvents such as tetrahydrofuran and chloroform, it is unfavorable in terms of environmental burden and safety and health of workers.

Secondary ion mass spectrometry methods such as dynamic secondary ion mass spectrometry (Dynamic SIMS method) and time-of-flight secondary ion mass spectrometry (TOF-SIMS method) are methods for directly analyzing the resin itself with additives attached. One example is the law. The Dynamic SIMS method is a method of continuously irradiating a solid sample with an ion beam to obtain the distribution of impurities in the depth direction. On the other hand, the TOF-SIMS method is a method of obtaining information on elements and molecules present on the surface of a solid sample by irradiating the solid sample with a smaller amount of ion beam than the Dynamic SIMS method (see Patent Documents 1 and 2). . Here, the principle of the TOF-SIMS method will be explained in detail.

When a solid sample is irradiated with an ion beam (primary ions) in a high vacuum, constituent components on the surface of the solid sample are released into the vacuum. Positively or negatively charged ions (secondary ions) generated in this process are focused in one direction by an electric field and detected at a fixed distance away. Secondary ions with various masses are generated depending on the composition of the solid sample surface, but in a constant electric field, ions with lower mass fly faster and ions with higher mass fly slower. Therefore, by measuring the time (flight time) from when the secondary ions are generated until they reach the detector, the mass of the generated secondary ions can be analyzed.

The TOF-SIMS method can detect components on the ppm order, that is, 10 ¹¹ atoms/cm ² , can be applied to both organic and inorganic materials, and can detect components present at a depth of 1 nm or less from the surface of a solid sample. It has the advantage of being measurable. However, since this method directly analyzes the solid sample itself, the resulting spectrum is likely to contain peaks from impurities. In addition, the ease with which constituent components become secondary ions (ionization efficiency) is easily affected by slight differences in the shape and charge of the solid sample surface, and the secondary ion intensity varies depending on the measurement location of the solid sample. Cheap. Furthermore, the secondary ion strength is susceptible to differences in the type and composition of the solid sample. Therefore, spectra obtained by the TOF-SIMS method tend to be complicated, and it takes time to obtain information about a solid sample only from the secondary ion intensity. Therefore, in order to obtain information about a solid sample (resin), a method of analyzing spectra obtained by the TOF-SIMS method using machine learning is being considered (see Non-Patent Document 1).

Japanese Patent Application Publication No. 2011-68005 Japanese Patent Application Publication No. 2007-156093

“TOF-SIMS image data fusion using multivariate analysis and TOF-SIMS spectral analysis using sparse modeling and machine learning”, Journal of Surface Analysis 25(2) 103-114 (2018)

As mentioned above, in order to obtain information about a solid sample (resin), methods of analyzing spectra obtained by TOF-SIMS using machine learning are being considered. However, in order to obtain information on the type of deposits (hereinafter sometimes referred to as "substance to be measured") attached to a solid sample (resin), spectra obtained by the TOF-SIMS method can be easily, quickly, and Until now, no consideration has been given to methods for highly accurate analysis.

Therefore, an object of the present invention is to provide an information processing device that can easily, quickly, and highly accurately analyze the type of a substance to be measured based on spectral information of a sample including a solid sample and a substance to be measured. Another object of the present invention is to provide a method and program for controlling the information processing apparatus.

The information processing device of the present invention provides a type of the substance to be measured in the sample, which is estimated by inputting spectral information of the solid sample and spectral information of the sample including the solid sample and the substance to be measured into a learning model. An information processing device comprising an information acquisition means for acquiring information regarding the spectral information, wherein the spectral information is acquired by secondary ion mass spectrometry.

In addition, the control method for an information processing apparatus of the present invention includes a method for controlling an information processing apparatus, which is estimated by inputting spectral information of a solid sample and spectral information of a sample including the solid sample and a substance to be measured into a learning model. A method for controlling an information processing apparatus, which includes an information acquisition step of acquiring information regarding the type of a substance to be measured, characterized in that the spectrum information is acquired by secondary ion mass spectrometry.

According to the present invention, there is provided an information processing device that can easily, quickly, and highly accurately analyze the type of a substance to be measured from spectral information of a sample including a solid sample and a substance to be measured, a method for controlling the information processing device, and The goal is to provide programs.

1 is a diagram showing the overall configuration of an information processing system including an information processing apparatus according to an embodiment. It is a figure which shows an example of the flowchart of the processing procedure regarding the generation of the learning model based on this embodiment. FIG. 7 is a diagram illustrating an example of a flowchart of a processing procedure for displaying information regarding types according to the present embodiment. An example of spectral information of a solid sample and a sample A including the solid sample and the substance A to be measured is shown. An example of spectrum information of a solid sample and a sample B including the solid sample and the substance B to be measured is shown. An example of spectrum information of a solid sample and a sample C including a solid sample and a substance C to be measured is shown. An example of spectrum information of a solid sample and a sample D including the solid sample and the substance D to be measured is shown. This is an example of a flowchart of a procedure for creating training data to which noise has been added. It is an example of a display of information regarding the type of a substance to be measured in an information processing device.

Hereinafter, modes for carrying out the present invention (embodiments) will be described with reference to the drawings. However, the scope of the present invention is not limited to each embodiment described below.

In the present invention, in order to easily, quickly, and highly accurately analyze the type of substance to be measured in the target sample from the spectrum information of the sample including the solid sample and the substance to be measured, the following information processing device is required. use The information processing device acquires information regarding the type of substance to be measured in the sample, which is estimated by inputting spectral information of the solid sample and spectral information of the sample including the solid sample and the substance to be measured into the learning model. It has information acquisition means. Further, this spectral information is obtained by secondary ion mass spectrometry.

This eliminates the need to prepare spectral information according to the type of substance to be measured in the sample and input it into the learning model, and even if the type of substance to be measured in the target sample is unknown, It becomes possible to obtain the following types of information. It can also be applied to spectrum analysis in secondary ion mass spectrometry, where the spectrum obtained tends to be complicated due to the influence of the surface shape, slight differences in charge, type, and composition of a solid sample.

(sample)
The sample includes a solid sample and a substance to be measured. The solid sample is preferably at least one selected from the group consisting of resin, silicon substrate, and copper foil, and resin is preferable. The resin is preferably composed of units derived from unsaturated carboxylic acid esters, and more preferably composed of units derived from alkyl (meth)acrylates. Here, the unit refers to a repeating unit derived from one monomer.

(Measurement target substance)
The substance to be measured is preferably an additive added to the solid sample or a deposit attached to the solid sample. Measured substances include plasticizers, thickeners, thinners, crystal nucleating agents, crystallization promoters, crystallization retarders, mold release agents, flame retardants, flame retardant aids, thermal decomposition inhibitors, and reinforcing agents. , conductive materials, antistatic agents, thermal conductive agents, antioxidants, weatherproofing agents, ultraviolet absorbers, antibacterial agents, coloring agents, gas barrier agents, antifogging agents, antireflective agents, antifouling agents, impurities, etc. . Among these, it is preferable that the substance to be measured is a plasticizer. Examples of plasticizers include surfactants.

It is preferable that the group of elements constituting the solid sample and the group of elements constituting the substance to be measured are the same. The element group constituting the solid sample and the element group constituting the substance to be measured preferably include at least one selected from the group consisting of carbon atoms, hydrogen atoms, oxygen atoms, and nitrogen atoms.

(Information regarding the type of substance to be measured)
Information regarding the name of the substance to be measured in this embodiment includes the name, abbreviation, chemical classification, structural formula, chemical formula, 3D structure diagram, etc. of the substance to be measured.

(Quantitative information on the substance to be measured)
In the present invention, it is possible to obtain not only information regarding the type of the substance to be measured in the target sample but also quantitative information from the spectrum information of the sample including the solid sample and the substance to be measured. The quantitative information may be at least one type selected from the group consisting of the content of the substance to be measured in the solid sample, the concentration of the substance to be measured in the solid sample, and the presence or absence of the substance to be measured in the solid sample. preferable. In addition, the quantitative information is selected from the group consisting of the ratio of the concentration or amount of the substance to be measured in the solid sample to the reference amount of the substance to be measured, and the ratio of the content or concentration of the substance to be measured in the solid sample. It is preferable that it is at least one type.

(spectrum information)
The spectral information in this embodiment is acquired by secondary ion mass spectrometry. Examples of the secondary ion mass spectrometry include the Dynamic SIMS method and the TOF-SIMS method. Among these, it is preferable to use the TOF-SIMS method as the secondary ion mass spectrometry method.

(information processing system, information processing device)
Next, the information processing system in this embodiment will be explained using FIG. 1. FIG. 1 is a diagram showing the overall configuration of an information processing system including an information processing apparatus according to this embodiment.

The information processing system includes an information processing device 10, a database 22, and an analysis device 23. The information processing device 10 and the database 22 are communicably connected to each other via a communication means. In this embodiment, the communication means is constituted by a LAN (Local Area Network) 21. Further, the information processing device 10 and the analysis device 23 are connected by standard communication means such as USB (Universal Serial Bus). Note that the LAN may be a wired LAN, a wireless LAN, or a WAN (Wide Area Network). Further, the USB may be a LAN.

The database 22 manages spectrum information obtained through analysis by the analyzer 23. Further, the database 22 manages learning models (trained models) generated by a learning model generation unit 42, which will be described later. The information processing device 10 acquires spectrum information and learning models managed in the database 22 via the LAN 21.

(learning model)
The learning model in this embodiment is a regression learning model, and one generated by machine learning such as deep learning can be used. A machine learning algorithm that uses training data to perform learning and is constructed to make appropriate predictions is referred to here as a learning model. There are various types of machine learning algorithms used in learning models. For example, deep learning using neural networks can be used. A neural network is composed of an input layer, an output layer, and a plurality of hidden layers, and each layer is connected by a calculation formula called an activation function. When using labeled training data (output corresponding to input), the coefficients of the activation function are determined so that the relationship between input and output holds. By determining coefficients using multiple pieces of training data, it is possible to generate a learning model that can predict outputs for inputs with high accuracy.

(Analysis equipment)
The analyzer 23 is a device for analyzing samples, solid samples, and the like. The analysis device 23 corresponds to an example of analysis means. Note that, as described above, in this embodiment, the information processing device 10 and the analysis device 23 are communicably connected. However, the analysis device 23 may be provided inside the information processing device 10, or the information processing device 10 may be provided inside the analysis device 23. Furthermore, the analysis result (spectral information) may be transferred from the analysis device 23 to the information processing device 10 via a recording medium such as a nonvolatile memory.

The analyzer 23 in this embodiment can be used as long as it can acquire spectrum information by secondary ion mass spectrometry. Examples of devices that can acquire spectral information using secondary ion mass spectrometry include time-of-flight secondary ion mass spectrometers (TOF-SIMS). Hereinafter, TOF-SIMS will be explained in detail.

In TOF-SIMS, when a sample is irradiated with an ion beam (primary ions), elements and molecules are released from the surface of the solid sample. Examples of primary ions include Ga ⁺ , In ⁺ , Au ⁺ , Bi ⁺ and the like. These primary ions are shaped into a pulse with a width of 0.7 nsec or less and a repetition frequency of several kHz to 50 kHz, and are irradiated onto the sample. The secondary ions extracted from the sample by the extraction electric field fly at a constant speed in the electric field-free space between the grid and the detector, and then are detected by a time-of-flight mass spectrometer. Since the flight speed at that time differs depending on the mass of the secondary ion, mass analysis is performed based on the difference in arrival time to the time-of-flight mass spectrometer. Note that the reflection time-of-flight mass spectrometer uses a reflection flight (drift) tube to perform mass spectrometry based on the time it takes to fly over a distance of 2 m.

The information processing device 10 includes a communication IF 31, a ROM 32, a RAM 33, a storage section 34, an operation section 35, a display section 36, and a control section 37 as its functional configuration.

The communication IF (Interface) 31 is realized by, for example, a LAN card and a USB interface card. Communication IF 31 controls communication between external devices (for example, database 22 and analysis device 23) and information processing device 10 via LAN 21 and USB. The ROM (Read Only Memory) 32 is realized by a nonvolatile memory or the like, and stores various programs and the like. A RAM (Random Access Memory) 33 is realized by a volatile memory or the like, and temporarily stores various information. The storage unit 34 is realized by, for example, an HDD (Hard Disk Drive), and stores various information. The operation unit 35 is realized by, for example, a keyboard or a mouse, and inputs instructions from the user into the device. The display unit 36 is realized by, for example, a display, and displays various information to the user. The operation unit 35 and the display unit 36 provide a function as a GUI (Graphical User Interface) under the control of the control unit 37.

(control unit)
The control unit 37 is realized by, for example, at least one CPU (Central Processing Unit), and centrally controls the processing in the information processing device 10. The control unit 37 includes a spectrum information acquisition unit 41, a learning model generation unit 42, a learning model acquisition unit 43, an estimation unit 44, an information acquisition unit 45, and a display control unit 46 as its functional configuration.

(Spectral information acquisition unit 41)
The spectral information acquisition unit 41 acquires the analysis result of the target sample including the solid sample and the substance to be measured, specifically, the spectrum information of the target sample from the analyzer 23 . Note that the spectrum information of the target sample may be acquired from the database 22 in which analysis results are stored in advance. Similarly, spectral information of the solid sample and spectral information of the sample including the solid sample and the substance to be measured are acquired. The spectral information of this solid sample is the spectral information when a single solid substance exists. The spectral information of a sample containing a solid sample and a substance to be measured is, for example, multiple types of samples containing different amounts of the substance to be measured, such as 0.2%, 0.4%, etc. This refers to the spectral information obtained from each sample.

Then, the spectrum information acquisition unit 41 outputs the spectrum information of the acquired sample to the estimation unit 44. Further, the acquired sample spectral information and the solid sample spectral information are output to the learning model generation section 42 .

(Learning model generation unit 42)
The learning model generation unit 42 generates teacher data using the solid sample acquired by the spectral information acquisition unit 41 and the spectral information of the sample. The learning model generation unit 42 then executes deep learning using the teacher data to generate a learning model. A detailed explanation regarding the generation of teacher data and the generation of a learning model will be described later. The learning model generation unit 42 then outputs the generated learning model to the learning model acquisition unit 43. Note that the learning model generation unit 42 may output the generated learning model to the database 22.

(Learning model acquisition unit 43)
The learning model acquisition unit 43 acquires the learning model generated by the learning model generation unit 42. Note that if the learning model is stored in the database 22, the learning model acquisition unit 43 acquires the learning model from the database 22. The learning model acquisition unit 43 then outputs the acquired learning model to the estimation unit 44.

(Estimation unit 44)
The estimation unit 44 learns information regarding the type of substance to be measured contained in the target sample by inputting the spectrum information of the target sample acquired by the spectrum information acquisition unit 41 into the learning model acquired by the learning model acquisition unit 43. Let the model estimate. Then, the estimation unit 44 outputs information regarding the estimated type to the information acquisition unit 45. The estimator 44 corresponds to an example of an estimator that estimates information regarding the type of substance to be measured by inputting spectral information of the target sample into a learning model.

(Information acquisition unit 45)
The information acquisition unit 45 acquires information regarding the type estimated by the learning model. That is, the information acquisition unit 45 corresponds to an example of an information acquisition unit that acquires information regarding the type of the substance to be measured, which is estimated by inputting the spectrum information of the target sample into the learning model. The information acquisition unit 45 then outputs the acquired information regarding the type to the display control unit 46.

(Display control unit 46)
The display control unit 46 causes the display unit 36 to display the information regarding the type acquired by the information acquisition unit 45. The display control unit 46 corresponds to an example of display control means.

Note that at least a part of each unit included in the control unit 37 may be realized as an independent device. Alternatively, each may be realized as software that realizes the functions. In this case, the software that implements the function may operate on a server via a network such as a cloud. In this embodiment, it is assumed that each part is realized by software in a local environment.

Furthermore, the configuration of the information processing system shown in FIG. 1 is merely an example. For example, the storage unit 34 of the information processing device 10 may have the function of the database 22, and the storage unit 34 may hold various information.

Next, the processing procedure for generating a learning model in this embodiment will be explained using FIG. 2. FIG. 2 is a diagram illustrating an example of a flowchart of a processing procedure regarding generation of a learning model according to the present embodiment.

(S201: Analyze a single solid sample and a sample containing a solid sample and any one of the substances to be measured A, B, C, or D)
First, the analyzer 23 analyzes a single solid sample and a sample containing a solid sample and a substance to be measured (step S1). The analysis conditions may be selected as appropriate from the viewpoint of sensitivity, analysis time, etc. When analyzing a sample containing a solid sample and a substance to be measured, the concentration of the substance to be measured is varied in several ways. The number required varies depending on the properties of the substance, but it is generally preferable to change three or more points. When there are multiple types of substances to be measured, it is preferable to analyze them individually, but if the signals of the substances to be measured can be sufficiently separated, they may be measured simultaneously.

The analysis device 23 then outputs the acquired spectrum information to the information processing device 10. The information processing device 10 receives spectrum information from the analysis device 23 and stores it in the RAM 33 or storage section 34. The spectrum information acquisition unit 41 acquires the spectrum information held in this way.

Note that, as described above, the database 22 may hold the spectrum information that is the analysis result. In this case, the spectrum information acquisition unit 41 acquires spectrum information from the database 22. Moreover, the timing at which the analyzer 23 analyzes the test substance may be any timing as long as it is executed before the selection of spectrum information in step S3.

(S202: Generate teacher data)
The learning model generation unit 42 generates teacher data using the spectrum information of the solid sample obtained in step S201 and the sample including the solid sample and the substance to be measured. At that time, teacher data is generated based on the first spectrum information and the second spectrum information. Hereinafter, the first spectrum information and the second spectrum information will be explained in detail.

The first spectral information refers to spectral information of a single solid sample, spectral information of a sample including a solid sample and a substance to be measured, and spectral information created using a noise spectrum.

The second spectral information is created from the spectral information of samples with different concentrations of the substance to be measured, extracted from the first spectral information group. Specifically, the concentration of the substance to be measured is included between the concentrations of the two extracted first spectrum information. Hereinafter, the created spectrum will be referred to as an artificial spectrum, and the method for creating this spectrum will be referred to as concentration interpolation.

Next, a method for creating training data with noise added will be described. The spectral information consists of the ion intensity measured for each fragment or the ion intensity normalized by dividing by the sum of the ion intensities of fragments within a predetermined range of m/z, and the value is the molecular weight m It is measured for each m/z, which is calculated by dividing the ion valence by the ion valence z.

Even if the ratio of the ion intensity in the spectral information of the substance to be measured to the total of the fragment ion intensities with m/z of 10 or more in the spectral information of the solid sample and the sample containing the substance to be measured is 7×10 ⁻⁵ or less good. Even if the ratio of the ion intensity of the spectral information of the substance to be measured is as small as 7×10 ^-5 or less, by using the information processing device of the present invention, information regarding the type of substance to be measured can be obtained simply, quickly, and with high precision. becomes possible. Preferably, the ratio is less than 7×10 ⁻⁵ .

In this embodiment, in the spectrum information of a single solid sample, each ion intensity is multiplied by an arbitrary value between 0.1 and 5.0. The spectrum multiplied by this arbitrary value is referred to as a noise spectrum, and this multiplication operation is referred to as noise addition. Next, among the spectrum information of the sample including the solid sample and the substance to be measured, the average value of the value multiplied by the ion intensity having the same m/z value, and the average value of the value multiplied by the ion intensity and an arbitrary value are calculated. value, or the largest value may be used as the newly created ion strength. This operation is performed for all m/z and used as first spectrum information.

In generating the training data, either density interpolation or noise addition may be performed first.

(S203: Generate learning model)
Using the generated training data, machine learning is performed according to a predetermined algorithm to construct a learning model (step S203). As a specific learning method, for example, general machine learning methods such as a neural network or a support vector machine may be used. Further, as a deep learning method with multiple hidden layers, DNN (deep neural network), CNN (convolutional neural network), etc. may be used. If there are multiple types of substances to be measured, a learning model is generated for each substance. The learning model generation unit 42 then stores the generated learning model in the RAM 33, the storage unit 34, or the database 22.

As described above, a learning model that estimates information regarding the type of substance to be measured contained in the target sample is generated based on the spectral information of the solid sample alone and the sample including the solid sample and the substance to be measured.

Next, a method of displaying information regarding types will be explained. FIG. 3 is a diagram illustrating an example of a flowchart of a processing procedure for displaying information regarding types according to the present embodiment.

(S301: Analyze target sample)
The analyzer 23 analyzes a target sample including a solid sample and a substance to be measured, and acquires spectrum information of the target sample. The analysis conditions are the same as those in step 201 described above. The analysis device 23 then outputs the acquired spectrum information to the information processing device 10. The information processing device 10 receives spectrum information from the analysis device 23 and stores it in the RAM 33 or storage section 34. The spectrum information acquisition unit 41 acquires the spectrum information held in this way. Note that, as described above, the database 22 may hold the spectrum information that is the analysis result. In this case, the spectrum information acquisition unit 41 acquires spectrum information from the database 22. Moreover, the timing at which the analyzer 23 analyzes the sample may be any timing as long as it is executed before the estimation of the information regarding the type in step S302.

(S302: Estimate information regarding type)
The learning model acquisition unit 43 acquires a learning model stored in the RAM 33, the storage unit 34, or the database 22. Then, the estimating unit 44 inputs the spectrum information of the target sample acquired in step S301 into the acquired learning model to estimate information regarding the type of the substance to be measured contained in the target sample. Furthermore, if necessary, the estimating unit 44 converts the information regarding the estimated type into a format to be displayed on the display unit 36. The format displayed on the display unit 36 may be a concentration such as g/L or mol/L, or a ratio to a reference amount (standard amount). If the values estimated by the learning model are in these display formats, there is no need to convert them. Then, the information acquisition unit 45 acquires information regarding the estimated type from the estimation unit 44 and stores it in the RAM 33 or the storage unit 34.

(S303: Display information regarding type)
The display control unit 46 causes the display unit 36 to display information regarding the type of substance to be measured contained in the target sample, which was estimated by the learning model in step 302. At that time, the information may be organized and displayed in a graph format or a table format. FIG. 7 shows an example of a screen (window) displayed on the display unit 36. Furthermore, the level may be displayed according to the numerical value of the estimated concentration, such as "high" or "low".

The present invention provides a system or device with a program that implements one or more of the functions of the embodiments described above via a network or a storage medium, and one or more processors in the computer of the system or device reads and executes the program. This can also be achieved by processing. It can also be realized by a circuit (for example, ASIC) that realizes one or more functions.

<Example>
The present invention will be explained in more detail below by giving Examples and Comparative Examples. Note that the present invention is not limited to the following examples.

(Example 1)
[Preparation of solid sample]
49.5 g each of methyl methacrylate as a monomer and trimethylolpropane methacrylate (both manufactured by Tokyo Kasei Kogyo) as a crosslinking agent were mixed. Furthermore, 1 g of Omnirad 819 (manufactured by IGM Resins B.V.) as a polymerization initiator was added to the mixed solution. Next, 1 g was taken from the obtained solution and stirred for 5 minutes using a mixer (Awatori Rentaro AR-100, manufactured by Shinky). Then, 0.3 g was taken out from the stirred solution, sandwiched between fluororesin films (Neoflon PFA film, manufactured by Daikin Industries), and irradiated with a mercury lamp (UL750 manufactured by HOYA) for 5 minutes from a distance of 30 cm from the fluororesin film. As a result, a photocurable resin was obtained, which was used as a solid sample.

[Preparation of sample containing solid sample and substance to be measured]
[Preparation of sample A]
To 1 g of the stirred liquid obtained in the preparation of the solid sample, 0.001 g of diisononyl phthalate (the following structural formula A, manufactured by Fujifilm Wako Pure Chemical Industries, Ltd.), which is the substance A to be measured, was added as a plasticizer, and the mixture was added to the solid sample. Sample A containing 0.1% of substance A to be measured was prepared.

Note that the method for obtaining a photocurable resin from a solution obtained by adding a plasticizer is the same method as for producing a solid sample.

Furthermore, by changing the amount of plasticizer added, the target substance A was added to the solid sample by 0.2%, 0.4%, 0.6%, 0.8%, 1.0%, and 10%, respectively. A sample containing .0% was prepared. A sample was produced in the same manner as Sample A, except that the amount of plasticizer added was changed.

[Preparation of sample B]
Sample B was prepared in the same manner as Sample A, except that the plasticizer was changed to 2-isopropylthioxanthone (Structural Formula B below, manufactured by Tokyo Kasei Kogyo), which is the substance to be measured B. At that time, by changing the amount of plasticizer added, 0.1%, 0.2%, 0.4%, 0.6%, 0.8%, Samples containing 1.0% and 10.0% were prepared.

[Preparation of sample C]
Sample C was prepared in the same manner as sample A, except that the plasticizer was changed to acetylenol E100 (the following structural formula C, manufactured by Kawaken Fine Chemicals), which is the substance C to be measured. At that time, by changing the amount of plasticizer added, the measured substance C was added to the solid sample by 0.1%, 0.2%, 0.4%, 0.6%, 0.8%, respectively. Samples containing 1.0% and 10.0% were prepared.

[Preparation of sample D]
Sample D was produced in the same manner as Sample A, except that the plasticizer was changed to merocyanine (Structural Formula D below, FDB-009, manufactured by Yamada Chemical), which is the substance D to be measured. At that time, by changing the amount of plasticizer added, the measured substance D was added to the solid sample by 0.1%, 0.2%, 0.4%, 0.6%, 0.8%, respectively. Samples containing 1.0% and 10.0% were prepared.

[Measurement by TOF-SIMS method]
The prepared solid sample and the sample containing the solid sample and the substance to be measured were each cut into a 1 cm square, and measured by the TOF-SIMS method under the following conditions.
Measurement model: TOF-SIMS5 (manufactured by ION-TOF)
Primary ion species: Bi ³⁺
Primary ion current: 0.4pA (Pulse current)
Primary ion energy: 25keV
Measurement mode: Bunching Mode
Detection range: 1~800amu
Electron gun: Charge correction electron gun Measurement vacuum: 7×10 ^-8 Pa
Measurement area: 300μm x 300μm
Number of measurement points: 128 x 128 points within the measurement area Integration: 16 times measurement mode Negative ion Measurement number: 6 times or more for each sample The spectrum groups shown in FIGS. 4 to 7 were obtained by measuring at least 6 times for each sample. FIG. 4 is an example of spectrum information of a solid sample and a sample A that includes the solid sample and the substance A to be measured. FIG. 5 is an example of spectrum information of sample B including a solid sample and measurement target substance B. FIG. 6 is an example of spectrum information of a sample C including a solid sample and a substance C to be measured. FIG. 7 is an example of spectrum information of a sample D including a solid sample and a substance D to be measured.

(Comparative example 1)
The four operators shown in Table 1, who have been working on TOF-SIMS for various years, contained 0.1%, 0.4%, or 0.8% of the target substances A, B, C, and D. The substance to be measured in the sample was identified using the spectrum groups of the sample (6 sets each, 72 sets in total). At that time, spectral groups (6 sets each, 24 sets in total) of samples containing 10.0% of any one of the measurement target substances A, B, C, and D were used as reference. Although all of the substances to be measured were correctly identified, as shown in Table 1, it was found that it took more than one hour to manually identify the substances to be measured.

(Example 2)
The method for creating the teacher data will be explained based on the flowchart of FIG. 8.

[Creating training data by adding noise]
It was created using the spectrum information measured in Example 1 in the following manner.

The ion intensities of the spectra (total of 6 sets) of a single solid sample were compared for each m/z, and the ion intensity with the highest value was extracted. After combining the ion intensities with the highest values from m/z 10 to 700, noise is generated by multiplying each m/z by an arbitrary value randomly selected between 0.1 and 5.0. A spectrum was created.

Next, the ion intensities of this noise spectrum and the spectrum of a single solid sample were compared for each m/z, and the ion intensity having the highest value was extracted. An artificial spectrum was created by combining the ion intensity groups having the highest values with m/z of 10 to 700.

The above operations were performed 10 times for each of the six sets of spectra of single solid samples, and a total of 60 sets of training data consisting of artificial spectra that did not contain the substance to be measured were created. Note that any numerical value is updated to a new value each time (step: S401).

Further, the ion intensities were compared for each m/z between the noise spectrum and the spectrum of sample A containing 0.2% of the substance A to be measured, and the ion intensity having the highest value was extracted. Then, an artificial spectrum was created by combining the ion intensity groups having the highest values from m/z of 10 to 700. This process was performed 10 times for each of the six sets of spectra of sample A containing 0.2% of the target substance A, and a total of 60 sets of training data consisting of artificial spectra of sample A containing 0.2% of the target substance A were obtained. created (step: S402).

Next, for the spectra containing 0.6% and 1.0% of the target substance A, respectively, we performed the same work as creating training data consisting of the artificial spectrum of sample A containing 0.2% of the target substance A. , 60 pairs of teacher data were created for each.

As described above, a total of 240 sets of training data were created from the first spectrum group (steps: S403 and S404).

Furthermore, the spectra containing 0.2%, 0.6%, and 1.0% of any one of the target substances B, C, and D, respectively, are also composed of artificial spectra of sample A containing the target substance A. We performed the same work as creating the training data. As a result, 60 sets of training data were created for each.

From the above, a total of 780 sets of teacher data were obtained in creating the teacher data by adding noise.

[Creating training data by density interpolation]
One spectrum was extracted from each of the training data (60 sets in total) consisting of the spectrum of a single solid sample and the training data (60 sets in total) consisting of the spectrum of sample A containing 0.2% of target substance A. . Furthermore, using calculation formulas (1) to (3) for each m/z, the spectrum (artificial spectrum) of sample A containing 0.05%, 0.1%, and 0.15% of substance A to be measured is calculated. Created.

Here, I ₀ is the ion intensity of the training data consisting of the spectrum of a single solid sample, and I _n is the ion intensity of the spectrum containing n% of the substance A to be measured. The above work was performed on 60 sets of training data consisting of the spectrum of a single solid sample and 60 sets of training data containing 0.2% of target substance A, and 60 x 60 x 3 = 10,800 sets of training data were used. Data was obtained (step: S405).

In the same way, from the teacher data group containing 0.2% of the substance A to be measured and the group of teacher data containing 0.6% of the substance A to be measured, the substance A to be measured is 0.3%, 0.4%, 0. A total of 10,800 sets of teacher data containing .5% each were created.

In addition, using the same method, from the teacher data group containing 0.6% of the target substance A and the teacher data group containing 1.0% of the target substance A, the target substances were added to 0.7%, 0.8%, A total of 10,800 sets of teacher data containing 0.9% each were created. As described above, a total of 32,400 sets of training data were created from the second spectrum group (steps: S406 and S407).

Furthermore, for the substances to be measured B, C, and D, the same work as the creation of teacher data by concentration interpolation for sample A containing the substance to be measured was performed, and 32,400 sets of teacher data were created for each.

A learning model was generated by machine learning using 130,320 sets of teacher data, which are a combination of 720 sets of sample teacher data created by adding noise and 129,600 sets of sample teacher data created by density interpolation, as teacher data. A fully connected neural network was used as a machine learning method, and a relu function and a softmax function were used as activation functions. The mean squared error was used as the loss function, and Adam was used as the optimization algorithm. In order to obtain sufficient identification accuracy, it was necessary to repeat calculations for about 1000 epochs.

Next, we prepared six sets of spectra each for samples containing 0.1%, 0.4%, and 0.8% of any of the measurement target substances A, B, C, and D, and applied them to the obtained learning model. The method was applied to predict the type of substance to be measured contained in the sample. That is, the classification probabilities of each substance to be measured calculated using the softmax function were compared, and the substance to be measured with the highest probability was determined by machine learning, and the probability was taken as the reliability. Furthermore, this prediction was made by a person who had no experience with TOF-SIMS. As shown in Table 2, even a person with no experience or analytical skills was able to quickly (in just 5 seconds) identify the substance to be measured with high accuracy.

(Example 3)
A learning model was created in the same manner as in Example 2, except that noise was not added in Example 2. Specifically, it is as shown below.

A total of 72 sets of spectra, 6 sets each of the spectrum of the solid sample alone and the spectrum of the sample containing 0.2%, 0.6%, and 1.0% of any one of the target substances A, B, C, and D, respectively. The spectrum was used as first spectrum information.

[Creating training data by density interpolation]
One spectrum was extracted from each of six sets of spectra of the solid sample alone and six sets of spectra of sample A containing 0.2% of substance A to be measured. Furthermore, using the above calculation formulas (1) to (3) for each m/z, the spectra of sample A containing 0.05%, 0.1%, and 0.15% of the substance A to be measured (artificial spectrum), respectively. It was created.

This work was performed in a round-robin manner on 6 sets of spectra of a single solid sample and 6 sets of spectra of sample A containing 0.2% of substance A to be measured, and 6×6×3=108 pieces of training data were obtained.

Using the same method, from 6 sets of spectra containing 0.2% of target substance A and 6 sets of spectra containing 0.6% of target substance A, 0.3% and 0.4% of target substance A were obtained, respectively. A total of 108 sets of teacher data containing 0.5% were created.

In addition, using the same method, from 6 sets of spectra containing 0.6% of target substance A and 6 sets of spectra containing 1.0% of target substance A, 0.7% and 0.8% of the target substance were obtained, respectively. , a total of 108 pieces of training data containing 0.9% were created. From the above, a total of 324 sets of second spectrum information were set.

Furthermore, for the substances to be measured B, C, and D, the same work as the creation of teacher data by concentration interpolation for sample A containing the substance to be measured A was performed, and a total of 1296 sets of teacher data were created.

As described above, a learning model was generated by performing machine learning using 1368 pairs of first and second spectrometers as training data. A fully connected neural network was used as a machine learning method, and a relu function and a softmax function were used as activation functions. The mean squared error was used as the loss function, and Adam was used as the optimization algorithm. In order to obtain sufficient identification accuracy, it was necessary to repeat calculations for about 1000 epochs.

Next, we prepared six sets of spectra each for samples containing 0.1%, 0.4%, and 0.8% of any of the measurement target substances A, B, C, and D, and applied them to the obtained learning model. The method was applied to predict the type of substance to be measured contained in the sample. That is, the classification probabilities of each substance to be measured calculated using the softmax function were compared, and the substance to be measured with the highest probability was determined by machine learning, and the probability was taken as the reliability. Furthermore, this prediction was made by a person who had no experience with TOF-SIMS. As shown in Table 3, even a person with no experience or analytical skills was able to quickly (in just 5 seconds) identify the substance to be measured with high accuracy.

(Comparative example 2)
A learning model was created in the same manner as in Example 3, except that the training data was not created by density interpolation in Example 3. Specifically, it is as shown below.

6 sets of spectra of a single solid sample, 6 sets of spectra of sample A containing 0.2% of target substance A, 6 sets of spectra of sample A containing 0.6% of target substance A, and 1 set of spectra of sample A containing 0.6% of target substance A. Six sets of spectra of sample A containing .0% were used as training data.

Furthermore, 18 sets of training data were obtained for each of the measurement target substances B, C, and D in the same manner as for the measurement target substance A.

Using these 72 sets as training data, machine learning was performed to generate a learning model. A fully connected neural network was used as a machine learning method, and a relu function and a softmax function were used as activation functions. The mean squared error was used as the loss function, and Adam was used as the optimization algorithm. In order to obtain sufficient identification accuracy, it was necessary to perform repeated calculations for about 200 epochs.

Next, we prepared six sets of spectra each for samples containing 0.1%, 0.4%, and 0.8% of any of the measurement target substances A, B, C, and D, and applied them to the obtained learning model. The method was applied to predict the type of substance to be measured contained in the sample. That is, the classification probabilities of each substance to be measured calculated using the softmax function were compared, and the substance to be measured with the highest probability was determined by machine learning, and the probability was taken as the reliability. As shown in Table 4, the substance to be measured could not be identified with high accuracy.

Claims (21)

Information acquisition means for acquiring information regarding the type of the substance to be measured in the sample, estimated by inputting spectral information of the solid sample and spectral information of the sample including the solid sample and the substance to be measured into a learning model. An information processing device having:
The spectral information is a spectrum indicating fragment ion intensity for each m/z obtained by secondary ion mass spectrometry,
The learning model includes a spectrum of the solid sample alone, a spectrum of the solid sample containing the substance to be measured, an artificial spectrum in which noise is added to the fragment ion intensity for each m/z of the spectrum of the solid sample alone, and the An information processing device characterized in that the information processing device is generated based on at least an artificial spectrum in which noise is added to the fragment ion intensity for each m/z of the spectrum of the solid sample containing the substance to be measured. The information processing device according to claim 1, wherein the secondary ion mass spectrometry is time-of-flight secondary ion mass spectrometry. 3. The information on the type of the substance to be measured is at least one type selected from the group consisting of a name, an abbreviation, a chemical classification, a structural formula, a chemical formula, and a three-dimensional structure diagram of the substance to be measured. The information processing device described. The information processing device according to any one of claims 1 to 3, wherein the solid sample includes at least one selected from the group consisting of a resin, a silicon substrate, and a copper foil. The information processing device according to claim 4, wherein the solid sample contains resin. The information processing device according to claim 5, wherein the resin has a unit derived from an unsaturated carboxylic acid ester. The substance to be measured is a plasticizer, thickener, thinner, crystal nucleating agent, crystallization promoter, crystallization retardant, mold release agent, flame retardant, flame retardant aid, thermal decomposition inhibitor, reinforcing agent. , conductive materials, antistatic agents, thermal conductive agents, antioxidants, weatherproofing agents, ultraviolet absorbers, antibacterial agents, colorants, gas barrier agents, antifogging agents, antireflective agents, antifouling agents, and impurities. The information processing device according to any one of claims 1 to 6, wherein the information processing device is at least one type selected from the following. The information processing device according to claim 7, wherein the substance to be measured is a plasticizer. The information processing device according to claim 8, wherein the plasticizer is a surfactant. The information processing apparatus according to any one of claims 1 to 9, further comprising a display control unit that causes a display unit to display information regarding the type of the acquired substance to be measured. Any one of claims 1 to 10, wherein the learning model is further generated based on a plurality of spectral information that is a spectrum of a solid sample containing a substance to be measured, the concentrations of the substance to be measured with respect to the solid sample being different from each other. The information processing device according to item 1. Claims 1 to 10, wherein the learning model is generated based on spectral information of a single solid sample and spectral information of a plurality of samples including a solid sample and a substance to be measured, each having a different concentration of the substance to be measured relative to the solid sample. The information processing device according to any one of the above. 13. The learning model is further generated based on spectral information of a sample including a solid sample and a substance to be measured whose concentration is different from the concentration of the substance to be measured used in the spectral information of the plurality of samples. The information processing device described. 14. The information processing device according to claim 1, wherein a group of elements constituting the solid sample and a group of elements constituting the substance to be measured are the same. 2. The element group constituting the solid sample and the element group constituting the substance to be measured include at least one selected from the group consisting of carbon atoms, hydrogen atoms, oxygen atoms, and nitrogen atoms. 15. The information processing device according to any one of 14. Any one of claims 1 to 15, wherein the ratio of the ion intensity in the spectral information of the substance to be measured to the total intensity of fragment ions having m/z of 10 or more in the spectral information of the sample is 7×10 ^-5 or less. The information processing device described in section. Information regarding the type of the substance to be measured in the sample and quantitative information estimated by inputting spectral information of the solid sample and spectral information of the sample including the solid sample and the substance to be measured into a learning model. An information processing device having an information acquisition means for acquiring
The spectral information is a spectrum indicating fragment ion intensity for each m/z obtained by secondary ion mass spectrometry,
The learning model includes a spectrum of the solid sample alone, a spectrum of the solid sample containing the substance to be measured, an artificial spectrum in which noise is added to the fragment ion intensity for each m/z of the spectrum of the solid sample alone, and the An information processing device characterized in that the information processing device is generated based on at least an artificial spectrum in which noise is added to the fragment ion intensity for each m/z of the spectrum of the solid sample containing the substance to be measured. 18. The information processing apparatus according to claim 1, wherein reliability of information regarding the type of the measurement substance is acquired. an information acquisition step of acquiring information regarding the type of the substance to be measured in the sample, estimated by inputting spectral information of the solid sample and spectral information of the sample including the solid sample and the substance to be measured into a learning model; A method for controlling an information processing device, the method comprising:
The spectral information is a spectrum indicating fragment ion intensity for each m/z obtained by secondary ion mass spectrometry,
The learning model includes a spectrum of the solid sample alone, a spectrum of the solid sample containing the substance to be measured, an artificial spectrum in which noise is added to the fragment ion intensity for each m/z of the spectrum of the solid sample alone, and the A method for controlling an information processing device, characterized in that the control method for an information processing device is generated based on at least an artificial spectrum in which noise is added to the fragment ion intensity for each m/z of the spectrum of the solid sample containing a substance to be measured. Information regarding the type of the substance to be measured in the sample and quantitative information estimated by inputting spectral information of the solid sample and spectral information of the sample including the solid sample and the substance to be measured into a learning model. A method for controlling an information processing device, comprising an information acquisition step of acquiring
The spectral information is a spectrum indicating fragment ion intensity for each m/z obtained by secondary ion mass spectrometry,
The learning model includes a spectrum of the solid sample alone, a spectrum of the solid sample containing the substance to be measured, an artificial spectrum in which noise is added to the fragment ion intensity for each m/z of the spectrum of the solid sample alone, and the A method for controlling an information processing device, characterized in that the control method for an information processing device is generated based on at least an artificial spectrum in which noise is added to the fragment ion intensity for each m/z of the spectrum of the solid sample containing a substance to be measured. A program that causes a computer to function as each means of the information processing apparatus according to any one of claims 1 to 20 .

Images