JP7288345B2 - Particle size distribution measuring device, particle size distribution measuring method, and particle size distribution measuring device program - Google Patents

Description

The present invention relates to a particle analyzer that irradiates a sample containing particle groups in a dispersion medium with light and analyzes the characteristics of particles or particle groups based on the scattered light.

In order to control the quality of products manufactured using powder, the particle size distribution of powder is sometimes measured. For such measurements, a laser diffraction/scattering type particle size distribution measuring device for measuring the particle size distribution based on the Mie scattering theory as disclosed in Patent Document 1 is used.

By the way, the Mie scattering theory is based on the premise that the shape of particles is spherical, and the particle size parameter x = 2πr /λ = πd /λ (r: particle radius, d: particle diameter, λ: light source wavelength), this theory holds when x≈1.

Therefore, when measuring a non-spherical sample, or when the x≈1 relationship gradually deviates and approaches the lower detection limit of a laser diffraction/scattering particle size distribution analyzer, it becomes difficult to measure based on theory. .

In addition, the information of the measured scattering pattern is used only for calculating the particle size distribution, and there is a possibility that useful information that may appear in the scattering pattern cannot be sufficiently extracted.

JP 2015-7546 A

The present invention has been made in view of the above-described problems, and enables calculation of the particle size distribution even when x≈1 does not hold for the particle size parameter x, and information other than the particle size distribution from the scattering pattern. It is an object of the present invention to provide a particle size distribution measuring device capable of extracting

That is, the particle size distribution measuring apparatus according to the present invention includes a light source that emits light to a cell containing a sample composed of a particle group that is a dispersion medium and a dispersoid, and a detector that detects light scattered by the sample. and a particle size distribution calculator for calculating a particle size distribution of a group of particles contained in the sample based on the output signal of the detector, wherein the particle size distribution calculator a particle size distribution model generation unit, wherein a particle size distribution model generation unit uses a particle size distribution model, wherein at least a light scattering pattern by a particle group is used as an explanatory variable and a particle size distribution of the particle group is used as an objective variable, and the particle size distribution model is generated by machine learning; a particle size distribution output unit for outputting a particle size distribution of a group of particles contained in the sample from the scattering pattern indicated by the output signal of the detector based on a model.

In another aspect of the particle size distribution measuring apparatus according to the present invention, a light source for emitting light to a cell containing a sample composed of a particle group as a dispersion medium and a dispersoid, and light scattered in the sample. A particle size distribution measuring device comprising a detector that detects light and a particle size distribution calculator that calculates the particle size distribution of a group of particles contained in the sample based on the output signal of the detector, A model storage unit for storing a particle size distribution model in which the particle size distribution calculator uses at least the light scattering pattern by the particle group generated by machine learning as an explanatory variable and the particle size distribution of the particle group as at least the objective variable. and a particle size distribution output unit for outputting the particle size distribution of the particle group contained in the sample from the scattering pattern indicated by the output signal of the detector based on the particle size distribution model.

In such a case, the measured scattering pattern is input to the particle size distribution model generated by machine learning to obtain the particle size distribution. Particle size distribution can be obtained even when the shape is non-spherical or when the measurement conditions are such that the particle size parameter x deviates from x≈1.

In addition, since the particle size distribution model is created by machine learning, it is possible to take into account the influence of information other than the particle size included in the scattering parameters, for example, and to obtain a more realistic particle size distribution. Become.

In order to easily grasp the change in the particle size distribution in real time and to easily compare the relationship between the change in the measurement conditions and the particle size distribution, the particle system distribution output unit outputs the particle size distribution corresponding to each measurement. and further includes a display section for displaying a list of each particle size distribution output from the particle size distribution output section on one screen.

In order to correct the effect of the particle shape on the calculation result of the particle size distribution in a short time without performing complicated calculations, and to obtain the ratio of each shape in the particle group, etc., the particle size The distribution model generation unit performs machine learning using the light scattering pattern by the particle group having a non-spherical shape and the particle size distribution corresponding thereto as teacher data, and the machine learning unit aims at the particle shape and particle size distribution. Any configuration may be used as long as it is configured to generate a particle size distribution model to be used as a variable.

For areas where it is difficult to obtain actual values only by theoretical calculations related to particle size distribution, particle size distribution is obtained based on a particle size distribution model generated by machine learning, and for areas where theoretical calculations easily match actual results In order to ensure a predetermined measurement accuracy regardless of the range of particle diameters by performing normal calculations, the particle size distribution model generation unit generates light from a group of particles having a particle size distribution within a predetermined range. , and the corresponding particle size distribution as teacher data for machine learning.

When the particle size distribution in the predetermined range has a particle size parameter x = 2πr / λ = πd / λ (r: particle diameter radius, d: particle diameter diameter, λ: light source wavelength), 0.05 ≤ x ≤ 0.5 If the particle size distribution satisfies the requirements, the time required for building a model for calculating the particle size distribution can be shortened compared to the conventional model building by simulation. However, it should be added that the above range is a rough guide because it easily fluctuates depending on the temperature of the light source wavelength and the number of digits used for the pi.

That is, under conditions where the particle size parameter x deviates from x ≈ 1, it is practically impossible to calculate the theoretical scattering intensity by, for example, solving Maxwell's equation. It is conceivable to calculate the light intensity that will be emitted and build a model for calculating the particle size distribution. However, building a model by such a simulation and checking the validity of the model require an enormous amount of time. On the other hand, if a particle size distribution model is created by machine learning within the range of the particle size parameter x as described above, it becomes possible to construct an appropriate model in a short time even in a region where model construction by simulation is difficult.

By creating a particle size distribution model by machine learning, for example, the benefits of shortening the calculation time required for model construction can be obtained. ), the particle size distribution in the predetermined range is 10 nm or more and 100 nm or less.

Different particle size distribution measuring devices may output different particle size distributions even if the same sample is measured. In order to grasp such instrumental errors and to convert the results obtained with each particle size distribution measuring device into the results of measurement with another particle size distribution measuring device, the particles The particle size distribution calculator uses the scattering pattern or the particle size distribution as an explanatory variable, and machine-learns an instrumental error model in which the instrumental error between the measured values of the same sample measured by itself and another particle size distribution measuring device is used as the objective variable. and the particle size distribution output unit generates the particle size distribution of the particles contained in the sample from the scattering pattern indicated by the output signal of the detector based on the instrumental error model. Any configuration may be used as long as it is configured to output With such a system, even if an old particle size distribution measuring device is replaced with a new one, by outputting a particle size distribution corrected for instrumental errors, the measurement results obtained up to that point can be compared. It is possible to maintain the continuity of

When calculating the particle size distribution based on the scattering pattern and a given theoretical formula, for example, the refractive index of the sample is required, but it may be difficult to obtain the refractive index of the sample clearly. In such a case, calibration work is performed in which the operator performs calculation of the particle size distribution multiple times while appropriately changing the refractive index, and searches for the most likely result. The selection of calibration parameters such as the refractive index in such calculation of the particle size distribution is based on the operator's experience, and is laborious and time-consuming, and is very difficult for inexperienced operators. In order to solve such a problem, the particle size distribution calculator uses a scattering pattern or a particle size distribution as an explanatory variable and generates a calibration parameter model using a calibration parameter including at least a refractive index as an objective variable by machine learning. a calibration parameter model generation unit; a calibration parameter model; a calibration parameter output unit for outputting calibration parameters to be set based on the measured scattering pattern or particle size distribution; and a calibration parameter output from the calibration parameter output unit. , a predetermined theoretical calculation formula, and a theoretical calculation unit for calculating the particle size distribution based on the output of the detector.

For example, if the scattering pattern is affected by contamination in the cell, or if an incorrect particle size distribution is output, the particle size distribution analyzer will automatically detect that it is in an abnormal state. In order to notify the operator, an abnormal state model generation unit for generating an abnormal state model using at least a scattering pattern as an explanatory variable and an abnormal state of the particle size distribution measuring apparatus as an objective variable by machine learning; It is sufficient that it further includes an abnormal state model and an abnormality output unit that outputs an abnormality that has occurred based on the measured scattering pattern.

For example, in order to obtain a particle size distribution in a short period of time from the scattering pattern indicated by the signal output from the detector, light is emitted to a cell containing a sample consisting of particles that are a dispersion medium and a dispersoid. A particle size distribution measuring method using a particle size distribution measuring device comprising: a light source for detecting light scattered by a sample; wherein the particle size distribution calculating step uses at least the light scattering pattern by the particle group as an explanatory variable, and the particle size distribution of the particle group as at least an objective variable. a particle size distribution model generating step of generating a particle size distribution model by machine learning; and a particle size distribution output step of outputting the above.

In order to obtain the same effect as the particle size distribution measuring device according to the present invention by simply updating the program used in the existing particle size distribution measuring device, it is necessary to obtain a particle group that is a dispersion medium and a dispersoid. A program for use in a particle size distribution measuring device comprising: a light source for emitting light to a cell containing a sample, and a detector for detecting light scattered by the sample, wherein the detector The computer functions as a particle size distribution calculator that calculates the particle size distribution of the particle group contained in the sample based on the output signal of the particle size distribution calculator. A particle size distribution model generating unit for generating a particle size distribution model using at least a scattering pattern as an explanatory variable and having at least a particle size distribution of a particle group as an objective variable by machine learning; and a particle size distribution output unit for outputting the particle size distribution of the particle group contained in the sample from the scattering pattern indicated by the output signal of (1).

The program for the particle size distribution measuring apparatus may be electronically distributed, or may be recorded on a program recording medium such as a CD, DVD, flash memory, or the like.

Thus, the particle size distribution measuring apparatus according to the present invention generates a particle size distribution model in which at least the scattering pattern generated by machine learning is used as an explanatory variable and at least the particle size distribution is used as an objective variable. Since the particle size distribution is obtained by inputting the scattering pattern into the particle size distribution model, the condition may deviate from the particle size parameter x≈1. It is also possible to add information other than the particle size distribution contained in the scattering pattern by machine learning, extract and output correction and other information.

1 is a schematic diagram showing the overall configuration of a particle size distribution measuring device according to a first embodiment of the present invention; FIG. FIG. 2 is a functional block diagram of the particle size distribution measuring device according to the first embodiment; 4 is a flow chart showing steps related to generation of a particle size distribution model for the particle size distribution measuring device according to the first embodiment. 4 is a flow chart showing steps related to outputting the particle size distribution of the particle size distribution measuring apparatus according to the first embodiment. The functional block diagram of the particle size distribution measuring device according to the second embodiment. 9 is a flow chart showing steps related to generation of an instrumental error model of the particle size distribution measuring device according to the second embodiment. 10 is a flow chart showing steps related to outputting the particle size distribution of the particle size distribution measuring apparatus according to the second embodiment. The functional block diagram of the particle size distribution measuring device according to the third embodiment. 10 is a flow chart showing steps related to generation of a calibration parameter model for the particle size distribution measuring device according to the third embodiment. 9 is a flow chart showing steps related to the output of the particle size distribution of the particle size distribution measuring device according to the third embodiment. The functional block diagram of the particle size distribution measuring device according to the fourth embodiment. 10 is a flow chart showing steps related to generation of an abnormal state model of the particle size distribution measuring device according to the fourth embodiment. 10 is a flow chart showing steps related to an abnormal state output of the particle size distribution measuring apparatus according to the fourth embodiment. The functional block diagram of the particle size distribution measuring device according to another embodiment.

A particle size distribution measuring device 100 according to a first embodiment of the present invention will be described with reference to FIGS. 1 to 4. FIG. In addition, suppose that the same code|symbol is attached|subjected to the corresponding thing about each component demonstrated in each embodiment.

This particle size distribution measuring apparatus 100 measures the particle size distribution of a particle group contained in a sample by static light scattering (SLS). That is, the particle size distribution measuring apparatus 100 outputs the particle size distribution from the scattering pattern generated when the sample is irradiated with laser light. Here, the scattering pattern in the first embodiment is measured as the angular dependence of the scattered light intensity.

Specifically, the particle size distribution measuring apparatus 100 includes a laser light source LS emitting laser light, a cell C containing a sample, and a plurality of detectors D1, D2, and D3 arranged around the cell C. and a computer COM. The computer COM includes a CPU, a memory, an A/D/D/A converter, input/output means, a display, and the like. By collaborating, a particle size distribution calculator 1 that calculates the particle size distribution based on part or all of the output of each detector D1, D2, and D3, and a display unit that displays the calculated particle size distribution on the display It functions as 6.

The detection mechanism D outputs the scattering angle and the intensity of light detected at that scattering angle. The intensity of each scattered light detected by each detector of the detection mechanism D is stored in the computer as vector data.

The particle size distribution calculator 1 calculates the particle size distribution based on the particle size distribution model generated by machine learning and the scattering pattern measured by the detection mechanism D in at least a part of the particle size range. be. Here, the particle size distribution model in the first embodiment uses the light scattering pattern by the particle group as an explanatory variable, and the particle size distribution of the particle group and the shape distribution of the particle group as objective variables. By inputting the measured scattering pattern to such a particle size distribution model, the corresponding particle size distribution and shape distribution are output. In other words, the particle size distribution calculator 1 is configured to output the particle size distribution by pattern recognition of the scattering pattern using AI in at least a part of the particle size range, without performing calculations based on theoretical calculation formulas. ing.

As shown in the functional block diagram of FIG. 2, the particle size distribution calculator 1 comprises at least a theoretical calculation unit 2, a learning data storage unit 3, a particle size distribution model generation unit 4, and a particle size distribution output unit 5.

The theoretical calculation unit 2 calculates the particle size distribution from the scattering pattern measured using a theoretical calculation formula based on, for example, the Rayleigh scattering theory or the Mie scattering theory, as in the conventional case. In the theoretical calculation unit 2, iterative calculations are performed until a predetermined result is obtained. The theoretical calculation unit 2 also calculates the shape distribution of the particle group based on, for example, finite element analysis from the scattering pattern.

The learning data storage unit 3 stores a large amount of data used to generate the particle size distribution model. In the first embodiment, the learning data storage unit 3 stores a large number of data pairs of the scattering pattern measured by the particle size distribution measuring device 100 and the particle size distribution and shape distribution calculated by the theoretical calculation unit 2. It is That is, the learning data storage unit 3 stores a large number of teacher data for generating the particle size distribution model. Moreover, in the first embodiment, the shape of the particles contained in the sample is not limited to spherical, but may be non-spherical. Note that the teacher data can be generated by actually measuring the scattering pattern using a sample with a known particle shape and particle diameter, for example.

The particle size distribution model generating unit 4 generates a particle size distribution measurement model by machine learning based on teacher data. Here, an existing algorithm can be used as the learning algorithm, and various techniques such as neural network and deep learning can be used. In the first embodiment, the teacher data used to generate the particle size distribution model is limited to a part of the particle size range. The procedure for generating the particle size distribution model will be described below with reference to the flowchart of FIG.

The particle size distribution model generation unit 4 first determines whether the data stored in the learning data storage unit 3 has a particle size distribution within a predetermined range (step S11). Here, the predetermined range is the range between the particle size to which the Rayleigh scattering theory is applicable and the particle size to which the Mie scattering theory is applicable. In other words, when the particle size parameter x = 2πr/λ = πd/λ (r: particle diameter radius, d: particle diameter, λ: light source wavelength), the predetermined range is 0.05 ≤ x ≤ 0.5 is defined as the range of particle size distribution that satisfies More specifically, when the laser light emitted from the laser light source LS is blue and the wavelength λ is, for example, 450 nm or a value in the vicinity thereof, the predetermined range is set as the particle size distribution range of 10 nm or more and 100 nm or less. That is, the predetermined range defines a range in which the relationship between the theoretical calculation and the actual scattering pattern and particle size distribution is difficult to match.

When the range of the particle size distribution is within the predetermined range, the particle size distribution model generator 4 uses the data as teacher data (step S12). On the other hand, the particle size distribution model generator 4 does not use the data with the particle size distribution outside the predetermined range as the teacher data (step S13).

Of the data stored in the learning data storage unit 3, the particle size distribution model generating unit 4 generates a particle size distribution model by machine learning, using only a part of the data with a particle size within a predetermined range as teacher data (step S14).

Next, the particle size distribution output section 5 and the display section 6 shown in FIG. 2 will be described. The particle size distribution output unit 5 calculates the particle size distribution and shape distribution of the particle group contained in the sample from the scattering pattern indicated by the output signal obtained from the detection mechanism D, based on the particle size distribution model generated by machine learning. Output. The particle size distribution output unit 5 inputs the scattering pattern to the particle size distribution model each time the scattering pattern is measured, and outputs the corresponding particle size distribution and shape distribution.

The display unit 6 is configured to display a list of the particle size distribution and shape distribution output from the particle size distribution output unit 5 or the particle size distribution and shape distribution calculated by the theoretical calculation unit 2 on one screen. .

A procedure for calculating the particle size distribution and the shape distribution will be described below with reference to the flow chart of FIG.

The particle size distribution output unit 5 inputs the scattering pattern to the particle size distribution model generated by machine learning each time the scattering pattern is measured, and outputs the corresponding particle size distribution and shape distribution (step ST11). .

Next, the particle size distribution output unit 5 determines whether or not the output particle size distribution is within the above-described predetermined range (step ST12). When the particle size distribution is within the predetermined range, the particle size distribution output unit 5 adopts the particle size distribution and the shape distribution obtained from the particle size distribution model, and outputs them to the display unit 6 (step ST13 ). On the other hand, when the particle size distribution is out of the predetermined range, the particle size distribution output unit 5 causes the theoretical calculation unit 2 to perform calculation based on the measured scattering pattern. Then, the particle size distribution and shape distribution calculated by the theoretical calculation section 2 are output to the display section 6 (step ST14).

The display unit 6 displays a list of either the result output from the particle size distribution output unit 5 based on the particle size distribution model or the result output from the theoretical calculation unit 2 (step ST15).

With the particle size distribution measuring apparatus 100 of the first embodiment configured as described above, the particle size distribution and the shape are based on the particle size distribution model generated by machine learning for the range where the theoretical calculation and the result are difficult to match. distribution can be obtained.

Therefore, it is possible to accurately obtain the actual particle size distribution and shape distribution in a predetermined range, even though the calculation is performed in a shorter time than in the conventional method. That is, since modeling is performed by limiting the teacher data that is the basis of the particle size distribution model to those that indicate particle sizes within a predetermined range, it is difficult for errors to be included in the model generated by machine learning. .

In addition, since the particle size distribution and the shape distribution are obtained from the theoretical calculation in the same manner as in the conventional case, the measurement accuracy can be ensured even in such a range.

In addition, since the particle size distribution and the shape distribution are output by pattern recognition of the scattering pattern by machine learning, even if the shape of the particles is non-spherical, the effect on the particle size distribution can be corrected and obtained.

A modification of the first embodiment will be described.

In the first embodiment, when the particle size is within a predetermined range, a particle size distribution model is generated by machine learning, and the particle size distribution and shape distribution are obtained based on the model. A particle size distribution model may be generated in step 3, and the particle size distribution and shape distribution may be output from the scattering pattern based on the particle size distribution model.

In addition, the particle size distribution model may use at least the scattering pattern as an explanatory variable, and another parameter may be added to the explanatory variable. Also, the objective variable may be limited to the particle size distribution only.

Next, a particle size distribution measuring device 100 of a second embodiment will be described with reference to FIGS. 5 to 7. FIG. In the following description, portions different from the first embodiment will be described.

The particle size distribution measuring apparatus 100 of the second embodiment is configured to model the instrumental error between different apparatuses by machine learning, and correct or convert the calculated particle size distribution.

As shown in the functional block diagram of FIG. 5 , the particle size distribution calculator 1 of the second embodiment includes an instrumental error model generation section 7 , an instrumental error output section 8 and a theoretical calculation section 2 .

The instrumental error model generation unit 7 uses machine learning to generate an instrumental error model using the scattering pattern as an explanatory variable and the instrumental error between the measured values of the same sample measured by itself and another particle size distribution measuring apparatus as the objective variable. do. Here, in the learning data storage unit 3, the scattering patterns measured by the own particle size distribution measuring apparatus and another particle size distribution measuring apparatus for the same sample are stored as teaching data. The procedure for generating an instrumental error model will be described below with reference to the flow chart of FIG.

First, the scattering pattern of a certain sample is measured by the own particle size distribution measuring device 100 (step S21).

Next, the scattering pattern of the same sample as in step S21 is measured by another particle size distribution measuring device (step S22). Steps S21 and S22 are repeated many times while changing the particle diameter distribution of the particle group contained in the sample, and many teaching data are created.

Based on a large amount of obtained teacher data, the instrumental error model generation unit 7 inputs its own scattering pattern, and creates a model that outputs the corresponding scattering pattern in another particle size distribution measuring apparatus as an instrumental error of the measured value. A difference model is generated (step S23).

Next, the instrumental error output section 8 and the theoretical calculation section 2 will be described with reference to the flow chart of FIG.

The theoretical calculation unit 2 calculates the particle size distribution by theoretical calculation based on the scattering pattern measured by the detection mechanism D (step ST21).

The instrumental error output unit 8 inputs the scattering pattern measured by the detection mechanism D to the generated instrumental error model, and outputs the scattering pattern that would have been measured by another particle size distribution measuring apparatus. output (step ST22).

Next, the theoretical calculation unit 2 performs a theoretical calculation based on the corresponding scattering pattern in the other particle size distribution measuring device output from the instrumental error output unit 8 to obtain particles that are considered to be output in the other particle size distribution measuring device 100. A diameter distribution is calculated (step ST23).

The display unit 6 displays a list of the particle size distribution, which is the result of its own measurement, and the particle size distribution, which is the result of measurement by another particle size distribution measuring device (step ST24).

According to the particle size distribution measuring apparatus 100 of the second embodiment, it is possible to simultaneously display the difference in measurement results between a certain apparatus and another apparatus based on the instrumental error model. Therefore, it is possible to easily grasp differences in measurement results due to instrumental errors within the same manufacturer and differences between manufacturers.

A modification of the second embodiment will be described.

For the instrumental error model, the explanatory variable may be the scattering pattern measured by itself, and the objective variable may be the difference in particle size distribution. In this case, the particle size distribution calculated from the scattering pattern can be converted to the measurement results of another particle size distribution measuring apparatus by correcting the difference in the particle size distribution output from the instrumental error model. can. In addition, instead of calculating the particle size distribution from the scattering pattern by theoretical calculation, the particle size distribution is output from the scattering pattern by a particle size distribution model generated by machine learning, and the result is output from the instrumental error model. It may be corrected by the difference in diameter distribution. Additionally, the explanatory variable of the instrumental error model may be the particle size distribution.

Next, a particle size distribution measuring device 100 of a third embodiment will be described with reference to FIGS. 8 to 10. FIG. In the following description, portions different from the first embodiment will be described.

The particle size distribution measuring apparatus 100 of the third embodiment uses machine learning to model calibration parameters such as the refractive index that are set when the particle size distribution is calculated from the scattering pattern by theoretical calculation, and automatically obtains appropriate values. is set.

As shown in the functional block diagram of FIG. 8, the particle size distribution calculator 1 of the third embodiment includes a calibration parameter model generation section, a calibration parameter output section 10, and a theoretical calculation section 2.

The configuration parameter generator uses machine learning to generate a calibration parameter model having a scattering pattern as an explanatory variable and a calibration parameter including at least a refractive index as an objective variable.

As shown in the flowchart of FIG. 9, at the time of model generation, the scattering pattern, the refractive index actually set for calculating the particle size distribution for the scattering pattern by theoretical calculation, and the refractive index determined by the operator Teacher data paired with the appropriateness result of the rate are collected (step S31). A large number of teacher data collected in this way are stored in the learning data storage unit 3 . Further, the suitability result is "adequate" when the operator determines that the calculated particle size distribution is appropriate as the actual measurement result, and "fail" when the operator determines that it is not appropriate.

Based on the collected teacher data, the calibration parameter model generation unit generates, by machine learning, a calibration parameter model that outputs the refractive index corresponding to the input scattering pattern (step S32).

As shown in the flowchart of FIG. 10, during measurement, the calibration parameter output unit 10 inputs the scattering pattern measured by the detection mechanism D into the calibration parameter model, and outputs the corresponding refractive index to the theoretical calculation unit 2. (Step ST31).

Theoretical calculation unit 2 calculates the particle size distribution by theoretical calculation based on the set refractive index and the measured scattering pattern (step ST32).

The display unit 6 displays a list of particle size distributions calculated for each scattering pattern measurement (step ST33).

As described above, according to the particle size distribution measuring apparatus 100 of the third embodiment, it is possible to automate setting of calibration parameters such as a refractive index, which is conventionally set by trial and error based on experience and intuition by an operator.

Moreover, since a refractive index suitable for each scattering pattern is selected from the beginning based on teacher data, there is no need to calculate the particle size distribution multiple times while changing the refractive index, and the time required for calculation can be greatly reduced.

A modification of the third embodiment will be described.

The calibration parameter is not limited to the refractive index, and may be other parameters. For example, the concentration, viscosity, etc. of the sample may affect the calculation result of the particle size distribution, and it may be difficult to obtain these exact values.

For calibrated parameter models, the explanatory variable may be the particle size distribution. Alternatively, both the scattering pattern and the particle size distribution may be used as explanatory variables.

Next, a particle size distribution measuring device 100 of a fourth embodiment will be described with reference to FIGS. 11 to 13. FIG. In the following description, portions different from the first embodiment will be described.

In the particle size distribution measuring device 100 of the fourth embodiment, the relationship between the scattering pattern and the abnormal state in the particle size distribution measuring device 100 is modeled by machine learning, and where the abnormality occurs from the scattering pattern is detected. I am making it possible.

Specifically, the computer COM is configured to function as an abnormal state model generation unit 11 and an abnormality output unit 12 separately from the particle size distribution calculator 1 .

The abnormal state model generation unit 11 generates an abnormal state model using at least the scattering pattern as an explanatory variable and the abnormal state of the particle size distribution measuring apparatus 100 as an objective variable by machine learning.

As shown in the flowchart of FIG. 12, at the time of model generation, teacher data paired with a scattering pattern and an abnormal state of the particle size distribution measuring apparatus 100 when the scattering pattern is being measured are collected and learned. It is stored in the data storage unit 3 (step S41). Here, the abnormal condition is the result of judgment by the operator, and includes various conditions such as no abnormality, abnormality such as contamination of the cell, and abnormality of the optical system.

The abnormal state model generating unit 11 generates, by machine learning, an abnormal state model that outputs an abnormal state corresponding to the input scattering pattern based on the teacher data (step S42).

As shown in the flowchart of FIG. 13, the abnormality output unit 12 inputs the scattering pattern measured by the detection mechanism D to the generated abnormal state model, and determines whether there is no abnormality or an abnormal state that has occurred. is output (step ST41).

The display unit 6 displays the state of the particle size distribution measuring device 100 output from the abnormality output unit 12 (step ST42).

As described above, with the particle size distribution measuring apparatus 100 of the fourth embodiment, information regarding an abnormal state, which is information other than the particle size distribution appearing in the scattering pattern, can be obtained using an abnormal state model generated by machine learning. can be done. Then, it becomes possible to notify the operator of the abnormal state that has occurred and prompt him to take countermeasures.

A modification of the fourth embodiment will be described.

The explanatory variable of the abnormal state model may be the particle size distribution instead of the scattering pattern. Alternatively, both the scattering pattern and the particle size distribution may be used as explanatory variables. As for the objective variable, a correction coefficient may be set for calculating the correct particle size distribution in an abnormal state as well as in an abnormal state.

Other embodiments will be described.

The particle size distribution measuring device 100 itself need not be configured to generate a particle size distribution model by machine learning as shown in the first embodiment. Specifically, as shown in FIG. 14, a model storage unit 41 is provided instead of the particle size distribution model generation unit 4 of the first embodiment, and the particle size distribution output unit 5 is stored in the model storage unit 41. A particle size distribution may be output from the measured scattering pattern with reference to a size distribution model. Here, the particle size distribution model stored in the model storage unit 41 is generated by machine learning, for example, before shipment of the particle size distribution measuring device 100 from the factory. Even with such a configuration, substantially the same effects as in the first embodiment can be enjoyed.

In each embodiment, machine learning generates each model using teacher data, but each model may be generated by unsupervised learning. The learning algorithm may also be appropriately selected according to the purpose.

Although the particle size distribution measuring device based on the static light scattering method has been described in each embodiment, the present invention may be applied to a particle size distribution measuring device based on the dynamic light scattering method (DLS). In this case, the scattering pattern is not vector data indicating the intensity of the scattered light with respect to the scattering angle, but vector data indicating the temporal change in the intensity of the signal output from the detector, or vector data indicating the autocorrelation function with respect to the delay time. Become.

In addition, as long as it does not contradict the gist of the present invention, various modifications of the embodiments and combinations of parts of the embodiments may be performed.

100 Particle size distribution measuring device LS Laser light source D Detection mechanism D1, D2, D3 Detector 1 Particle size distribution calculator 2 Theoretical calculation unit 3 Learning data storage unit 4 Particle size distribution model generation unit 41 Model storage unit 5 Particle size distribution output unit 6 Display unit 7 Instrument error model generation unit 8 Instrumental error output unit 9 ... calibration parameter model generation unit 10 ... calibration parameter output unit 11 ... abnormal state model generation unit 12 ... abnormality output unit

Claims (12)

a light source for emitting light to a cell containing a sample consisting of a group of particles as a dispersion medium and a dispersoid; a detector for detecting light scattered by the sample; A particle size distribution measuring device comprising a particle size distribution calculator for calculating a particle size distribution of a group of particles contained in a sample,
The particle size distribution calculator,
a particle size distribution model generation unit for generating a particle size distribution model using machine learning, in which at least the light scattering pattern of the particle group is used as an explanatory variable and the particle size distribution of the particle group is used as at least the objective variable;
a theoretical calculation unit that calculates a particle size distribution based on a predetermined theoretical calculation formula;
a particle size distribution output unit that outputs the particle size distribution of the particle group contained in the sample from the scattering pattern indicated by the output signal of the detector;
The particle size distribution output unit is
A particle size distribution is calculated from the particle size distribution model,
when the particle size distribution calculated by the particle size distribution model is within a predetermined range, outputting the particle size distribution obtained from the particle size distribution model;
A particle size distribution measuring device for outputting the particle size distribution calculated by the theoretical calculation unit when the particle size distribution calculated by the particle size distribution model is out of the predetermined range. The predetermined range is a particle diameter that satisfies 0.05 ≤ x ≤ 0.5 when the particle size parameter x = 2πr / λ = πd / λ (r: particle diameter radius, d: particle diameter diameter, λ: light source wavelength) 2. The particle size distribution measuring device according to claim 1, which is a distribution range. The particle size distribution output unit is configured to output a corresponding particle size distribution for each measurement,
3. The particle size distribution measuring apparatus according to claim 1, further comprising a display section for displaying a list of each particle size distribution output from the particle size distribution output section on one screen. The particle size distribution model generation unit performs machine learning using a light scattering pattern by a particle group having a non-spherical shape and a particle size distribution corresponding thereto as teacher data,
4. The particle size distribution measurement according to any one of claims 1 to 3, wherein the particle size distribution model generation unit is configured to generate a particle size distribution model having particle shape and particle size distribution as objective variables. Device. 5. Any one of claims 1 to 4, wherein the particle size distribution model generation unit is configured to perform machine learning using only a light scattering pattern by a particle group having a particle size distribution within a predetermined range and a particle size distribution corresponding thereto as teacher data. The particle size distribution measuring device according to any one of the above. 6. The particle size distribution measuring apparatus according to claim 5, wherein when the light emitted from the light source is blue, the particle size distribution in the predetermined range is a particle size distribution of 10 nm or more and 100 nm or less. The particle size distribution calculator,
An instrumental error model that uses a scattering pattern or particle size distribution as an explanatory variable and uses machine learning to generate an instrumental error model that uses the instrumental error between measured values of the same sample measured by itself and another particle size distribution analyzer as an objective variable. a generator;
7. An instrumental error output unit for outputting a scattering pattern measured by another particle size distribution measuring device from the scattering pattern indicated by the output signal of the detector based on the instrumental error model. The particle size distribution measuring device according to any one of the above. The particle size distribution calculator,
a calibration parameter model generation unit that uses machine learning to generate a calibration parameter model having a scattering pattern or particle size distribution as an explanatory variable and a calibration parameter including at least a refractive index as an objective variable;
a calibration parameter output unit that outputs calibration parameters to be set based on the calibration parameter model and the measured scattering pattern or particle size distribution;
8. Any one of claims 1 to 7, further comprising a theoretical calculation unit for calculating a particle size distribution based on the calibration parameters output from the calibration parameter output unit, a predetermined theoretical calculation formula, and the output of the detector. The particle size distribution measuring device described. an abnormal state model generation unit that generates an abnormal state model by machine learning, in which at least the scattering pattern is used as an explanatory variable and the abnormal state of the particle size distribution measuring apparatus is used as an objective variable;
9. The particle size distribution measuring apparatus according to any one of claims 1 to 8, further comprising the abnormal state model and an abnormality output unit that outputs an abnormality occurring based on the measured scattering pattern. a light source for emitting light to a cell containing a sample consisting of a group of particles as a dispersion medium and a dispersoid; a detector for detecting light scattered by the sample; A particle size distribution measuring device comprising a particle size distribution calculator for calculating a particle size distribution of a group of particles contained in a sample,
The particle size distribution calculator,
a model storage unit for storing a particle size distribution model having at least a light scattering pattern by a particle group generated by machine learning as an explanatory variable and having at least a particle size distribution of the particle group as an objective variable;
a theoretical calculation unit that calculates a particle size distribution based on a predetermined theoretical calculation formula;
a particle size distribution output unit that outputs the particle size distribution of the particle group contained in the sample from the scattering pattern indicated by the output signal of the detector;
The particle size distribution output unit is
A particle size distribution is calculated from the particle size distribution model,
when the particle size distribution calculated by the particle size distribution model is within a predetermined range, outputting the particle size distribution obtained from the particle size distribution model;
A particle size distribution measuring device for outputting the particle size distribution calculated by the theoretical calculation unit when the particle size distribution calculated by the particle size distribution model is out of the predetermined range. A particle size distribution measuring device is used that includes a light source that emits light to a cell containing a sample containing a group of particles that are a dispersion medium and a dispersoid, and a detector that detects light scattered by the sample. A particle size distribution measuring method comprising:
A particle size distribution calculating step of calculating a particle size distribution of a group of particles contained in the sample based on the output signal of the detector;
The particle size distribution calculation step includes
a particle size distribution model generation step of generating a particle size distribution model using machine learning, in which at least the light scattering pattern by the particle group is used as an explanatory variable and the particle size distribution of the particle group is used as at least the objective variable;
A theoretical calculation step of calculating a particle size distribution based on a predetermined theoretical calculation formula;
a particle size distribution output step of outputting the particle size distribution of the particle group contained in the sample from the scattering pattern indicated by the output signal of the detector;
The particle size distribution output step includes:
A particle size distribution is calculated from the particle size distribution model,
when the particle size distribution calculated by the particle size distribution model is within a predetermined range, outputting the particle size distribution obtained from the particle size distribution model;
A particle size distribution measuring method, wherein when the particle size distribution calculated by the particle size distribution model is outside the predetermined range, the particle size distribution calculated in the theoretical calculation step is output. Used in a particle size distribution measuring device equipped with a light source for emitting light to a cell containing a sample containing a group of particles as a dispersion medium and a dispersoid, and a detector for detecting light scattered by the sample. A program that
causing a computer to exhibit a function as a particle size distribution calculator for calculating the particle size distribution of a group of particles contained in the sample based on the output signal of the detector,
The particle size distribution calculator,
a particle size distribution model generation unit for generating a particle size distribution model using machine learning, in which at least the light scattering pattern of the particle group is used as an explanatory variable and the particle size distribution of the particle group is used as at least the objective variable;
a theoretical calculation unit that calculates a particle size distribution based on a predetermined theoretical calculation formula;
a particle size distribution output unit that outputs the particle size distribution of the particle group contained in the sample from the scattering pattern indicated by the output signal of the detector;
The particle size distribution output unit is
A particle size distribution is calculated from the particle size distribution model,
when the particle size distribution calculated by the particle size distribution model is within a predetermined range, outputting the particle size distribution obtained from the particle size distribution model;
A program for a particle size distribution measuring device, which outputs the particle size distribution calculated by the theoretical calculation unit when the particle size distribution calculated by the particle size distribution model is out of the predetermined range.

Images