JP7451306B2 - Non-destructive structural analysis equipment, non-destructive structural inspection equipment and non-destructive structural analysis methods - Google Patents

Description

Embodiments of the present invention relate to a non-destructive structural analysis device, a non-destructive structural inspection device using the same, and a non-destructive structural analysis method.

Various non-destructive testing methods are known as techniques for measuring the distribution of substances and internal information inside structures.

Typical techniques include measurement using radiography using X-rays to create a two-dimensional transmission image, or computed tomography (CT) to create a three-dimensional image. Other methods include radiography using a neutron beam source, pulsed neutron imaging using a pulsed neutron source, neutron scattering, diffraction method, two-dimensional image measurement using cosmic ray muon transmittance, and muon tomography using muon scattering. A large number of methods exist.

These methods inject specific radiation (photons or particles) into a material as a probe, and the radiation attenuates as it loses energy within the material, or is scattered due to interaction with the particles that make up the material. This is a method of indirectly obtaining information about substances by measuring phenomena.

These methods are used in a wide range of fields, including diagnosis using X-rays and CT in the medical field, and evaluation of the soundness of structures and composition of materials in the industrial field.

In structural analysis methods using radiation, it is possible to predict the structure by analyzing the interaction between the probe particle and the substance inside the substance, so the Monte Carlo method is used as a simulation method to reproduce measured values. is widely used.

The Monte Carlo method is a method that uses random numbers to analyze the behavior of a large number of particles in a material, and can reproduce particle behavior in actual measurements with high accuracy. Therefore, as a simulation model, by accurately reproducing the shape and composition of the substance to be measured, and by reproducing conditions such as the type and energy of the probe particles, measurement results from various methods can be obtained with extremely high precision. It can be reproduced with.

These simulation techniques are used to examine measurement conditions in the preparation stage before measuring actual objects, and to evaluate consistency between measurement results and predicted values. Particularly under complex measurement conditions, it is difficult to interpret measurement results. Therefore, in order to confirm whether the desired measurement results are obtained, detailed analysis of the measurement results by comparing the measured values and simulations is required.

JP2013-231700

As mentioned above, the non-destructive testing method using radiation as a probe can reproduce measurement results with high accuracy through Monte Carlo simulation. However, in order to perform Monte Carlo simulation, it is necessary to create a model that reproduces the measurement system in detail. For this reason, there has been a problem in that predictions cannot be made by simulation if information about the measurement target cannot be obtained in advance.

Therefore, an object of the embodiments of the present invention is to obtain information about a measurement target even under conditions where the shape or composition of the measurement target is unknown.

In order to achieve the above-mentioned object, the non-destructive structural analysis device according to the present embodiment is a non-destructive structural analysis device that obtains material information of the target object from the results of measurement of the target object using radiation. A random material model that divides a simulation space that simulates a measurement region in which an object exists into a plurality of three-dimensional cells , and creates a plurality of random material models in which the material information is randomly set in each of the plurality of three -dimensional cells. a creation unit; a simulation execution unit that executes a simulation that reproduces the non-destructive inspection measurement for each of the plurality of random material models; and an actual measurement reproduction value that simulates the result of the non-destructive inspection measurement from the result of each of the simulations. A measured value calculation unit that calculates A predictor creation unit that creates a predictor with substance information as a target variable ; and a structure analysis unit that uses the predictor to obtain the substance information of the object based on the measurement results of the object. It is characterized by

1 is a block diagram showing the configuration of a non-destructive structural inspection device and its non-destructive structural analysis device according to a first embodiment; FIG. FIG. 3 is a conceptual division diagram showing cell division for creating a random material model in the non-destructive structural analysis apparatus according to the first embodiment. FIG. 3 is a conceptual model diagram showing an example of a random material model in the non-destructive structural analysis apparatus according to the first embodiment. FIG. 2 is a block diagram showing the configuration of a measurement system in the non-destructive structural inspection apparatus according to the first embodiment. FIG. 2 is a three-dimensional diagram showing an example object by the non-destructive structural analysis apparatus according to the first embodiment. FIG. 2 is a flow diagram showing the procedure of a non-destructive structural analysis method according to the first embodiment. FIG. 2 is a three-dimensional diagram showing an example of a random material model of an example using the non-destructive structural analysis apparatus according to the first embodiment. FIG. 3 is a three-dimensional diagram showing measurement results of an example using the non-destructive structural analysis apparatus according to the first embodiment. FIG. 2 is a block diagram showing the configuration of a measurement system in a non-destructive structural inspection apparatus according to a second embodiment. FIG. 3 is a block diagram showing the configuration of a measurement system in a non-destructive structural inspection apparatus according to a third embodiment. It is a block diagram showing the composition of the non-destructive structure inspection device and its non-destructive structure analysis device concerning a 4th embodiment. FIG. 3 is a block diagram showing the configuration of a measurement system in a non-destructive structural inspection apparatus according to a fourth embodiment. It is a graph which shows an example of the material weight prediction result by the non-destructive structure inspection apparatus based on 4th Embodiment. Non-destructive structural inspection equipment

DESCRIPTION OF THE PREFERRED EMBODIMENTS A non-destructive structural analysis apparatus and a non-destructive structural analysis method according to embodiments of the present invention will be described below with reference to the drawings. Here, parts that are the same or similar to each other are given the same reference numerals, and redundant explanations will be omitted.

[First embodiment]
FIG. 1 is a block diagram showing the configuration of a non-destructive structural inspection device 300 and its non-destructive structural analysis device 100 according to the first embodiment. The nondestructive structure inspection device 300 includes a nondestructive structure analysis device 100 that analyzes the structure of the object 5 within the measurement region 10 and a measurement system 200 that measures the measurement region 10 to be measured. In this embodiment, as will be described later, the measurement system 200 is exemplified using a measurement device 210 that performs muon scattering measurement.

The non-destructive structural analysis apparatus 100 includes a predictor relation calculation section 110, a storage section 120, a predictor 130, a structural analysis section 140, an input section 150 that receives external input, and an output section 160 that outputs to the outside.

The predictor relationship calculation unit 110 includes a random material model creation unit 111, a simulation execution unit 112, a measured value calculation unit 113, and a predictor creation unit 114.

First, the random material model creation section 111 will be explained. FIG. 2 is a conceptual division diagram showing cell division for creating a random material model.

The random material model creation unit 111 divides the simulation space 10a that simulates the measurement region 10 into a plurality of cells or voxels (hereinafter referred to as three-dimensional cells) 11, and arranges a specified composition in each three-dimensional cell 11. . Here, the cells may be divided at equal intervals in each axial direction, or evenly spaced in each axial direction, but the intervals in each axial direction may be different. Furthermore, the intervals may not be equal. FIG. 2 shows an example of division into 7 rows x 6 columns. Although FIG. 2 shows an example of two-dimensional division for convenience of illustration, if the measurement area 10 is a three-dimensional system, it is often divided three-dimensionally into a plurality of three-dimensional cells.

FIG. 3 is a conceptual model diagram showing an example of a random material model. FIG. 3 shows two types of three-dimensional cells 11 in the measurement area 10: white three-dimensional cells are air, and black three-dimensional cells are iron. Note that at this time, a plurality of compositions may be used as the iron density. In addition, although FIG. 3 shows two cases of air and iron, if multiple types of substances are included in the measurement area 10, multiple types of substances can be set by setting all the main substances included in the system. The distribution of substances can be reproduced. Hereinafter, information such as material type or density classification of the same material in each three-dimensional cell 11 will be referred to as material information in the three-dimensional cell 11.

The random material model creation unit 111 creates a large number of simulation models in which iron is randomly distributed in three-dimensional cells. When the number of three-dimensional cells in the measurement area 10 is n, and the number of types of substances and densities is m, there are m ⁿ combinations. In other words, the number N of random material models is m ⁿ at most.

However, as will be described later, since the predictor 130 is created by machine learning, the number N of random material models does not need to be the maximum m ⁿ pieces described above. Here, the number of models to be created is arbitrary, but in order to create the predictor 130 with sufficient accuracy, it is preferable to create 1,000 or more models. If the three-dimensional cells to be set are sufficiently fine and the number of models to be created is sufficiently large, most of the iron distribution conditions that can be established in the measurement region can be covered by the method of creating a random material model.

Next, the simulation execution unit 112 will be explained.

The simulation execution unit 112 reproduces the interaction between the muons and the object 5 in the measurement region 10 and the resulting behavior of the muons by Monte Carlo simulation for each random material model created by the random material model creation unit 111. Run a simulation.

Here, in general, interactions between the object 5 and radiation include attenuation of electromagnetic waves such as X-rays and gamma rays and particle beams such as neutron beams and muons, and scattering and diffraction of X-rays, neutron beams, and muons. be.

At this time, simulation conditions such as the configuration, characteristics, and arrangement of the measurement system 200 and conditions for muon irradiation onto the measurement region 10 are received in advance as external inputs by the input unit 150 and stored and stored in the storage unit 120.

The simulation execution unit 112 executes a simulation that reproduces the measurement by the measurement system 200 for all the prepared random material models under this simulation condition.

Next, the measured value calculation section 113 will be explained. The measured value calculation unit 113 calculates the measured value by reproducing the muon scattering measurement under the muon scattering measurement system, based on the behavior of the muon due to the interaction between the muon and the object 5 simulated by the simulation execution unit 112. calculate. That is, data of the muon trajectory at the detector position is output, similar to the actual measurement. Since this data has the same format as the actual measurement value, it is called the actual measurement reproduction value. When N different random material models are created, N different simulations are executed to reproduce the measurement, and N different actual measurement reproduction values are output. Note that in the case of only muon scattering measurement, there are N types, but when applying multiple M different measurement methods, there are N sets of data sets with M actual measurement reproduction values for one random material model. Created.

Next, the predictor creation unit 114 will be explained. The predictor creation unit 114 creates a function (Y=f(X)) with the measured value as the explanatory variable X and the target value as the objective variable Y. Here, the target value is, for example, model information of a random material model corresponding to the measured value X, which is an explanatory variable, in the case of regression analysis to be described later.

The predictor creation unit 114 creates a function f(X) based on machine learning. Below, examples of regression analysis and class classification are shown as cases of supervised learning in machine learning.

First, the case of regression analysis will be explained. In regression analysis, the objective variable Y is numerical information possessed by a random material model, and the explanatory variable X is an actual measurement reproduction value.

In simple regression analysis in which there is one type of explanatory variable X, a target value y is predicted from a certain variable x. In addition, in multiple regression analysis with a plurality of m explanatory variables, a function is created to predict the target value of y from m variables x1, x2 to xm.

When using regression analysis, by setting the objective variable Y as a continuous quantity such as the weight or average density of iron in the measurement area, the information held by the model can be directly extracted from the muon scattering data, which is the explanatory variable X. You can guess.

Next, the case of class classification will be explained. In class classification, for example, we deal with a binary classification problem of whether the composition of a certain three-dimensional cell is air or iron. Alternatively, if there are multiple substances set, or if the same substance has multiple density levels, it becomes a multi-class classification problem. SVM (support vector machine) is a typical method for class classification.

By performing classification, it is determined whether each three-dimensional cell is iron or air. By performing this classification, the distribution of iron in the three-dimensional cell can be predicted from the measured values. Alternatively, in the case of other class classification, the distribution of substances in the three-dimensional cell can be predicted based on the classification according to the class.

In addition to the above methods, machine learning and deep learning methods can be used to classify materials in three-dimensional cells, or to apply prediction methods to estimate feature values such as density and material content in measurement systems. , it is possible to create a predictor that inputs measured values and outputs model information created using a random material model.

Next, the storage unit 120 will be explained. The storage unit 120 includes a model storage unit 121 , a simulation condition storage unit 122 , and a measurement value calculation result storage unit 123 .

First, the model storage section 121 stores and stores all the random material models created by the random material model creation section 111. Each random material model is stored as a set of material information within each three-dimensional cell 11 in a system of three-dimensional cells 11 as conceptually shown in FIG.

Next, the simulation condition storage unit 122 stores and stores the simulation conditions accepted by the input unit 150 as external input. The simulation conditions include information such as the dimensions of the three-dimensional cell 11 in addition to the configuration, characteristics, and arrangement of the measurement system 200 as described above, and muon irradiation conditions to the measurement region 10.

Next, the measured value calculation result storage unit 123 stores actual measurement reproduction values, which are the calculation results of the measured values at the detector positions by the measured value calculation unit 113, in a form that is associated with each random material model. Note that the measurement value calculation result storage unit 123 may further store and store calculation results etc. by the simulation execution unit 112.

Next, the predictor 130 will be explained. As described above, the predictor 130 has a function (Y=f(X)) created by the predictor creation unit 114 with the measured value as the explanatory variable X and the target value as the objective variable Y. ing. As described above, this function possessed by the predictor 130 is made possible by machine learning to associate the measured value (explanatory variable X) with the objective variable Y under the conditions of the measurement system 200 used. This is a function that has

The measurement results obtained by the measurement system 200 are accepted by the input unit 150 and input to the predictor 130. The predictor 130 uses the measurement result as the explanatory variable X and uses the function f(X) to output the objective variable Y, that is, the material information in the three-dimensional cell 11 corresponding to the measurement result.

Next, the structure analysis unit 140 uses the predictor 130 created by the predictor creation unit 114 to input the measurement results measured by the measurement system 200 received via the input unit 150 into the function f(X). , we obtain the material structure with the objective variable Y.

Next, the measurement system 200 will be explained. FIG. 4 is a block diagram showing the configuration of the measurement system 200 in the non-destructive structural inspection apparatus 300.

In the first embodiment, a measurement device 210 using a muon scattering method is used as the measurement system 200.

The muon scattering method is a method for analyzing the structure of materials by measuring muon scattering when cosmic ray muons pass through materials. In muon scattering, the scattering angle θ ₀ changes depending on the radiation length X ₀ specific to the atomic number of the substance, as shown in the following equation (1).

θ ₀ = [13.6/(β _c p)]・z・√(x/X ₀ ) [1+0.038ln(x/X ₀ )]
...(1)
Note that β _c , p, and z are the velocity, momentum, and charge of the muon, respectively, and x is the distance that the muon passes through the substance.

Utilizing this characteristic, by analyzing the statistical value of the magnitude of the muon scattering angle θ ₀ and the scattering position, it is possible to measure the three-dimensional distribution of the substance within the measurement region.

Two muon detectors 211 and 212 are required for measurement using the muon scattering method. That is, the muon detector 211 detects the muon trajectory before the muon enters the object 5, and the muon detector 212 detects the muon trajectory after the muon is scattered when passing through the object 5. Measure.

Here, regarding the measurement area 10, in order to prevent radiation from entering the measurement area 10 from the outside other than the controlled radiation used for measurement, the periphery of the measurement area 10 is shielded by a shielding structure (not shown). has been done. Note that instead of this, muons from space may be used as radiation for measurement. In this case, the conditions for muon irradiation are the conditions for irradiation from space.

The signal processing unit 213 receives the signals from the muon detector 211 and the muon detector 212, performs amplification, etc., and calculates the difference in the direction of the muon trajectory before and after the muon is incident on the object 5. Calculate as scattering angle θ ₀ . The calculated muon scattering angle θ ₀ is output to the input section 150.

Finally, the output unit 160 generates and displays data for three-dimensional imaging based on the results analyzed by the structure analysis unit 140.

Next, a nondestructive structural analysis method using the nondestructive structural inspection apparatus 300 configured as described above will be described. FIG. 5 is a three-dimensional diagram showing an example object 5 by the non-destructive structural analysis apparatus 100 according to the present embodiment. The object 5 of the example is a sphere housed in an aluminum container having a cubic outer shape. The simulation space 10a, which simulates the measurement area in which the object 5 is installed, is three-dimensionally divided into 200 parts in each direction of the X-axis, Y-axis, and Z-axis, and is divided into 8×10 ⁶ three-dimensional cells. It is divided into. Examples will be described below according to the steps of the non-destructive structural analysis method according to this embodiment.

FIG. 6 is a flow diagram showing the procedure of the non-destructive structural analysis method according to the first embodiment. FIG. 6 shows a predictor creation step S10, a measurement step S20 of the object 5, and a structure analysis step S30.

The non-destructive structural analysis method includes a predictor creation related step S10 and a structural analysis step S30.

First, the predictor creation related step S10 will be explained. The predictor creation related step S10 includes a random material model creation step (S11), a simulation execution step (S12), an actual measurement value reproduction step (S13), and a predictor creation step (S14).

In the random material model creation step (S11), the random material model creation section 111 randomly sets a model on the simulation space 10a divided into a plurality of three-dimensional cells 11 based on conditions that can be established in the measurement region 10. Create a large number of simulation models, that is, random material models. These created random material models are stored and stored in the model storage unit 121.

FIG. 7 is a three-dimensional diagram showing an example of a random material model according to an embodiment of the non-destructive structural analysis apparatus 100. For each three-dimensional cell divided into 200 cells in each axial direction, the material information was classified into air and iron. In the example of the random material model shown in FIG. 7, colored cells indicate cells classified as iron, and uncolored cells indicate cells classified as air.

In the simulation execution step (S12), the simulation execution unit 112 executes a simulation that reproduces the measurement by the measurement system 200 for all random material models. That is, for each created random material model, a Monte Carlo simulation is performed to reproduce the interaction between the muons and the object 5 in the measurement region 10 and the resulting behavior of the muons. At this time, the simulation is executed based on the simulation conditions accepted by the input unit 150 and stored in the simulation condition storage unit 122.

In the actual measurement value reproduction step (S13), the measurement value calculation unit 113 calculates the measurement value by reproducing the muon scattering measurement under the muon scattering measurement system, based on each result simulated by the simulation execution unit 112. Calculate the actual measured reproduction value. As a result, N types of actually measured reproduction values are obtained for N types of random material models. Note that in the case where there are M types of measurement methods, when calculating actual measurement reproduction values for these M types, N sets of data sets having M actual measurement reproduction values are created. These actually measured reproduction values are stored and stored in the measured value calculation result storage section 123.

In the predictor creation step (S14), the predictor creation unit 114 uses a function (Y=f(X)) with the measured value as the explanatory variable X and the target value as the objective variable Y, based on machine learning. Create. Here, as the measured value serving as the explanatory variable X, an actual measured reproduction value calculated by the measured value calculation section 113 and stored in the measured value calculation result storage section 123 is used.

When machine learning is supervised learning and regression analysis is used, the objective variable Y can be set as a continuous quantity such as the weight or average density of iron in the measurement area.

Further, when the machine learning is supervised learning and class classification is used, the distribution of substances in the three-dimensional cell can be set as the objective variable Y based on the classification according to the class.

Next, the step S20 of measuring the object 5 will be explained.

Muon scattering in the object 5 is measured by a measuring device 210 that uses a muon scattering method as the measuring system 200 . That is, the muon detector 211 measures the muon trajectory before the muon enters the object 5, and the muon detector 212 measures the muon trajectory after the muon is scattered as it passes through the object 5. . These results are accepted by the signal processing unit 213, the difference in the direction of the muon trajectory before and after the incidence on the object 5 is calculated as the muon scattering angle θ ₀ , and the calculated muon scattering angle θ ₀ is input to the input unit 150. is output to.

Next, the structural analysis step S30 will be explained.

First, the predictor 130 predicts material information based on the measurement results of the object 5 by the measurement system 200 (step S31). That is, the predictor 130 receives as input the actual measured value obtained by measurement by the measurement system 200. Since the actual measurement value by the measurement system 200 and the actual measurement reproduction value calculated by the measurement value calculation unit 113 are in the same format, the predictor 130 calculates the most suitable value predicted from the actual measurement value in the same way as when inputting the actual measurement reproduction value. Outputs material model information.

Next, the structure analysis unit 140 analyzes the material structure based on the prediction result by the predictor 130 (step S32).

Through the above procedure, even if the composition and structure of the object 5 meet unknown conditions, internal information such as the composition and structure can be output as a predicted value from the measured values by the measurement system 200.

FIG. 8 is a three-dimensional diagram showing measurement results of an example using the non-destructive structural analysis apparatus according to the present embodiment. Since the distribution of iron is learned at the time of creation of the predictor 130, the aluminum container does not appear in the three-dimensional image, and only the iron ball is imaged.

That is, regarding the learned distribution of iron, the position, shape, size, etc. of the iron ball are reproduced compared to the actual object 5 in FIG.

As described above, according to the present embodiment, information on the shape and composition of the object 5 to be measured can be obtained even under conditions where the shape and composition are unknown.

[Second embodiment]
FIG. 9 is a block diagram showing the configuration of a measurement system in a non-destructive structural inspection apparatus according to the second embodiment. The second embodiment is a modification of the first embodiment.

In the first embodiment, a non-destructive structural inspection device and a non-destructive structural analysis method were shown using the muon scattering method as an example, but in the second embodiment, neutron radiography was used as the measurement system 200. The case of measuring device 220 is shown.

The measurement device 220 includes a neutron generator 221, a neutron detector 222, and a signal processing section 223.

The neutron generator 221 is, for example, a nuclear reactor such as a research nuclear reactor, or an accelerator.

Neutrons generated from the neutron generator 221 are irradiated onto the target object 5, which is a sample in the measurement area 10, and the neutrons that are incident on the target object 5 and attenuated or scattered within the target object 5 are measured by the neutron detector 222. do.

The output of the neutron detector 222 is received by the signal processing unit 223, amplified, AD converted, and input to the input unit 150 (FIG. 1) of the non-destructive structural analysis apparatus 100.

On the other hand, in the nondestructive structural analysis apparatus 100, similarly to the first embodiment, the simulation execution unit 112 irradiates the object 5 with neutrons and calculates the behavior of the neutrons that are attenuated or scattered by passing through the object 5. The simulation is performed, and the measured value calculation unit 113 calculates the actual measurement reproduction value. The predictor creation unit 114 performs machine learning based on the combination of the random material model and the actual measured reproduction value, and creates the predictor 130.

As described above, this embodiment is similar to the first embodiment in that the predictor 130 outputs a random material model corresponding to the measurement results based on the measurement results by the neutron detector 222.

As described above, even under conditions where the shape and composition of the object 5 to be measured are unknown, information can be obtained by irradiation with neutrons. Furthermore, although radiography using neutrons has been shown as an example, measurements can be made with the same configuration even when the probe is changed to X-rays or gamma rays.

[Third embodiment]
FIG. 10 is a block diagram showing the configuration of a measurement system in a non-destructive structural inspection apparatus according to the third embodiment. The third embodiment is a modification of the second embodiment.

In the second embodiment, neutron radiography is used to measure neutrons attenuated or scattered in a sample. However, in the third embodiment, neutron activation analysis is used as the measurement system 200. The case of the measuring device 230 shown in FIG.

The measurement device 230 includes a neutron generator 221, a neutron detector 222, a gamma ray detector 232, and a signal processing section 233. The neutron generator 221 and the neutron detector 222 are the same as those in the second embodiment.

The gamma ray detector 232 measures gamma rays generated by activation of the object 5 with respect to incident neutrons. Since the generated gamma ray spectrum differs depending on the composition within the substance, the type of substance contained in the measurement region can be estimated from the gamma ray spectrum.

In the nondestructive structural analysis apparatus 100, similarly to the first embodiment, the simulation execution unit 112 executes a simulation of the behavior in which the target object 5 is irradiated with neutrons and becomes radioactive, and the measured value calculation unit 113 calculates the actual measurement reproduction value. The predictor creating unit 114 performs machine learning based on the combination of the random material model and the actual measurement reproduction value, and creates the predictor 130.

Note that, using a similar system, a non-destructive structural inspection can be performed by a method of injecting neutrons and measuring fission neutrons generated in the object 5. Since the amount of neutrons generated in the object 5 corresponds to the amount of fissile material contained in the object 5, this method is effective when evaluating the amount of nuclear material in the object 5.

[Fourth embodiment]
FIG. 11 is a block diagram showing the configuration of a non-destructive structural inspection device 300a and its non-destructive structural analysis device 100a according to the fourth embodiment.

The non-destructive structural analysis device 100a of the non-destructive structural inspection device 300a in the fourth embodiment is a modification of the first embodiment. The nondestructive structural analysis apparatus 100a in the fourth embodiment has two predictors 130, that is, a first predictor 131 and a second predictor 132.

The measurement system 200 in the fourth embodiment is a measurement device 240 that combines muon scattering measurement and neutron measurement.

FIG. 12 is a block diagram showing the configuration of a measurement system in a non-destructive structural inspection apparatus according to the fourth embodiment. The measurement device 240 as the measurement system 200 of the fourth embodiment is a combination of the measurement device 220 in the second embodiment and the measurement device 230 in the third embodiment. That is, the measurement device 240 is a measurement system that combines muon scattering measurement and neutron measurement, and includes muon detectors 211 and 212, a neutron generator 221, a neutron detector 222, and a signal processing section 243. In addition to the combination of muon scattering measurement and neutron measurement, it is also possible to combine X-ray or gamma ray measurement, or add a combination of these.

In the non-destructive structural analysis apparatus 100a, the simulation execution unit 112 first performs a simulation of the behavior of muon scattering measurement, and the measured value calculation unit 113 calculates the actual measurement reproduction value. The predictor creation unit 114 performs machine learning based on the combination of the random material model and the measured reproduction value, and creates the first predictor 131. Next, the simulation execution unit 112 executes a simulation of the behavior of neutron measurement, and the measurement value calculation unit 113 calculates an actual measurement reproduction value. The predictor creation unit 114 performs machine learning based on the combination of the random material model and the actual measured reproduction value, and creates the second predictor 132. Note that this order may be reversed or may be performed in parallel.

The prediction result obtained by the first predictor 131 and the prediction result obtained by the second predictor 132 are collated, and, for example, by averaging the two, the prediction result as the non-destructive structural analysis device 100a is calculated. will be done.

Although FIG. 11 shows a case where two predictors 130 are created based on two measurement techniques, the same applies to the case of three or more measurement techniques.

In addition, although the case has been described above in which the predictors 130 are created independently for each of a plurality of measurement methods and prediction is performed by each predictor 130, the present invention is not limited to this.

That is, it is possible to create a predictor 130 that integrates multiple methods. When creating a predictor 130 that integrates multiple methods, the predictor creation unit 114 performs multiple regression analysis instead of simple regression analysis to calculate the results from multiple explanatory variables X (X1,...,Xm). Output one objective variable Y.

Different measurement methods yield different information, so the more measurement methods that are combined, the more accurate structural analysis becomes possible.

FIG. 13 is a graph showing an example of the material weight prediction result by the non-destructive structure inspection apparatus 300a according to the fourth embodiment. FIG. 13 is a graph of the weight of uranium in a material containing a mixture of iron and uranium, where the horizontal axis is the true _UO2 weight [kg], and the vertical axis is the predicted value of the true _UO2 weight [kg]. shows.

FIG. 13 shows the results predicted by combining three methods: muon scattering, neutron count rate measurement, and gamma ray measurement.

Based on the results of such evaluation, the present embodiment can also be applied to the purpose of extracting the amount of some of the substances in a state where a plurality of substances are mixed.

[Other embodiments]
Although the embodiments of the present invention have been described above, the embodiments are presented as examples and are not intended to limit the scope of the invention. For example, in the embodiment, a measurement method using radiation is shown as an example, but the measurement method is not limited to this, and other non-destructive measurement methods such as an electromagnetic method or a method using ultrasonic waves may be used. There may be.

Moreover, the features of each embodiment may be combined. Further, the embodiments can be implemented in various other forms, and various omissions, substitutions, and changes can be made without departing from the gist of the invention.

The embodiments and their modifications are included within the scope and gist of the invention as well as within the scope of the invention described in the claims and its equivalents.

5... Object, 10... Measurement region, 10a... Simulation space, 11... Three-dimensional cell, 100... Non-destructive structural analysis device, 110... Predictor relation calculation section, 111... Random material model creation section, 112... Simulation execution section , 113... Measured value calculation section, 114... Predictor creation section, 120... Storage section, 121... Model storage section, 122... Simulation condition storage section, 123... Measured value calculation result storage section, 130... Predictor, 140... Structure Analysis section, 150... Input section, 160... Output section, 200... Measurement system, 210... Measurement device, 211, 212... Muon detector, 213... Signal processing section, 220... Measurement device, 221... Neutron generator, 222... Neutron detector, 223...Signal processing section, 230...Measuring device, 232...Gamma ray detector, 233...Signal processing section, 240...Measuring device, 243...Signal processing section, 300...Non-destructive structure inspection device

Claims (11)

A non-destructive structural analysis device that obtains material information of an object from the results of measurement of the object using radiation,
A random method in which a simulation space simulating a measurement region in which the object exists is divided into a plurality of three-dimensional cells , and a plurality of random material models are created in which the material information is randomly set in each of the plurality of three -dimensional cells. Material model creation department,
a simulation execution unit that executes a simulation that reproduces non-destructive inspection measurements for each of the plurality of random material models;
a measurement value calculation unit that calculates an actual measurement reproduction value that simulates the result of the non-destructive inspection measurement from each result of the simulation;
A predictor that uses the actual measured reproduction value as an explanatory variable and the substance information of the random material model as an objective variable, using machine learning using a plurality of sets of the plurality of random material models and the corresponding actual measurement reproduction values. a predictor creation unit that creates
a structural analysis unit that uses the predictor to obtain the material information of the object based on the measurement results of the object;
A non-destructive structural analysis device comprising: The non-destructive structural analysis according to claim 1, wherein the simulation execution unit reproduces a non-destructive inspection method by Monte Carlo simulation using the random material model created by the random material model creation unit. Device. The predictor creation unit includes:
creating a plurality of data sets including the random material model and the corresponding actual measurement reproduction value;
The random material model is the objective variable,
With the actual measured reproduction value as the explanatory variable,
Perform machine learning using regression analysis or class classification,
The non-destructive structural analysis device according to claim 1 or 2, characterized in that: According to any one of claims 1 to 3, the radiation is an electromagnetic wave such as an X-ray or a gamma ray, or a particle beam of at least one of a neutron beam, an electron beam, a muon beam, and a proton beam. non-destructive structural analysis equipment. The non-destructive structural analysis device according to any one of claims 1 to 4,
a measuring device that performs the measurement of the object;
A non-destructive structural inspection device comprising : 6. The non-destructive structural inspection device according to claim 5 , wherein the measuring device has a structure that shields radiation that enters the measurement region from the outside in addition to the radiation for measurement . The nondestructive method according to claim 5 or 6, wherein the measurement device includes at least one of a radioisotope, an accelerator, or a nuclear fusion device, depending on the type of the radiation for measurement. Structural inspection equipment. 7. The nondestructive structural inspection device according to claim 5 , wherein the measurement device measures cosmic ray muons that are naturally generated in the atmosphere as muons . 9. The measuring device measures attenuation of any of electromagnetic waves such as X-rays and gamma rays, and particle beams such as neutron beams and muons, which pass through the measurement region. The non-destructive structural inspection device according to item (1) . 9. The measuring device measures scattering and diffraction of any of X-rays , neutron beams, muons, etc. passing through the measurement region. non-destructive structural inspection equipment. A non-destructive structural analysis method for obtaining material information of an object from the results of measurement of the object using radiation, the method comprising:
A random material model creation unit divides a simulation space simulating the measurement region in which the target object exists into a plurality of three-dimensional cells, and randomly sets the material information in each of the plurality of three-dimensional cells. a random material model creation step for creating a material model;
a simulation execution step in which the simulation execution unit executes a simulation to reproduce non-destructive inspection measurements for each of the plurality of random material models;
a measurement value calculation step in which the measurement value calculation unit calculates an actual measurement reproduction value that simulates the result of the non-destructive inspection measurement from each result of the simulation;
The predictor creation unit uses machine learning using the plurality of sets of the random material model and the corresponding actual measurement reproduction values to set the actual measurement reproduction value as an explanatory variable and the substance information of the random material model as an objective variable. a predictor creation step of creating a predictor for
After the predictor creation step, a prediction step in which the structure analysis unit predicts the material information based on the measurement results of the object using the predictor;
A non-destructive structural analysis method characterized by having the following .

Images