Sample composition estimation method and device, learning method and recording medium

Abstract

The present invention provides a sample composition estimation method and its device, a learning method and a recording medium. The sample composition estimation method comprises the following steps: obtaining a spectrum of a sample measured by a wavelength dispersion type X-ray analysis device (100); determining an analysis target element of the sample and an input wavelength range corresponding to the analysis target element; and inputting a spectrum of the input wavelength range in the spectrum of the sample into a first learning completion model (44) to estimate the chemical bonding state of the analysis target element in the sample.

Description

Technical Field

[0001] The present invention relates to a sample component estimation method, a sample component estimation device, a learning method, and a non-transitory computer-readable recording medium. Background Art

[0002] In X-ray analysis apparatuses such as an electron probe microanalyzer (EPMA), there is known a method of using a characteristic X-ray spectrum emitted from a sample in order to estimate the chemical bonding state of elements constituting the sample.

[0003] When using a characteristic X-ray spectrum to estimate the chemical bonding state of the element to be analyzed, the following phenomenon is used: the shape of the spectrum waveform such as the peak wavelength and half-peak width of the spectrum and the intensity ratio between multiple characteristic X-ray types (peaks) change depending on the chemical bonding state. In other words, the chemical bonding state of the element to be analyzed is estimated by comparing the above parameters between the spectrum waveform of a compound of known composition and the spectrum waveform of the sample to be analyzed. This method is generally called "state analysis".

[0004] For example, Japanese Patent Application Laid-Open No. 2003-75376 discloses a technique for analyzing the chemical bonding state of tungsten based on its Mα line and Mβ line, and Japanese Patent Application Laid-Open No. 2014-228307 discloses a technique for analyzing the chemical bonding state of aluminum based on the intensities of its Kα line and sKα line and the half-value width of its Kβ line. Summary of the invention

[0005] When analyzing the state by analyzing the spectrum waveform, the following processing is usually performed. Initially, the baseline is obtained from the spectrum waveform of the analysis object and the baseline is subtracted to obtain the peak wavelength, peak intensity and half-value width. At this time, the spectrum waveform is also measured for the sample with a known composition, and the same processing is performed. Then, the spectrum waveforms of the two are overlapped and displayed on the graph to compare the shape of the waveforms, and the above parameters (peak wavelength, peak intensity and half-value width) are compared. In the case of multiple peaks, their intensity ratios are also compared.

[0006] However, the above-mentioned processing is very complicated for users. In addition, the parameters to be focused on are different depending on the element to be analyzed and its compound, so analytical knowledge and experience are required, which may not be easy for all users. Moreover, not all users have samples with known compositions.

[0007] The present invention is made to solve the above-mentioned problems, and an object of the present invention is to provide a technology that can simply and easily estimate the chemical bonding state of the analysis target element in the sample when the state analysis of the sample is performed using the spectrum measured by an X-ray analyzer.

[0008] A sample composition estimation method involved in one embodiment of the present invention includes the following steps: obtaining a spectrum of a sample measured by a wavelength dispersion type X-ray analysis device; determining the analysis target element of the sample and the input wavelength range corresponding to the analysis target element; and inputting the spectrum of the input wavelength range in the spectrum of the sample into a first learning completion model to estimate the chemical bonding state of the analysis target element in the sample.

[0009] A sample composition estimation device involved in one embodiment of the present invention comprises: a spectrum acquisition unit, which acquires the spectrum of the sample measured by a wavelength dispersion type X-ray analysis device; an input unit, which determines the analysis target element of the sample and the input wavelength range corresponding to the analysis target element; a compound species identification unit, which inputs the spectrum of the input wavelength range in the spectrum of the sample into a first learning completion model to estimate the chemical bonding state of the analysis target element in the sample; and a display unit, which displays the estimated chemical bonding state.

[0010] A non-transitory computer-readable recording medium according to one embodiment of the present invention records a sample composition estimation program executed by a computer. The sample composition estimation program causes the computer to execute the following steps: obtaining a spectrum of a sample measured by a wavelength dispersion type X-ray analysis device; determining an analysis target element of the sample and an input wavelength range corresponding to the analysis target element; and inputting a spectrum of the input wavelength range in the X-ray spectrum of the sample into a first learning completion model to estimate the chemical bonding state of the analysis target element in the sample.

[0011] A learning method involved in one embodiment of the present invention is used to generate the above-mentioned first learning completion model and second learning completion model, and the learning method includes the following steps: making a first learning completion model by using a group of spectra of compound species with known compositions and the composition of the compound species as learning processing of teacher data; and making a second learning completion model by using the following group as learning processing of teacher data: a group of a first spectrum containing high-order spectral line peak components and a shaped spectrum with the high-order spectral line peak components removed from the first spectrum; and a group of a second spectrum not containing high-order spectral line peak components and a shaped spectrum with the high-order spectral line peak components not removed from the second spectrum.

[0012] A non-transitory computer-readable recording medium according to one embodiment of the present invention records a learning program executed by a computer. The learning program is a learning program for generating the first learning completed model and the second learning completed model, and the learning program causes the computer to execute the following steps: the first learning completed model is produced by using a spectrum of a compound species with a known composition and a composition of the compound species as a learning process of teacher data; and the second learning completed model is produced by using a group of the following groups as learning processes of teacher data: a group of a first spectrum containing a high-order spectral line peak component and a shaped spectrum with the high-order spectral line peak component removed from the first spectrum; and a group of a second spectrum not containing a high-order spectral line peak component and a shaped spectrum with the high-order spectral line peak component not removed from the second spectrum.

[0013] The foregoing and other objects, features, aspects and advantages of the present invention will become more apparent from the following detailed description of the present invention which can be understood in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 1 is a diagram showing the overall configuration of an analysis system including a sample component estimation device according to the present embodiment.

[0015] Figure 2 This is a block diagram showing an example of the functional configuration of a sample component estimation device.

[0016] Figure 3 This is a table showing an example of the relationship between elements and input wavelength ranges.

[0017] Figure 4 This is a diagram schematically showing an example of a spectrum generator model.

[0018] Figure 5 This is a diagram schematically showing an example of a classifier model.

[0019] Figure 6 This is a block diagram showing an example of the functional configuration of the learning unit.

[0020] Figure 7 This is a diagram for explaining the procedure of the learning process of the spectrum generator model executed by the learning process unit.

[0021] Figure 8 This is a diagram for explaining the procedure of the learning process of the classifier model executed by the learning process unit.

[0022] Fig. 9 This is a diagram showing an example of a characteristic X-ray spectrum of a sample measured by EPMA.

[0023] Fig.10 This is a diagram showing an example of a characteristic X-ray spectrum emitted from an iron compound species.

[0024] Fig.11 is a flowchart for explaining the process of the learning process of the model of this embodiment.

[0025] Fig.12 is a flowchart for explaining the process of the identification process of the chemical bonding state of the element to be analyzed in the specimen of this embodiment.

[0026] Fig.13 is a diagram showing an example of the identification result displayed on the display unit.

[0027] Fig.14 is a diagram showing the overall structure of the first modification example of the analysis system according to this embodiment.

[0028] Fig.15 is a diagram showing the overall structure of the second modification example of the analysis system according to this embodiment. Detailed implementation mode

[0029] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In addition, the same or corresponding parts in the drawings are denoted by the same reference numerals, and their descriptions will not be repeated.

[0030] <Structure of the analysis system>

[0031] Figure 1 is a diagram showing the overall structure of an analysis system including a specimen composition estimation device according to this embodiment. The specimen composition estimation device according to this embodiment is a device for analyzing the state of a specimen, and is configured to estimate the chemical bonding state of the element to be analyzed contained in the specimen using the characteristic X-ray spectrum of the specimen obtained by the X-ray analysis device.

[0032] <Structure of the X-ray analysis device>

[0033] The X-ray analysis device has a wavelength dispersive spectrometer (WDS: Wavelength Dispersive Spectrometer), for example, an electron probe microanalyzer (EPMA: Electron Probe Micro Analyzer) that irradiates an electron beam onto the specimen. In addition, the X-ray analysis device is not limited to the EPMA, and may also be a fluorescence X-ray analysis device that irradiates an X-ray onto the specimen and then disperses the characteristic X-ray by the WDS.

[0034] Refer to Figure 1EPMA 100 includes an electron gun 1, a deflection coil 2, an objective lens 3, a sample stage 4, a sample stage driving unit 5, and a plurality of spectrometers 6a and 6b. In addition, EPMA 100 also includes a control unit 10, a data processing unit 11, and a deflection coil control unit 8. The electron gun 1, the deflection coil 2, the objective lens 3, the sample stage 4, and the spectrometers 6a and 6b are disposed in a measurement chamber (not shown), and during the measurement of X-rays, the measurement chamber is exhausted to a vacuum state.

[0035] The electron gun 1 is an excitation source for generating electron beams E to be irradiated toward a sample S on a sample stage 4. The beam current of the electron beams E can be adjusted by controlling a condenser lens (not shown). The deflection coil 2 forms a magnetic field using a drive current supplied from a deflection coil control unit 8. The magnetic field formed by the deflection coil 2 can deflect the electron beams E.

[0036] The objective lens 3 is disposed between the deflection coil 2 and the sample S placed on the sample stage 4, and reduces the electron beam E passing through the deflection coil 2 to a micro diameter. The electron gun 1, the deflection coil 2, and the objective lens 3 constitute an irradiation device for irradiating the sample with electron beams. The sample stage 4 is a stage for placing the sample S, and is configured to be movable in the vertical direction and in the horizontal plane by the sample stage driving unit 5.

[0037] By driving the sample stage 4 by the sample stage driving unit 5 and / or driving the deflection coil 2 by the deflection coil control unit 8, the irradiation position of the electron beam E on the sample S can be two-dimensionally scanned. When the scanning range is relatively narrow, the deflection coil 2 is used for scanning, and when the scanning range is relatively wide, the sample stage 4 is moved for scanning.

[0038] The spectrometers 6a and 6b are devices for detecting characteristic X-rays emitted from the sample S irradiated with the electron beam E. Figure 1 In the example of , two spectrometers 6a and 6b are shown, but the number of spectrometers is not limited thereto, and may be one or more than three. The structures of the spectrometers are the same except for the spectrometer crystal, and each spectrometer is sometimes referred to as simply "splitter 6" below.

[0039] The spectrometer 6a includes a spectroscopic crystal 61a, a detector 63a, and a slit 64a. The irradiation position of the electron beam E on the sample S, the spectroscopic crystal 61a, and the detector 63a are located on a Rowland circle not shown in the figure. Through a driving mechanism not shown in the figure, the spectroscopic crystal 61a is tilted while moving on the straight line 62a, and the detector 63a rotates as shown in the figure in accordance with the movement of the spectroscopic crystal 61a in such a way that the incident angle of the characteristic X-ray to the spectroscopic crystal 61a and the emission angle of the diffracted X-ray meet the Bragg diffraction condition. In this way, the wavelength scanning of the characteristic X-ray emitted from the sample S can be performed.

[0040] The spectrometer 6b includes a spectroscopic crystal 61b, a detector 63b, and a slit 64b. The structure of the spectrometer 6b is the same as that of the spectrometer 6a except for the spectroscopic crystal, so it will not be described again. In addition, the structure of each spectrometer is not limited to the above structure, and various known structures can be adopted.

[0041] The control unit 10 includes a CPU (Central Processing Unit) 12, a memory 13, and a communication interface (I / F) 14. The memory 13 includes a ROM (Read Only Memory) and a RAM (Random Access Memory) not shown in the figure. The CPU 12 expands the program stored in the ROM in the RAM, etc. and executes the program. The program stored in the ROM is a program that describes the processing process of the control unit 10. Various tables (mappings) used in various operations are also stored in the ROM. The control unit 10 performs various processes in the EPMA 100 according to these programs and tables. Regarding the processing, it is not limited to software-based, and it can also be executed using dedicated hardware (electronic circuits). Such software is sometimes stored in a flash memory (not shown) in advance. In addition, the software is sometimes stored in a non-transitory recording medium and circulated as a program product.

[0042] In addition, the recording medium is not limited to DVD-ROM, CD-ROM, FD (Flexible Disk), hard disk, and can also be a medium that carries the program in a fixed manner, such as a magnetic tape, a cassette tape, an optical disk (MO (Magnetic Optical Disc: magneto-optical disk) / MD (Mini Disc: mini optical disk) / DVD (Digital Versatile Disc: digital versatile disk)), an optical card, a mask ROM, an EPROM (Electronically Programmable Read-Only Memory: Electrically Programmable Read-Only Memory), an EEPROM (Electronically Erasable Programmable Read-Only Memory: Electrically Erasable Programmable Read-Only Memory), a semiconductor memory such as a flash ROM, etc. In addition, the recording medium is a non-transitory medium that can read the program, etc. on a computer.

[0043] The program mentioned here includes not only programs that can be directly executed by the CPU, but also programs in source program form, compressed programs, encrypted programs, and the like.

[0044] The software is sometimes provided as a program product that can be downloaded through a so-called information provider connected to the Internet. Such software is temporarily stored in the flash memory after being read out from its storage medium by an IC card reader / writer (not shown) or other reading device, or downloaded via the communication I / F 14. The software is read out from the flash memory by the CPU 12 and stored in the flash memory in the form of an executable program. The CPU 12 executes the program.

[0045] The communication I / F 14 is connected to a communication network such as the Internet, and the EPMA 100 exchanges data with external devices via the communication I / F 14. The external devices include a sample component estimation apparatus 200.

[0046] Although not shown in the figure, the data processing unit 11 also includes a CPU, a memory, and an input / output buffer. The data processing unit 11 creates a characteristic X-ray spectrum (hereinafter also referred to as spectrum data) of the analysis object. The data processing unit 11 may be integrally formed with the control unit 10.

[0047] The deflection yoke control unit 8 controls the drive current supplied to the deflection yoke 2 according to the instruction from the control unit 10. By controlling the drive current according to a predetermined drive current pattern (magnitude and change speed), the irradiation position of the electron beam E can be scanned on the sample S at a desired scanning speed.

[0048] <Structure of Sample Composition Estimation Device>

[0049] (Hardware Configuration of Sample Composition Estimation Device)

[0050] The sample component estimation device 200 includes a CPU 20, a memory 22, a communication I / F 24, an operation unit 26, and a display unit 28. The memory 22 includes a ROM and a RAM (not shown).

[0051] The CPU 20 expands the program stored in the ROM in the RAM or the like and executes the program. The program stored in the ROM includes a program (sample composition estimation program) that describes the processing procedures of the sample composition estimation device 200. Various tables (mappings) used in various operations are also stored in the ROM. The sample composition estimation device 200 uses the characteristic X-ray spectrum (spectral data) acquired by the EPMA 100 in accordance with these programs and tables to perform processing for estimating the chemical bonding state of the elements contained in the sample S. The estimation processing is not limited to being based on software, and can also be performed using dedicated hardware (electronic circuits).

[0052] The communication I / F 24 is connected to a communication network such as the Internet. The sample component estimation apparatus 200 exchanges data with external devices including the EPMA 100 via the communication I / F 24 .

[0053] The operation unit 26 is an input device for the user to provide various instructions to the sample component estimation device 200, and is composed of, for example, a mouse or a keyboard. The display unit 28 is an output device for providing various information to the user, and is composed of, for example, a display having a touch panel that the user can operate. In addition, the touch panel can also be used as the operation unit 26.

[0054] (Functional Structure of Sample Composition Estimation Device)

[0055] Figure 2 : is a block diagram showing an example of the functional configuration of the sample component estimation device 200 .

[0056] Reference Figure 2 As main functional structures, the sample component estimation device 200 includes an input unit 31, a spectrum acquisition unit 30, a spectrum shaping unit 32, a compound species identification unit 34, a display control unit 36, and a display unit 28. The sample component estimation device 200 also includes learning units 38 and 42, a spectrum generator model 40, and an identifier model 44. These functions can be realized, for example, by the CPU 20 of the sample component estimation device 200 executing a program stored in the memory 22. In addition, some or all of these functions can also be configured to be realized by hardware.

[0057] The input unit 31 receives information related to the element to be analyzed. Specifically, the input unit 31 receives information related to the element to be analyzed from the operation unit 26 (see Figure 1 ) When information indicating the analysis target element specified by the user is obtained, the input wavelength range of the spectrum data is determined according to the specified analysis target element. The "input wavelength range" is equivalent to the wavelength range of the characteristic X-ray spectrum required to estimate the chemical bonding state of the analysis target element contained in the sample S. The input wavelength range varies depending on the analysis target element. Figure 3 is a table showing an example of the relationship between elements and input wavelength ranges. Figure 3 , the input wavelength ranges of iron (Fe), silicon (Si), and aluminum (Al) are exemplified.

[0058] like Figure 3As shown, for each element, there is a peak of the 1st-order spectral line that should be paid attention to in order to estimate the chemical bonding state. For iron, it is the Lα spectral line and the Lβ spectral line. The input wavelength range is set to include the wavelength of the 1st-order spectral line that should be paid attention to. In the case of iron, the input wavelength range is set to 1.69nm~1.81nm. In the case of silicon, the 1st-order spectral lines that should be paid attention to are the sKα3 spectral line and the sKα4 spectral line, and the input wavelength range is set to 0.705nm~0.710nm. For aluminum, the 1st-order spectral lines that should be paid attention to are the sKα3 spectral line and the sKα4 spectral line, and the input wavelength range is set to 0.825nm~0.832nm.

[0059] In the sample composition estimation device 200, Figure 3 The table shown can be preset and stored in the memory 22. Based on this, when the input unit 31 obtains the analysis target element information, it can refer to Figure 3 The input unit 31 provides information indicating the determined input wavelength range to the spectrum acquisition unit 30.

[0060] The spectrum acquisition unit 30 communicates with the communication interface 24 (see Figure 1 ) receives the characteristic X-ray spectrum (spectrum data) of the sample S from the EPMA 100, and receives information indicating the input wavelength range from the input unit 31. In addition, the acquisition end of the spectrum data is arbitrary. Therefore, in addition to being configured to acquire spectrum data from the EPMA 100, the spectrum acquisition unit 30 may also be configured to acquire spectrum data stored in an external storage device or a server (not shown) provided on the Internet.

[0061] The spectrum acquisition unit 30 extracts spectrum data of an input wavelength range from the spectrum data of the sample S to acquire the spectrum data. For example, when the element to be analyzed is iron, the spectrum acquisition unit 30 acquires spectrum data of a wavelength range of 1.69 nm to 1.81 nm. The spectrum acquisition unit 30 outputs the acquired spectrum data to the spectrum shaping unit 32.

[0062] The spectrum shaping unit 32 shapes the acquired spectrum data. Specifically, the spectrum shaping unit 32 removes high-order spectrum line peak components from the spectrum data using the spectrum generator model 40, thereby generating shaped spectrum data. The high-order spectrum line peak components refer to the peak components of high-order diffraction lines (so-called high-order spectrum lines) contained in the characteristic X-ray spectrum within the input wavelength range. The spectrum shaping unit 32 outputs the generated shaped spectrum data to the compound species identification unit 34.

[0063] The spectrum generator model 40 is a model that has been subjected to a learning process by the learning unit 38 (a learned model). Figure 4Schematically shows an example of the spectrum generator model 40. As the spectrum generator model 40, for example, Figure 4 The U-shaped neural network shown (so-called U-NET).

[0064] Figure 4 The U-shaped neural network shown is that in the convolution layer of the downstream path, the waveform feature is extracted from the input spectrum data. Then, in the deconvolution layer of the upstream path, the spectrum data is restored to its original size in a manner that keeps the waveform feature. In addition, in the upstream path, the data of the downstream layers with the same data size are merged step by step from the deep layer, thereby being able to restore the overall position information while keeping the local features of the waveform intact.

[0065] Return to Figure 2 The compound species identification unit 34 identifies which compound species of the sample S corresponds to among a plurality of compound species containing the analyte element by analyzing the shaped spectrum data. Specifically, the compound species identification unit 34 identifies the compound species of the sample S using the identifier model 44 .

[0066] The classifier model 44 is a learned model that has been subjected to the learning process by the learning unit 42 . Figure 5 4 is a diagram schematically showing an example of a recognizer model 44. As the recognizer model 44, a convolutional neural network (CNN) is typically used. CNN is mainly composed of a convolution layer, a pooling layer, and a fully connected layer. A general CNN has the following structure: after alternately stacking convolution layers and pooling layers, several fully connected layers are stacked.

[0067] The shaped spectrum data is input to the identifier model 44. The convolution layer and the pooling layer extract waveform features from the input shaped spectrum data. The fully connected layer identifies the chemical bonding state (i.e., compound species) of the analysis target element based on the extracted waveform features and outputs the identification result.

[0068] Generally speaking, in a model such as CNN, a learning process is performed in advance using teacher data. In the present embodiment, for example, a set of spectrum data of an input wavelength range preset for each analysis target element and a compound species corresponding to the spectrum data can be used as teacher data to enable the identifier model 44 to learn. The learning process of the model will be described later.

[0069] Return to Figure 2The compound species identification unit 34 inputs the shaped spectrum data to the identifier model 44, and obtains the calculation results of the probabilities corresponding to the plurality of compound species output from the identifier model 44. The compound species identification unit 34 outputs the obtained calculation results of the probabilities as the identification results to the display control unit 36. The display control unit 36 ​​displays the obtained identification results on the display unit 28.

[0070] (Functional structure of the Learning Department)

[0071] exist Figure 2 In the functional structure shown, the spectrum generator model 40 is a learning model that has been subjected to the learning process in the learning unit 38. The spectrum generator model 40 corresponds to an embodiment of the "second learning model", and the learning unit 38 corresponds to an embodiment of the "second learning unit". The identifier model 44 is a learning model that has been subjected to the learning process in the learning unit 42. The identifier model 44 corresponds to an embodiment of the "first learning model", and the learning unit 42 corresponds to an embodiment of the "first learning unit". Figures 6 to 8 , to briefly describe the functional structure of the learning unit 38 and the learning unit 42.

[0072] Figure 6 3 is a block diagram showing an example of the functional configuration of the learning unit 38 and the learning unit 42 .

[0073] Reference Figure 6 The learning unit 38 includes a learning data acquisition unit 50, a preprocessing unit 52, a learning processing unit 54, and an output unit 56. These functions are realized by, for example, the CPU 20 of the sample component estimation device 200 executing a program stored in the memory 22. In addition, some or all of these functions may be realized by hardware.

[0074] The learning data acquisition unit 50 acquires spectrum data of a plurality of compound species containing the element to be analyzed. The plurality of spectrum data includes spectrum data having high-order spectrum line peak components. The learning data acquisition unit 50 acquires spectrum data of an input wavelength range corresponding to the element to be analyzed as learning spectrum data for each of the plurality of spectrum data.

[0075] The pre-processing unit 52 removes the high-order spectral line peak components from the learning spectrum data having the high-order spectral line peak components. Specifically, the pre-processing unit 52 fits the high-order spectral line peak parts using standard function waveforms such as Gauss or Lorentz or their synthetic functions. The pre-processing unit 52 subtracts the fitted waveform data from the original spectrum data, thereby obtaining shaped spectrum data with the high-order spectral line peak components removed.

[0076] Furthermore, by measuring the input wavelength range using a peak height analyzer (PHA), the learning data acquisition unit 50 can obtain spectrum data without high-order line peak components.

[0077] The learning processing unit 54 performs a learning process of the spectrum generator model 40 using the spectrum data obtained by removing the high-order spectrum line peak components by the pre-processing unit 52 and the spectrum data without the high-order spectrum line peak components as correct data. Figure 7 1 is a diagram for explaining the procedure of the learning process of the spectrum generator model 40 executed by the learning processing unit 54 .

[0078] like Figure 7 As shown, when the learning spectrum data input to the spectrum generator model 40 includes high-order spectral line peak components, the learning processing unit 54 uses the learning spectrum data as input data, and uses spectrum data obtained by removing the high-order spectral line peak components from the learning spectrum data as correct data to cause the spectrum generator model 40 to learn. On the other hand, when the learning spectrum data input to the spectrum generator model 40 does not include high-order spectral line peak components, the learning processing unit 54 uses the learning spectrum data as input data, and uses the learning spectrum data itself as correct data to cause the spectrum generator model 40 to learn. The learning processing unit 54 obtains the error (loss) between the output data of the spectrum generator model 40 when the learning spectrum data is input and the correct data, and optimizes the spectrum generator model 40 to reduce the error.

[0079] In this way, in the learning unit 38, the teacher data is used to perform the learning process of the spectrum generator model 40, so that the spectrum generator model 40 can generate shaped spectrum data in which the high-order spectrum line peak components are removed from the input spectrum data. Figure 7 As shown, the teacher data is composed of the following groups: a group of spectrum data containing high-order spectral line peak components and its shaped spectrum data; and a group of spectrum data not containing high-order spectral line peak components and shaped spectrum data without removing the high-order spectral line peak components from the spectrum data.

[0080] Return to Figure 6 The learning unit 42 includes a learning data acquisition unit 60 and a learning processing unit 62. The learning data acquisition unit 60 acquires spectrum data of a plurality of compound species of known composition. Each of the plurality of spectrum data is shaped spectrum data from which high-order spectrum line peak components have been removed. The learning data acquisition unit 60 acquires spectrum data of an input wavelength range corresponding to the element to be analyzed as learning spectrum data for each of the plurality of spectrum data.

[0081] The learning processing unit 62 performs a learning process of the classifier model 44 using the learning spectrum data. Figure 84 is a diagram for explaining the process of learning the discriminator model 44 performed by the learning processing unit 62. Figure 8 As shown, the learning processing unit 62 uses the acquired multiple spectrum data as input data, and uses the label values ​​of the compound species corresponding to the multiple spectrum data as correct data to learn the discriminator model 44. At this time, the learning processing unit 62 obtains the error (loss) between the output data of the discriminator model 44 when the learning spectrum data is input and the correct data, and optimizes the discriminator model 44 to reduce the error.

[0082] In this way, in the learning unit 42, the teacher data is used to perform the learning process of the identifier model 44, so that the identifier model 44 can calculate the probability of which of the multiple chemical species is matched based on the input spectrum data. Figure 8 As shown, the teacher data is composed of a combination of spectrum data of a compound species with a known composition and the composition of the compound species.

[0083] <Example>

[0084] Next, an example of sample composition estimation processing using the sample composition estimation device 200 according to this embodiment is described. In this example, the analysis target element is iron, and the chemical bonding state of iron contained in the sample is estimated. Assume the following case: As the chemical bonding state of iron, iron (Fe), iron oxide (wüstite (FeO), hematite (Fe 2 O 3 ), magnetite (Fe 3 O 4 )) and iron sulfide (FeS 2 ) were identified in total of 5 compounds.

[0085] Fig. 9 : is a diagram showing an example of a characteristic X-ray spectrum of a sample measured by EPMA 100. Fig. 9 In the figure, the horizontal axis represents the wavelength of the characteristic X-ray spectrum, and the vertical axis represents the signal intensity of the characteristic X-ray spectrum. Fig. 9 , a characteristic X-ray spectrum in the wavelength range of 1.69 nm to 1.81 nm is shown. The characteristic X-ray spectrum includes the first-order lines of the Lα line and the Lβ line.

[0086] like Fig. 9 As shown in the figure, between the Lα line and the Lβ line, there are high-order diffraction lines (high-order lines) of the Kα line. The high-order lines of the Kα line are close to the Lα line and the Lβ line, so it is not appropriate to evaluate the waveform of the 1st-order line. Therefore, it is necessary to remove the high-order lines of the Kα line from the characteristic X-ray spectrum.

[0087] Fig.10This is a diagram showing an example of a characteristic X-ray spectrum emitted from an iron compound. Fig.10 In, with Fig. 9 Likewise, a characteristic X-ray spectrum in the wavelength range of 1.69 nm to 1.81 nm is shown.

[0088] exist Fig.10 In the figure, iron (Fe), magnetite (Fe 3 O 4 ) and hematite (Fe 2 O 3 ) When comparing these spectra, it can be seen that the peak wavelength, half-peak width and other waveform shapes of the spectrum and the peak intensity change depending on the chemical bonding state of iron.

[0089] exist Fig.10 In the figure, the characteristic X-ray spectra of two samples (sample A, sample B) are shown in an overlapping manner corresponding to the characteristic X-ray spectra of the compound species with a known composition. When the characteristic X-ray spectra are used to estimate the chemical bonding state of each sample, the shape of the waveform and the change in peak intensity that are different according to the chemical bonding state are used. In other words, the chemical bonding state of iron contained in each sample can be estimated by comparing these parameters between the characteristic X-ray spectrum of the known composition and the characteristic X-ray spectrum of the sample. This method is generally referred to as a "state analysis" method.

[0090] The sample composition estimation process according to this embodiment is composed of the following processes: using a characteristic X-ray spectrum of a known composition to learn a model (spectrum generator model 40 and identifier model 44); and using the learned model to identify the chemical bonding state of iron in the sample S. Fig.11 and Fig.12 To illustrate the process of each treatment.

[0091] (Model learning process)

[0092] Fig.11 This is a flowchart for explaining the procedure of the learning process based on the model of this embodiment. Fig.11 The flowchart is mainly composed of the learning unit 38 and the learning unit 42 (refer to Figure 2 )implement.

[0093] Reference Fig.11 The learning unit 38 obtains spectrum data of multiple compound species containing the analysis target element (iron) as learning spectrum data through step S100. The learning unit 38 obtains spectrum data of the input wavelength range (1.69nm to 1.81nm) corresponding to the analysis target element (iron) for each spectrum data. The acquired spectrum data contains peaks of the Lα spectrum line and the Lβ spectrum line (refer to Fig. 9). In addition, the spectrum data includes peaks of high-order spectral lines of Kα spectral lines.

[0094] The learning unit 38 removes the high-order spectral line peak components from the acquired learning spectrum data through step S110. In step S110, the learning unit 38 fits the wavelength range in the spectrum data that corresponds to the high-order spectral line peak of the Kα spectrum line with a standard waveform based on the Gauss function or the Lorentz function or its synthetic function. The learning unit 38 subtracts the waveform data obtained by fitting from the original spectrum data, thereby removing the high-order spectral line peak components of the Kα spectrum line.

[0095] The learning unit 38 also obtains a plurality of spectrum data not containing high-order spectrum line peak components as spectrum data for learning in step S120. In step S120, the learning unit 38 obtains a plurality of spectrum data not containing high-order spectrum line peak components for the above-mentioned plurality of compound species by screening the input energy of the X-ray signal by a wave height analyzer (PHA).

[0096] When learning spectrum data with high-order spectral line peak components removed and learning spectrum data without high-order spectral line peak components are obtained through steps S100 to S120, the learning unit 38 enters step S130 and uses these learning spectrum data to perform learning processing of the spectrum generator model 40. In step S130, the learning unit 38 regards the multiple spectrum data with high-order spectral line peak components removed through step S110 as correct data for the multiple spectrum data obtained through step S100. For the multiple spectrum data without high-order spectral line peak components obtained through step S120, the multiple spectrum data themselves are regarded as correct data. The learning unit 38 uses these correct data to perform learning processing of the spectrum generator model 40.

[0097] Next, the learning unit 38 inputs the multiple spectrum data (i.e., spectrum data containing high-order spectrum line peak components) acquired in step S100 to the spectrum generator model 40 after learning. Thus, multiple shaped spectrum data with high-order spectrum line peak components removed can be obtained from the spectrum generator model 40. The learning unit 38 outputs the multiple shaped spectrum data output from the spectrum generator model 40 to the learning unit 42.

[0098] The learning unit 42 acquires a plurality of shaped spectrum data as learning data. The learning unit 42 uses the label values ​​of the compound species corresponding to the acquired plurality of shaped spectrum data as correct data in step S150 to perform a learning process of the classifier model 44. Thus, a learned classifier model 44 can be obtained.

[0099] (Chemical bonding state recognition processing)

[0100] Fig.12This is a flowchart for explaining the procedure of the identification process of the chemical bonding state of the analysis target element in the sample S according to the present embodiment. Fig.12 The flowchart is composed of a spectrum acquisition unit 30, a spectrum shaping unit 32, a compound species identification unit 34 and a display control unit 36 ​​(refer to Figure 2 )implement.

[0101] Reference Fig.12 In step S10 , the spectrum acquisition unit 30 acquires the spectrum data of the sample S from the EPMA 100 (or an external storage device).

[0102] The input unit 31 refers to the analysis target element specified by the user in step S20. Figure 3 The input unit 31 provides information indicating the determined input wavelength range to the spectrum acquisition unit 30.

[0103] In step S30 , the spectrum acquisition unit 30 extracts spectrum data in the input wavelength range (1.69 nm to 1.81 nm) corresponding to the analysis target element (iron) from the spectrum data of the sample S and acquires the spectrum data. The spectrum acquisition unit 30 outputs the acquired spectrum data to the spectrum shaping unit 32 .

[0104] The spectrum shaping unit 32 inputs the acquired spectrum data of the sample S to the spectrum generator model 40 in step S40. The spectrum shaping unit 32 acquires the shaped spectrum data output from the spectrum generator model 40. The spectrum shaping unit 32 outputs the acquired shaped spectrum data to the compound species identification unit 34.

[0105] The compound species identification unit 34 inputs the acquired shaped spectrum data to the identifier model 44 in step S50. The compound species identification unit 34 acquires the calculation result of the probability of matching each of the plurality of compound species output from the identifier model 44. The compound species identification unit 34 outputs the acquired calculation result of the probability to the display control unit 36 ​​as the identification result.

[0106] The display control unit 36 ​​displays the acquired recognition result on the display unit 28 in step S60 . Fig.13 FIG. 2 is a diagram showing an example of the recognition result displayed on the display unit 28. Fig.13 As shown, the probability of which of the plurality of compound species the sample S corresponds to is shown in a table on the display unit 28. The table shows the probability for each compound species. Fig.13 In the example, among the five iron compounds, magnetite (Fe 3 O 4 ) has the highest probability. Therefore, the user can judge that the sample S is related to magnetite (Fe 3 O4 ) is likely to be consistent.

[0107] As described above, according to the sample composition estimation device according to the present embodiment, the user does not need to perform complicated analysis condition settings and spectrum waveform analysis, and thus can simply and easily estimate the chemical bonding state of elements in the sample.

[0108] [Other structural examples]

[0109] Next, other configuration examples of the sample component estimation device and the analysis system according to the present embodiment will be described.

[0110] (1) In the above-mentioned embodiment, the sample component estimation device 200 is described as having a configuration including the learning units 38 and 42 that perform learning processing for the spectrum generator model 40 and the identifier model 44. However, Fig.14 As shown, a configuration may be adopted in which the learning process is executed in a learning device 300 provided outside the sample component estimation device 200 .

[0111] Fig.14 2 is a diagram showing the overall configuration of a first modified example of the analysis system according to the present embodiment. Fig.14 , according to the analysis system of the first modification example and Figure 1 The difference from the analysis system shown is that the learning device 300 is provided. The learning device 300 has a communication I / F (not shown) and can exchange data with the sample component estimation device 200 via the communication I / F.

[0112] The learning device 300 includes a learning unit 38 and a learning unit 42. The learning unit 38 and the learning unit 42 include Figure 6 The learning device 300 can perform a learning process of the spectrum generator model 40 and the identifier model 44 and provide the learned spectrum generator model 40 and the identifier model 44 to the sample component estimation device 200 .

[0113] In the first modification, the sample component estimation device 200 includes a transfer learning unit 46 instead of the learning unit 38 and the learning unit 42. The transfer learning unit 46 is configured to perform transfer learning, which is to transfer the knowledge of the discriminator model 44 learned by the learning device 300 to learn a new model. For example, when the compound species of the analysis target element specified by the user has not been learned in the discriminator model 44, the transfer learning unit 46 performs the learning process using the discriminator model 44 provided by the learning device 300 as an initial value. In this way, the discriminator model 44 specifically for the analysis target element of each user can be constructed.

[0114] (2) Fig.152 is a diagram showing the overall configuration of a second modified example of the analysis system according to the present embodiment. Fig.15 , according to the analysis system of the second modification example and Figure 1 Compared with the analysis system shown in FIG. 1 , the spectrum data acquisition terminal of the sample component estimation apparatus 200 is different. The sample component estimation apparatus 200 can acquire the spectrum data of the sample S from a server 310 connected to a communication network such as the Internet 330 .

[0115] In the second modification, a plurality of X-ray analysis devices 100 are connected to the Internet 330. The spectrum data of the sample acquired by each X-ray analysis device 100 is accumulated in the server 310 via the Internet 330. The learning device 300 is configured to be connected to the server 310, and to perform the learning process of the spectrum generator model 40 and the identifier model 44 using the plurality of spectrum data accumulated in the server 310. The learning device 300 saves the learned spectrum generator model 40 and the identifier model 44 to the server 310. The server 310 is configured to manage the spectrum data acquired by the plurality of X-ray analysis devices 100, and to manage the learned spectrum generator model 40 and the identifier model 44.

[0116] The sample composition estimation device 200 can obtain the learned spectrum generator model 40 and the discriminator model 44 by accessing the server 310 via the Internet 330. When the sample composition estimation device 200 obtains the spectrum data of the sample from the server 310, it uses the spectrum generator model 40 and the discriminator model 44 to perform the estimation process of the chemical bonding state of the analysis target element contained in the sample (see Fig.12 ).

[0117] In the second modification, the learning device 300 can enhance the learning model by adding analysis target elements and / or compound species using spectrum data appropriately stored in the server 310. Thus, the server 310 can provide the model enhanced by the learning device 300 to the sample composition estimation device 200.

[0118] [Way]

[0119] Those skilled in the art will appreciate that the above-described multiple exemplary embodiments are specific examples of the following aspects.

[0120] (Item 1) A method for estimating sample composition according to one embodiment includes the following steps: obtaining a spectrum of a sample measured by a wavelength dispersion type X-ray analyzer; determining an analysis target element of the sample and an input wavelength range corresponding to the analysis target element; and inputting a spectrum of the input wavelength range in the spectrum of the sample into a first learning completion model to estimate a chemical bonding state of the analysis target element in the sample.

[0121] According to the sample composition estimation method described in Item 1, when the sample state is analyzed using the spectrum of the sample measured by an X-ray analyzer, the user does not need to set complicated analysis conditions and analyze the spectrum, so the chemical bonding state of the element to be analyzed can be simply and easily estimated.

[0122] (Item 2) The sample composition estimation method described in Item 1 further includes the following steps: obtaining a shaped spectrum by removing high-order line peak components from the spectrum in the input wavelength range. The step of estimating the chemical bonding state includes the following steps: inputting the shaped spectrum into the first learning completion model.

[0123] According to the sample composition estimation method described in Item 2, the state analysis of the sample is performed using the shaped spectrum from which high-order line peak components are removed, thereby making it possible to estimate the chemical bonding state of the analysis target element in the sample with high accuracy.

[0124] (Item 3) In the sample component estimation method described in Item 2, the step of acquiring a shaped spectrum includes the following steps: inputting a spectrum in an input wavelength range to a second learned model, and outputting a shaped spectrum from the second learned model.

[0125] According to the sample component estimation method described in Item 3, the user does not need to set conditions for removing high-order line peak components, and thus the shaped spectrum can be acquired simply and easily.

[0126] (Item 4) In the sample component estimation method described in Items 1 to 3, the first learned model is created by a learning process using a combination of a spectrum of a compound species with a known composition and the composition of the compound species as teacher data.

[0127] According to the sample component estimation method described in Item 4, the accuracy of the state analysis can be improved by optimizing the first learning completed model using the teacher data.

[0128] (Item 5) In the sample component estimation method described in Item 3, the second learning completion model is produced by using the following group as learning processing of teacher data: a group of a first spectrum including a high-order spectral line peak component and a shaped spectrum in which the high-order spectral line peak component is removed from the first spectrum; and a group of a second spectrum not including a high-order spectral line peak component and a shaped spectrum in which the high-order spectral line peak component is not removed from the second spectrum.

[0129] According to the sample component estimation method described in Item 5, the second learned model is optimized using teacher data, thereby making it possible to accurately generate a shaped spectrum.

[0130] (Item 6) In the sample component estimation method described in Item 2, the step of obtaining a shaped spectrum includes the following steps: subtracting waveform data obtained by fitting the high-order spectral line peak component from the spectrum of the input wavelength range, thereby obtaining a shaped spectrum.

[0131] According to the sample component estimation method described in Item 6, a shaped spectrum can be accurately generated.

[0132] (Item 7) A sample composition estimation device involved in one embodiment comprises: a spectrum acquisition unit, which acquires the spectrum of the sample measured by a wavelength dispersion type X-ray analysis device; an input unit, which determines the analysis object element of the sample and the input wavelength range corresponding to the analysis object element; a compound species identification unit, which inputs the spectrum of the input wavelength range in the spectrum of the sample into a first learning completion model to estimate the chemical bonding state of the analysis object element in the sample; and a display unit, which displays the estimated chemical bonding state.

[0133] According to the sample composition estimation device described in Item 7, when using the spectrum of the sample measured by the X-ray analyzer to perform sample state analysis, the user does not need to set complicated analysis conditions and analyze the spectrum, so the chemical bonding state of the element to be analyzed can be simply and easily estimated.

[0134] (Item 8) The sample component estimation device described in Item 7 further includes a first learning unit that creates a first learned model by performing a learning process using a combination of a spectrum of a compound species with a known composition and the composition of the compound species as teacher data.

[0135] According to the sample component estimation device described in Item 8, the accuracy of the state analysis can be improved by optimizing the first learned model using the teacher data.

[0136] (Item 9) The sample component estimation device described in Item 7 or 8 further includes a spectrum shaping unit that removes high-order line peak components from the spectrum in the input wavelength range to obtain a shaped spectrum. The compound species identification unit inputs the shaped spectrum into the first learned model.

[0137] According to the sample composition estimation device described in Item 9, by performing state analysis of the sample using the shaped spectrum from which high-order line peak components have been removed, the chemical bonding state of the analysis target element in the sample can be estimated with high accuracy.

[0138] (Item 10) The sample component estimation device described in Item 9 further includes a second learning unit that creates a second learned model by performing a learning process using the following group as teacher data: a group of a first spectrum including a high-order spectral line peak component and a shaped spectrum in which the high-order spectral line peak component is removed from the first spectrum; and a group of a second spectrum not including a high-order spectral line peak component and a shaped spectrum in which the high-order spectral line peak component is not removed from the second spectrum. The spectrum shaping unit is configured to input a spectrum in an input wavelength range into the second learned model and output a shaped spectrum from the second learned model.

[0139] According to the sample component estimation device described in Item 10, the second learned model is optimized using the teacher data, thereby making it possible to accurately generate a shaped spectrum.

[0140] (Item 11) A sample composition estimation program involved in one method causes a computer to perform the following steps: obtaining a spectrum of a sample measured by a wavelength dispersion type X-ray analysis device; determining the analysis target element of the sample and the input wavelength range corresponding to the analysis target element; and inputting the spectrum of the input wavelength range in the X-ray spectrum of the sample into a first learning completion model to estimate the chemical bonding state of the analysis target element in the sample.

[0141] According to the sample composition estimation procedure described in Item 11, when the sample state is analyzed using the spectrum of the sample measured by an X-ray analyzer, the user does not need to set complicated analysis conditions and analyze the spectrum, so the chemical bonding state of the element to be analyzed can be simply and easily estimated.

[0142] (Item 12) The sample composition estimation procedure described in Item 11 further includes the following steps: obtaining a shaped spectrum by removing high-order line peak components from the spectrum in the input wavelength range. The step of obtaining the shaped spectrum includes the following steps: inputting the spectrum in the input wavelength range into the second learning completion model, and outputting the shaped spectrum from the second learning completion model. The step of estimating the chemical bonding state includes the following steps: inputting the shaped spectrum into the first learning completion model.

[0143] According to the sample composition estimation procedure described in Item 12, the state analysis of the sample is performed using the shaped spectrum from which the high-order line peak components are removed, thereby being able to estimate the chemical bonding state of the analysis target element in the sample with high accuracy. In addition, since the user does not need to set the conditions for removing the high-order line peak components, the shaped spectrum can be obtained simply and easily.

[0144] (Item 13) A learning method for generating a first learning completed model and a second learning completed model used in the sample component estimation method described in Item 3, the learning method comprising: producing a first learning completed model by using a group of spectra of compound species with known compositions and the composition of the compound species as learning processing of teacher data; and producing a second learning completed model by using the following group as learning processing of teacher data: a group of a first spectrum including a high-order spectral line peak component and a shaped spectrum with the high-order spectral line peak component removed from the first spectrum; and a group of a second spectrum not including a high-order spectral line peak component and a shaped spectrum with the high-order spectral line peak component not removed from the second spectrum.

[0145] According to the learning method described in Item 13, the accuracy of state analysis can be improved by optimizing the first learned model using the teacher data. Also, the shaped spectrum can be accurately generated by optimizing the second learned model using the teacher data.

[0146] (Item 14) A learning program for generating a first learning completed model and a second learning completed model used in the sample composition estimation program described in Item 12, the learning program causing a computer to perform the following steps: producing a first learning completed model by using a group of a spectrum of a compound species with a known composition and the composition of the compound species as learning processing of teacher data; and producing a second learning completed model by using the following group as learning processing of teacher data: a group of a first spectrum containing a high-order spectral line peak component and a shaped spectrum with the high-order spectral line peak component removed from the first spectrum; and a group of a second spectrum not containing a high-order spectral line peak component and a shaped spectrum with the high-order spectral line peak component not removed from the second spectrum.

[0147] According to the learning program described in Item 14, the accuracy of state analysis can be improved by optimizing the first learned model using the teacher data. Also, the shaped spectrum can be accurately generated by optimizing the second learned model using the teacher data.

[0148] In addition, regarding the above-mentioned embodiment and modification examples, it is considered from the beginning of the application that the configurations described in the embodiment can be appropriately combined within the range that no inconvenience or contradiction occurs, including combinations not described in the specification.

[0149] Although the embodiments of the present invention have been described, it should be understood that the embodiments disclosed this time are illustrative in all aspects and are not restrictive. The scope of the present invention is indicated by the claims, and it is intended to include all modifications within the meaning and scope equivalent to the claims.

Claims

1. A method for estimating sample composition, The following steps are involved: preparing a first learning completion model, the first learning completion model inputting spectrum data and outputting probabilities that the spectrum data and the chemical species are respectively consistent; acquiring a sample spectrum of the sample measured by a wavelength dispersive X-ray analyzer; Determining an analysis target element in the sample and an input wavelength range corresponding to the analysis target element; Extracting spectrum data of the input wavelength range from the sample spectrum to obtain a partial spectrum of the sample spectrum; as well as by inputting the partial spectrum within the input wavelength range in the sample spectrum into the first learned model, thereby estimating the chemical bonding state of the analysis target element in the sample, The method further comprises the following steps: obtaining a shaped spectrum by removing high-order spectral line peak components from the partial spectrum within the input wavelength range, The step of estimating the chemical bonding state of the element to be analyzed in the sample includes the steps of: inputting the shaped spectrum into the first learned model, The step of acquiring a shaped spectrum includes the following steps: inputting the partial spectrum within the input wavelength range into a second learned model, and outputting the shaped spectrum from the second learned model.

2. The sample composition estimation method according to claim 1, in, The first learned model is created by performing a learning process using a combination of a spectrum of a compound species with a known composition and the composition of the compound species as teacher data.

3. The sample composition estimation method according to claim 1 or 2, in, The second learning completion model is produced by using the following group as learning processing of teacher data: a group of a first spectrum including the high-order spectral line peak component and the shaped spectrum with the high-order spectral line peak component removed from the first spectrum; and a group of a second spectrum not including the high-order spectral line peak component and the shaped spectrum without removing the high-order spectral line peak component from the second spectrum.

4. The sample composition estimation method according to claim 1 or 2, in, The step of acquiring a shaped spectrum includes the following steps: subtracting waveform data obtained by fitting the high-order spectral line peak components from the partial spectrum within the input wavelength range, thereby acquiring the shaped spectrum.

5. A sample composition estimation device comprising: a preparation unit that prepares a first learning model that receives spectrum data as input and outputs probabilities that the spectrum data corresponds to chemical compounds; a spectrum acquisition unit that acquires a spectrum of a sample measured by a wavelength dispersion type X-ray analyzer; An input unit that determines an analysis target element of the sample and an input wavelength range corresponding to the analysis target element; a compound species identification unit that inputs a spectrum in the input wavelength range among the spectrum of the sample into the first learned model to estimate a chemical bonding state of the analysis target element in the sample; a display unit that displays the estimated chemical bonding state; as well as a spectrum shaping unit, wherein the spectrum shaping unit obtains a shaped spectrum by removing high-order spectral line peak components from the spectrum in the input wavelength range, wherein the compound species identification unit inputs the shaped spectrum into the first learned model, The spectrum shaping unit is configured to input the spectrum in the input wavelength range to a second learned model, and output the shaped spectrum from the second learned model.

6. The sample composition estimation device according to claim 5, in, The method further includes a first learning unit that creates the first learned model by performing a learning process using a combination of a spectrum of a compound species with a known composition and the composition of the compound species as teacher data.

7. The sample composition estimation device according to claim 5 or 6, in, The sample component estimation device also has a second learning unit, which produces the second learning completion model by using the following group as learning processing for teacher data: a group of a first spectrum including the high-order spectral line peak component and the shaped spectrum with the high-order spectral line peak component removed from the first spectrum; and a group of a second spectrum not including the high-order spectral line peak component and the shaped spectrum without removing the high-order spectral line peak component from the second spectrum.

8. A non-transitory computer-readable recording medium having recorded thereon a sample component estimation program executed by a computer, the sample component estimation program being configured to cause the computer to execute the following steps: preparing a first learning completion model, the first learning completion model inputting spectrum data and outputting probabilities that the spectrum data and the chemical species are respectively consistent; acquiring a sample spectrum of the sample measured by a wavelength dispersive X-ray analyzer; Determining an analysis target element in the sample and an input wavelength range corresponding to the analysis target element; Extracting spectrum data of the input wavelength range from the sample spectrum to obtain a partial spectrum of the sample spectrum; as well as by inputting the partial spectrum within the input wavelength range in the sample spectrum into the first learned model, thereby estimating the chemical bonding state of the analysis target element in the sample, The sample component estimation program is used to cause the computer to further perform the following steps: obtaining a shaped spectrum by removing high-order spectral line peak components from the partial spectrum within the input wavelength range, The step of acquiring the shaped spectrum comprises the following steps: inputting the partial spectrum within the input wavelength range into a second learning completion model, outputting the shaped spectrum from the second learning completion model, The step of estimating the chemical bonding state of the analysis target element in the sample includes the step of inputting the shaped spectrum into the first learned model.

9. A learning method for generating the first learned model and the second learned model used in the sample composition estimation method according to claim 1, wherein the learning method The following steps are involved: The first learning completed model is produced by a learning process using a combination of a spectrum of a compound species with a known composition and the composition of the compound species as teacher data; as well as The second learning completion model is produced by using the following group as learning processing of teacher data: a group of a first spectrum including the high-order spectral line peak component and the shaped spectrum with the high-order spectral line peak component removed from the first spectrum; and a group of a second spectrum not including the high-order spectral line peak component and the shaped spectrum without removing the high-order spectral line peak component from the second spectrum.

10. A non-transitory computer-readable recording medium having recorded thereon a learning program for generating the first learned model and the second learned model used in the sample composition estimation program stored in the non-transitory computer-readable recording medium according to claim 8, in, The learning program causes the computer to perform the following steps: The first learning completed model is produced by a learning process using a combination of a spectrum of a compound species with a known composition and the composition of the compound species as teacher data; as well as The second learning completion model is produced by using the following group as learning processing of teacher data: a group of a first spectrum including the high-order spectral line peak component and the shaped spectrum with the high-order spectral line peak component removed from the first spectrum; and a group of a second spectrum not including the high-order spectral line peak component and the shaped spectrum without removing the high-order spectral line peak component from the second spectrum.

Images