CN115127965B - Inversion method and system for particle size distribution of mixed particle system - Google Patents
Inversion method and system for particle size distribution of mixed particle system Download PDFInfo
- Publication number
- CN115127965B CN115127965B CN202211038130.5A CN202211038130A CN115127965B CN 115127965 B CN115127965 B CN 115127965B CN 202211038130 A CN202211038130 A CN 202211038130A CN 115127965 B CN115127965 B CN 115127965B
- Authority
- CN
- China
- Prior art keywords
- particle size
- size distribution
- inversion
- particle
- scattering
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000002245 particle Substances 0.000 title claims abstract description 263
- 238000009826 distribution Methods 0.000 title claims abstract description 126
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000005311 autocorrelation function Methods 0.000 claims abstract description 48
- 238000002296 dynamic light scattering Methods 0.000 claims abstract description 36
- 238000009499 grossing Methods 0.000 claims abstract description 36
- 230000000694 effects Effects 0.000 claims abstract description 22
- 239000012798 spherical particle Substances 0.000 claims abstract description 15
- 238000012360 testing method Methods 0.000 claims description 83
- 238000011156 evaluation Methods 0.000 claims description 49
- 230000005684 electric field Effects 0.000 claims description 28
- 238000004088 simulation Methods 0.000 claims description 23
- 238000012549 training Methods 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 4
- 230000002547 anomalous effect Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 abstract description 6
- 238000000149 argon plasma sintering Methods 0.000 abstract description 2
- 239000011859 microparticle Substances 0.000 abstract 1
- 238000013528 artificial neural network Methods 0.000 description 21
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000013101 initial test Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 239000013618 particulate matter Substances 0.000 description 2
- 241000894006 Bacteria Species 0.000 description 1
- 241000700605 Viruses Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000005314 correlation function Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000003344 environmental pollutant Substances 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 244000005700 microbiome Species 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 231100000719 pollutant Toxicity 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 239000000725 suspension Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
- G01N15/02—Investigating particle size or size distribution
- G01N15/0205—Investigating particle size or size distribution by optical means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
- G01N15/02—Investigating particle size or size distribution
- G01N15/0205—Investigating particle size or size distribution by optical means
- G01N2015/0238—Single particle scatter
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
- G01N15/02—Investigating particle size or size distribution
- G01N2015/0277—Average size only
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Chemical & Material Sciences (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Pathology (AREA)
- Immunology (AREA)
- Dispersion Chemistry (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a particle size distribution inversion method and system for a mixed particle system, and belongs to the field of light scattering micro particle detection. The method comprises the following steps: calculating the average particle size of a sample to be detected under a plurality of scattering angles by an accumulative method according to a light intensity autocorrelation function of the sample to be detected with unknown particle size distribution measured by a plurality of scattering angles, inputting the average particle size into a particle size distribution inversion regression network of a hybrid particle system trained by an optimal smoothing factor, and finally obtaining a particle size distribution curve of the sample to be detected, wherein the optimal smoothing factor is obtained by evaluating and comparing the particle size distribution inversion regression network inversion effects of the hybrid particle system according to the noise level of a calibrated dynamic light scattering system. The method is suitable for spherical particle systems, non-spherical particle systems and mixed particle systems with various particles with different properties and forms, has high inversion speed of particle size distribution, good effect and non-contact, and can meet various requirements under actual detection environments.
Description
Technical Field
The invention belongs to the field of light scattering tiny particle detection, and particularly relates to a particle size distribution inversion method and system for a mixed particle system.
Background
The particles are tiny, discrete solids, liquid drops or air bubbles, and can also be vital viruses, bacteria, microorganisms and the like, and the scale ranges from sub-nanometer to millimeter magnitude. Since physical, chemical, biological and other characteristics of the particle group have important influences on the quality of the product, the production process, efficiency, safety and the like, more and more industries need to detect the characteristics of the particle group. The influence of the physical properties of the particle group on the behavior thereof is manifold, including particle size distribution, state distribution, shape distribution, density distribution, ductility, conductivity, microstructure and morphology, and the particle size distribution information of the particle group, i.e. the distribution rule of all particle sizes, is closely related to the efficient use of energy, the monitoring of atmospheric pollutants, the research and development of new materials, the detection of medical samples, and the like, and has become an important research topic in the field of practical measurement, and has been widely paid attention to and researched by domestic and foreign scholars.
The dynamic light scattering technology is an important method for measuring the particle size distribution of nano-scale and submicron-scale particle groups, and the particle size distribution of the particle groups is mainly obtained by inverting a correlation function of scattered light. The method is a label-free measuring method and has the advantages of high response speed, high accuracy and the like. However, in a certain particle suspension sample, the solution of the particle size distribution is obtained by inverting the particle size distribution through the light intensity autocorrelation function obtained by the dynamic light scattering technology, particularly, the particle size distribution inverted through the noisy light intensity autocorrelation function belongs to a ill-conditioned problem, and the solution is seriously deviated due to slight disturbance of original data. Therefore, the inversion of the particle size distribution by solving the ill-conditioned equation is difficult and very inaccurate in the conventional regularization inversion algorithm, and the inversion of the particle size distribution can only be performed for the spherical particle system. The method is limited by the traditional regularized inversion algorithm, and the dynamic light scattering method cannot accurately analyze the particle size distribution of non-spherical particle systems and mixed particle systems with various particles with different properties and shapes at present.
The neural network method not only avoids the problem of knowing the ill-conditioned equation, but also can carry out targeted training according to the property and the form of the inversion object a priori. Therefore, not only can more accurate particle size distribution be obtained, but also the inversion of the particle size distribution can be carried out for spherical particle systems, non-spherical particle systems and mixed particle systems.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a particle size distribution inversion method and system for a hybrid particle system, and aims to solve the problem that a reliable inversion method is lacked under the conditions of a non-spherical particle system, a hybrid particle system with various particles with different properties and forms and the like.
In order to achieve the above object, in a first aspect, the present invention provides a method for inverting a particle size distribution of a hybrid particle system, where there are a plurality of particulate matters with different properties and morphologies, the method including: s1, measuring a light intensity autocorrelation function of a sample to be measured with unknown particle size distribution at a plurality of scattering angles by using a multi-angle dynamic light scattering system; s2, calculating the average particle size of the sample to be measured under a plurality of scattering angles by an accumulative method according to the light intensity autocorrelation function of the sample to be measured with unknown particle size distribution measured by the plurality of scattering angles; s3, inputting the average particle size of the sample to be detected under a plurality of scattering angles into a trained particle size distribution inverse regression network of the hybrid particle system to obtain a particle size distribution curve of the sample to be detected with unknown particle size distribution, wherein the trained particle size distribution inverse regression network of the hybrid particle system adopts an optimal smooth factor; the determination of the optimal smoothing factor comprises the following steps: the method comprises the following steps of (1) calibrating the noise level of the dynamic light scattering system; (2) Simulating and calculating the scattering light intensity fractions corresponding to the particles with various forms and various particle sizes in the mixed particle system simulation sample under a plurality of scattering angles, wherein the simulation calculation process is carried out by adjusting the particle size distribution, the form and the refractive index of the particles, and then further calculating to obtain a noise-free electric field autocorrelation function under a plurality of scattering angles; (3) Multiplying the calibrated noise level by the noise distribution with the same actual noise type, and adding the result to the noise-free electric field autocorrelation function under a plurality of scattering angles to obtain a corresponding noise-containing electric field autocorrelation function; (4) The method comprises the steps of adopting a noisy electric field autocorrelation function under a plurality of scattering angles, using the average particle size of a simulation sample obtained by calculation through an accumulative method as network input, adopting corresponding particle size distribution of the simulation sample as network output, and comparing inversion effect evaluation values of a mixed particle system particle size distribution inversion regression network trained by different smoothing factors in a prior particle size inversion range; (5) And determining an optimal smoothing factor according to the inversion effect evaluation value.
Preferably, the step (1) comprises: (11) Measuring the light intensity autocorrelation function of a calibration sample with known particle size distribution at a plurality of scattering angles by using a multi-angle dynamic light scattering system; (12) Calculating inversion particle size distribution under different test noise levels according to the light intensity autocorrelation function of the calibration sample with known particle size distribution measured by a plurality of scattering angles; (13) Comparing the inverted particle size distribution under different test noise levels with the known particle size distribution, and determining the optimal interval of the test noise level according to the evaluation value of the inversion result; (14) Determining two or three optimal test noise levels by adopting a successive dichotomy in the optimal test noise level interval according to inversion result evaluation values corresponding to the test noise levels; (15) And weighting the optimal test noise level to obtain the noise level of the calibrated dynamic light scattering system.
It should be noted that the noise level of the dynamic light scattering system is calibrated by adopting the successive dichotomy, which is beneficial to enabling the test noise level to quickly and accurately approach the noise level of the real dynamic light scattering system and reducing the time consumption of the noise level calibration process.
Preferably, step (12) comprises: (121) Simulating and calculating the scattering light intensity fraction corresponding to the particles with each particle size in the simulated sample of the quasi-calibration sample under a plurality of scattering angles, wherein the simulation calculation process is carried out by adjusting the particle size distribution of a particle system, and then further calculating to obtain a noise-free electric field autocorrelation function under a plurality of scattering angles; (122) Multiplying each test noise level by the noise distribution with the same actual noise type, and adding the result with the noise-free electric field autocorrelation functions under a plurality of scattering angles to obtain a plurality of groups of corresponding noise-containing electric field autocorrelation functions; (123) The method comprises the steps of adopting a noise-containing electric field autocorrelation function under a group of a plurality of scattering angles corresponding to a test noise level in the middle of each test noise level, calculating the average particle size of a simulation sample through an accumulative method to be used as network input, adopting the particle size distribution of the corresponding simulation sample to be used as network output, and comparing inversion effect evaluation values of a noise level calibration regression network of a dynamic light scattering system trained by different smoothing factors in a prior particle size inversion range; (124) Determining an optimal smoothing factor under the test noise level according to the inversion effect evaluation value; (125) Respectively training to obtain a noise level calibration regression network of the dynamic light scattering system by adopting the noise-containing electric field autocorrelation functions under a plurality of scattering angles in each group and adopting the optimal smoothing factor determined in the step (124); (126) According to the light intensity autocorrelation function of the calibration sample with known particle size distribution measured by a plurality of scattering angles, the average particle size of the calibration sample under a plurality of scattering angles is calculated by an accumulative method, and the average particle size is input to the noise level calibration regression network of each trained dynamic light scattering system to invert the particle size distribution under each test noise level.
It should be noted that, the optimal smoothing factor under the test noise level in the middle of the test noise levels is adopted for training the dynamic light scattering system noise level calibration regression network under each test noise level, so as to invert the particle size distribution under each test noise level.
Preferably, if the optimal 2 test noise levels are selected, the calibrated system noise level is calculated as follows:
wherein, a and b are respectively the optimal 2 groups of evaluation values in the inversion result evaluation values of all the test noise levels, and x and y are respectively the test noise levels corresponding to a and b;
if the optimal 3 test noise levels are selected, the calibrated system noise level is calculated according to the following formula:
wherein, a, b and c are respectively the optimal 3 groups of evaluation values in the inversion result evaluation values of all the test noise levels, and x, y and z are respectively the test noise levels corresponding to a, b and c.
It should be noted that, in the present invention, the weighted optimal noise levels are obtained by using the specific gravity of the inversion result evaluation values, which is beneficial to fully using the inversion result evaluation values under the test noise levels close to the noise levels of the real dynamic light scattering system, and reducing the influence of the inherent noise randomness on the final calibration result.
Preferably, the step (2) is specifically as follows: if the inversion object of the particle size distribution is a spherical particle system, mie scattering is selected to be used, and if the inversion object of the particle size distribution is a non-spherical particle system, an anomalous diffraction integral theory is selected to be used, and the scattering light intensity fraction corresponding to the particles with the particle sizes in the particle system is calculated; if the inversion object of the particle size distribution is a mixed particle system with various particle matters with different properties and shapes, various theories are used for calculating the shapes of the mixed particle system and the scattering light intensity fractions corresponding to the particle matters with various particle sizes, and then the scattering light intensity fractions are superposed according to the particle sizes, and the electric field autocorrelation function is obtained through calculation.
It should be noted that the invention adopts different theories to calculate the scattering light intensity fractions corresponding to the particles with various particle sizes in the particle system, which is beneficial to fully utilizing the prior knowledge of the properties, the forms and the like of the particles and purposefully calculating the training sample of the neural network according to the difference of the inversion objects.
Preferably, the inversion effect evaluation value calculation formula in step (4) is as follows:
wherein,
is the particle size of the particles in the particle system,
for the training of the particle size distribution curve of the particle system used,
and obtaining a particle system particle size distribution curve by network inversion, wherein N is the number of sampling points in an inversion range, and M is the number of inversion objects in a prior particle size inversion range.
It should be noted that the inversion effect evaluation formula is optimized, and the requirement that multiple theoretical particle size distribution curve inversion effects with multi-parameter changes can be comprehensively evaluated by the requirement index of neural network training is met.
Preferably, in the step (5), an abscissa corresponding to a minimum value in a curve with the smoothing factor as an abscissa and the inversion effect evaluation index as an ordinate is determined as an optimal smoothing factor; if the curve does not have a minimum value, the smoothing factor range is modified until the curve has a minimum value.
In order to achieve the above object, in a second aspect, the present invention provides a system for inverting a particle size distribution of a mixed particle system, including: a processor and a memory; the memory is for storing a computer program or instructions; the processor is adapted to execute the computer program or instructions in the memory such that the method of the first aspect is performed.
Generally, compared with the prior art, the technical scheme conceived by the invention has the following beneficial effects:
the invention provides a particle size distribution inversion method and a particle size distribution inversion system for a hybrid particle system, which are characterized in that the actual noise level of a dynamic light scattering system is calibrated, the calibrated noise level is multiplied by the noise distribution with the same actual noise type, and then the multiplied noise level is superposed with noise-free electric field autocorrelation functions under a plurality of scattering angles to obtain corresponding noise-containing electric field autocorrelation functions, the noise-containing electric field autocorrelation functions under the plurality of scattering angles are adopted, the average particle size of a simulation sample obtained by calculation through an accumulative method is taken as network input, the particle size distribution of the corresponding simulation sample is taken as network output, the inversion effect evaluation values of particle size distribution inversion regression networks of the hybrid particle system trained by different smoothing factors in the prior particle size inversion range are compared, and the optimal smoothing factor is determined; and inputting the average particle size of the sample to be detected under a plurality of scattering angles, which is obtained by calculation through an accumulative method, into the particle size distribution inversion regression network of the hybrid particle system trained by the optimal smoothing factor to obtain the particle size distribution of the sample to be detected with unknown particle size distribution. The inversion of the particle size distribution adopts a mode based on neural network training, so the method is suitable for various complex practical application conditions such as spherical particle systems, non-spherical particle systems and mixed particle systems with various particles with different properties and forms, has high inversion speed of the particle size distribution, good effect and non-contact, and can meet various requirements under the actual detection environment. Once the network parameters are determined through the processes of calibration, training and the like, the rapid inversion of millisecond magnitude can be realized, and the method is more suitable for the requirements of practical application of industrial flow line production, medical multi-sample detection instruments and the like compared with the inversion time of few minutes in the traditional inversion algorithm.
Drawings
Fig. 1 is a flow chart of an inversion method of particle size distribution of a hybrid particle system provided by the present invention.
Fig. 2 is a flow chart of the noise level calibration of the dynamic light scattering system provided by the present invention.
FIG. 3 is a flow chart of simulation of the scattering intensity fraction of the mixed particle system according to the present invention.
Fig. 4 is a schematic diagram of a neural network training process provided by the present invention.
FIG. 5 is a flow chart of training a neural network and optimizing a smoothing factor according to the present invention.
Fig. 6 is a schematic diagram of an inversion process of a neural network provided by the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Fig. 1 is a flow chart of an inversion method of particle size distribution of a hybrid particle system provided by the present invention. As shown in fig. 1, the method comprises the steps of:
(1) And calibrating the noise level of the dynamic light scattering system.
And measuring a light intensity autocorrelation function of a calibration sample with known particle size distribution, calculating inverted particle size distribution under different test noise levels, comparing the inverted particle size distribution with the known particle size distribution, and calibrating the noise level of the real dynamic light scattering system by adopting a successive dichotomy in the optimal interval of the test noise level.
The step (1) of calibrating the noise level of the dynamic light scattering system comprises the following steps:
(11) The light intensity autocorrelation function of a calibration sample of known particle size distribution is measured at several scattering angles using a multi-angle dynamic light scattering system.
The light emitted by the laser is irradiated to a sample cell of a multi-angle dynamic light scattering system for placing a calibration sample with known particle size distribution, the multi-angle scattered light is received by a photon counter, and a digital correlator calculates the light intensity autocorrelation function curve of the calibration sample according to the received photon fluctuation.
(12) Calculating inversion particle size distribution under different test noise levels by using a dynamic light scattering system noise level calibration regression network, comparing with the known particle size distribution, selecting the optimal 2 test noise levels from inversion result evaluation under multiple groups of test noise levels as test noise level optimal intervals, dividing the test noise level optimal intervals successively, repeating the test until the difference between the test noise level obtained after dividing by two and the test noise level before dividing by two is less than the initial minimum test noise level, or the inversion result evaluation of the test noise level obtained after dividing by two is more than the inversion result evaluation of the test noise level before dividing by 2, and terminating dividing by two.
Fig. 2 is a flow chart of the noise level calibration of the dynamic light scattering system provided by the present invention. As shown in fig. 2, selecting initial test noise levels as 1.00%, 0.10%, 0.01%, combining the measured light intensity autocorrelation function curve of the calibrated sample particle system, calculating inversion particle size distributions under different test noise levels using a dynamic light scattering system noise level calibration regression network, comparing with the known particle size distributions, evaluating the inversion results, selecting the optimal 2 test noise levels as the optimal test noise level interval, successively bisecting the optimal test noise level interval, repeating the above test until the difference between the test noise level obtained after bisection and the test noise level before bisection is less than the initial minimum test noise level, or the inversion result evaluation of the test noise level obtained after bisection is greater than the inversion result evaluation of the 2 test noise levels before bisection, and terminating bisection. In this embodiment, the inversion result evaluation is measured by using the relative error between the peak value of the inverted particle size distribution curve and the peak value of the theoretical particle size distribution curve.
(13) And if the dichotomy stopping criterion is that the difference value between the test noise level obtained after the dichotomy and the test noise level before the dichotomy is less than the initial minimum test noise level, selecting the optimal inversion result evaluation value of 2 groups of test noise levels, and if the dichotomy stopping criterion is that the inversion result evaluation value of the test noise level obtained after the dichotomy is greater than the inversion result evaluation values of 2 test noise levels before the dichotomy, selecting the optimal inversion result evaluation value of 3 groups of test noise levels for calculating and calibrating the real noise level of the dynamic light scattering system.
In this embodiment, if the difference between the test noise level obtained after the bisection and the test noise level before the bisection is less than the initial minimum test noise level, the bisection is terminated, and then the optimal 2 test noise levels are selected, and the calibrated system noise level is calculated according to the following formula:
wherein, a and b are respectively the optimal 2 groups of evaluation values in the inversion result evaluation values of all the test noise levels, and x and y are respectively the test noise levels corresponding to a and b.
If the binary division is terminated because the inversion result evaluation value of the test noise level obtained after the binary division is greater than the inversion result evaluation value of the test noise level 2 before the binary division, the optimal 3 test noise levels are selected, and the calibrated system noise level is calculated according to the following formula:
wherein, a, b and c are respectively the optimal 3 groups of evaluation values in the inversion result evaluation values of all the test noise levels, and x, y and z are respectively the test noise levels corresponding to a, b and c.
(2) The intermixed particles are simulated by the fraction of scattered light.
FIG. 3 is a flow chart of simulation of the scattering intensity fraction of the mixed particle system according to the present invention. As shown in fig. 3, if the inversion target of the particle size distribution is a mixed particle system with various types and forms of particulate matter, the scattering light intensity fractions corresponding to the various types and forms of the mixed particle system and the particulate matter with various particle sizes are calculated by using various theories at the same time; if the inversion object of the particle size distribution is a spherical particle system, selecting and using Mie scattering and other theories to calculate the scattering light intensity fraction corresponding to the particles with each particle size in the particle system; if the inversion object of the particle size distribution is a non-spherical particle system, the scattering light intensity fraction corresponding to the particles with the particle sizes of the particle system is calculated by using theories such as anomalous diffraction integral and the like. And then, respectively superposing the simulation calculation according to each particle size to obtain the scattering light intensity fraction corresponding to the particulate matters with each shape and each particle size in the simulation sample of the mixed particle system under a plurality of scattering angles, and further calculating to obtain a noise-free electric field autocorrelation function under a plurality of scattering angles.
(3) Training the neural network and optimizing the smoothing factor.
Fig. 4 is a schematic diagram of a neural network training process provided by the present invention. The neural network used in this embodiment is a GRNN (generalized recurrent neural network), and other neural networks may also be used. As shown in fig. 4, the training process of the neural network is as follows: multiplying the noise level calibrated in the step (1) by the noise distribution with the same actual noise type, adding the result with the noise-free electric field autocorrelation function under a plurality of scattering angles obtained in the step (2) to obtain a corresponding noise-containing electric field autocorrelation function, calculating the average particle size of the obtained simulation sample by an accumulative method as network input, adopting the corresponding simulation sample particle size distribution as network output, and comparing the inversion effect evaluation value of the particle size distribution inversion regression network of the mixed particle system trained by different smoothing factors with the particle size distribution curve of the simulation sample in the prior particle size inversion range, thereby determining the optimal smoothing factor of the neural network under the condition.
FIG. 5 is a flow chart of training a neural network and optimizing a smoothing factor according to the present invention. As shown in FIG. 5, the neural network is trained in the prior particle size inversion range according to the calibrated system noise level and the calculated electric field autocorrelation function, and the range of the smoothing factor of the initial test can be set as
And comparing the evaluation indexes of the inversion effect of the mixed particle system particle size distribution inversion regression network trained by different smoothing factors on the inversion object in the prior particle size inversion range
：
Wherein,
is the particle size of the particles in the particle system,
for the training of the particle size distribution curve of the particle system used,
obtaining a particle system particle size distribution curve for network inversion, wherein N is the number of sampling points in an inversion range, and M is the number of inversion objects in a prior particle size inversion range。
The smaller the value is, the closer the particle system particle size distribution curve obtained by inversion is to the particle system particle size distribution curve used in training, and the better the inversion effect is.
If the curve taking the smoothing factor as the abscissa and the inversion effect evaluation index as the ordinate has no minimum value, modifying the magnitude of the upper and lower limits of the smoothing factor range until the evaluation index curve has the minimum value, and determining the optimal smoothing factor of the neural network under the condition as the value of the abscissa smoothing factor corresponding to the minimum value of the evaluation index curve.
(4) And measuring experimental data of the dynamic light scattering system.
Fig. 6 is a schematic diagram of an inversion process of a neural network according to the present invention. As shown in fig. 6, in the inversion process of the neural network, a multi-angle dynamic light scattering system is used to measure the light intensity autocorrelation function of the sample to be measured with unknown particle size distribution at a plurality of scattering angles. The light emitted by the laser irradiates to a sample cell of a multi-angle dynamic light scattering system for placing a sample to be detected with unknown particle size distribution, the multi-angle scattered light is received by a photon counter, and a digital correlator calculates a light intensity autocorrelation function curve of the sample to be detected according to the received photon fluctuation.
(5) And (4) performing inverse calculation on the particle size distribution of the mixed particles.
As shown in fig. 6, in the inversion process of the neural network, an electric field autocorrelation function is calculated according to the light intensity autocorrelation function of the sample to be measured with unknown particle size distribution measured at a plurality of scattering angles in step (4), and the average particle size of the sample to be measured at a plurality of scattering angles is further calculated by an accumulative method; and then designing a neural network-based particle size distribution inverse regression network of the hybrid particle system, and inputting the average particle size of the sample to be detected under a plurality of scattering angles into the particle size distribution inverse regression network of the hybrid particle system trained by the optimal smoothing factor to obtain a particle size distribution curve of the sample to be detected with unknown particle size distribution.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (7)
1. A method for inverting the particle size distribution of a hybrid particle system, wherein a plurality of particles with different properties and forms exist in the hybrid particle system, is characterized by comprising the following steps:
s1, measuring a light intensity autocorrelation function of a sample to be measured with unknown particle size distribution at a plurality of scattering angles by using a multi-angle dynamic light scattering system;
s2, calculating the average particle size of the sample to be measured under a plurality of scattering angles by an accumulative method according to the light intensity autocorrelation function of the sample to be measured with unknown particle size distribution measured by the plurality of scattering angles;
s3, inputting the average particle size of the sample to be detected under a plurality of scattering angles into a trained particle size distribution inverse regression network of the hybrid particle system to obtain a particle size distribution curve of the sample to be detected with unknown particle size distribution, wherein the trained particle size distribution inverse regression network of the hybrid particle system adopts an optimal smooth factor;
the determination of the optimal smoothing factor comprises the following steps:
(1) Calibrating the noise level of the dynamic light scattering system;
(2) Simulating and calculating the scattering light intensity fractions corresponding to the particles with various forms and various particle sizes in the mixed particle system simulation sample under a plurality of scattering angles, wherein the simulation calculation process is carried out by adjusting the particle size distribution, the form and the refractive index of the particles, and then further calculating to obtain a noise-free electric field autocorrelation function under a plurality of scattering angles;
(3) Multiplying the calibrated noise level by the noise distribution with the same actual noise type, and adding the result to the noise-free electric field autocorrelation function under a plurality of scattering angles to obtain a corresponding noise-containing electric field autocorrelation function;
(4) The method comprises the steps of adopting a noisy electric field autocorrelation function under a plurality of scattering angles, using the average particle size of a simulation sample obtained by calculation through an accumulative method as network input, adopting corresponding particle size distribution of the simulation sample as network output, and comparing inversion effect evaluation values of a mixed particle system particle size distribution inversion regression network trained by different smoothing factors in a prior particle size inversion range;
(5) Determining an optimal smoothing factor according to the inversion effect evaluation value;
the step (1) comprises the following steps:
(11) Measuring the light intensity autocorrelation function of a calibration sample with known particle size distribution at a plurality of scattering angles by using a multi-angle dynamic light scattering system;
(12) Calculating inversion particle size distribution under different test noise levels according to the light intensity autocorrelation function of the calibration sample with known particle size distribution measured by a plurality of scattering angles;
(13) Comparing the inverted particle size distribution under different test noise levels with the known particle size distribution, and determining the optimal interval of the test noise level according to the evaluation value of the inversion result;
(14) Determining two or three optimal test noise levels by adopting a successive dichotomy in the optimal test noise level interval according to inversion result evaluation values corresponding to the test noise levels;
(15) And weighting the optimal test noise level to obtain the calibrated noise level of the dynamic light scattering system.
2. The method of claim 1, wherein step (12) comprises:
(121) Simulating and calculating the scattering light intensity fraction corresponding to the particles with each particle size in the simulated sample of the quasi-calibration sample under a plurality of scattering angles, wherein the simulation calculation process is carried out by adjusting the particle size distribution of a particle system, and then further calculating to obtain a noise-free electric field autocorrelation function under a plurality of scattering angles;
(122) Multiplying each test noise level by the noise distribution with the same actual noise type, and adding the result with the noise-free electric field autocorrelation functions under a plurality of scattering angles to obtain a plurality of groups of corresponding noise-containing electric field autocorrelation functions;
(123) The method comprises the steps of adopting a noise-containing electric field autocorrelation function under a group of a plurality of scattering angles corresponding to a test noise level in the middle of each test noise level, calculating the average particle size of a simulation sample through an accumulative method to be used as network input, adopting the particle size distribution of the corresponding simulation sample to be used as network output, and comparing inversion effect evaluation values of a noise level calibration regression network of a dynamic light scattering system trained by different smoothing factors in a prior particle size inversion range;
(124) Determining an optimal smoothing factor under the test noise level according to the inversion effect evaluation value;
(125) Respectively training to obtain a dynamic light scattering system noise level calibration regression network by adopting noise-containing electric field autocorrelation functions under a plurality of scattering angles in each group and adopting the optimal smoothing factor determined in the step (124);
(126) According to the light intensity autocorrelation function of the calibration sample with known particle size distribution measured by a plurality of scattering angles, the average particle size of the calibration sample under a plurality of scattering angles is calculated by an accumulative method, and the average particle size is input to the noise level calibration regression network of each trained dynamic light scattering system to invert the particle size distribution under each test noise level.
3. The method of claim 1, wherein if the optimal 2 test noise levels are selected, the calibrated system noise level is calculated as follows:
wherein, a and b are respectively the optimal 2 groups of evaluation values in the inversion result evaluation values of all the test noise levels, and x and y are respectively the test noise levels corresponding to a and b;
if the optimal 3 test noise levels are selected, the calibrated system noise level is calculated according to the following formula:
wherein, a, b and c are respectively the optimal 3 groups of evaluation values in the inversion result evaluation values of all the test noise levels, and x, y and z are respectively the test noise levels corresponding to a, b and c.
4. The method according to claim 1, wherein the step (2) is specifically as follows:
if the inversion object of the particle size distribution is a spherical particle system, mie scattering is selected to be used, and if the inversion object of the particle size distribution is a non-spherical particle system, an anomalous diffraction integral theory is selected to be used, and the scattering light intensity fraction corresponding to the particles with the particle sizes of the particle system is calculated;
if the inversion object of the particle size distribution is a mixed particle system with various particles with different properties and forms, various theories are used for calculating the forms of the mixed particle system and the scattering light intensity fractions corresponding to the particles with various particle sizes at the same time, and then the scattering light intensity fractions are superposed according to the particle sizes, and the electric field autocorrelation function is obtained through calculation.
5. The method of claim 1, wherein the inversion effect evaluation value in step (4) is calculated as follows:
wherein,
is the particle size of the particles in the particle system,
for the training of the particle size distribution curve of the particle system used,
and obtaining a particle system particle size distribution curve by network inversion, wherein N is the number of sampling points in an inversion range, and M is the number of inversion objects in a prior particle size inversion range.
6. The method according to claim 1, wherein in the step (5), an abscissa corresponding to a minimum value in a curve having the smoothing factor as an abscissa and the inversion effect evaluation index as an ordinate is determined as an optimal smoothing factor; if the curve does not have a minimum value, the smoothing factor range is modified until the curve has a minimum value.
7. An inversion system of particle size distribution of a mixed particle system, comprising: a processor and a memory;
the memory is for storing a computer program or instructions;
the processor is for executing the computer program or instructions in the memory, causing the method of any of claims 1-6 to be performed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211038130.5A CN115127965B (en) | 2022-08-29 | 2022-08-29 | Inversion method and system for particle size distribution of mixed particle system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211038130.5A CN115127965B (en) | 2022-08-29 | 2022-08-29 | Inversion method and system for particle size distribution of mixed particle system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115127965A CN115127965A (en) | 2022-09-30 |
| CN115127965B true CN115127965B (en) | 2022-12-02 |
Family
ID=83387430
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211038130.5A Active CN115127965B (en) | 2022-08-29 | 2022-08-29 | Inversion method and system for particle size distribution of mixed particle system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115127965B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115753532B (en) * | 2023-01-04 | 2023-05-12 | 武汉格蓝若智能技术股份有限公司 | Transformer oil granularity online analysis method and system based on light intensity time domain characteristics |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110595962A (en) * | 2019-09-29 | 2019-12-20 | 山东理工大学 | Non-negative TSVD dynamic light scattering inversion method for self-adaptive sampling of particle size distribution |
| JP2020190508A (en) * | 2019-05-23 | 2020-11-26 | 株式会社堀場製作所 | Particle size distribution measuring device, particle size distribution measurement method, and program for particle size distribution measuring device |
| CN113552031A (en) * | 2021-08-30 | 2021-10-26 | 山东理工大学 | Dynamic light scattering measurement method for ultralow-concentration suspended particles |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170260574A1 (en) * | 2016-01-15 | 2017-09-14 | Juan Noveron | Rapid Gene Sensors from Carbon Nanotube-DNA Systems |
| NL2020138B1 (en) * | 2017-12-20 | 2019-06-26 | Inprocess Ip B V | Method and apparatus for in-process particle size determination of colloidal suspensions under flow. |
| EP3521806A1 (en) * | 2018-02-06 | 2019-08-07 | Malvern Panalytical Limited | Multi-angle dynamic light scattering |
-
2022
- 2022-08-29 CN CN202211038130.5A patent/CN115127965B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2020190508A (en) * | 2019-05-23 | 2020-11-26 | 株式会社堀場製作所 | Particle size distribution measuring device, particle size distribution measurement method, and program for particle size distribution measuring device |
| CN110595962A (en) * | 2019-09-29 | 2019-12-20 | 山东理工大学 | Non-negative TSVD dynamic light scattering inversion method for self-adaptive sampling of particle size distribution |
| CN113552031A (en) * | 2021-08-30 | 2021-10-26 | 山东理工大学 | Dynamic light scattering measurement method for ultralow-concentration suspended particles |
Non-Patent Citations (3)
| Title |
|---|
| A Neural Network Model for Estimating the Particle Size Distribution of Dilute Latex from Multiangle Dynamic Light Scattering Measurements;Luis M. Gugliotta等;《Particle & Particle Systems Characterization》;20091231;第26卷;第41-52页 * |
| 多角度动态光散射加权贝叶斯反演算法;刘玲等;《计算物理》;20191130;第36卷(第6期);第673-681页 * |
| 多角度动态光散射粒度分布递归非负Phillips-Twomey算法;李蕾等;《激光与光电子学进展》;20151231;第52卷;第032901-1至032901-9页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115127965A (en) | 2022-09-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101398311B (en) | Repeat dynamic measurement data processing method based on grey system theory | |
| CN105444923A (en) | Mechanical temperature instrument error prediction method based on genetic-algorithm optimized least square support vector machine | |
| CN115127965B (en) | Inversion method and system for particle size distribution of mixed particle system | |
| CN106383072A (en) | Method for synchronous measurement of optical constant and particle size distribution of spherical particles based on multi-angle light scattering-transmission process | |
| CN116562171A (en) | Error assessment method for online measurement of temperature and humidity | |
| Uijlenhoet et al. | Analytical solutions to sampling effects in drop size distribution measurements during stationary rainfall: Estimation of bulk rainfall variables | |
| Larasati et al. | The relationship between data skewness and accuracy of Artificial Neural Network predictive model | |
| CN117669376A (en) | Inversion method and system for particle size distribution of liquid impurity particle system | |
| Ding et al. | Dynamic evolutionary model based on a multi-sampling inherited hapfnn for an aluminium electrolysis manufacturing system | |
| CN110889207A (en) | System combination model credibility intelligent evaluation method based on deep learning | |
| CN116882324B (en) | Data-driven sliding flow near-wall area nonlinear constitutive flow field modeling method | |
| Xiao et al. | Impact of damping amplitude on chaos detection reliability of the improved 0–1 test for oversampled and noisy observations | |
| Sunori et al. | Prediction of Air Pollutant PM 10 using Various SVM Models | |
| CN109101759A (en) | A kind of parameter identification method based on forward and reverse response phase method | |
| Vega et al. | Particle size distribution by combined elastic light scattering and turbidity measurements. A novel method to estimate the required normalization factor | |
| CN112307670A (en) | Design method of pressurized water reactor core parameter prediction model based on bagging integrated neural network | |
| CN106772306B (en) | A kind of detection method and server of object | |
| CN111062118A (en) | Multilayer soft measurement modeling system and method based on neural network prediction layering | |
| Aslam et al. | Moving average control chart under neutrosophic statistics | |
| Su et al. | Online Fusion Method for Homogeneous Multi-sensor Based on Improved Fuzzy Clustering | |
| Tu et al. | Metal-loss defect depth inversion in oil and gas pipelines based on Bayesian regularisation neural network | |
| Zhai et al. | A directed grey incidence model based on panel data | |
| Bamrungsetthapong et al. | Designing of Double Acceptance Sampling Plan for Zero-inflated and Over-dispersed Data Using Multi-objective Optimization | |
| Li et al. | A study on the UKFNN-based online detection of effluent COD in water sewage treatment | |
| Yong-Hong et al. | Soft sensor modeling based on GD-FNN for microbial fermentation process |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |