JP2019184249A - Particle generation source analysis system, particle generation source analysis method, and program - Google Patents

Abstract

To evaluate adequately a generation source of particulate substances.SOLUTION: A particle generation source analysis device 300 constructs a classifier that classifies fine particles either as those from one or multiple predetermined generation sources or as those from generation sources other than the one or multiple predetermined generation sources (neither from the one or multiple generation sources) on the basis of the attributes of fine particles including their components (constituent elements) obtained using sample particles that are the fine particles collected at the one or multiple predetermined generation sources. Then, using the classifier, the particle generation source analysis device 300 obtains information on the generation source of the fine particles whose generation source is unknown.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a particle generation source analysis system, a particle generation source analysis method, and a program, and is particularly suitable for use in analyzing a generation source (discharge source) of particulate matter.

  Fine particles having a diameter (particle diameter) of 10 μm or less are called suspended particulate matter (SPM). It is known that suspended particulate matter floats in the atmosphere for a long time and enters the human body to affect the human body. In particular, there is a concern that fine particles having a particle size of 2.5 μm or less (this particle is called PM2.5) strongly influences the human body. Therefore, a technique for specifying the generation source of such particulate matter is required. This is because it is possible to take measures to reduce the influence of particulate matter by specifying the source of particulate matter. There are various sources of particulate matter in the atmosphere, including natural sources such as soil and the ocean, and human sources such as factories and automobiles. Accordingly, in order to accurately know the generation source of the particulate matter group, it is required to analyze the generation source for each particulate matter.

As this type of technology, there are technologies described in Non-Patent Documents 1 and 2.
In Non-Patent Document 1, element mapping is performed by time-of-flight secondary ion mass spectrometry (TOF-SIMS) on a region where particles are present. Next, the shape of each particulate matter is specified by performing binarization processing on all ion images in this region. Then, a mass spectrum is created from the ions detected for each particulate matter, and cluster analysis is performed by the Ward method using this mass spectrum. Then, the composition of each cluster is estimated to identify the origin of each cluster.

  Non-Patent Document 2 discloses the following technique. First, the X-ray spectrum of the particulate matter is measured. Then, the characteristic X-ray intensity for each element is obtained from the X-ray spectrum, and cluster analysis is performed by the Median method based on the characteristic X-ray intensity. At that time, a logarithmic value of a value obtained by dividing the characteristic X-ray intensity of each element by the total intensity is obtained as a chemical composition index representing the composition of each particle. The chemical composition index thus obtained is standardized so that the average is 0 (zero) and the standard deviation is 1 for each element, and then cluster analysis is performed. And the typical shape of the particulate matter contained in each cluster and the generation source of the particulate matter classified into each cluster by the X-ray spectrum are estimated.

JP 2017-72593 A

Norihito Mayama, 9 others, “Development of origin analysis method for fine particles using high-resolution time-of-flight secondary ion mass spectrometry”, Proceedings of the 52nd Joint Lecture on Vacuum, J. Vac. Soc. Jpn, Vol. 55, no. 3, 2012 Takenobu Suyasu, 5 others, "Analysis of particle-by-grain origin of suspended particulate matter using an electron microanalyzer", Journal of the Chemical Society of Japan, 1996, (5), p. 500-507

  However, in the techniques described in Non-Patent Documents 1 and 2, cluster analysis is performed. That is, in the techniques described in Non-Patent Documents 1 and 2, a mass spectrum or an X-ray spectrum is obtained for each particulate matter collected during a specific period (for example, a specific day), and the mass spectrum or The generation source of each particulate matter is analyzed by clustering each particulate matter using the X-ray spectrum. Note that clustering refers to dividing a set of data to be classified into subsets based on similarity.

Therefore, the result of clustering may vary depending on what particulate matter is collected and clustered. As a result, the analysis result of the particulate matter generation source can also change. That is, for example, when the particulate matter collected on a specific day is used, and when the particulate matter collected on two days, the first day and the next day, is used as a result of clustering. Can change. The accuracy of clustering also depends on what particulate matter is collected and clustered.
As described above, the conventional technique has a problem that it is not easy to appropriately evaluate the source of the particulate matter.

  This invention is made | formed in view of the above problems, and it aims at enabling it to evaluate the generation source of a particulate matter appropriately.

  The particle generation source analysis system of the present invention is a particle generation source analysis system that analyzes a generation source of fine particles existing in the atmosphere, and uses sample particle attribute information that is information based on attributes of a plurality of sample particles. Classifier construction means for deriving classifier information that is information for configuring a classifier, analysis target particle attribute information that is information based on the attribute of one or a plurality of analysis target particles, and the classifier information Source analysis means for deriving information on the generation source of the analysis target particle based on the result of classification by the classifier, and the sample particle is predetermined. Fine particles collected by a generation source, the predetermined number of generation sources is one or more, the analysis target particle is a fine particle whose generation source is unknown, and the attribute is a fine particle of The classifier includes, based on the information based on the attributes of the fine particles, the individual fine particles belonging to at least one of the predetermined generation sources, and the predetermined generation sources. It is characterized in that it is classified into either one of those not belonging to any of the above.

  The particle generation source analysis method of the present invention is a particle generation source analysis method for analyzing a generation source of fine particles existing in the atmosphere, using sample particle attribute information which is information based on attributes of a plurality of sample particles, A classifier construction step for deriving classifier information which is information for configuring a classifier, analysis target particle attribute information which is information based on the attribute of one or a plurality of analysis target particles, and the classifier information; And a source analysis step of deriving information on the source of the particle to be analyzed based on the result of classification by the classifier, and the sample particles are predetermined. Fine particles collected by a generation source, the predetermined number of generation sources is one or more, the analysis target particle is a fine particle whose generation source is unknown, and the attribute is a fine particle Constituent elements of The classifier includes, based on the information based on the attribute of the fine particles, each of the fine particles belonging to at least one of the predetermined generation sources and any of the predetermined generation sources. It is characterized in that it is classified into either one of those that do not belong.

  The program of the present invention causes a computer to function as each means of the particle generation source analysis system.

  According to the present invention, it is possible to appropriately evaluate the source of particulate matter.

FIG. 1 is a diagram illustrating an example of a configuration of a particle generation source analysis system. FIG. 2 is a diagram illustrating an example of a mass spectrum. FIG. 3 is a diagram illustrating an example of the relationship between average signal strength and data No. FIG. 4 is a flowchart for explaining an example of processing of the particle generation source analyzer (classifier construction unit) when constructing a classifier. FIG. 5 is a flowchart for explaining an example of processing of the particle generation source analysis device (generation source analysis unit) when analyzing the generation source of the analysis target particles. FIG. 6 is a diagram showing an example of a rejection rate in each generation source in a table format.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the present embodiment, the particulate matter is referred to as fine particles as necessary. In this embodiment, fine particles having a diameter (particle diameter) of 10 μm or less are measured.

<Overview>
In the present embodiment, a plurality of fine particles are collected at one or a plurality of predetermined generation sources, and for each of the collected fine particles, a component (constituent element), a signal intensity of m / z, and a particle diameter are set. taking measurement. Here, as a predetermined generation source, a factory is mentioned, for example. When determining a generation source from an area where a plurality of factories are gathered, for example, the plurality of factories can be determined as a plurality of generation sources. Further, a plurality of locations (for example, a chimney or a flue) in a certain factory can be defined as a plurality of generation sources. A combination of these sources can be defined as a plurality of sources. The predetermined generation source is not limited to a factory, and may be, for example, an automobile. However, the predetermined generation source is preferably a generation source (artificial generation source) other than the natural generation source.

  Although details will be described later, in the present embodiment, a classifier is used by using information on components (constituent elements), m / z signal intensity, and particle diameter as information on fine particles whose source is known. To construct. Based on the information based on the attribute of the fine particle whose source is unknown, the classifier determines whether the fine particle belongs to at least one of the predetermined source and none of the predetermined source. It is for classifying either. In this embodiment, an unknown generation source of fine particles is estimated using this classifier.

<Particle generation source analysis system>
FIG. 1 is a diagram illustrating an example of a configuration of a particle generation source analysis system.
In FIG. 1, the particle generation source analysis system of the present embodiment includes a classifier 100, an analysis device 200, and a particle generation source analysis device 300.

<< Classifier 100 >>
The classifier 100 sucks a gas containing fine particles having various particle sizes, classifies the fine particles according to the particle size, and collects the classified fine particles by the filter 110 installed in the classifier 100. As the classifier 100, it is realizable by using the virtual impactor of patent document 1, for example. Moreover, the filter described in patent document 1 can also be used for the filter 110 set in the classifier 100, for example. As described above, the classifier 100 can be realized by a known technique, and thus detailed description thereof is omitted.

<< Analyzer 200 >>
The analysis apparatus 200 is an apparatus for performing single particle analysis. That is, the analyzer 200 is an apparatus for obtaining information including the component (constituent element), the m / z signal intensity, and the particle diameter as information on the fine particles collected by the classifier 100. is there. The analyzer 200 is realized by using, for example, the analyzer described in Patent Document 1. Patent Document 1 describes using an FIB-EB-TOF-SIMS device as an example of an analysis device. The FIB-EB-TOF-SIMS apparatus corresponds to an FIB-TOF-SIMS (FIB-Time of Flight-Secondary Ion Mass Spectrometry) apparatus mounted with an electron beam irradiator.

  In FIG. 1, the analysis apparatus 200 includes a chamber 210, a sample stage 215, an electron beam irradiator 220, a focused ion beam irradiator 230, a mass spectrometer 240, and a secondary electron detector 250. The sample stage 215 is provided in the chamber 210.

As shown in FIG. 1, the filter 110 in which the fine particles P are collected on the surface is fixed on the sample stage 215.
A mass spectrometer 240 is disposed substantially vertically above the sample stage 215. In addition, the electron beam irradiator 220 and the focused ion beam irradiator 230 can be used to irradiate the electron beam and the focused ion beam to the same spot on the surface of the fine particle 260 at any height of the sample stage 215, respectively. The sample plate 215 is disposed at a position that is symmetrical with respect to the sample table 215 with the 240 therebetween. For example, the electron beam irradiator 220 and the focused ion beam irradiator 230 are arranged so as to irradiate the fine particles P with a beam from a direction in which the dip angle is 45 ° with respect to the horizontally placed sample stage 215. .

  When performing single particle analysis, the chamber 210 is first evacuated. The fine particles P on the filter 110 are irradiated while scanning with an electron beam from the electron beam irradiator 220. Secondary electrons generated from the fine particles P by the electron beam irradiation are detected by the secondary electron detector 250, and an SEM image of the fine particles P (a reflected electron image of the SEM) is generated.

  The analyzer 200 is equipped with a function of automatically recognizing the fine particles P based on the difference in composition contrast between the surface of the filter 110 and the fine particles P to be analyzed in the SEM image. When the SEM image is generated, one fine particle P to be analyzed is automatically identified from the plurality of fine particles P on the filter 110 based on the SEM image. At this time, based on the SEM image, the particle diameter of the specified fine particle P that is the analysis target can be measured. Note that one fine particle P to be analyzed may be manually specified by appropriately moving the sample stage 215 while observing the SEM image.

  Then, SIMS analysis is performed on the identified fine particles P. That is, a focused ion beam is irradiated from the focused ion beam irradiator 230, and secondary ions generated from the fine particles P by the irradiation of the focused ion beam are detected by the mass spectrometer 240.

  The mass spectrometer 240 is a time-of-flight mass spectrometer, and by detecting the accelerated secondary ions with the secondary ion detector 241 provided therein, the mass-to-charge ratio of the secondary ions is determined from the time of flight. m / z can be measured. In a time-of-flight mass spectrometer, positive ions and negative ions are measured separately (positive mode / negative mode). Based on the measurement result, the mass spectrum of the fine particles P can be obtained.

  When mass analysis by the mass spectrometer 240 is completed for one particle P, another particle P on the filter 110 is selected based on the SEM image, and SIMS analysis is similarly performed on the other particle P. Do. By repeating this process, the component for each fine particle P, the signal intensity of m / z, and the particle diameter can be obtained for the collected particle group of fine particles P.

  In this embodiment, when creating a classifier, a plurality of fine particles P are collected from one or more predetermined generation sources, and an SEM image and a secondary ion are collected for each of the collected fine particles P. The mass to charge ratio m / z is measured by the analyzer 200. On the other hand, when estimating the generation source of the fine particles whose source is unknown using the classifier, the fine particles P are collected at an arbitrary place where the fine particles P whose generation source is to be estimated exist, and the collected fine particles P For each of these, measurement of the SEM image and the mass-to-charge ratio m / z of the secondary ions is performed by the analyzer 200.

  As described above, the analyzer described in Patent Document 1 can be used as the analyzer 200. Thus, since the analyzer 200 can be realized by a known technique, a detailed description thereof will be omitted.

<< Particle generation source analyzer 300 >>
The particle generation source analysis apparatus 300 has a function of constructing a classifier using information on fine particles whose generation source is known (information including components (constituent elements), m / z signal intensity, and particle diameter), and this classification. And a function of estimating the (unknown) generation source of fine particles using a vessel. The hardware of the particle generation source analysis device 300 is realized by using, for example, an information processing device having a CPU, ROM, RAM, HDD, and various interfaces, or dedicated hardware. Hereinafter, an example of a functional configuration of the particle generation source analyzer 300 will be described.

The particle generation source analysis apparatus 300 includes a classifier construction unit 310 and a generation source analysis unit 320. The classifier construction unit 310 is a part that realizes a function of constructing a classifier, and the source analysis unit 320 is a part that realizes a function of estimating a generation source of fine particles whose source is unknown. In the following description, fine particles whose source is known are referred to as sample particles as necessary, and fine particles whose source is unknown are referred to as analysis target particles as necessary.
(Classifier construction unit 310)
The classifier construction unit 310 includes a data preprocessing unit 311, a dimension compression unit 312, a classifier information deriving unit 313, and a classifier information storage unit 314.

((Data preprocessing unit 311))
The data preprocessing unit 311 removes the mass spectrum that becomes noise from the converted mass spectrum and the function of converting the mass spectrum of the sample particles (fine particles P whose generation source is known) into a mass spectrum that contributes to the creation of the classifier. With functions.
The data preprocessing unit 311 includes an image acquisition unit 311a, a particle size derivation unit 311b, a spectrum acquisition unit 311c, a spectrum reconstruction unit 311d, a spectrum normalization unit 311e, a spectrum normalization unit 311f, and a spectrum average value derivation. Part 311g and noise removing part 311h.

[Image Acquisition Unit 311a]
The image acquisition unit 311a acquires SEM images for the plurality of sample particles obtained by the analyzer 200. Here, sample particles are collected at each of one or more predetermined sources. Therefore, at least one SEM image is obtained for each of one or more predetermined generation sources. A larger number of sample particles is preferable because the accuracy of the classifier can be increased.

[Particle size deriving unit 311b]
The particle size deriving unit 311b derives the size of each sample particle included in the SEM image based on the SEM image acquired by the image acquiring unit 311a. In the present embodiment, the number of pixels in which the fine particles P are present in the SEM image is the size of the fine particles P. The particle size deriving unit 311b derives the size of each sample particle included in the SEM image for each of all the SEM images acquired by the image acquiring unit 311a.

[Spectrum acquisition unit 311c]
The spectrum acquisition unit 311c acquires the mass-to-charge ratio m / z of secondary ions for each sample particle and the signal intensity detected at the m / z obtained by the analyzer 200 as a mass spectrum. As described above, the mass spectrometer 240 selects the fine particles P included in the SEM image, and measures the mass-to-charge ratio m / z of the secondary ions with respect to the selected fine particles P. Therefore, the fine particles P included in the SEM image, the mass-to-charge ratio m / z of the secondary ions with respect to the fine particles P, and the signal intensity are in a mutually correlated state. As described above, at least one SEM image is obtained for each of one or more predetermined generation sources. Therefore, the spectrum acquisition unit 311c measures the mass-to-charge ratio m / z of secondary ions for each sample particle included in these SEM images. When there are a plurality of mass spectra belonging to one fine particle P, all these mass spectra are integrated to obtain the mass spectrum of the fine particle P.

[Spectrum reconstruction unit 311d]
The spectrum reconstruction unit 311d derives data characterizing each sample particle based on the mass spectrum of the secondary ions for each sample particle acquired by the spectrum acquisition unit 311c. This data is extracted, for example, as m / z detected in any of the fine particles P (for example, S / N is more than 3) and the signal intensity detected at this m / z. That is, the spectrum reconstruction unit 311d reconstructs a mass spectrum from the significant m / z and the signal intensity detected at this m / z among the mass spectra acquired by the spectrum acquisition unit 311c. FIG. 2 is a diagram illustrating an example of a result of reconstructing a mass spectrum by extracting and arranging extracted m / z according to a negative mode and a positive mode. In this embodiment, 52 m / z was extracted for each of the negative mode and the positive mode. Therefore, in this example, the dimension of data is 104 dimensions. As described above, in this embodiment, the ratio m / z is measured in each of the negative mode and the positive mode. In the example shown in FIG. 2, the extracted m / z is sequentially numbered 1, 2,..., And the numbers in the range 1 to 52 correspond to the measurement results in the negative mode, and 53 to 104. The number in the range corresponds to the measurement result in the positive mode. Note that the information including the component (constituent element) of the fine particle P is specified by m / z.

[Spectrum normalization unit 311e]
In the present embodiment, since a 104-dimensional mass spectrum is measured, a 104-dimensional vector having a signal intensity value at a peak (convex upward) of the mass spectrum as an element is obtained. In the mass spectrum, the value of the element corresponding to the region where the signal intensity value is 0 (zero) is 0 (zero). In the following description, such a vector obtained from sample particles will be referred to as a sample particle intensity vector as necessary. Each element of the sample particle intensity vector is referred to as an intensity value as necessary.

  Depending on the size of the sample particles, the level of the signal intensity of the mass spectrum changes. Therefore, the spectrum normalization unit 311e normalizes the intensity value in each mass spectrum reconstructed by the spectrum reconstruction unit 311d with the size of the sample particle that is the measurement target of the mass spectrum (intensity value in the mass spectrum). Is changed according to the size of the sample particle that is the measurement target of the mass spectrum). Specifically, the spectrum normalization unit 311e divides the intensity value in the mass spectrum by the size of the sample particle that is the measurement target of the mass spectrum. Note that, as described above, in the present embodiment, the number of pixels in which fine particles are present in the SEM image is the size of the fine particles. In this way, sample particle intensity vectors normalized by the size of the fine particles are obtained for each of the sample particles.

[Spectrum standardization unit 311f]
The sample particle intensity vector is a 104-dimensional vector, but the value level varies for each dimension (each element). Therefore, the spectrum standardization unit 311f standardizes each sample particle intensity vector normalized by the size of the fine particles so that the average is 0 (zero) and the standard deviation is 1 in all dimensions (elements). . Note that variance may be used instead of the standard deviation.

Assuming that a variable for specifying a sample particle intensity vector is i (this variable i is also a variable for specifying a sample particle and a mass spectrum), a sample particle intensity vector x _i normalized by the size of a fine particle is expressed by the following ( 1) It represents with a formula. In the formula (1), T represents transposition (this is the same in other formulas). In this embodiment, the column vector is expressed as the following equation (2), and the row vector is expressed as the following equation (3).

Further, j is a variable for specifying the dimension (element) of the sample particle intensity vector x _i normalized by the size of the fine particles, and N _teacher is the number of the sample particle intensity vectors x _i (that is, N _teacher pieces). It is assumed that a mass spectrum is measured for each sample particle). Then, the average value μ _j and the standard deviation σ _{j for} each element of the sample particle intensity vector x _i normalized by the size of the fine particles are expressed by the following expressions (4) and (5), respectively.

The spectrum normalization unit 311f calculates the average μ _j by performing the calculations of the expressions (4) and (5) for the element x _ij of the same dimension of the sample particle intensity vector x _i normalized by the size of the fine particles. The value of the element x _ij is standardized so that 0 (zero) and the standard deviation σ _j are 1. The spectrum standardization unit 311f individually performs such correction for each of the 104 dimensions. In this way, each of the sample particle intensity vectors x _i normalized by the size of the fine particles is standardized. The vector z _i obtained in this way is expressed by the following equations (6) and (7).

[Spectrum average value deriving unit 311g]
The spectrum average value deriving unit 311g, for each of the vectors z _i derived by the spectrum standardizing unit 311f, the average value of the dimension (element) corresponding to the negative mode and the value of the dimension (element) corresponding to the positive mode. The average value of is derived.
FIG. 3 is a diagram illustrating an example of an average value of dimensions (elements) corresponding to the negative mode and an average value of dimensions (elements) corresponding to the positive mode in each vector z _i . The average signal intensity on the vertical axis in FIG. 3 represents an average value of dimensions (elements) corresponding to the negative mode or an average value of dimensions (elements) corresponding to the positive mode. Data No. on the horizontal axis in FIG. 3 represents a number that identifies the vector z _i (that is, the sample particle). FIG. 3A is a diagram showing the average signal intensity in all the vectors z _i (sample particles), and FIG. 3B is an enlarged view of the data Nos. 1 to 100 in FIG. FIG. 2A is a mass spectrum for the sample particle with data No. 7 in FIG. 3, and FIG. 2B is a mass spectrum for the sample particle with data No. 66 in FIG.

[Noise removal unit 311h]
For each of the vectors z _i derived by the spectrum standardization unit 311f, the noise removal unit 311h calculates the average value of each dimension (element) corresponding to the negative mode and the value of each dimension (element) corresponding to the positive mode. If at least one of the average value exceeds the threshold (> 0), the vector z _i is excluded from the processing target as noise. The threshold value is determined in advance by trial and error so that, for example, only the vector z _i that causes noise can be excluded as much as possible.

In FIG. 3B, when the threshold value is 5, the vector z _i (sample particle) having the data numbers of 34, 66, 73, and 82 is excluded from the processing target. As shown in FIG. 2B, the mass spectrum for the sample particle with data No. 66 is greatly oscillated in the region corresponding to the negative mode, and is found to be noise. As shown in FIG. 3B, the average signal intensity for the vector z _i (sample particle) with data No. 66 exceeds the threshold value of 5. For this reason, the vector z _i (sample particle) having the data No. 66 is excluded from the processing target.

In this case, the noise removal unit 311h instructs the spectrum standardization unit 311f to re-derived the vector z _i . Upon receipt of this instruction, the spectrum standardization unit 311f includes samples other than the sample particle intensity vector corresponding to the vector z _i removed from the processing target as noise among the sample particle intensity vectors x _i normalized by the size of the fine particles. A standardized version of each is re-derived as a vector z _i in the same manner as described above. Then, the spectrum standardization unit 311f outputs the re-derived vector z _i to the dimension compression unit 312.

On the other hand, for each of the vectors z _i derived by the spectrum standardization unit 311f, the average value of each dimension (element) corresponding to the negative mode, and the average value of each dimension (element) corresponding to the positive mode, If any of the above does not exceed the threshold (> 0), the noise removal unit 311h does not instruct the spectrum standardization unit 311f to instruct the re-derivation of the vector z _i described above, and the vector z derived by the spectrum standardization unit 311f. _i is directly output to the dimension compression unit 312.

((Dimension compression unit 312))
In the mass spectrum, there are components (constituent elements) that have the same value for many fine particles P regardless of the generation source. In this case, even if the dimension (element) corresponding to such a component of the vector z _i is not considered, the accuracy of classification in the classifier is not greatly affected. Therefore, in the present embodiment, the dimension compression unit 312 performs a principal component analysis on the vector z _i output from the noise removal unit 311h, and based on the result of the principal component analysis, the dimension of the vector z _i is calculated. Compress (reduce the order of the vector z _i ).

When principal component analysis is performed on the vector z _i output from the noise removing unit 311h, principal component vectors w _{1 to} w ₁₀₄ shown in the following equation (8) are obtained.

Then, for each of the vectors z _i output from the noise removing unit 311h, the first principal component y _i1 to the 104th principal component y _i104 are obtained as shown in the following equation (9).

The dimension compression unit 312 includes principal component vectors w ₁ to w _s and first principal components y _i1 to _s-th principal among principal component vectors w _{1 to} w ₁₀₄ and first principal components y _i1 to 104th principal components y _i104. Select component y _is . For example, s is determined in advance by trial and error so that the accuracy of classification in the classifier is as high as possible. Note that the same value is used as the value of s for each of the vectors z _i output from the noise removing unit 311h.
From the vector z _i output from the noise removing unit 311h as described above, as shown in the following equation (10), the feature vector y _i having the first principal component to the s-th principal component of the vector z _i as elements. Is obtained. That is, among the information held by the original vector z _i, while retaining the information that contributes to the construction of classifier as possible, feature vector y _i which has been compressed in the low-dimensional can be obtained than the original vector z _i.

((Classifier information deriving unit 313))
A feature vector y _i is obtained for each sample particle whose source is known (variable i is a variable that identifies the sample particle). One source is _defined as one class, and the class is _denoted as C _k (k = 1, 2,..., N). The number of sample particles belonging to each class C _k is denoted as N _k (N _k is also the number of feature vectors y _i belonging to each class C _k ). In addition, the total number of (different from each other) sources of all the sample particles is expressed as n (the same source is not counted twice when n is counted). Then, the feature vector y _i ^(k) belonging to each class C _k is expressed as the following equation (11).

Classifier information deriving unit 313, for each class C _k, derives the center of gravity b _k feature vector y _i belonging to the class C ^{_{k (k).}} In the present embodiment, a case where the center of gravity b _k is an average value will be described as an example. Then, the classifier information deriving unit 313 derives the center of gravity b _k of the feature vector y _i ^(k) by the following equation (12).

The center of gravity b _k is not limited to an average value, and a representative value such as a median value may be used, for example.
The classifier information deriving unit 313 derives the variance var _k of the feature vector y _i ^(k) belonging to the class C _k for each class C _k . As the variance var _k , for example, a variance var _k ^(EUC) based on the Euclidean distance or a variance var _k ^(MAH) based on the Mahalanobis distance can be employed. The variance var _k ^(EUC) based on the Euclidean distance is expressed by the following equation (13), and the variance var _k ^(MAH) based on the Mahalanobis distance is expressed by the following equation (14).

In Equation (14), S _k is a variance covariance matrix obtained from the feature vector y _i ^(k) belonging to the class C _k . The portion of (y _i ^(k) −b _k ) S _k ⁻¹ (y _i ^(k) −b _k ) ^T corresponds to the square of the Mahalanobis distance. Thus, the Mahalanobis distance includes division by the variance-covariance matrix. For this reason, when the value spread (degree of variation ⁾ in the feature vector y _i ^(k) differs depending on the dimension, the Mahalanobis distance can be treated more fairly than the Euclidean distance.
As described above, the center of gravity b _k and the variance var _k are obtained for each of the classes C _k (sources).

((Classifier information storage unit 314))
The classifier information storage unit 314 stores the centroid b _k and variance var _k of each class C _k derived by the classifier information deriving unit 313.

(Source analysis unit 320)
The source analysis unit 320 includes an image acquisition unit 321, a particle size derivation unit 322, a mass-to-charge ratio acquisition unit 323, a spectrum reconstruction unit 324, a spectrum normalization unit 326, a dimension compression unit 327, and a discriminant function derivation. A unit 328, a source derivation unit 329, and a display unit 330.

((Image acquisition unit 321))
The image acquisition unit 321 acquires an SEM image for the analysis target particle obtained by the analysis apparatus 200. As described above, the analysis target particle is a fine particle whose generation source is unknown. Further, the number of analysis target particles included in this SEM image may be one or plural. When there are a plurality of analysis target particles included in the SEM image, for each of the plurality of analysis target particles, a particle size deriving unit 322, a mass-to-charge ratio acquisition unit 323, a spectrum reconstruction unit 324, and a spectrum normalization are performed. The processing of the unit 325, the spectrum normalization unit 326, the dimension compression unit 327, the discriminant function derivation unit 328, the source derivation unit 329, and the display unit 330 is performed individually.

((Particle size deriving unit 322))
The particle size deriving unit 311b derives the size of the analysis target particle included in the SEM image based on the SEM image acquired by the image acquiring unit 311a. As described above, in the present embodiment, the size of the fine particles P is the number of pixels in which the fine particles P are present in the SEM image.

((Spectrum acquisition unit 323))
The spectrum acquisition unit 323 acquires the mass-to-charge ratio m / z of secondary ions with respect to the analysis target particle and the signal intensity detected at the m / z obtained by the analyzer 200 as a mass spectrum.
((Spectrum reconstruction unit 324))
The spectrum reconstruction unit 324 reconstructs a mass spectrum from the significant m / z and the signal intensity detected at the m / z among the mass spectrum acquired by the spectrum acquisition unit 323.

((Spectrum normalization unit 325))
As described above, since a 104-dimensional mass spectrum is measured in this embodiment, a 104-dimensional vector having a signal intensity value at a peak (convex upward) of the mass spectrum as an element is obtained. In the following description, such a vector obtained from the analysis target particle is referred to as an analysis target particle intensity vector as necessary. Each element of the analysis target particle intensity vector is referred to as an intensity value as necessary.

  The spectrum normalization unit 325 normalizes the intensity value in the mass spectrum derived by the spectrum reconstruction unit 324 with the size of the analysis target particle that is the measurement target of the mass spectrum (the intensity value in the mass spectrum is It is changed according to the size of the particle to be analyzed, which is the measurement target of the mass spectrum). Specifically, the spectrum normalization unit 325 divides the intensity value in the mass spectrum by the size of the analysis target particle that is the measurement target of the mass spectrum. The analysis target particle intensity vector X normalized by the size of the analysis target particle in this way is expressed by the following equation (15).

((Spectrum standardization unit 326))
Since the classifier (the center of gravity b _k and the variance var _{k of} each class C _k ) is constructed based on the sample particle intensity vector, the analysis target particle intensity vector needs to be handled with reference to the sample particle intensity vector. Therefore, the spectrum standardization unit 326 derives the vector Z by calculating the following equation (16) using the average value μ _j and the standard deviation σ _j derived by the spectrum standardization unit 311f. That is, the spectrum standardization unit 326 derives the values of the elements X _{1 to} X ₁₀₄ of the analysis target particle intensity vector X normalized by the size of the analysis target particle by the spectrum normalization unit 325 by the spectrum standardization unit 311f. A vector Z having a value normalized based on the average value μ _j and the standard deviation σ _j as an element is derived.

((Dimension compression unit 327))
The dimension compression unit 327 calculates the following equation (17) using the vector Z derived by the spectrum standardization unit 326 and the principal component vectors w _{1 to} w _s derived by the dimension compression unit 312. Thus, a feature vector Y having the first principal component to the s-th principal component of the vector Z as elements is derived.

((Discriminant function deriving unit 328))
The discriminant function deriving unit 328 uses the feature vector Y derived by the spectrum standardization unit 326 and the centroid b _k and variance var _{k of} each class C _k stored by the classifier information storage unit 314 to The value of the discriminant function h _k (Y) for class C _k is derived. The value of the discriminant function h _k (Y) for a certain class C _k is a value serving as an index for discriminating whether or not the feature vector Y is similar to the feature vector y _i ^(k) belonging to the class C _k. It is.

When the classifier information deriving unit 313 derives the variance var _k ^(EUC) based on the Euclidean distance ⁽ see equation (13)), the discriminant function h _k (Y) is expressed by the following equation (18). The On the other hand, when the classifier information deriving unit 313 derives the variance var _k ^(MAH) based on the Mahalanobis distance (see equation (14)), the discriminant function h _k (Y) is expressed by the following equation (19): expressed.

In this way, the equations (18) and (19) express the square of the distance between the feature vector Y of the particle to be analyzed and the centroid b _k of the feature vector y _i ^(k) of the sample particle as the variance var _k ^{(EUC )} , Var _k ^(MAH) . The variances var _k ^(EUC) and var _k ^(MAH) represent the average of the scattering from the centroid b _k of the feature vector y _i ^(k) , and correspond to the size of the class C _k (in data space). Accordingly, the discriminant function h _k (Y) is a scale corresponding to the size of the class C _k , and how far the feature vector Y of the particle to be analyzed is from the center of gravity b _k of the feature vector y _i ^(k) of the sample particle. Indicates whether or not In the examples shown in the equations (18) and (19), when the value of the discriminant function h _k (Y) is large, the feature vector Y of the particle to be analyzed is the centroid b _k of the feature vector y _i ^(k) of the sample particle. Indicates that you are away from In addition, the value of the discriminant function h _k (Y) in each class C _k can be compared as it is. That is, when towards the value of the discriminant function h _k (Y) at a certain class C _k is greater than the value of the discriminant function h _k (Y) in the class C _k another certain feature vector Y of the analyzed particles It will be similar to the latter class C _k than the former class C _k.

((Source Deriving Unit 329))
When the discriminant function derivation unit 328 derives the values of the discriminant function h _k (Y) for all classes C _k , the source derivation unit 329 calculates the discriminant function h _k ( It is determined whether or not the value of Y) exceeds a threshold value T.

When the value of the discriminant function h _k (Y) for the class C _k does not exceed the threshold T, the generation source deriving unit 329 determines that the feature vector Y of the analysis target particle is similar to the class C _k , and A generation source corresponding to C _k is set as a candidate generation source of the analysis target particle. When the generation source candidate of the analysis target particle is obtained, the generation source deriving unit 329 outputs information on the generation source candidate to the display unit 330. When a plurality of candidates are obtained as the generation source candidates of the analysis target particle, the generation source deriving unit 329 derives the distance between the feature vector Y of the analysis target particle and the centroid of the plurality of candidates or an index of the distance, respectively. Then, it is possible to output information on one candidate that minimizes the distance or the index of the distance to the display unit 330. As the distance index, for example, the square of the Euclidean distance or the square of the Mahalanobis distance can be used. This is preferable because it is not necessary to calculate the square root, but the distance itself (for example, the Euclidean distance or the Mahalanobis distance) may be used. However, it is not always necessary to do this, and the generation source deriving unit 329 may output information on the plurality of candidates to the display unit 330.

In addition, when the value of the discriminant function h _k (Y) for all classes C _k derived by the discriminant function deriving unit 328 exceeds the threshold value T, the source derivation unit 329 rejects the feature vector Y of the analysis target particle. . That is, the generation source derivation unit 329 determines that the generation source of the analysis target particle is a generation source different from the one or more predetermined generation sources, and displays information indicating that on the display unit 330. Output.

  Here, if the threshold value T is excessively increased, the generation source of the analysis target particle is easily determined to be a predetermined generation source. Therefore, in fact, there is a possibility that even a particle to be analyzed of a generation source different from a predetermined generation source may be erroneously determined to be a predetermined generation source. On the other hand, if the threshold value T is too small, it is easy to determine that the generation source of the analysis target particle is a generation source different from the predetermined generation source. Therefore, in fact, there is a possibility that even a particle to be analyzed of a predetermined generation source may be erroneously determined as a generation source different from the predetermined generation source. For example, the threshold value T is determined in advance from such a viewpoint by trial and error so that the misjudgment rate becomes as small as possible.

  Similar to the data preprocessing unit 311, the source analysis unit 320 also uses the average value of each dimension (element) corresponding to the negative mode and the positive mode for the vector Z derived by the spectrum standardization unit 326. It may be determined whether or not at least one of the average values of the dimensions (elements) corresponding to 1 exceeds a threshold value (> 0). For the vector Z derived by the spectrum standardization unit 326, at least one of the average value of each dimension (element) corresponding to the negative mode and the average value of each dimension (element) corresponding to the positive mode When the value exceeds the threshold value (> 0), the generation source deriving unit 329 determines that the vector Z (particle to be analyzed) is noise, and the noise is included in the mass spectrum, so that the source of the particle to be analyzed is specified. Information indicating that it cannot be performed may be output to the display unit 330.

As described above, in the present embodiment, the processing logic performed by the discriminant function deriving unit 328 and the source deriving unit 329 is a classifier. That is, by inputting the feature vector Y derived by the spectrum standardization unit 326 and the centroid b _k and variance var _{k of} each class C _k stored in the classifier information storage unit 314 to this logic, The analysis target particles are classified into at least one of the predetermined generation sources or a generation source different from the predetermined generation source, and information indicating the result is obtained.

((Display unit 330))
The display unit 330 displays the information output from the source derivation unit 329.

<Operation flowchart>
Next, an example of processing of the particle source analysis device 300 (classifier construction unit 310) when constructing a classifier will be described with reference to the flowchart of FIG.
First, in step S401, the image acquisition unit 311a acquires SEM images for a plurality of sample particles obtained by the analyzer 200.
Next, in step S402, the particle size deriving unit 311b derives the size of each sample particle included in the SEM image acquired in step S401.

Next, in step S403, the spectrum acquisition unit 311c acquires a mass spectrum for each sample particle.
Next, in step S404, the spectrum reconstruction unit 311d reconstructs the mass spectrum for each sample particle acquired in step S403.

Next, in step S405, the spectrum normalization unit 311e normalizes the intensity value in each mass spectrum reconstructed in step S404 with the size of the sample particle that is the measurement target of the mass spectrum, and the size of the fine particle. in deriving the sample particle strength vectors x _i which are standardized.
Next, in step S406, the spectrum standardization unit 311f standardizes each of the sample particle intensity vectors x _i normalized by the size of the fine particles by performing the calculations of equations (4) to (7). Deriving a vector z _i .

Next, in step S407, the spectrum average value deriving unit 311g obtains the average value of the dimension (element) value corresponding to the negative mode and the dimension corresponding to the positive mode for each of the vectors z _i derived in step S406. The average value of the (element) values is derived.
Next, in step S408, the noise removal unit 311h corresponds to the average value of each dimension (element) for each measurement mode (the average value of each dimension (element) corresponding to the negative mode and the positive mode. It is determined whether or not there is a vector z _i (at least one of the average value of each dimension (element)) exceeding a threshold (> 0) (one or more). As a result of this determination, if there is no vector z _i (there is no vector) in which the average value of each dimension (element) for each measurement mode exceeds the threshold value, the noise removal unit 311h determines the vector z _i (in step S406). Are all output to the dimension compression unit 312. And a process abbreviate | omits step S409 and progresses to step S410 mentioned later. On the other hand, if there is a vector z _{i in} which the average value of each dimension (element) for each measurement mode exceeds the threshold, the process proceeds to step S409.

In step S409, the noise removing unit 311h excludes the vector z _i whose average value of each dimension (element) for each measurement mode exceeds the threshold from the processing target. Then, the spectrum standardization unit 311f standardizes each of the sample particle intensity vectors normalized by the size of the fine particles, other than the sample particle intensity vector corresponding to the vector z _i removed from the processing target, to obtain the vector z Re-derived _i . Then, the spectrum standardization unit 311f outputs the re-derived vector z _i to the dimension compression unit 312. Then, the process proceeds to step S410.
In step S410, the dimension compressing unit 312 performs principal component analysis on the vector z _i output from the noise removing unit 311h, and based on the result of the principal component analysis, the principal component vectors w _{1 to} w ₁₀₄ and the first The principal component vectors w _{1 to} w _s and the first principal component y _i1 to the sth principal component y _is selected from the principal components y _i1 to 104th principal component y _i104 , and the first principal component to the first principal component of the vector z _i are selected. The feature vector y _i having the s principal component as an element is derived.

Next, in step S411, the classifier information deriving unit 313 derives the centroid b _k and the variance var _{k of the} feature vector y _i ^(k) belonging to the class C _k for each class C _k ((12 ) Formula to (14) formula).
Finally, in step S412, the classifier information storage unit 314 stores the centroid b _k and variance var _k of each class C _k derived in step S411.

  Next, an example of processing of the particle generation source analysis apparatus 300 (generation source analysis unit 320) when analyzing the generation source of the analysis target particle will be described with reference to the flowchart of FIG. Here, a case where there is one analysis target particle to be processed will be described as an example. For example, when there are a plurality of analysis target particles included in the SEM image, the processes of steps S502 to S515 may be performed for each of the analysis target particles.

First, in step S501, the image acquisition unit 321 acquires an SEM image for the analysis target particle.
Next, in step S502, the particle size deriving unit 311b derives the size of the analysis target particle included in the SEM image acquired in step S501.

Next, in step S503, the spectrum acquisition unit 323 acquires a mass spectrum for the analysis target particle.
Next, in step S504, the spectrum reconstruction unit 324 reconstructs the mass spectrum for the analysis target particle acquired in step S503.

Next, in step S505, the spectrum normalization unit 325 normalizes the intensity value in the mass spectrum derived in step S504 with the size of the analysis target particle that is the measurement target of the mass spectrum, and uses the size of the fine particle. A normalized analysis target particle intensity vector X is derived.
Next, in step S506, the spectrum standardization unit 326 calculates the equation (16) using the average value μ _j and the standard deviation σ _j derived in step S406 or S409, thereby analyzing the particle size vector X to be analyzed. To derive a vector Z.

Next, in step S507, the dimension compressing unit 327 uses the vector Z derived by the spectrum standardizing unit 326 and the principal component vectors w _{1 to} w _s derived by the dimension compressing unit 312 to By performing the calculation, a feature vector Y whose elements are the first principal component to the s-th principal component of the vector Z is derived.
Next, in step S508, the discriminant function deriving unit 328 selects one unselected class C _k from among the predetermined classes C _k (generation sources).

Next, in step S509, the discriminant function deriving unit 328 uses equation (13) using the feature vector Y derived in step S507, the centroid b _k and the variance var _{k of} the class C _k selected in step S508. Alternatively, the value of the discriminant function h _k (Y) for the class C _k selected in step S508 is derived by calculating the equation (14). The center of gravity b _k and the variance var _k of the class C _k are stored in step S412 of the flowchart of FIG.

Next, in step S510, the source deriving unit 329 determines whether or not the value of the discriminant function h _k (Y) for the class C _k selected in step S508 exceeds a threshold value T. As a result of this determination, if the value of the discriminant function h _k (Y) for the class C _k selected in step S508 exceeds the threshold value T, the process skips step S511 and proceeds to step S512 described later. On the other hand, if the value of the discriminant function h _k (Y) for the class C _k selected in step S508 does not exceed the threshold value T, the process proceeds to step S511.

In step S511, the generation source deriving unit 329 sets the generation source corresponding to the class C _k selected in step S508 as the generation source candidate of the analysis target particle to be processed. Then, the process proceeds to step S512.
In step S512, the generation source deriving unit 329 determines whether or not all of the predetermined classes C _k (generation sources) have been selected. As a result of the determination, if all of the class C _k a predetermined is not selected, the process returns to step S508. Then, the processes in steps S508 to S512 are repeatedly executed until it is specified whether or not the generation sources corresponding to all the predetermined classes C _k are the generation target candidates of the analysis target particles to be processed. The

When all classes C _k predetermined is selected, the process, the program goes on to Step S513. In step S513, the generation source deriving unit 329 determines whether or not there are (one or more) generation source candidates for the analysis target particles to be processed. As a result of the determination, if there is a candidate for the generation source of the analysis target particle to be processed, the process proceeds to step S514. In step S514, the display unit 330 displays information on generation source candidates of the analysis target particles to be processed. As described in the section of ((source generation unit 329)), when a plurality of candidates are obtained as generation source candidates for the analysis target particle, the generation source derivation unit 329 displays the feature vector of the analysis target particle. A distance or an index of distance between Y and the centroids of the plurality of candidates can be derived, and information on one candidate that minimizes the distance or the index of distance can be output to the display unit 330.
On the other hand, if there is no candidate for the generation source of the analysis target particle to be processed, the process proceeds to step S515. In step S515, the display unit 330 displays information indicating that the generation source of the analysis target particle to be processed is a generation source different from the predetermined generation source.

<Example>
Next, an example of the particle generation source analysis system of this embodiment will be described. In this embodiment, in the area where seven factories A to G are gathered, factories A to G are used as known sources, and sample particles collected at factories A to G are used to classify the classifiers (for each class C _k . The centroid b _k and variance var _k ) were constructed. At that time, the dimension compression unit 312 selects the first principal component y _i1 to the ninth principal component y _i9 . In addition, the variance var _k ^(MAH) (Equation (14) ⁾ based on the Mahalanobis distance is used as the variance var _k of the feature vector y _i ^{(k), and} the equation (19) is used as the discriminant function h _k (Y). It was. In addition, 4.0 is used as the threshold T.

  After the classifier is constructed as described above, the generation source of the fine particles collected at each location shown in FIG. 6 is analyzed by the particle generation source analyzer 300, and the generation source obtained from the result and the analysis target particle We investigated whether it matched with the actual source of. That is, originally, a fine particle whose source is unknown is set as an analysis target particle, and the source is analyzed by the particle source analysis device 300. In this embodiment, in order to confirm the performance of the particle source analysis device 300, The fine particles collected at the factories A to G are also analyzed by the particle generation source analyzer 300 as analysis target particles. The number of fine particles collected at each location in FIG. 6 was 500 to 3000, respectively. An example of the result is shown in FIG.

  In FIG. 6, the boundary H indicates a region in which the wind direction is from the inside to the outside of the area among the boundaries of the area where the factories A to G gather. The boundary I refers to an area in the direction in which the wind direction is from the outside to the inside of the area among the boundaries of the area where the factories A to G are gathered. External J and K refer to the source outside the area where factories A to G gather. Exhaust gases unrelated to the area where the factories A to G gather are discharged from the external J and K.

Further, the rejection rate refers to any source (factory A to factory) determined in advance as a result of analysis by the particle source analysis device 300 among a plurality of fine particles collected at a certain (one) collection point. G) also represents the percentage of the number of fine particles that did not become a generation source candidate as a percentage.
As described above, in this embodiment, the classifier is constructed using the sample particles collected in the factories A to G. Therefore, for the fine particles collected in the factories A to G, the smaller the rejection rate, the higher the accuracy of analysis by the particle source analysis device 300. On the other hand, for fine particles collected at different locations from the factories A to G (boundaries H to I, external J to K), the accuracy of analysis by the particle source analysis device 300 is higher as the rejection rate is larger. Become.

  Referring to FIG. 6, the fine particles collected at factories A to G have a small rejection rate, and the fine particles collected at locations different from factories A to G (boundaries H to I, external J to K). Shows that the rejection rate is increasing. Since the wind direction at the boundary H is a direction from the inside to the outside of the area, and the wind direction at the boundary I is a direction from the outside to the inside of the area, the factories A to G are discharged at the boundary H compared to the boundary I. There is a high possibility of collecting fine particles as a source. The low rejection rate at the boundary H corresponds to this.

  In addition, the rejection rate in the factories A to G is originally 0 (zero), but if the threshold value T is determined so that the rejection rate in the factories A to G becomes 0 (zero), in the particle source analysis apparatus 300 Actually, the generation source of the analysis target particles (boundary HI, external J to K) collected in a place (boundary HI, external J to K) different from the factory A to G is the factory A to The possibility of misjudging to be G increases. For this reason, it is preferable to determine the threshold value T so as to satisfy both the reduction of the rejection rate in the factories A to G and the increase of the rejection rate in the boundaries H to I and the external J to K. However, when it is not desired to overlook the fine particles whose emission sources are the factories A to G, the threshold value T is increased, and conversely, when the fine particles whose emission sources are the boundaries H to I and the external J to K are not desired to be missed. The threshold value T can be appropriately determined according to the purpose, such as reducing the threshold value T. Thus, the rejection rate in factories A to G does not have to be 0 (zero).

<Summary>
As described above, in the present embodiment, the particle source analysis apparatus 300 uses sample particles that are fine particles collected by one or more predetermined generation sources, and uses the component (constituent element) as an attribute of the fine particles. ) From the attribute including the at least one of one or more predetermined generation sources and a generation source other than the one or more predetermined generation sources (predetermined one Or a classifier that is classified into either one of a plurality of generation sources). Then, the particle generation source analysis apparatus 300 obtains information on the generation source of fine particles whose generation source is unknown using a classifier. Therefore, it is possible to suppress the analysis result from changing every time the generation source of fine particles whose generation source is unknown is analyzed. Therefore, the generation source of fine particles can be appropriately evaluated. In addition, the fine particles are classified into at least one of one or more predetermined generation sources and one of the generation sources other than the one or more predetermined generation sources, and all generations are classified. Do not specify individual sources. Since the number of generation sources of fine particles collected in the general environment is enormous, it is not possible to specify all these generation sources. Suppose that Therefore, for example, when the source to be known is specified, it is simply determined how many of the plurality of fine particles collected in the general environment have the source as the source. be able to.

<Modification>
In the present embodiment, the case where the analysis apparatus 200 performs measurement up to the SEM image and the mass-to-charge ratio m / z of the secondary ions has been described as an example. However, this is not always necessary. For example, the analyzer 200 derives the particle diameter of the fine particles from the SEM image and derives the mass spectrum from the mass-to-charge ratio m / z of the secondary ions. At least one of them may be performed.

  In the present embodiment, the case where the classifier 100 is a virtual impactor has been described as an example. However, as described in Patent Document 1, the classifier 100 is not limited to a virtual impactor, and may be a portable impactor or the like. In addition, fine particles may be collected by the methods described in Non-Patent Documents 1 and 2.

  Moreover, in this embodiment, the case where the analyzer 200 is a FIB-EB-TOF-SIMS device has been described as an example. However, for each fine particle P, if the component, the signal intensity of m / z, the particle diameter, etc. can be measured, information for determining the component (constituent element) of the fine particle P and the attribute of the component (constituent element) can be measured. The analysis apparatus 200 is not limited to the FIB-EB-TOF-SIMS apparatus. For example, as described in Patent Document 1, an SEM-EDX (SEM-Energy Dispersive X-ray Spectroscopy) apparatus may be used as the analysis apparatus 200. Further, as the analysis apparatus 200, a FIB-EB-TOF-SIMS apparatus and a SEM-EDX apparatus may be combined. In addition, as the analyzer 200, EPMA (Electron Probe Micro Analyzer) may be used.

Further, in the present embodiment, the spectrum average value deriving unit 311g corresponds to the average value of the dimension (element) value corresponding to the negative mode and the positive mode for each of the vectors z _i derived by the spectrum standardizing unit 311f. The case of deriving the average value of the dimensions (elements) to be performed has been described as an example. However, a representative value such as a median value other than the average value may be used.
Further, in the present embodiment, the classifier information deriving unit 313, for each class C _k, and the case of deriving the variance var _k feature vector y _i belonging to the class C _k ^(k) is described as an example . However, the standard deviation may be derived instead of the variance.
Further, in the present embodiment, the case where the classifier construction unit 310 and the generation source analysis unit 320 are realized by the same device (particle generation source analysis device 300) has been described as an example. However, the classifier construction unit 310 and the source analysis unit 320 may be realized by separate devices.

The embodiment of the present invention described above can be realized by a computer executing a program. Further, a computer-readable recording medium in which the program is recorded and a computer program product such as the program can also be applied as an embodiment of the present invention. As the recording medium, for example, a flexible disk, a hard disk, an optical disk, a magneto-optical disk, a CD-ROM, a magnetic tape, a nonvolatile memory card, a ROM, or the like can be used.
In addition, the embodiments of the present invention described above are merely examples of implementation in carrying out the present invention, and the technical scope of the present invention should not be construed as being limited thereto. Is. That is, the present invention can be implemented in various forms without departing from the technical idea or the main features thereof.

<Relationship with Claims>
Below, an example of the correspondence between a claim and an embodiment is explained. Note that the description of the claims is not limited to the description of the embodiment, as described in the modification.
(Claim 1)
The classifier construction unit is realized by using the classifier construction unit 310, for example.
The attribute of the sample particle is realized, for example, by information including the component (constituent element) of the fine particle P specified by the mass-to-charge ratio m / z of the secondary ions.
The sample particle attribute information is realized by, for example, a feature vector y _i .
The source analysis means is realized by using the source analysis unit 320, for example.
The analysis target particle attribute information is realized by, for example, the feature vector Y.
The classifier information is realized by, for example, the centroid b _k and the variance var _k of the feature vector y _i ^(k) belonging to each class C _k .
The information on the generation source of the analysis target particle is, for example, a generation source different from the generation source information of the analysis target particle to be processed and the generation source in which the generation source of the analysis target particle to be processed is predetermined. This is realized by information indicating this.
The classifier is realized by, for example, logic of processing performed by the discriminant function deriving unit 328 and the generation source deriving unit 329.
(Claim 2)
The first normalization unit is realized by using, for example, a spectrum normalization unit 311e.
The vector whose element is a value based on the attribute of the sample particle is realized by a sample particle intensity vector, for example.
The information indicating the size of the sample particle is realized by, for example, the number of pixels in which the sample particle exists in the SEM image.
The first standardization unit is realized by using, for example, the spectrum standardization unit 311f.
The vector normalized by the first normalization means, and the representative value and the variation value of the element corresponding to each other for the plurality of sample particles are, for example, a sample particle intensity vector This is realized by the average value μ _j and standard deviation σ _{j for} each element.
The second normalization means is realized, for example, by using the spectrum normalization unit 325.
The vector whose element is a value based on the attribute of the analysis target particle is realized by, for example, an analysis target particle intensity vector.
The information indicating the size of the analysis target particle is realized by, for example, the number of pixels in the SEM image where the analysis target particle exists.
The second standardization means is realized by using, for example, the spectrum standardization unit 326.
(Claim 3)
The first dimension compressing means is realized by using, for example, the dimension compressing unit 312.
The principal component vector of the vector is realized by, for example, principal component vectors w _{1 to} w ₁₀₄ . The principal components of the vector are realized by, for example, the first principal component y _i1 to the 104th principal component y _i104 .
Selecting a predetermined number of principal component vectors and principal components from the principal components includes, for example, selecting the principal component vectors w ₁ to w _s from the principal component vectors w _{1 to} w ₁₀₄ , This _is realized by selecting the first principal component y _i1 to the s-th principal component y _is from the component y _i1 to the 104th principal component y _i104 .
A vector whose element is the value of the selected principal component is realized by, for example, a feature vector y _i (Equation (10)).
The second dimension compression unit is realized by using, for example, the dimension compression unit 327.
The principal component vectors selected by the first dimension compression means are realized by using, for example, principal component vectors w _{1 to} w _s .
A vector whose element is a value based on the attribute of the particle to be analyzed is realized by using a vector Z (Equation (16)), for example.
Conversion of a vector whose element is a value based on the attribute of the particle to be analyzed into a principal component is realized by, for example, Expression (17), and a vector whose element is the value of the principal component is, for example, a feature vector Y It is realized by.
(Claim 4)
The classifier information deriving means is realized by using the classifier information deriving unit 313, for example.
Same centroid and the variation value of the vector for each of said predetermined the collected the sample particles at the source, for example, dispersing the centroid b _k feature vector y _i ^(k) belonging to each class C _k It is realized by var _k .
(Claim 5)
The evaluation means is realized by using a discriminant function deriving unit 328, for example.
The source analysis unit is realized by using, for example, the source derivation unit 329.
(Claim 8)
The value of the signal intensity indicating the peak in the mass spectrum of the sample particle is realized by, for example, a value based on the element value (intensity value) of the sample particle intensity vector.
The value of the signal intensity indicating a peak in the mass spectrum of the analysis target particle is realized by, for example, a value based on the value (intensity value) of the element of the analysis target particle intensity vector.
(Claim 9)
The spectrum representative value deriving unit is realized by using, for example, a spectrum average value deriving unit 311g.
The noise removing unit is realized by using, for example, a noise removing unit 311h.

  100: Classifier, 200: Analyzer, 300: Particle source analyzer, 310: Classifier construction unit, 311: Data preprocessing unit, 312: Dimension compression unit, 313: Classifier information deriving unit, 314: Classifier Information storage unit, 320: generation source analysis unit, 321: image acquisition unit, 322: particle size deriving unit, 323: spectrum acquisition unit, 324: spectrum reconstruction unit, 325: spectrum normalization unit, 326: spectrum standardization unit 327 : Dimension compression unit, 328: Discriminant function deriving unit, 329: Generation source deriving unit, 330: Display unit

Claims (11)

A particle source analysis system for analyzing a source of fine particles existing in the atmosphere,
Classifier construction means for deriving classifier information that is information for configuring a classifier using sample particle attribute information that is information based on attributes of a plurality of sample particles;
The analysis target particle attribute information, which is information based on the attribute of one or a plurality of analysis target particles, and the classifier information are input to the classifier, and the analysis target is based on the classification result by the classifier A source analysis means for deriving information on the source of the particles,
The sample particles are fine particles collected at a predetermined source,
The predetermined number of sources is one or more,
The analysis target particles are fine particles whose source is unknown,
The attribute includes a constituent element of the fine particle,
From the information based on the attribute of the fine particles, the classifier belongs to at least one of the predetermined generation sources and belongs to any of the predetermined generation sources. A particle generation source analysis system, characterized in that it is classified into one of the following. The classifier construction unit performs normalization for each of the plurality of sample particles by normalizing a vector having elements based on values based on the attribute of the sample particles using information representing the size of the sample particles. 1 standardization means,
The vector normalized by the first normalization means, wherein a representative value and a variation value of the element corresponding to each other of the vector for the plurality of sample particles are derived, and the derived representative value and First standardization means for performing, for each element, standardizing the value of the element using the variation value;
The source analysis means includes: a second normalization means for normalizing a vector having a value based on an attribute of the analysis target particle as an element using information representing a size of the analysis target particle;
Standardizing the value of the element of the vector of the analysis target particle normalized by the second normalization unit using the representative value and the variation value derived by the first standardization unit, A second standardization means for each element;
The sample particle attribute information includes information based on a vector standardized by the first standardization means,
The particle generation source analysis system according to claim 1, wherein the analysis target particle attribute information includes information based on a vector standardized by the second standardization unit. The classifier construction unit is a vector whose elements are values based on the attributes of the sample particles, and performs a principal component analysis on the vector for each of the plurality of sample particles, whereby a principal component of the vector A vector and a principal component are derived, a predetermined number of principal component vectors and principal components are selected from the derived principal component vector and principal components, and a feature vector that is a vector having the selected principal component value as an element is obtained. A first dimensional compression means for deriving;
The source analysis means performs conversion to a principal component of a vector whose element is a value based on the attribute of the analysis target particle using the principal component vector selected by the first dimension compression means, and A second dimension compressing means for deriving a feature vector that is a vector having the principal component value as an element;
The sample particle attribute information includes a feature vector derived by the first dimension compression means,
The particle generation source analysis system according to claim 1, wherein the analysis target particle attribute information includes a feature vector derived by the second dimension compression unit. The sample particle attribute information includes a vector whose element is a value based on the attribute of the sample particle,
The analysis target particle attribute information includes a vector whose element is a value based on the attribute of the analysis target particle,
The classifier construction means derives, as the classifier information, the centroid and variation value of the vector for each of the sample particles collected at the same predetermined source as the classifier information. The particle generation source analysis system according to any one of claims 1 to 3, further comprising classifier information deriving means for each of the generation sources. The source analysis unit includes the vector included in the analysis target particle attribute information, the centroid and the variation value derived for the same predetermined source by the classifier information deriving unit. Using an evaluation means for evaluating the distance between the vector and the center of gravity for at least one of the predetermined sources;
Based on the result of evaluation by the evaluation means, the analysis target particle belongs to at least one of the predetermined generation sources, and does not belong to any of the predetermined generation sources The particle generation source analysis system according to claim 4, further comprising generation source analysis means for classifying the generation source into any one of the following.   6. The particle generation source analysis system according to claim 5, wherein the evaluation unit individually performs the evaluation for all of the predetermined generation sources.   The particle generation source analysis system according to claim 5 or 6, wherein the distance is a Euclidean distance or a Mahalanobis distance. The vector whose element is a value based on the attribute of the sample particle is a vector whose element is a signal intensity value indicating a peak in the mass spectrum of the sample particle,
The vector having the value based on the attribute of the analysis target particle as an element is a vector having a signal intensity value indicating a peak in the mass spectrum of the analysis target particle as an element. The particle source analysis system according to claim 1. The sample particle attribute information is a mass spectrum of the sample particle, and includes information based on a mass spectrum obtained in each of a negative mode and a positive mode,
The classifier construction means includes a spectrum representative value deriving means for deriving a representative value of the mass spectrum obtained in the negative mode and a representative value of the mass spectrum obtained in the positive mode.
Based on the representative value of the mass spectrum obtained in the negative mode and the representative value of the mass spectrum obtained in the positive mode, the mass not used for deriving the classifier information among the mass spectra of the sample particles Noise removing means for specifying a spectrum,
The classifier information is derived using a mass spectrum other than a mass spectrum that is not used for deriving the classifier information, among the mass spectra of the sample particles. The particle source analysis system described. A particle source analysis method for analyzing a source of fine particles existing in the atmosphere,
A classifier construction step of deriving classifier information that is information for configuring a classifier using sample particle attribute information that is information based on attributes of a plurality of sample particles;
The analysis target particle attribute information, which is information based on the attribute of one or a plurality of analysis target particles, and the classifier information are input to the classifier, and the analysis target is based on the classification result by the classifier A source analysis step for deriving information on the source of the particles,
The sample particles are fine particles collected at a predetermined source,
The predetermined number of sources is one or more,
The analysis target particles are fine particles whose source is unknown,
The attribute includes a constituent element of the fine particle,
From the information based on the attribute of the fine particles, the classifier belongs to at least one of the predetermined generation sources and belongs to any of the predetermined generation sources. A particle generation source analysis method, characterized in that the particle generation source is classified into any one of the following.   A program that causes a computer to function as each unit of the particle generation source analysis system according to any one of claims 1 to 9.

Images