X-ray analysis device and peak prediction program product

Abstract

In the present invention, regardless of the type of X-ray detector and the circuit structure of the preamplifier, the peak value of the X-ray signal is measured at high speed and high precision even in high-count X-ray measurement. The present invention has: an excitation source; an X-ray detector; a preamplifier that outputs an analog signal; an AD converter that converts the analog signal into a digital signal; a signal detection unit that detects the incident timing; a waveform conversion unit that converts the digital signal into a step wave including a rising portion and a flat portion before and after the rising portion; a waveform shaping unit that generates a shaped wave including a step or a peak; a peak measurement unit that measures the peak value based on the incident timing and the step, etc., a learning unit that uses the incident timing to obtain a part of the step wave including the rising portion and generates a learning model that has learned these correlations using learning data including a group of multiple obtained parts and peaks; and a peak prediction unit that uses the incident timing to obtain a part of the step wave from the newly converted step wave and calculates the predicted peak value based on the obtained part of the step wave and the learning model.

Description

Technical Field

[0001] The present invention relates to an X-ray analysis device and a peak prediction program. Background Art

[0002] As a device for measuring elements contained in a sample or the concentration of the elements, there is known an X-ray analysis device that analyzes elements contained in a sample based on X-rays generated when excitation rays such as primary X-rays, electron beams, and particle beams are irradiated to the sample.

[0003] The X-ray analysis device first detects X-rays emitted from a sample that is excited or reflected by irradiation of primary X-rays, etc., using an X-ray detector. The integrated amount of charge output from the X-ray detector is converted into a voltage signal corresponding to the integrated amount by a preamplifier and input to a waveform shaping digital filter. The waveform shaping digital filter shapes the voltage signal into a pulse signal composed of, for example, a trapezoidal wave. The peak value of the pulse signal corresponds to the energy of the X-ray, and the X-ray energy spectrum can be obtained by counting the pulse signal of each peak value (see Patent Documents 1, 2, and 3, Non-Patent Document 1).

[0004] X-rays generated from the sample enter the X-ray detector at random time intervals. Therefore, by setting the shaping time of the waveform shaping digital filter to be shorter and shortening the rise time and fall time of the pulse signal, it is possible to detect X-rays that enter the X-ray detector without omission.

[0005] However, if the shaping time is set to be short, the electronic noise caused by the series noise increases, and the measured value of the peak value is deviated, thereby reducing the energy resolution of the X-ray energy spectrum.

[0006] Therefore, a radiation detection device is disclosed in Patent Document 4, which uses a noise removal filter to smooth the flat portion of a step-shaped digital voltage signal except for the step portion caused by X-ray incidence. The noise removal filter can suppress the degradation of energy resolution caused by setting the shaping time of the waveform shaping digital filter to be short.

[0007] In addition, Patent Document 5 discloses a method and apparatus for modeling X-ray signal waveform data using a function and obtaining a peak value of an output signal corresponding to the X-ray energy by performing function-based fitting. By performing fitting instead of the above-mentioned waveform shaping, it is possible to measure the peak value at a high speed and reduce the omission of X-rays.

[0008] Prior art literature

[0009] Patent Literature

[0010] Patent Document 1: Japanese Patent Application Laid-Open No. 7-333346

[0011] Patent Document 2: Japanese Patent Application Laid-Open No. 10-318946

[0012] Patent Document 3: Japanese Patent Application Publication No. 2012-168124

[0013] Patent Document 4: International Publication No. 2017 / 187972

[0014] Patent Document 5: Japanese Patent Application No. 2017-512297

[0015] Non-patent literature

[0016] Non-patent literature 1: VT Jordanov, GF Knoll, ACHuber, and JA Patazis, Nucl Instr. and Meth. A 353 (1994) 261-264. Summary of the invention

[0017] Problems to be solved by the invention

[0018] In the above-mentioned Patent Document 4, a noise removal filter is used to smooth the flat portion from which the step portion caused by X-ray incidence is removed. However, when the X-ray incidence count rate increases, the step portion caused by X-ray incidence increases, while the flat portion decreases. As a result, the noise removal effect of the smoothing is reduced. As a result, the deviation of the measured value of the peak value increases, and the energy resolution of the X-ray energy spectrum deteriorates.

[0019] According to the above-mentioned Patent Document 5, the X-ray signal waveform data is modeled using a function, and the peak value of the output signal corresponding to the X-ray energy is obtained by performing fitting based on the function. On the other hand, the model of the X-ray signal waveform data depends on the type of X-ray detector and the circuit structure of the preamplifier, etc. Therefore, the user needs to construct a new model for each X-ray detector and preamplifier.

[0020] The present invention has been completed in view of the above-mentioned problems, and its purpose is to provide an X-ray signal processing device and a peak prediction program, which can measure the peak value of the X-ray signal at high speed and high precision even when measuring high-count X-rays regardless of the type of X-ray detector and the circuit structure of the preamplifier.

[0021] Means for solving problems

[0022] (1) An X-ray analysis device according to one aspect of the present invention is characterized in that it comprises: an excitation source that irradiates a sample with an excitation line; an X-ray detector that detects X-rays generated from the sample and generates an electric charge corresponding to the energy of the X-rays; a preamplifier that outputs an analog signal representing a time change of a voltage corresponding to the generation of the electric charge; an AD converter that converts the analog signal into a digital signal; a signal detection unit that detects the incident timing of the X-rays from the digital signal to the X-ray detector; a waveform conversion unit that converts the digital signal into a step wave including a flat portion before a rise, a rise portion, and a flat portion after a rise; and a waveform shaping unit that shapes the step wave. A step wave is shaped to generate a shaped wave including steps or peaks; a peak measurement unit, which measures the peak value based on the incident timing and the steps or peaks of the shaped wave; a learning unit, which uses the incident timing to obtain a part of the step wave including the rising part, and uses learning data including a plurality of groups of the obtained parts of the step wave and the peaks to generate a learning model that has learned the correlation between the part of the step wave and the peak; and a peak prediction unit, which uses the incident timing to obtain a part of the step wave from the step wave newly converted by the waveform conversion unit, and calculates a predicted peak value based on the obtained part of the step wave and the learning model.

[0023] (2) In the above-mentioned method of the present disclosure, it is characterized in that a part of the step wave contained in the learning data is a part of the rising part, and the peak prediction unit uses the incident timing to obtain a part of the rising part from the step wave newly converted by the waveform conversion unit, and calculates the predicted peak value based on the obtained part of the rising part and the learning model.

[0024] (3) In the above aspect of the present disclosure, the X-ray analysis device further includes a selection unit that selects and outputs either the peak value or the predicted peak value.

[0025] (4) In the above aspect of the present disclosure, the selection unit is characterized in that the selection unit selects and outputs either the peak value or the predicted peak value based on the incident count rate of X-rays incident on the X-ray detector.

[0026] (5) In the above aspect of the present disclosure, the X-ray analysis device further includes a determination unit that determines whether to perform re-learning based on the peak value and the predicted peak value.

[0027] (6) In the above aspect of the present disclosure, the determination unit determines whether to perform re-learning based on a difference between the peak value and the predicted peak value and a preset first threshold value.

[0028] (7) In the above-mentioned method of the present disclosure, it is characterized in that the X-ray analysis device also has a counting unit, which counts each of the peaks or the predicted peaks to generate a histogram, and the judgment unit judges whether to perform re-learning based on the difference between the dispersion of the peaks in the peaks of the histogram and the dispersion of the predicted peaks and a pre-set second threshold.

[0029] (8) In the above aspect of the present disclosure, the sample used when acquiring the learning data is a sample that does not contain an element to be analyzed.

[0030] (9) In the above aspect of the present disclosure, the sample is graphite or acrylic resin.

[0031] (10) In the above-mentioned method of the present disclosure, it is characterized in that the learning model is constructed using a regression algorithm of machine learning.

[0032] (11) In the above-mentioned method of the present disclosure, it is characterized in that the learning model is constructed using a decision tree algorithm.

[0033] (12) A peak prediction program according to another aspect of the present disclosure is executed by an information processing device used in an X-ray analysis device, the X-ray analysis device comprising: an excitation source that irradiates a sample with an excitation line; an X-ray detector that detects X-rays generated from the sample and generates an electric charge corresponding to the energy of the X-rays; a preamplifier that outputs an analog signal representing a time variation of a voltage corresponding to the generation of the electric charge; an AD converter that converts the analog signal into a digital signal; a signal detection unit that detects the incident timing of the X-rays from the digital signal to the X-ray detector; a waveform conversion unit that converts the digital signal into a step wave including a flat portion before a rise, a rise portion, and a flat portion after a rise; and a waveform shaping unit that performs a waveform shaping on the step wave. shaping, and generating a shaped wave including a step or a peak; and a peak measurement unit, which measures the peak value based on the incident timing and the step or the peak of the shaped wave, wherein the peak prediction program is characterized in that the program causes the information processing device to perform the following steps: a learning step, which uses the incident timing to obtain a part of the step wave including the rising part, and uses learning data including a group of multiple obtained parts of the step wave and the peak to generate a learning model that has learned the correlation between the part of the step wave and the peak; and a peak prediction step, which uses the incident timing to obtain a part of the step wave from the step wave newly converted by the waveform conversion unit, and calculates a predicted peak value based on the obtained part of the step wave and the learning model.

[0034] Effects of the Invention

[0035] According to the present invention, regardless of the type of X-ray detector and the circuit configuration of the preamplifier, it is possible to measure the peak value of an X-ray signal at high speed and with high accuracy even when measuring a high number of X-ray counts. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a diagram schematically showing an X-ray analysis device according to an embodiment of the present invention.

[0037] Figure 2 This is a diagram schematically showing an X-ray signal processing device according to an embodiment of the present invention.

[0038] Figure 3 It is a diagram for explaining the waveform shape of X-ray signal data.

[0039] Figure 4 This is a diagram for explaining learning data.

[0040] Figure 5 This is a diagram for explaining the learning section.

[0041] Figure 6 This is a flowchart showing the generation of learning data.

[0042] Figure 7 is a flowchart representing the learning method.

[0043] Figure 8 is a flow chart showing a method for calculating a predicted peak value.

[0044] Fig. 9 This is a schematic diagram of a learned model of a decision tree used in this embodiment.

[0045] Fig.10 It is a diagram showing MnKα line energy spectra generated using the conventional example and the present embodiment.

[0046] Fig.11 This is a diagram showing MnKα line energy spectra generated using the conventional example and the present embodiment when the processing time is long.

[0047] Fig.12 This is a diagram showing a MnKα line spectrum generated using this embodiment.

[0048] Fig.13 It is a diagram showing the full width at half maximum of each processing time in a conventional example and in the present embodiment.

[0049] Fig.14 It is a diagram showing the full width at half maximum of each processing time in a conventional example and in the present embodiment. DETAILED DESCRIPTION

[0050] Hereinafter, preferred modes for carrying out the present invention (hereinafter referred to as embodiments) will be described. Figure 1 1 is a diagram schematically showing an X-ray analysis device 100 of the present invention. Figure 2 FIG. 1 is a diagram schematically showing an X-ray signal processing device 110 according to the present invention.

[0051] The X-ray analysis device 100 has a sample stage 102, an excitation source 104, and an X-ray signal processing device 110. The X-ray signal processing device 110 has an X-ray detector 106, a processing unit 108, a judgment unit 112, and a learning unit 114. The learning unit 114 and the judgment unit 112 are functionally realized by an information processing device used in the X-ray analysis device 100, and the information processing device is, for example, a personal computer.

[0052] The sample stage 102 is provided with a sample. The sample is an object used to obtain learning data or an object for elemental analysis. The excitation source 104 irradiates the sample with excitation rays. Specifically, for example, the excitation source 104 uses a collimator (not shown) to narrow the X-rays generated by an X-ray tube (not shown), and uses a filter (not shown) to adjust the incident X-ray spectrum distribution, and then irradiates the excitation rays as X-rays to the surface of the sample. X-rays are generated from the sample irradiated with the excitation rays.

[0053] The X-ray detector 106 detects X-rays generated from the sample and generates electric charges corresponding to the energy of the X-rays. Specifically, for example, the X-ray detector 106 generates electric charges corresponding to the energy of the X-rays by detecting X-rays generated by irradiating the sample with excitation lines. The X-ray detector 106 is, for example, a semiconductor detector such as a Si (Li) detector or an SDD detector. The X-ray detector 106 may be a scintillation detector or a proportional counter tube. In addition, a structure may be adopted in which the X-rays generated from the sample are split using a spectroscopic crystal (not shown) and detected using the X-ray detector 106.

[0054] like Figure 2 As shown, the processing unit 108 includes a preamplifier 202, an AD converter 204, a signal detection unit 206, a waveform conversion unit 208, a waveform shaping unit 210, a peak measurement unit 212, a selection unit 214, a counting unit 216 and a peak prediction unit 218.

[0055] The preamplifier 202 outputs an analog signal representing a time change of a voltage corresponding to the generated charge. Specifically, for example, the preamplifier 202 accumulates the charge input from the X-ray detector 106 in a capacitor (not shown), integrates the charge input from the X-ray detector 106 over time, and outputs a step-shaped voltage signal (hereinafter referred to as a step wave). In addition, the preamplifier 202 converts the step wave into a differential wave (see FIG. 1 ) by a differential circuit (not shown). Figure 3 (a)), and outputs it to the AD converter 204 at the subsequent stage. In addition, the preamplifier 202 may also be a structure that does not have a differential circuit and outputs a voltage signal as a step wave to the AD converter 204 at the subsequent stage. In addition, below, data representing the analog signal output by the preamplifier 202 and data representing each signal generated based on the analog signal ( Figure 3 The waveform data shown in (a) to (e) are collectively referred to as X-ray signal data.

[0056] The AD converter 204 converts the analog signal into a digital signal. Specifically, the AD converter 204 converts the voltage signal of the analog signal output by the preamplifier 202 into a digital signal that can be processed by the subsequent signal detection unit 206 (see Figure 3 (b)).

[0057] The signal detection unit 206 detects the incident timing of X-rays from the digital signal to the X-ray detector 106. Specifically, the signal detection unit 206 shapes the digital signal using a short waveform shaping time, and performs signal detection of the X-rays incident on the detector. For example, the signal detection unit 206 shapes the digital signal into a function wave such as a trapezoidal wave, a triangle wave, or a Gaussian wave with a small time width using a short waveform shaping time. Figure 3 (c) is a diagram showing an example of a triangular wave shaped by the signal detection unit 206. The waveform shaping time is, for example, less than 50 ns. The peak value of the shaped waveform (shaped wave) is approximately proportional to the energy of the detected X-ray. The signal detection unit 206 sets a predetermined threshold value for the peak value of the shaped wave to detect the incident timing of the X-ray signal.

[0058] The waveform converter 208 converts the digital signal into a step wave including a flat portion before rising, a rising portion, and a flat portion after rising. Specifically, the waveform converter 208 performs processing for removing attenuation corresponding to the time constant of the differential circuit in the preamplifier 202 on the digital signal converted by the A / D converter 204, and converts the differential wave of the digital signal into a step wave. For example, the waveform converter 208 uses a correction coefficient for eliminating the attenuation amount due to the time constant of the differential circuit, and performs conversion by delaying, adding, subtracting, multiplying, integrating, etc. the digital signal. Figure 3(d) is a diagram showing an example of a step wave converted by the waveform converter 208. In addition, when the preamplifier 202 is configured not to include a differentiating circuit, the processing unit 108 may also be configured not to include the waveform converter 208.

[0059] The waveform shaping unit 210 shapes the step wave and generates a shaped wave including steps or peaks. Specifically, the waveform shaping unit 210 shapes each step included in the step wave into a function wave such as a trapezoidal wave, a triangular wave, a Gaussian wave, etc. having a peak value corresponding to the height of the step. For example, Figure 3 (e) is a diagram showing an example of a trapezoidal wave shaped by the waveform shaping unit 210. The waveform shaping time is, for example, 0.5 to 2 μs. The peak value of the shaped waveform is proportional to the energy of the detected X-ray.

[0060] The peak measurement unit 212 measures the peak value based on the incident timing and the step or peak of the shaped wave. Specifically, for example, the peak measurement unit 212 is a multi-channel analyzer. The peak measurement unit 212 calculates the peak values ​​of the flat parts at the bottom and top of the trapezoidal wave, respectively, based on the incident timing detected by the signal detection unit 206. Furthermore, the peak measurement unit 212 measures the peak value by calculating the difference between the peak values ​​at the bottom and the top. In the case where the shaped wave is a triangular wave or a Gaussian wave, the peak measurement unit 212 calculates the peak value of the triangular wave or the Gaussian wave and the peak value of the flat part before and after the peak value, respectively, based on the incident timing. Furthermore, the peak measurement unit 212 measures the peak value by calculating the difference between the peak value and the peak value of the flat part. The peak measurement unit 212, for example, converts the peak value into an energy value per 10 eV for identification.

[0061] The selection unit 214 selects and outputs either a peak value or a predicted peak value. Specifically, for example, the selection unit 214 selects and outputs either a peak value or a predicted peak value based on the incident count rate of X-rays to the X-ray detector 106. When the incident count rate is higher than a predetermined threshold, if the processing time (the waveform shaping time of the waveform shaping unit 210) is shortened in order to reduce the omission of X-rays, the deviation of the peak value becomes larger, and the energy resolution of the X-ray energy spectrum deteriorates. On the other hand, even if the processing time is set shorter, the deviation of the predicted peak value does not change significantly, and the energy resolution of the X-ray energy spectrum is also better ( Fig.13 and Fig.14 ). Therefore, the selection unit 214 selects and outputs the predicted peak value when the incident count rate is higher than the predetermined threshold value, and selects and outputs the peak value when the incident count rate is lower than the predetermined threshold value. In this way, the X-ray energy can be measured at high speed and high precision, and the omission of X-rays can be reduced. The selection unit 214 can make the selection according to the instructions of the software, or according to the instructions of the user.

[0062] The counting unit 216 counts each peak value or predicted peak value to generate a histogram. Specifically, the counting unit 216 counts each channel corresponding to the energy of the X-ray based on the peak value measured by the peak measurement unit 212 or the predicted peak value calculated by the peak prediction unit 218. In addition, the counting unit 216 generates a histogram indicating the acquisition frequency of each energy of the acquired X-ray.

[0063] The peak prediction unit 218 uses the incident timing to obtain a portion of the step wave from the step wave newly converted by the waveform conversion unit 208, and calculates a predicted peak value based on the obtained portion of the step wave and a learning model (described later). The method of calculating the predicted peak value will be described later.

[0064] The learning unit 114 uses the incident timing to obtain a portion of the step wave including the rising portion, and uses learning data including a plurality of sets of the obtained portions of the step wave and the peak value to generate a learning model that has learned the correlation between the portion of the step wave and the peak value. Specifically, using the representation Figure 4 An example of learning data will be described using a diagram shown in FIG.

[0065] As described above, the step wave includes a plurality of rising portions and includes flat portions before and after each rising portion. Figure 4 The waveform shown on the left side of is X-ray signal data representing a rising portion and flat portions before and after the rising portion contained in the step wave. The learning unit 114 uses the starting position of the rising portion as the incident timing, and obtains X-ray signal data (X1 to XN) of the flat portion before the rising portion and the rising portion after the flat portion (which can also be a part of the rising portion). At this time, each data point of X1 to XN is obtained so that the position of the incident timing in the obtained X-ray signal data is fixed. That is, the position of the data point X1 in the step wave is set based on the position of the incident timing.

[0066] In addition, the learning unit 114 obtains the peak value (y) measured based on the shaped wave corresponding to the incident timing. The learning unit 114 obtains the X-ray signal data (X1 to XN) as a feature amount and the peak value as a target variable as a set. The learning data includes a plurality of such combinations. The learning unit 114 generates a learning model of the correlation between a part of the learned step wave and the peak value using the learning data.

[0067] In addition, the part of the step wave included in the learning data is not limited to the above as long as it includes at least a part of the rising part. For example, the part of the step wave included in the learning data may also be waveform data of the rising part and the flat part after the rising part. In addition, the part of the step wave included in the learning data may also be only a part of the rising part. The learning and learning model performed by the learning unit 114 will be described later.

[0068] The determination unit 112 determines whether to perform re-learning based on the peak value and the predicted peak value. The determination by the determination unit 112 will be described in detail later.

[0069] Next, use Figure 5 The details of the learning unit 114 are described below. The learning unit 114 includes a learning data acquisition unit 502, a learning model generation unit 504, and a learning model output unit 506. The learning data acquisition unit 502 acquires learning data including a group of a portion of a plurality of step waves having a rising portion and a peak value using the incident timing.

[0070] The learning model generation unit 504 generates a learned learning model using the learning data. Specifically, the learning model is established using a regression algorithm of machine learning. For example, a learning model is constructed using a decision tree algorithm (see Fig. 9 ). The generated learned learning model is output to the peak prediction unit 218 through the learning model output unit 506.

[0071] Furthermore, the learning unit 114 obtains the judgment result of whether to re-learn from the judgment unit 112. When the learning unit 114 obtains the judgment result of the intention to re-learn, the learning unit 114 obtains new learning data again. Then, the learning unit 114 generates the learned learning model using the new learning data and outputs it to the peak prediction unit 218.

[0072] Figure 6 is a flowchart representing the generation of learning data. First, a sample is measured (S602). Specifically, first, the sample is placed on the sample stage 102. The sample is a sample containing an element to be measured. X-rays (here, fluorescent X-rays) are generated from the sample irradiated with excitation lines by the excitation source 104. The X-ray detector 106 detects the X-rays generated from the element to be measured in the sample. In addition, the sample may also be a sample that does not contain the element to be measured. Specifically, for example, the sample may be graphite or acrylic resin. In this case, the X-ray detector 106 detects continuous X-rays generated from the sample.

[0073] Next, the learning data acquisition unit 502 acquires X-ray signal data for learning (S604). Specifically, the learning data acquisition unit 502 uses the incident timing to acquire a portion including the rising portion from the step wave output from the waveform conversion unit 208 as X-ray signal data. Here, a plurality of X-ray signal data are acquired. In addition, the X-ray signal data may be only a portion of the rising portion.

[0074] Next, the peak measurement unit 212 measures the peak value (S606). Specifically, for example, the peak measurement unit 212 measures the peak value of the step wave corresponding to each X-ray signal data obtained in S604. At this time, the shaping time of the waveform shaping unit 210 is set so that the dispersion (dispersion, standard deviation, variation count, etc.) of the peak value becomes smaller. The waveform shaping unit 210 uses the set shaping time to generate a trapezoidal wave, i.e., a shaped wave. The peak measurement unit 212 uses the shaped trapezoidal wave to measure the peak value. The waveform shaping time when performing trapezoidal shaping is, for example, about 1 to 2 μs.

[0075] Next, the learning data acquisition unit 502 combines the X-ray signal data and the peak value measured based on the X-ray signal data to generate learning data (S608). Specifically, for example, the learning data acquisition unit 502 acquires the X-ray signal data acquired in S604 and the peak value corresponding to the X-ray signal data, and combines them to generate learning data. At this time, the X-ray signal data acquired based on the same incident timing corresponds to the peak value.

[0076] Figure 7 It means to use Figure 6 Flowchart of a learning method for learning data generated in the flow shown in FIG. First, the learning data acquisition unit 502 acquires learning data from the processing unit 108 ( S702 ). Specifically, the learning data acquisition unit 502 acquires the learning data generated in S608 .

[0077] Next, the learning model generation unit 504 performs learning using a machine learning algorithm (S704) and generates a learned learning model (S706). Specifically, for example, the machine learning algorithm is a decision tree. The learning model generation unit 504 uses a decision tree for a group of X-ray signal data and its peak value, i.e., learning data, to learn the correlation between the X-ray signal data and the peak value. Thus, a learned learning model including a decision tree for segmentation conditions and a segmentation threshold required for predicting a peak value from the X-ray signal data is generated. The learning model output unit 506 outputs the generated learned learning model to the peak prediction unit 218 (S708).

[0078] In addition, the machine learning algorithm is not limited to the decision tree, but can also be a random forest and gradient boosting algorithm based on the decision tree. In addition, the machine learning algorithm can also be a linear regression, k-nearest neighbor method, neural network, support vector regression and other algorithms.

[0079] Figure 81 is a flowchart showing a method for predicting a peak value using a learning model. First, a sample is measured (S802). The sample is an unknown sample to be measured. X-rays are generated from the sample irradiated with excitation rays by the excitation source 104. The X-ray detector 106 detects the X-rays generated from the sample. This step is the same as S602 except that the sample is different.

[0080] Next, the peak prediction unit 218 obtains X-ray signal data (S804). Specifically, the peak prediction unit 218 uses the incident timing to obtain a portion including the rising portion from the step wave output from the waveform conversion unit 208 as the X-ray signal data. At this time, each data point X1 to XN is obtained so that the position of the incident timing in the obtained X-ray signal data is the same as the position of the incident timing in the X-ray signal data included in the learning data. In addition, in the case where the X-ray signal data included in the learning data only includes a portion of the rising portion, the X-ray signal data obtained by the peak prediction unit 218 also only includes a portion of the rising portion.

[0081] Next, the peak prediction unit 218 calculates the predicted peak value ( S806 ). Specifically, the peak prediction unit 218 inputs the X-ray signal data acquired in S804 into the learned model generated in S706 . Fig. 9 : is a schematic diagram of the learning model used in this embodiment. The learning model is a decision tree including the conditions for segmentation of X-ray signal data and the threshold value for segmentation. In addition, the peak prediction unit 218 obtains the learned learning model in S708. When the X-ray signal data is input to the peak prediction unit 218, the learning model estimates the peak value by classifying the X-ray signal data and outputs the predicted peak value. Then, the peak prediction unit 218 outputs the predicted peak value to the counting unit 216 (S808). Through the above steps, the predicted peak value of each X-ray signal data is obtained.

[0082] In the present embodiment, the learning unit 114 may be configured to perform re-learning according to the judgment of the judgment unit 112. For example, the judgment unit 112 may also determine whether to perform re-learning based on the difference between the peak value and the predicted peak value and a pre-set first threshold value. Specifically, the judgment unit 112 may also compare the X-ray energy spectrum obtained from the peak value with any peak contained in the X-ray energy spectrum obtained from the predicted peak value, and if the energy difference between the peaks is greater than 20 eV, it is determined to perform re-learning. Here, the judgment criterion is not limited to the peak energy of the histogram, and may also be set for each peak value and predicted peak value. Specifically, for example, the judgment criterion may also be an error indicator such as the absolute mean error, mean square error, and determination coefficient between the peak value and the predicted peak value.

[0083] In addition, the determination unit 112 may also determine whether to perform relearning based on the difference between the dispersion of the peak value in the peak of the histogram and the dispersion of the predicted peak value and a preset second threshold value. Specifically, for example, the determination unit 112 may also compare an X-ray energy spectrum obtained from the peak value with an arbitrary peak included in the X-ray energy spectrum obtained from the predicted peak value, and if the difference between the full width at half maximum (FWHM) of the two peaks is greater than 10 eV, it is determined that relearning is to be performed.

[0084] Figure 10 to Figure 12 This is a diagram showing the energy spectrum of the MnKα line obtained by the learned model according to the present embodiment. Figures 10 to 12 The energy spectrum of the MnKα line shown is measured with a SDD detector from a sealed line source. 55 Here, the sealed line source 55 The X-ray emitted by Fe is an X-ray that simulates the fluorescent X-ray generated by the sample. The X-ray signal data used for learning and peak prediction is obtained by detecting 55 A portion of a step wave generated by X-rays emitted from Fe. The measured rise time (width of the rise portion) of the step wave is about 0.2 μs. Fig.10 The X-ray signal data in includes the flat part before the rise, the rise part, and the flat part after the rise, and the processing time is 0.6μs. Fig.10 In the figure, the energy spectrum of the MnKα line of the prior art is shown as a comparative example. The waveform shaping unit 210 performs trapezoidal shaping, and the histogram of the peak value measured by the peak value measuring unit 212 is represented as the prior art. In addition, the time width from the rise to the fall of the trapezoidal wave is used as the processing time of the prior art. The processing time of the prior art is 0.65μs. In addition, Fig.11 It means that in order to Fig.10 A diagram showing measurement results of the conventional example and the present embodiment when the processing time is long for comparison. Fig.11 The processing time of the conventional example is 2.25 μs, and the processing time of the present embodiment is 2 μs. Fig.10 As shown in the conventional example, by setting the processing time to be shorter, Fig.11 Compared with the results of the prior art, the energy of the MnKα line peak becomes lower and the full width at half maximum (FWHM) becomes larger. On the other hand, in the present embodiment, even if the processing time is shortened, the MnKα line peak energy does not change, and a MnKα line spectrum with a smaller full width at half maximum (FWHM) than the prior art is obtained. In addition, Fig.12 The X-ray signal data in contains only a part of the rising part, and the processing time is 0.1μs. Fig.12As shown, although the peak prediction was performed using a processing time of 0.1 μs which is shorter than 0.2 μs which is the rise time of the step wave, the Gaussian shape was hardly destroyed and a MnKα line spectrum with a small full width at half maximum (FWHM) was obtained.

[0085] Fig.13 and Fig.14 : This is a diagram showing the full width at half maximum (FWHM) of the MnKα line spectrum at each processing time obtained by the learned model of this embodiment and the conventional method. The waveform shaping unit 210 performs trapezoidal shaping and shows the energy resolution obtained by the histogram of the peak value measured by the peak measurement unit 212 as a conventional example. In addition, the time width from the rise to the fall of the trapezoidal wave is used as the processing time of the conventional example. Fig.13 The X-ray signal data used when the X-ray signal is used includes at least the flat part before the rise and a part of the rise. When the processing time is 0.1 to 0.2 μs, the X-ray signal data includes the flat part before the rise and a part of the rise. When the processing time is 0.4 to 2.0 μs, the X-ray signal data includes the flat part before the rise, the rise, and the flat part after the rise. On the other hand, when obtaining Fig.14 The X-ray signal data used when processing includes only the rising part. When the processing time is 0.05 to 0.1 μs, the X-ray signal data includes only a part of the rising part. When the processing time is 0.2 μs, the X-ray signal data includes all the rising parts. Fig.13 and Fig.14 As shown, it can be seen that the energy resolution of this embodiment is significantly improved compared with the conventional example, especially in the region where the processing time is 0.2 μs or less.

[0086] Fig.13 The X-ray signal data in at least includes a flat portion before the rise and a portion of the rise, and Fig.14 The X-ray signal data in the image only contains the rising part. When the X-ray incidence count rate increases, the rising part caused by the X-ray incidence increases and the flat part decreases. As a result, when processing including the flat part is performed, the amount of X-ray signal data that cannot be processed increases, resulting in the omission of X-rays. Fig.14 As shown, by using X-ray signal data including only the rising portion, compared with the case of using X-ray signal data including the flat portion, it is possible to further reduce the omission of X-rays and measure the peak value at a higher speed and with higher accuracy.

[0087] As described above, in the X-ray signal processing device 110 of the present embodiment, it is possible to measure the peak value at high speed and high precision compared to the conventional method of applying a waveform shaping filter. Thus, it is possible to reduce the omission of X-rays and to improve the throughput of the X-ray signal processing device 110. In addition, the present embodiment can also be applied to various X-ray detectors 106 and preamplifiers 202 by generating a learned learning model by the learning unit 114. Furthermore, by relearning as needed, it is possible to obtain a predicted peak value that is always optimal for the X-ray detector 106. In addition, by measuring the continuous X-rays generated from the sample and using them as learning data, it is not necessary to prepare a sample containing the element to be measured. Thus, the measurement for obtaining the learning data becomes simple, and at the same time, a predicted peak value can be obtained for the X-rays generated from the sample with unknown elemental composition.

[0088] The learning unit 114 and the judgment unit 112 of the present disclosure are configured using, for example, a CPU, a GPU, and a multi-core CPU, and the processing unit 108 after the signal detection unit 206 including the peak prediction unit 218 is configured using an FPGA (field-programmable gate array). The learned learning model is realized by the learning unit 114 performing information processing (learning step) according to a computer program (peak prediction program). The peak prediction unit 218 is constructed by writing the parameters stored in the learned learning model contained in the learning unit 114 into the FPGA constituting the processing unit 108. The peak prediction unit 218 performs information processing (peak prediction step) to obtain a predicted peak value.

[0089] As another structure, the structure after the signal detection unit 206 including the peak prediction unit 218 of the learning unit 114, the judgment unit 112, and the processing unit 108 can also be configured using FPGA. In this structure, the learning unit 114 configured inside the FPGA performs information processing (learning step) and generates a learned learning model. The obtained learned learning model is output to the peak prediction unit 218, and the peak prediction unit 218 performs information processing (peak prediction step) to obtain a predicted peak value.

[0090] The present invention is not limited to the above-mentioned embodiments, and various modifications are possible. The structure of the above-mentioned X-ray analysis device 100 is an example and is not limited thereto. It can also be replaced by a structure that is substantially the same as the structure shown in the above-mentioned embodiment, a structure that has the same effect, or a structure that achieves the same purpose.

[0091] For example, in Figure 1The X-ray analysis device is described as a fluorescent X-ray analysis device, but the X-ray analysis device is not limited to this. The X-ray analysis device may also be a scanning electron microscope (SEM), an electron probe microanalyzer (EPMA), or a proton X-ray fluorescence analyzer (PIXE) that performs analysis by irradiating a sample with an electron beam or a particle beam as an excitation line and detecting the generated X-rays.

[0092] Description of Reference Numerals

[0093] 100X-ray analysis device, 102 sample stage, 104 excitation source, 106X-ray detector, 108 processing unit, 110X-ray signal processing device, 112 judgment unit, 114 learning unit, 202 preamplifier, 204AD converter, 206 signal detection unit, 208 waveform conversion unit, 210 waveform shaping unit, 212 peak measurement unit, 214 selection unit, 216 counting unit, 218 peak prediction unit, 502 learning data acquisition unit, 504 learning model generation unit, 506 learning model output unit.

Claims

1. An X-ray analysis device, characterized in that: have: an excitation source, which irradiates the sample with excitation radiation; An X-ray detector that detects X-rays generated from the sample and generates charges corresponding to the energy of the X-rays; a preamplifier that outputs an analog signal representing a time variation of a voltage corresponding to generation of the charge; An AD converter converts the analog signal into a digital signal; a signal detection unit that detects an incident timing of X-rays from the digital signal to the X-ray detector; a waveform conversion unit that converts the digital signal into a step wave including a flat portion before rising, a rising portion, and a flat portion after rising; a waveform shaping section that shapes the step wave and generates a shaped wave including steps or peaks; a peak value measuring section that measures a peak value based on the incident timing and the step or peak of the shaped wave; a learning unit that uses the incident timing to acquire a portion of the step wave including the rising portion, and generates a learning model that has learned the correlation between the portion of the step wave and the peak value by using learning data including a plurality of sets of the acquired portions of the step wave and the peak value; and a peak prediction section that uses the incident timing to obtain a portion of the step wave from the step wave newly converted by the waveform conversion section, and calculates a predicted peak value based on the obtained portion of the step wave and the learning model, The part of the step wave included in the learning data is a part of the rising portion, The peak prediction unit obtains a portion of the rising portion from the step wave newly converted by the waveform conversion unit using the incident timing, and calculates a predicted peak value based on the obtained portion of the rising portion and the learning model.

2. The X-ray analysis device according to claim 1, characterized in that , The device further includes a selection unit that selects and outputs either the peak value or the predicted peak value.

3. The X-ray analysis device according to claim 2, characterized in that: The selection unit selects and outputs either the peak value or the predicted peak value based on an incident count rate of X-rays incident on the X-ray detector.

4. The X-ray analysis device according to any one of claims 1 to 3, characterized in that: The system further includes a determination unit that determines whether to perform re-learning based on the peak value and the predicted peak value.

5. The X-ray analysis device according to claim 4, characterized in that: The determination unit determines whether to perform re-learning based on a difference between the peak value and the predicted peak value and a preset first threshold value.

6. The X-ray analysis device according to claim 4, characterized in that: A counting unit is further provided, which counts each of the peak values ​​or the predicted peak values ​​to generate a histogram, The determination unit determines whether to perform re-learning based on a difference between a dispersion degree of the peak value in the peak of the histogram and a dispersion degree of the predicted peak value and a preset second threshold value.

7. The X-ray analysis device according to any one of claims 1 to 3, characterized in that: The sample used when acquiring the learning data is a sample that does not contain an element to be analyzed.

8. The X-ray analysis device according to claim 7, characterized in that: The sample is graphite or acrylic resin.

9. The X-ray analysis device according to any one of claims 1 to 3, characterized in that: The learning model is constructed using a regression algorithm of machine learning.

10. The X-ray analysis device according to claim 9, characterized in that: The learning model is constructed using a decision tree algorithm.

11. A peak prediction program product executed by an information processing device used in an X-ray analysis device, The X-ray analysis device comprises: an excitation source, which irradiates the sample with excitation radiation; An X-ray detector that detects X-rays generated from the sample and generates charges corresponding to the energy of the X-rays; a preamplifier that outputs an analog signal representing a time variation of a voltage corresponding to generation of the charge; An AD converter converts the analog signal into a digital signal; a signal detection unit that detects an incident timing of X-rays from the digital signal to the X-ray detector; a waveform conversion unit that converts the digital signal into a step wave including a flat portion before rising, a rising portion, and a flat portion after rising; a waveform shaping section that shapes the step wave and generates a shaped wave including steps or peaks; and a peak value measuring unit for measuring a peak value based on the incident timing and the step or peak of the shaped wave, The peak prediction procedure is characterized in that The program causes the information processing device to execute the following steps: a learning step of acquiring a portion of the step wave including the rising portion using the incident timing, and generating a learning model having learned the correlation between the portion of the step wave and the peak value using learning data including a plurality of sets of the acquired portions of the step wave and the peak value; and a peak prediction step of obtaining a portion of the step wave from the step wave newly converted by the waveform conversion unit using the incident timing, and calculating a predicted peak value based on the obtained portion of the step wave and the learning model, The part of the step wave included in the learning data is a part of the rising portion, In the peak prediction step, a part of the rising portion is obtained from the step wave newly converted by the waveform converter using the incident timing, and a predicted peak value is calculated based on the obtained part of the rising portion and the learning model.

Images