Spectral analysis device, spectral analysis method, spectral analysis program, learning device, learning method, and learning program

Abstract

The present invention provides a spectroscopic analysis device that accurately determines the concentration of a component to be measured from a measurement spectrum without determining the calibration line of each of the spectroscopic analysis devices, the spectroscopic analysis device being provided with: a spectrum generation unit (51) that generates the spectrum of light that has passed through a sample; a storage unit (52) for storing a reference calibration line created by the reference device (200) and a reference spectrum generated by the reference device (200) serving as a reference for the spectroscopic analysis device; a correction unit (53) that, on the basis of the result of comparison between a reference spectrum, which is the spectrum of the reference gas generated by the spectrum generation unit (51), corrects a measurement spectrum, which is the spectrum of the sample generated by the spectrum generation unit (51), or a reference calibration line; and a concentration calculation unit (54) that calculates the concentration of the component to be measured on the basis of the reference calibration line and the corrected measurement spectrum, or on the basis of the corrected reference calibration line and the measurement spectrum.

Description

Technical Field

[0001] The present invention relates to a spectroscopic analysis device, a spectroscopic analysis method, a spectroscopic analysis program, a learning device, a learning method, and a learning program. Background Art

[0002] In a Fourier transform type spectroscopic analyzer (FTIR spectroscopic analyzer), as shown in Patent Document 1, by performing multivariate analysis on the absorbance (or transmittance) spectrum of a sample, the concentration of a measurement target component in the sample is calculated.

[0003] Here, multivariate analysis is, for example, a method in which, when the absorbance spectrum of a sample is set to A(λ k ), and the concentration of the measurement target component X is set to c X , an appropriate constant coefficient v X，k is used, and it is obtained by the sum-of-products operation shown in the following Mathematical Formula 1. In addition, the following operation formula is an example, and may include quadratic and cubic higher-order terms related to A(λ k ), and may also include products of A(λ k ) with each other.

[0004] [Mathematical Formula 1] c X = ∑ k v X，k A(λ k )

[0005] Here, v X，k in the above operation formula is a constant parameter that varies depending on the measurement target component X, and is called a "calibration line coefficient".

[0006] Even for the same measurement target component, this calibration line coefficient may vary depending on each FITR analyzer. This is because, even for a gas having the same composition and concentration, the spectral shape output by the interferometer used in the FTIR analyzer varies from one unit to another. Therefore, in order to accurately calculate the concentration of the measurement target component in each FTIR analyzer, a calibration line is created for each FTIR analyzer.

[0007] However, since a calibration line is created for each FTIR analyzer, not only the labor cost and time for creating the calibration line increase, but also the cost of consumables such as gas used for creating the calibration line increases. Prior Art Documents

[0008] Patent Document 1: International Publication No. 2021 / 005900 Summary of the Invention

[0009] Therefore, in view of the above problems, the main object of the present invention is to accurately determine the concentration of the component to be measured in a high-precision manner based on the measured spectrum without determining the calibration lines of the respective spectroscopic analysis devices.

[0010] That is, the spectroscopic analysis device of the present invention irradiates light on a sample to analyze the component to be measured contained in the sample, and is characterized by comprising: a light irradiation unit that irradiates light on the sample; a light detector that detects the light transmitted through the sample to obtain a detection signal; a spectrum generation unit that generates a spectrum of the light transmitted through the sample based on the detection signal; a storage unit that stores a reference calibration line and a reference spectrum, the reference calibration line being made using a reference gas by a reference device that serves as a reference for the spectroscopic analysis device, and the reference spectrum being the spectrum of the reference gas generated by the reference device; a correction unit that compares the spectrum of the reference gas generated by the spectrum generation unit, i.e., the reference spectrum, with the reference spectrum, and corrects the spectrum of the sample generated by the spectrum generation unit, i.e., the measurement spectrum, or the reference calibration line based on the difference in the wave number or half-value width of their peaks; and a concentration calculation unit that calculates the concentration of the component to be measured based on the reference calibration line and the corrected measurement spectrum, or the corrected reference calibration line and the measurement spectrum.

[0011] According to such a spectroscopic analysis device, by comparing the spectrum of the reference gas generated by the spectrum generation unit, i.e., the reference spectrum, with the spectrum of the reference gas generated by the reference device, i.e., the reference spectrum, and correcting the spectrum of the sample generated by the spectrum generation unit, i.e., the measurement spectrum, or the reference calibration line based on the difference in the wave number or half-value width of their peaks, it is possible to accurately determine the concentration of the component to be measured in a high-precision manner without determining the calibration line of the spectroscopic analysis device. In addition, since it is not necessary to determine the calibration line for each spectroscopic analysis device, it is possible to reduce the labor cost and time for making the calibration line, and also reduce the cost of consumables such as gas for making the calibration line.

[0012] As a specific embodiment of the correction unit, preferably, the correction unit compares the reference spectrum with the reference spectrum and corrects the measurement spectrum or the reference calibration line based on at least two of the difference in the wave number of their peaks, the difference in height, and the difference in half-value width.

[0013] As a more specific embodiment of the correction unit, preferably, the correction unit obtains a machine difference parameter and corrects the measurement spectrum or the reference calibration line based on the machine difference parameter, the machine difference parameter being composed of at least two of a wave number axis scaling ratio α obtained from the wave number of the peaks of the reference spectrum and the reference spectrum, an absorbance scaling ratio β obtained from one or both of the height and half-value width of their peaks, and a difference w in the half-value width obtained from the half-value width of their peaks. g constituted by.

[0014] Alternatively, it may be that the correction unit uses a machine learning model obtained in advance to estimate the instrumental difference parameter.

[0015] In addition, the spectral analysis method of the present invention uses a spectral analysis device that irradiates light on a sample to analyze a measurement target component contained in the sample. The spectral analysis method is characterized in that a reference calibration line of a reference gas used by a reference device that is a reference for the spectral analysis device, and a spectrum of the reference gas generated by the reference device are used to compare a reference spectrum, which is a spectrum of the reference gas generated by the spectral analysis device, and the reference spectrum, and based on a difference in a wave number or a half-value width of their peaks, a measurement spectrum, which is a spectrum of the sample generated by the spectral analysis device, or the reference calibration line is corrected. Based on the reference calibration line and the corrected measurement spectrum, or the corrected reference calibration line and the measurement spectrum, a concentration of the measurement target component is calculated.

[0016] Moreover, the spectral analysis program of the present invention is for a spectral analysis device that irradiates light on a sample to analyze a measurement target component contained in the sample. The spectral analysis program is characterized in that a computer functions as a spectrum generation unit, a storage unit, a correction unit, and a concentration calculation unit. The spectrum generation unit generates a spectrum of light transmitted through the sample. The storage unit stores a reference calibration line and a reference spectrum. The reference calibration line is made using a reference gas by a reference device that is a reference for the spectral analysis device. The reference spectrum is a spectrum of the reference gas generated by the reference device. The correction unit compares a reference spectrum, which is a spectrum of the reference gas generated by the spectrum generation unit, and the reference spectrum, and based on a difference in a wave number or a half-value width of their peaks, a measurement spectrum, which is a spectrum of the sample generated by the spectrum generation unit, or the reference calibration line is corrected. The concentration calculation unit calculates a concentration of the measurement target component based on the reference calibration line and the measurement spectrum corrected by the correction unit, or the reference calibration line corrected by the correction unit and the measurement spectrum.

[0017] The learning device of the present invention generates a learning model for obtaining the instrumental difference parameter from the reference spectrum. The learning device is characterized by including: a simulation unit that generates a plurality of virtual spectra based on each of the reference spectrum and a plurality of instrumental difference parameters; and a learning unit that generates the learned model based on a learning data set composed of the plurality of virtual spectra generated by the simulation unit and the plurality of instrumental difference parameters used to generate these virtual spectra.

[0018] In order to generate a machine learning model that takes into account the noise components superimposed on the reference spectrum, preferably, the simulation unit assigns nozzle components to each of the plurality of virtual spectra, and the learning unit generates the learned model based on a learning data set composed of the plurality of virtual spectra to which the noise components are assigned and the plurality of machine difference parameters used to generate these virtual spectra.

[0019] The machine learning method of the present invention generates a learning model for obtaining the machine difference parameters from the reference spectrum, characterized in that, based on each of the reference spectrum and a plurality of machine difference parameters, a plurality of virtual spectra are generated, and the learned model is generated based on a learning data set composed of the generated plurality of virtual spectra and the plurality of machine difference parameters used to generate these virtual spectra.

[0020] Furthermore, the learning program of the present invention is used to generate a learning model for obtaining the machine difference parameters from the reference spectrum, and the learning program is characterized in that it causes a computer to function as a simulation unit and a learning unit. The simulation unit generates a plurality of virtual spectra based on each of the reference spectrum and a plurality of machine difference parameters, and the learning unit generates the learned model based on a learning data set composed of the generated plurality of virtual spectra and the plurality of machine difference parameters used to generate these virtual spectra.

[0021] According to the present invention configured in this way, without obtaining the calibration lines of each spectroscopic analysis device, it is possible to accurately obtain the concentration of the component to be measured from the measured spectrum. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of a spectroscopic analysis device showing one embodiment of the present invention. Figure 2 It is a functional block diagram of an arithmetic processing device in the same embodiment. Figure 3 It is a graph showing the difference in absorbance spectra when measuring the same gas. Figure 4 It is a graph showing a method for calculating the machine difference parameter in the same embodiment. Figure 5 It is a graph showing a method for calculating the machine difference parameter in the same embodiment. Figure 6 It is a graph showing an example of a method for correcting a measured spectrum in the same embodiment. Figure 7 It is a flowchart showing the spectroscopic analysis method in the same embodiment. Figure 8 It is a functional block diagram of an arithmetic processing device in a modified embodiment. Figure 9It is a diagram showing the simulation of the simulation unit in the deformation implementation mode. Detailed implementation mode

[0023] <One implementation mode of the present invention> Next, an implementation mode of the spectroscopic analysis device of the present invention will be described with reference to the drawings. In addition, for any one of the following diagrams, for ease of understanding, appropriate omissions or exaggerated schematic depictions are made. The same reference numerals are assigned to the same constituent elements and the description is appropriately omitted.

[0024] <Device structure> The spectroscopic analysis device 100 of the present implementation mode irradiates light on the sample gas to analyze the measurement target component contained in the sample gas. As Figure 1 shown, it is an analysis device using Fourier transform infrared spectroscopy (FTIR) equipped with an infrared light source 1, an interferometer (spectral splitting unit) 2, a measurement cell 3, a light detector 4, an arithmetic processing device 5, etc.

[0025] The infrared light source 1 emits infrared light with a wide spectrum (continuous light containing light of multiple wavenumbers), for example, a tungsten iodide lamp or a high-brightness ceramic light source is used.

[0026] As shown in the same figure, the interferometer 2 uses a so-called Michelson interferometer, which includes a semi-transmissive semi-reflective mirror (beam splitter) 21, a fixed mirror 22, and a moving mirror 23. The light from the infrared light source 1 incident on the interferometer 2 is split into reflected light and transmitted light by the semi-transmissive semi-reflective mirror 21. One of the lights is reflected by the fixed mirror 22, and the other is reflected by the moving mirror 23, and then returns to the semi-transmissive semi-reflective mirror 21 and is combined and emitted from the interferometer 2. In addition, the infrared light source 1 and the interferometer (spectral splitting unit) 2 constitute a light irradiation unit 10 that irradiates light on the sample gas. The light irradiation unit 10 can have a structure that only includes the infrared light source 1, or can have a structure that includes a laser light source instead of the infrared light source.

[0027] The measurement cell 3 is a transparent cell into which the sample gas is introduced, and the light emitted from the interferometer 2 passes through the exhaust gas in the measurement cell 3 and is guided to the light detector 4. In addition, the spectroscopic analysis device 100 can also measure the sample gas in the atmosphere or the sample gas flowing in a flow path such as a pipe. In this case, the measurement cell 3 is not an essential structure.

[0028] The light detector 4 detects the infrared light passing through the exhaust gas and outputs its detection signal (light intensity signal) to the arithmetic processing device 5. The light detector 4 of the present implementation mode is equipped with an MCT (HgCdTe) detector, but it can also be a light detector having other infrared detection elements.

[0029] The arithmetic processing device 5 is a computer that includes: an analog circuit having a buffer, an amplifier, etc.; a digital circuit having a CPU, a memory, a DSP, etc.; and an A / D converter, etc. between them.

[0030] In addition, the arithmetic processing device 5 causes the CPU and its peripheral devices to cooperate in accordance with the spectral analysis program stored in the memory, thereby calculating a transmission spectrum representing the spectrum of the light transmitted through the sample gas based on the output value of the photodetector 4, calculating an absorbance spectrum from the transmission spectrum, and calculating the concentration of the measurement target component in the sample gas.

[0031] Moreover, the arithmetic processing device 5 of the present embodiment uses the reference calibration line of the reference device 200 (refer to Figure 1 ) to calculate the concentration of the measurement target component in the sample gas, and functions as a control unit 50, a spectrum generation unit 51, a storage unit 52, a correction unit 53, a concentration calculation unit 54, etc., as shown in Figure 2 . Hereinafter, the spectral analysis device 100 of the present embodiment is also referred to as the target device with respect to the reference device 200.

[0032] Here, the reference device 200 is a spectral analysis device that serves as a reference for the spectral analysis device 100 of the present embodiment, and the reference calibration line of the reference device 200 is a calibration line made using a reference gas in the reference device 200. In addition, in the reference device 200, when making the reference calibration line, a reference spectrum that is the spectrum of the reference gas is also generated.

[0033] In addition, the reference gas is a gas with a known composition and concentration. For example, a carbon monoxide (CO) gas with a specified concentration, a methane (CH4) gas with a specified concentration, or a nitric oxide (NO) gas with a specified concentration. Here, the reference gas can be a gas containing multiple components in addition to being a single-component gas. In addition, the types of the reference gas are not limited to the above types.

[0034] Hereinafter, each of the units 51 to 54 of the arithmetic processing device 5 of the spectral analysis device 100 of the present embodiment will be described.

[0035] The control unit 50 controls the operation of the spectral analysis device 100. For example, the control unit 50 controls the reciprocating movement of the moving mirror 23. In addition, the control unit 50 can be provided in the arithmetic processing device 5 or can be provided separately from the arithmetic processing device 5.

[0036] The spectrum generation unit 51 calculates a transmission spectrum based on the output value (detection signal) of the photodetector 4, and calculates an absorbance spectrum (absorption spectrum) from the transmission spectrum. Moreover, the spectrum generation unit 51 sends the calculated absorbance spectrum to the correction unit 53.

[0037] Here, in the spectral analysis device 100 of the present embodiment, the absorbance spectrum of the reference gas obtained by introducing the above-mentioned reference gas into the measurement cell 3 is referred to as the "reference spectrum". In addition, the absorbance spectrum of the sample gas obtained by introducing the sample gas into the measurement cell 3 is referred to as the "measurement spectrum".

[0038] The storage unit 52 stores the reference calibration line produced using the reference gas in the reference device 200 and the reference spectrum of the reference gas generated by the reference device 200.

[0039] The correction unit 53 corrects the measurement spectrum or the reference calibration line based on the comparison result between the reference spectrum and the reference spectrum. The correction unit 53 of the present embodiment corrects the measurement spectrum.

[0040] Here, for the spectrum measured by the FTIR analysis device, even when measuring the same gas, if the devices are different, there will be (1) wavenumber deviation, (2) difference in height (intensity change), and (3) difference in full width at half maximum (refer to Figure 3 ). In addition, these differences are caused by the adjustment of the optical system. For example, the wavenumber deviation is caused by the deviation of the scanning axis of the moving mirror 23 or the deviation of the infrared optical axis, etc. The difference in height (intensity change) is caused by the difference in the optical path length of the measurement cell 3, etc. The difference in full width at half maximum is caused by the expansion around the optical axis of the infrared light, etc.

[0041] These differences can be approximately expressed by the following relational expressions.

[0042] [Mathematical formula 2] A S : A function representing the reference spectrum of the target device (the spectral analysis device of the present embodiment) A M : A function representing the reference spectrum of the reference device (the spectral analysis device serving as the reference) κ k : The wavenumber value at the k-th wavenumber point α: The wavenumber axis scaling ratio between the reference spectrum and the reference spectrum β: The absorbance scaling ratio between the reference spectrum and the reference spectrum w g : The difference (difference) in the full width at half maximum between the reference spectrum and the reference spectrum F gauss (κ; w g ) = exp(-κ 2 / w g 2 ) The symbol * represents the convolution operation (convolution). The symbol *-1 represents a deconvolution operation (deconvolution).

[0043] In addition, in the above relational expression, when the half-value width of the reference spectrum is smaller than that of the reference spectrum, the first formula (Q1) of the former is adopted, and in the opposite case, the second formula (Q2) of the latter is adopted.

[0044] Specifically, the correction unit 53 compares the reference spectrum and the reference spectrum, and corrects the measurement spectrum based on the difference in at least two of the wave number, height, or half-value width of their peaks.

[0045] More specifically, the correction unit 53 obtains a machine difference parameter, which is composed of a wave number axis scaling ratio α obtained from the wave number of the peaks of the reference spectrum and the reference spectrum, an absorbance scaling ratio β obtained from one or both of their peak heights and half-value widths, and a half-value width difference w obtained from the half-value widths of their peaks g constituted by at least two of them. Then, the correction unit 53 corrects the measurement spectrum based on the machine difference parameters α, β, w g to correct the measurement spectrum.

[0046] Figure 4 and Figure 5 represents an example of the operation method of the machine difference parameter of the correction unit 53. Here, the correction unit 53 uses the calculated peak position of the gas component (for example, CO) of the reference gas obtained by simulation such as physical calculation, and obtains the wave number axis scaling coefficient α based on the calculated peak position and the wave number of the peak of the reference spectrum M . In addition, the correction unit 53 obtains the height H M，i and the half-value width w M，i of the i-th peak of the reference spectrum.

[0047] In addition, using the above calculated peak position, the wave number axis scaling coefficient α is obtained based on the calculated peak position and the wave number of the peak of the reference spectrum S . In addition, the correction unit 53 obtains the height H S，i and the half-value width w S，i of the i-th peak of the reference spectrum.

[0048] Furthermore, the correction unit 53 obtains the wave number axis scaling ratio α through the following formula according to the wave number axis scaling coefficient α M of the reference spectrum and the wave number axis scaling coefficient α S of the reference spectrum.

[0049] [Mathematical formula 3]

[0050] In addition, the correction unit 53 is based on the height H M，iand the full width at half maximum w M，i and the height H of the i-th peak of the reference spectrum S，i and the full width at half maximum w S， i, the absorbance scaling ratio β is obtained by the following formula. In addition, C2 is a constant (about 1.144) inherent in the design of the interferometer 2, and w A is the full width at half maximum obtained based on a physical constant specific to absorption depending on the component to be measured.

[0051] [Mathematical formula 4] Here, when w A is large, it can be approximated as follows.

[0052] [Mathematical formula 5] In addition, when w A is small, it can be approximated as follows.

[0053] [Mathematical formula 6]

[0054] Furthermore, the correction unit 53 calculates the difference w M， i in the full width at half maximum of the i-th peak of the reference spectrum and the full width at half maximum w S，i of the i-th peak of the reference spectrum by the following formula to obtain the difference w g of the full width at half maximum.

[0055] [Mathematical formula 7]

[0056] Next, the correction unit 53 uses the machine error parameters α, β, w g obtained in the above manner to correct the measurement spectrum as Figure 6 shown.

[0057] Specifically, the correction unit 53 multiplies the measurement spectrum by the absorbance scaling ratio β. Thereby, the height of the peak of the measurement spectrum is corrected. In addition, the correction unit 53 performs upsampling processing such as spline interpolation on the measurement spectrum multiplied by the absorbance scaling ratio β.

[0058] Next, the correction unit 53 corrects the full width at half maximum of the measurement spectrum. Here, when the full width at half maximum (w M ) of the reference spectrum is greater than the full width at half maximum (w S ) of the reference spectrum (w M > w S ), the correction unit 53 multiplies the measurement spectrum by exp(-κ 2 / wg 2 ) performs convolution (convolution operation). On the other hand, when the half-value width (w M ) of the reference spectrum is smaller than the half-value width (w S ) of the reference spectrum (w M < w S ), the correction unit 53 performs deconvolution (deconvolution operation) on the measurement spectrum and exp(-κ 2 / w g 2 ).

[0059] After that, the correction unit 53 resamples the measurement spectrum with the corrected half-value width by spline interpolation or the like at the wave number ακ k . Thus, the wave number deviation of the measurement spectrum is corrected.

[0060] Through the above processing, the measurement spectrum is corrected using the machine difference parameters α, β, w g . In addition, the correction operation of the measurement spectrum is not limited to the above order, and the height correction of the peak of the measurement spectrum, the half-value width correction of the measurement spectrum, and the wave number deviation correction of the measurement spectrum can also be in any arbitrary order.

[0061] The concentration calculation unit 54 calculates the concentration of the component to be measured based on the measurement spectrum corrected by the correction unit 53 and the reference calibration line.

[0062] Specifically, the concentration calculation unit 54 uses the corrected measurement spectrum A’(λ k ) and the reference calibration line (reference calibration line coefficient) v X，k , and calculates the concentration of the component to be measured c X through the following formula.

[0063] [Mathematical formula 8] c X = ∑ k v X，k A’(λ k ) Instead of calculating the concentration using the measurement spectrum corrected with the machine difference parameters, the above (mathematical formula 6) can also correct the reference calibration line coefficient. The correction of the measurement spectrum using the machine difference parameters can be expressed in the following manner using the matrix M determined by at least two of α, β, w g . Here, the matrix M is a matrix obtained by multiplying at least two of the matrix using α for correcting the wave number deviation, the matrix using β for correcting the intensity deviation, and the matrix using ω g for correcting the half-value width.

[0064] [Mathematical formula 9]

[0065] Using such a matrix M, the corrected calibration line coefficient v' is determined by the following formula X，k . Additionally, the superscript T represents the transpose of the matrix.

[0066] [Mathematical formula 10]

[0067] Thus, through the formula shown below, the concentration can be calculated using the reference calibration line coefficient corrected with the machine error parameter.

[0068] [Mathematical formula 11]

[0069] <Spectral analysis method> Next, with reference to Figure 7 a spectral analysis method using the spectral analysis device 100 of the present embodiment will be described.

[0070] (S0: Storage of reference calibration line, etc.) First, using the reference device 200, a reference calibration line and a reference spectrum of the reference gas as the spectrum of the reference gas are generated. The reference calibration line and the reference spectrum thus obtained are stored in the storage unit 52 of the spectral analysis device 100 (target device) of the present embodiment. Additionally, in the case of having multiple reference gases, a reference calibration line and a reference spectrum are generated for each reference gas and stored in the storage unit 52.

[0071] (S1: Calibration of spectral analysis device) The spectral analysis device 100 of the present embodiment is calibrated regularly. The calibration of this spectral analysis device 100 does not obtain the calibration line (calibration line coefficient), but obtains the machine error parameters α, β, w g .

[0072] (S1-1: Generation of reference spectrum) Then, the reference gas is introduced into the measurement cell 3 of the spectral analysis device 100 of the present embodiment for spectral analysis. Thus, the reference spectrum as the spectrum of the reference gas is generated by the spectrum generation unit 51. Here, it is not necessary to introduce the reference gas for all the components to be measured to generate the reference spectrum, and the reference gas can also be introduced for at least one of the components to be measured to generate the reference spectrum. In this case, for the components to be measured for which the reference gas is not introduced and not generated, it can be calculated using the reference spectrum of the component with a wavelength region close to that of the component to be measured for which the reference gas is introduced and generated, or it can be calculated using the average value of the reference spectra generated by introducing the reference gas.

[0073] (S1-2: Calculation of machine error parameter) If a reference spectrum is generated by the spectrum generation unit 51, the correction unit 53 calculates the machine difference parameters α, β, and w based on the reference spectrum and the reference spectrum stored in the storage unit 52. g The calculated machine difference parameters α, β, and w g are stored in a memory such as the storage unit 52.

[0074] (S2: Normal measurement) After obtaining the machine difference parameters α, β, and w by the above method, g a normal measurement of the sample gas is performed. (S2-1: Generation of measurement spectrum) Specifically, the sample gas is introduced into the measurement cell 3 of the spectroscopic analysis device 100 of the present embodiment for absorption analysis. Thus, the spectrum generation unit 51 generates a measurement spectrum that is the spectrum of the sample gas.

[0075] (S2-2: Correction of measurement spectrum) The correction unit 53 uses the machine difference parameters α, β, and w g to correct the measurement spectrum generated by the spectrum generation unit 51.

[0076] (S2-3: Calculation of concentration) Then, the concentration calculation unit 54 calculates the concentration of the component to be measured using the measurement spectrum corrected by the correction unit 53 and the reference calibration line (coefficient) stored in the storage unit 52.

[0077] <Effect of the present embodiment> According to the spectroscopic analysis device 100 of the present embodiment configured in this way, the reference spectrum of the reference gas generated by the spectrum generation unit 51 is compared with the spectrum of the reference gas generated by the reference device 200, i.e., the reference spectrum. Based on the difference in the wave number or half-value width of their peaks, the measurement spectrum of the sample generated by the spectrum generation unit 51 is corrected. Therefore, without obtaining the calibration line of the spectroscopic analysis device 100 as the target device, the concentration of the component to be measured can be accurately obtained. In addition, there is no need to obtain the calibration line for each spectroscopic analysis device 100, so the labor cost and time for making the calibration line can be reduced, and the cost of consumables such as gas for making the calibration line can also be reduced.

[0078] <Other embodiments> For example, in the above embodiment, the wave number axis scaling ratio α, absorbance scaling ratio β, and difference in half-value width w as the machine difference parameters g are used to correct the measurement spectrum, but it may also be a configuration in which at least two of them are used to correct the measurement spectrum.

[0079] In the described embodiment, the measured spectrum can be corrected by assuming that the half-value widths of the peaks are equal, or by assuming that the wavenumbers of the peaks are equal, or by assuming that the heights of the peaks are equal.

[0080] In addition, in the described embodiment, when obtaining the wavenumber axis scaling ratio α, the wavenumber axis scaling coefficient α M 、α S is used. This wavenumber axis scaling coefficient α M 、α S is obtained by using the calculated peak position of the gas component (such as CO) of the reference gas obtained by simulation such as physical calculation, but the wavenumber axis scaling ratio α can also be obtained according to the peak interval of the reference spectrum and the peak interval of the reference spectrum without using the calculated peak position.

[0081] Moreover, the correction unit can also use the previously obtained learned model to estimate at least two of the machine difference parameters α, β, w g . This learned model is a model that outputs the machine difference parameters α, β, w g by inputting the reference spectrum and is generated by the learning device through learning.

[0082] As a way of learning, machine learning can be used. As the machine learning device 300 for generating the above-mentioned learned model, as Figure 8 shown, it includes: a simulation unit 301 that generates a plurality of virtual spectra based on each of the reference spectrum or the standard spectrum (ideal spectrum) and a plurality of machine difference parameters; and a machine learning unit 302 that generates a machine learning model based on the learning data set composed of the plurality of virtual spectra generated by the simulation unit 301 and the plurality of machine difference parameters used to generate these virtual spectra.

[0083] Here, as Figure 9 shown, the simulation unit 301 has a pre-made machine difference influence model, and uses this machine difference influence model to produce a spectrum with machine differences (virtual spectrum) according to the reference spectrum or the standard spectrum and the machine difference parameters. Since the reference spectrum obtained by actual measurement contains noise components, in order to reproduce this noise component, Figure 9 the simulation unit 301 of

[0084] The gas analysis device of the described embodiment uses Fourier transform infrared spectroscopy (FTIR), but can also be applied to a spectroscopic analysis device that acquires a spectrum and calculates a concentration. For example, mid-infrared laser spectroscopy (QCL-IR) can be used, infrared laser absorption modulation method (IRLAM) can be used, and other laser absorption spectroscopy methods can also be used.

[0085] In addition, as long as it does not violate the gist of the present invention, various modifications or combinations of the embodiments can also be made. Industrial Applicability

[0086] According to the present invention, without obtaining the calibration lines of each spectroscopic analysis device, the concentration of the component to be measured can be accurately obtained from the measured spectrum with high precision. Explanation of Reference Numerals

[0087] 100 Spectroscopic analysis device (target device) 200 Reference device 51 Spectrum generation unit 52 Storage unit 53 Correction unit 54 Concentration calculation unit.

Claims

1. A spectral analysis device that irradiates light onto a sample to analyze a measurement target component contained in the sample, wherein the spectral analysis device is characterized by comprising: A light irradiation unit that irradiates light onto the sample; A light detector that detects the light transmitted through the sample to obtain a detection signal; A spectrum generation unit that generates a spectrum of the light transmitted through the sample based on the detection signal; A storage unit that stores a reference calibration line and a reference spectrum, the reference calibration line being made using a reference gas by a reference device that is a reference for the spectral analysis device, and the reference spectrum being the spectrum of the reference gas generated by the reference device; A correction unit that compares the spectrum of the reference gas generated by the spectrum generation unit, i.e., the reference spectrum, with the reference spectrum, and corrects the spectrum of the sample generated by the spectrum generation unit, i.e., the measurement spectrum, or the reference calibration line based on the difference in the wave number or the full width at half maximum of their peaks; And A concentration calculation unit that calculates the concentration of the measurement target component based on the reference calibration line and the corrected measurement spectrum, or the corrected reference calibration line and the measurement spectrum.

2. The spectral analysis device according to claim 1, wherein The correction unit compares the reference spectrum and the reference spectrum, and corrects the measurement spectrum or the reference calibration line based on at least two of the difference in the wave number of their peaks, the difference in height, and the difference in full width at half maximum.

3. The spectral analysis device according to claim 1 or 2, characterized in that, The correction unit calculates instrumental difference parameters, and corrects the measurement spectrum or the reference calibration line based on the instrumental difference parameters. The instrumental difference parameters are composed of at least two of a wavenumber axis scaling ratio α obtained from the wavenumbers of the peaks of the reference spectrum and the reference spectrum, an absorbance scaling ratio β obtained from one or both of the peak heights and half-value widths thereof, and a difference w in half-value widths obtained from the half-value widths of the peaks thereof. g are constituted by.

4. The spectral analysis device according to claim 3, characterized in that, The correction unit uses a pre-obtained learning model to estimate the machine difference parameter.

5. A spectral analysis method that uses a spectral analysis device that irradiates light onto a sample to analyze a measurement target component contained in the sample, wherein the spectral analysis method is characterized in that A reference calibration line and a reference spectrum are used, the reference calibration line being made using a reference gas by a reference device that is a reference for the spectral analysis device, and the reference spectrum being the spectrum of the reference gas generated by the reference device, The spectrum of the reference gas generated by the spectral analysis device, i.e., the reference spectrum, is compared with the reference spectrum, and the spectrum of the sample generated by the spectral analysis device, i.e., the measurement spectrum, or the reference calibration line is corrected based on the difference in the wave number or the full width at half maximum of their peaks, The concentration of the measurement target component is calculated based on the reference calibration line and the corrected measurement spectrum, or the corrected reference calibration line and the measurement spectrum.

6. A spectral analysis program for a spectral analysis device that irradiates light onto a sample to analyze a measurement target component contained in the sample, wherein the spectral analysis program is characterized in that It causes a computer to function as a spectrum generation unit, a storage unit, a correction unit, and a concentration calculation unit, The spectrum generation unit generates a spectrum of the light transmitted through the sample, The storage unit stores a reference calibration line and a reference spectrum, the reference calibration line being made using a reference gas by a reference device that is a reference for the spectral analysis device, and the reference spectrum being the spectrum of the reference gas generated by the reference device, The correction unit compares the spectrum of the reference gas generated by the spectrum generation unit, i.e., the reference spectrum, with the reference spectrum, and corrects the spectrum of the sample generated by the spectrum generation unit, i.e., the measurement spectrum or the reference calibration line, based on the difference in the wave number or the half-value width of their peaks. The concentration calculation unit calculates the concentration of the component to be measured based on the reference calibration line and the measurement spectrum corrected by the correction unit, or the reference calibration line corrected by the correction unit and the measurement spectrum.

7. A learning device that generates the learning model described in claim 4, wherein the learning device is characterized by comprising: a simulation unit that generates a plurality of virtual spectra based on each of the reference spectrum and a plurality of differential parameters; and a learning unit that generates the learned model based on a learning data set composed of the plurality of virtual spectra generated by the simulation unit and the plurality of differential parameters used to generate these virtual spectra.

8. The learning device according to claim 7, wherein the simulation unit assigns a nozzle component to each of the plurality of virtual spectra, the learning unit generates the learned model based on a learning data set composed of the plurality of virtual spectra to which the noise component is assigned and the plurality of differential parameters used to generate these virtual spectra.

9. A learning method that generates the learned model described in claim 4, wherein the learning method is characterized in that a plurality of virtual spectra are generated based on each of the reference spectrum and a plurality of differential parameters, the learned model is generated based on a learning data set composed of the plurality of generated virtual spectra and the plurality of differential parameters used to generate these virtual spectra.

10. A learning program for generating the learning model described in claim 4, wherein the learning program is characterized in that the computer functions as a simulation unit and a learning unit, the simulation unit generates a plurality of virtual spectra based on each of the reference spectrum and a plurality of differential parameters, the learning unit generates the learned model based on a learning data set composed of the plurality of generated virtual spectra and the plurality of differential parameters used to generate these virtual spectra.

Images