A Machine Learning-Optimized Method for Preparing Multi-Objective TiO2-Based Nanoparticles

Abstract

The present application relates to a kind of preparation method of nano TiO2-based particles satisfying multi-objective demand based on machine learning optimization, the method is synthesized by solvothermal method and surface modification under the experimental condition of algorithm model optimization, and the nano TiO2-based particles with monodisperse, low photocatalytic activity, according to experimental variables and product performance, corresponding database is established, experimental variables and product performance are processed by variable one-hot encoding, data normalization, correlation coefficient calculation and other calculation analysis methods, and the prediction model with test set and training set division, random forest model construction and optimization, prediction result verification, ideal experimental variable screening as framework is constructed.The present application satisfies multi-objective demand according to model prediction, and verifies that titanium dioxide-based cerium oxide heterojunction can be successfully used as monodisperse, low photocatalytic activity ultraviolet shielding agent.

Description

Technical Field

[0001] This invention relates to the field of nanocomposite dispersion preparation technology; more specifically, it relates to a method for preparing nano-TiO2-based particles that meet multiple objective requirements based on machine learning optimization. Background Technology

[0002] In recent years, with changes in atmospheric composition and the destruction of the ozone layer, ultraviolet (UV) radiation has increased dramatically, mainly including UVB (280-315nm) and UVA (315-400nm). Long-term exposure to UV radiation not only harms human health but also severely affects the performance of industrial materials, such as causing photoaging of organic polymers. Therefore, the development of UV protection materials, especially transparent UV-shielding coatings that can block most UV rays while allowing high transmission of visible light, is particularly important.

[0003] Traditional organic UV absorbers have been widely used, but they are self-consumed and migrate within polymers, resulting in poor long-term stability. Recent emerging UV-absorbing fillers, such as lignin and melanin, often have large particle sizes and dark colors, affecting visible light transparency. Therefore, inorganic metal oxides that can reflect, scatter, or absorb UV light without absorbing visible light are suitable choices. Titanium dioxide (TiO2), with its wide bandgap, possesses high physicochemical stability, is non-toxic, and inexpensive, and has been widely reported as a filler in the manufacture of UV-shielding composite materials. Rayleigh scattering at the interface between the inorganic component and the matrix polymer due to particle agglomeration and phase separation can make the composite material opaque. Therefore, it is necessary to control the particle size to below 40 nanometers and modify the particle surface to improve compatibility with the matrix polymer.

[0004] It is important to note that although titanium dioxide possesses excellent UV shielding properties, its photocatalytic activity during UV absorption limits its applications. Reactive oxygen species (ROS) generated during photocatalysis accelerate the degradation of the matrix polymer, thus shortening the lifespan of the hybrid material. To suppress the photocatalytic activity of titanium dioxide, a traditional method is to coat the surface of titanium dioxide nanoparticles with organic or inorganic layers. Organic layers of titanium dioxide are typically achieved by reacting surface hydroxyl groups with coupling agents or diblock and triblock copolymers; silica, alumina, and zirconium oxide can be used to construct inorganic surface layers. Constructing a surface layer establishes a barrier between titanium dioxide and the surrounding matrix, thereby reducing the photodegradation effect of titanium dioxide on the matrix polymer. However, a drawback of such methods is that the coating process is usually carried out when the titanium dioxide particles have already agglomerated. Agglomerated particles increase visible light scattering and reduce the transparency of the hybrid material. By introducing metal ions with ionic radii much larger than titanium ions during the synthesis of titanium dioxide, these metal ions will precipitate and accumulate on the titanium dioxide surface in the form of oxides, subsequently forming titanium dioxide-based metal oxide heterojunctions in situ. By designing a reasonable band gap structure, a cross-gap (Type I) heterojunction is constructed. Since photogenerated electrons and holes are concentrated on the same semiconductor with a small band gap, the recombination of electron-hole pairs will be promoted, the redox potential of charge carriers will be lowered, and the photocatalytic activity of titanium dioxide will be significantly weakened.

[0005] In developing similar monodisperse functional heterostructures, current methods primarily rely on empirical prediction and experimental trial and error. This requires extensive experimentation, consuming significant human and material resources, resulting in low R&D efficiency and high costs. The independent variables in the development of monodisperse functional heterostructures include the volume ratio of titanium source to water, reaction temperature, reaction time, type of metal salt, molar ratio of metal salt to titanium source, and type of surface modifier. The target product must simultaneously possess monodispersity, low photocatalytic activity, and UV shielding capability. Due to the numerous independent variables and target product performance parameters, the inability to directly quantify some parameters, and the correlations between parameters, correlation coefficient analysis and screening are necessary to reduce the complexity of the prediction model and reveal potential relationships between parameters. By introducing machine learning methods, a prediction model based on the random forest algorithm is constructed. The model is trained using processed and screened preliminary experimental data to discover the multi-objective independent variables corresponding to the target product that meet the expected product performance requirements. This provides direction for optimizing material preparation analysis, greatly improving efficiency and the comprehensive understanding of overall experimental variables. Summary of the Invention

[0006] One objective of this invention is to overcome the shortcomings of existing technologies and provide a method for preparing a monodisperse, low-photocatalytically active nano-titanium dioxide-based UV shielding agent that meets multiple objective requirements, based on machine learning optimization. The prepared nano-titanium dioxide-based metal oxide heterojunction simultaneously possesses monodispersity, low photocatalytic activity, and UV shielding capability. Machine learning is used to improve the development efficiency of monodisperse functional heterojunctions, comprehensively considering the influence of different reaction variables on various product properties, thereby optimizing the selection of precursors and the design of reaction conditions. The process conditions meet the effective boundary conditions for the preparation of nano-TiO2-based particles and a wide range of performance variations.

[0007] The second objective of this invention is to overcome the shortcomings of the prior art and provide an application of a monodisperse, low-photocatalytically active nano-titanium dioxide-based UV shielding agent as a UV absorber filler in a transparent UV-protective coating. This nano-titanium dioxide-based UV shielding agent has good compatibility with the matrix polymer and can meet the requirements for long-term transparent UV protection.

[0008] Based on this, the present invention mainly adopts the following technical solution: a method for preparing a monodisperse low photocatalytic activity nano-titanium dioxide-based ultraviolet shielding agent that meets multiple objective requirements based on machine learning optimization, characterized by including the following steps:

[0009] 1) Prepare a precursor solution for preparing nano-TiO2-based particles, synthesize wet aggregate precipitates of nano-TiO2-based particles by solvothermal method, centrifuge and wash the wet aggregate precipitates, add a surface modifier and disperse them in an organic solvent, and obtain nano-TiO2-based particles after post-treatment; then, prepare multiple groups of nano-TiO2-based particles with different properties by changing experimental variables; the post-treatment includes anti-solvent precipitation, centrifugation and freeze drying;

[0010] 2) Evaluate the product performance of multiple groups of nano-TiO2-based particles prepared in step 1). The experimental variables and corresponding product performance of multiple groups of nano-TiO2-based particles are combined into a dataset. Then, the experimental variables and product performance in the dataset are normalized and scaled to obtain the scaled dataset.

[0011] 3) Based on the scaled dataset, calculate the correlation coefficients between experimental variables and between product performances. Represent multiple experimental variables with correlation coefficients greater than 0.9 using one of the experimental variables to obtain dimensionality-reduced experimental variables; at the same time, represent multiple product performances with correlation coefficients greater than 0.9 using one of the product performances to obtain dimensionality-reduced product performances; thus, the dataset is dimensionality-reduced to obtain the dimensionality-reduced dataset.

[0012] 4) Construct an initialization model based on machine learning methods; train and optimize the initialization model by selecting experimental variables and product performance indicators of nano-TiO2-based particles to obtain a performance prediction model for nano-TiO2-based particles.

[0013] 5) Input the product performance of the required nano-TiO2-based particles into the nano-TiO2-based particle performance prediction model; the nano-TiO2-based particle performance prediction model outputs the corresponding dimension-reduced experimental variables; then, based on the dimension-reduced experimental variables output by the prediction model and the correlation coefficients of the experimental variables, perform dimension-upgrading to obtain all the experimental variables required for preparing nano-TiO2-based particles, and prepare nano-TiO2-based particles according to all the experimental variables.

[0014] As a preferred embodiment of the present invention, the precursor solution in step 1) includes a titanium source, a metal salt, water, a hydrolysis inhibitor, and an organic solvent.

[0015] The titanium source includes one or more of tetrabutyl titanate, tetraethyl titanate, isopropyl titanate, titanium tetrachloride, titanium oxysulfate, and titanium acetylacetone.

[0016] The metal salts include sodium chloride, copper chloride, manganese chloride, tin chloride, lanthanum chloride, ferric nitrate, cerium nitrate, europium nitrate, aluminum nitrate, zirconium nitrate, lanthanum nitrate, gallium nitrate, terbium nitrate, and samarium nitrate;

[0017] The hydrolysis inhibitors include one or more of nitric acid, hydrochloric acid, sulfuric acid, phosphoric acid, formic acid, acetic acid, salicylic acid, lactic acid, tartaric acid, and citric acid;

[0018] The organic solvent includes one or more of methanol, ethanol, isopropanol, n-butanol, tert-butanol, octanol, isocetyl alcohol, propylene glycol, benzyl alcohol, acetone, butanone, propylene glycol methyl ether acetate, tetrahydrofuran, dimethyl sulfoxide, chloroform, ethyl acetate, butyl acetate, benzene, toluene, xylene, heptane, n-hexane, cyclohexane, and petroleum ether.

[0019] As a preferred embodiment of the present invention, the experimental variables include the volume ratio of titanium source to water, reaction temperature, reaction time, type of metal salt, molar ratio of metal salt to titanium source, and type of surface modifier.

[0020] As a preferred embodiment of the present invention, the precursor solution in step 1) is homogenized by multi-stream feeding and rapid stirring, and the homogenization mixing time is controlled within 0.05-60 min; the nano-TiO2-based particles are a composite of titanium dioxide and other metal oxides; the solvothermal reaction temperature is 80-250℃, the solvothermal reaction pressure is 0.01-10 MPa, and the solvothermal reaction time is 1-72 h.

[0021] As a preferred embodiment of the present invention, the surface modifier mentioned in step 1) includes one or more of long-chain organic acids, phosphate esters, titanates and silane coupling agents;

[0022] The long-chain organic acid is a carboxylic acid having 12-24 carbon atoms, including lauric acid, myristic acid, palmitic acid, stearic acid, arachidic acid, oleic acid, and linoleic acid.

[0023] The phosphate esters include one or more of the following: 4-hydroxybutyl phosphate dihydrogen ester, 2-ethylhexyl phosphate dihydrogen ester, 9-octadecene-1-ol phosphate ester, isooctyl phosphate ester, dodecyl phosphate ester, alkyl diphenyl phosphate ester, tetradecyl phosphate ester, lauryl polyoxyethylene ether phosphate ester, and tridecyl polyoxyethylene ether phosphate ester.

[0024] The titanate comprises one or more of the following: isopropyl tris(dodecylbenzenesulfonyl) titanate, isopropyl triisostearate titanate, bis(acetylacetonyl) diisopropyl titanate, isopropyl tris(dioctyl pyrophosphate) titanate, isopropyl tris(dioctyl phosphate) titanate, isopropyl dioleoyl phosphate (dioctyl phosphate) titanate, di(octylphenol polyoxyethylene ether) phosphate, and tetraisopropyl di(dioctyl phosphite) titanate.

[0025] The silane coupling agent includes one or more of the following: vinyltrimethoxysilane, vinyltriethoxysilane, dodecyltrimethoxysilane, dodecyltriethoxysilane, hexadecyltrimethoxysilane, hexadecyltriethoxysilane, phenyltrimethoxysilane, γ-aminopropyltrimethoxysilane, 3-glycidyl etheroxypropyltrimethoxysilane, γ-methacryloyloxypropyltrimethoxysilane, N-β-aminoethyl-γ-aminopropyltriethoxysilane, methoxytriethylene glycol etherylpropyltrimethoxysilane, 11-mercaptoundecyltrimethoxysilane, 3-mercaptopropyltrimethoxysilane, and diethylphosphorylethyltriethoxysilane.

[0026] The organic solvent includes one or more of methanol, ethanol, isopropanol, n-butanol, tert-butanol, octanol, isocetyl alcohol, propylene glycol, benzyl alcohol, acetone, butanone, propylene glycol methyl ether acetate, tetrahydrofuran, dimethyl sulfoxide, dichloromethane, chloroform, ethyl acetate, butyl acetate, benzene, toluene, xylene, heptane, n-hexane, cyclohexane, and petroleum ether.

[0027] As a preferred embodiment of the present invention, in step 2), the product performance of the nano-TiO2-based particles is evaluated by dispersibility test, particle size distribution test, solvent compatibility test, photocatalytic activity test, ultraviolet shielding ability test and crystallinity test; the experimental variables and product performance in the dataset are normalized by using the maximum and minimum values ​​in the dataset to scale data of different orders of magnitude to the [0,1] interval.

[0028] As a preferred embodiment of the present invention, in step 3), the correlation coefficient between experimental variables for:

[0029]

[0030] Correlation coefficient between product properties

[0031]

[0032] Where x1 and x2 are two different experimental variables; z1 and z2 are two different product properties; and N is the amount of data.

[0033] As a preferred embodiment of the present invention, in step 4), the initialization model based on the machine learning method is a random forest model. The experimental variables and product performance indicators of nano-TiO2-based particles are randomly divided into training set and test set. The random forest model is trained and optimized using the training set, and the performance prediction model of nano-TiO2-based particles is evaluated and verified using the test set.

[0034] As a preferred embodiment of the present invention, in step 4), the evaluation indices of the performance prediction model for nano-TiO2-based particles include mean square error (MSE) and coefficient of determination (R²). 2 :

[0035]

[0036]

[0037] Where N is the amount of data. For the prediction result, y i This represents the actual results for the test set.

[0038] This invention also provides an application of the nano-TiO2-based particles prepared by the above method as a UV-absorbing filler in a transparent UV-protective coating. The method is characterized by thoroughly mixing the nano-TiO2-based particles dispersed in toluene with the fluorocarbon coating main agent, then adding a fluorocarbon coating curing agent and mixing uniformly again, using this mixture as a transparent UV-protective coating for scraping, spin coating, and spraying. Compared with the prior art, this invention has at least the following beneficial effects:

[0039] 1) The nano-TiO2-based particles prepared by the method of this invention are essentially nano-titanium dioxide-based metal oxide heterojunctions, which simultaneously possess monodispersity, low photocatalytic activity, and ultraviolet shielding capability. The particle size distribution of these nano-titanium dioxide-based metal oxide heterojunctions is narrow and can be controlled within the range of 10-200 nm, with small-particle samples having a diameter distribution in the range of 3-15 nm; the photocatalytic activity is reduced by more than 30 times compared to the control group of titanium dioxide.

[0040] 2) Because metal ions are precipitated and enriched in situ on the surface of primary titanium dioxide particles during the solvothermal reaction, the drawback of traditional coating methods, which coat already agglomerated particles, can be avoided. This lays the foundation for achieving monodispersity in subsequent surface modification. On the other hand, the in-situ formation of another metal oxide on the titanium dioxide surface facilitates the formation of a fully contacted interface, which is beneficial for adjusting the electronic structure of the heterojunction and promoting the transfer of charge carriers between them, thereby helping to regulate photocatalytic activity. By rationally designing and constructing a cross-gap (type I) heterojunction, since photogenerated electrons and holes are concentrated on the same semiconductor with a small band gap, electron-hole recombination will be promoted, the redox potential of charge carriers will decrease, and the photocatalytic activity of titanium dioxide will be significantly weakened.

[0041] 3) Improving the development efficiency of monodisperse functional heterojunctions by using machine learning methods, comprehensively considering the influence of different reaction variables on various properties of the product, thereby optimizing the screening of precursors and the design of reaction conditions, can save a lot of manpower and resources, and has inspirational significance for the design of other heterojunction functional materials.

[0042] 4) Because nano-TiO2-based particles exhibit monodispersity in organic solvents of varying polarities, they can be used as fillers in organic systems, such as organic coatings, organic encapsulating resins, and organic films. Furthermore, when used as fillers, nano-TiO2-based particles do not affect the transparency of the original system. These factors contribute to the product's wide range of applications and excellent performance. Attached Figure Description

[0043] Figure 1 This is a flowchart of the preparation model method for nano-TiO2-based particles that meet multiple objectives based on machine learning optimization, as described in an embodiment of the present invention.

[0044] Figure 2 The above is a heatmap of the correlation coefficients of the reaction variables in Example 1;

[0045] Figure 3 The above is a heat map showing the correlation coefficients of the product performance in Example 1.

[0046] Figure 4The following is a fitting diagram of the model's prediction results and experimental measurement results in Example 1, showing (a) lower limit of particle size, (b) photocatalytic activity, (c) ultraviolet shielding ability and (d) crystallinity.

[0047] Figure 5 The image shows a transmission electron microscope (TEM) image of the product under the predicted experimental conditions in Example 1.

[0048] Figure 6 This is a photograph of the dispersion of the product at a mass concentration of 5% under the predicted experimental conditions in Example 1.

[0049] Figure 7 This is a particle size distribution diagram of the product under the predicted experimental conditions in Example 1;

[0050] Figure 8 This is a comparison of the reaction rate parameters between the product under the predicted experimental conditions in Example 1 and the dye degradation experiment of untreated titanium dioxide.

[0051] Figure 9 This is a UV-Vis transmittance diagram of the composite coating of the product dispersion and fluorocarbon paint under the predicted experimental conditions in Example 1.

[0052] Figure 10 The X-ray diffraction patterns of the product under the predicted experimental conditions in Example 1 and untreated titanium dioxide are shown.

[0053] Figure 11 This is a high-resolution transmission electron microscope image of the titanium dioxide-based cerium oxide in Example 2. Detailed Implementation

[0054] The present invention will be further described and illustrated below with reference to specific embodiments. The embodiments described are merely examples of the content of this disclosure and do not limit the scope of the invention. The technical features of each embodiment in the present invention can be combined accordingly, provided that there is no mutual conflict.

[0055] Example 1

[0056] Reference Figure 1 This invention proposes a method for preparing a monodisperse, low-photocatalytic-activity nano-titanium dioxide-based ultraviolet shielding agent based on machine learning optimization, including but not limited to the following steps:

[0057] 1) Prepare a precursor solution for preparing nano-TiO2-based particles, synthesize wet aggregate precipitates of nano-TiO2-based particles by solvothermal method, centrifuge and wash the wet aggregate precipitates, add a surface modifier and disperse them in an organic solvent, and obtain nano-TiO2-based particles after post-treatment; then, prepare multiple groups of nano-TiO2-based particles with different properties by changing experimental variables; the post-treatment includes anti-solvent precipitation, centrifugation and freeze drying;

[0058] 2) Evaluate the product performance of multiple groups of nano-TiO2-based particles prepared in step 1). The experimental variables and corresponding product performance of multiple groups of nano-TiO2-based particles are combined into a dataset. Then, the experimental variables and product performance in the dataset are normalized and scaled to obtain the scaled dataset.

[0059] 3) Based on the scaled dataset, calculate the correlation coefficients between experimental variables and between product performances. Represent multiple experimental variables with correlation coefficients greater than 0.9 using one of the experimental variables to obtain dimensionality-reduced experimental variables; at the same time, represent multiple product performances with correlation coefficients greater than 0.9 using one of the product performances to obtain dimensionality-reduced product performances; thus, the dataset is dimensionality-reduced to obtain the dimensionality-reduced dataset.

[0060] 4) Construct an initialization model based on machine learning methods; train and optimize the initialization model by selecting experimental variables and product performance indicators of nano-TiO2-based particles to obtain a performance prediction model for nano-TiO2-based particles.

[0061] 5) Input the product performance of the required nano-TiO2-based particles into the nano-TiO2-based particle performance prediction model; the nano-TiO2-based particle performance prediction model outputs the corresponding dimension-reduced experimental variables; then, based on the dimension-reduced experimental variables output by the prediction model and the correlation coefficients of the experimental variables, perform dimension-upgrading to obtain all the experimental variables required for preparing nano-TiO2-based particles, and prepare nano-TiO2-based particles according to all the experimental variables.

[0062] In this embodiment, for step 1, a specific experimental scheme and operation instructions will be provided as follows:

[0063] Dissolve 8 mL of isopropyl titanate and 0.5 g of europium nitrate in 10 mL of ethanol solution and stir thoroughly until dissolved to form mixture A. Mix 0.5 mL of acetic acid, 0.6 mL of water and 5 mL of ethanol thoroughly to form mixture B. Under high-speed magnetic stirring, add mixture B evenly to mixture A in multiple feed streams to form a uniform and transparent precursor solution and let it mature for 30 min.

[0064] The above precursor solution was transferred to a 50 mL hydrothermal reactor, and the reaction temperature was 160 °C for 25 h.

[0065] The wet solid suspension was centrifuged at 8000 rpm, washed with ethanol, and centrifuged three times. 1 g of the treated wet solid was added to 10 mL of toluene, and 0.06 g of dodecyltrimethoxysilane was added. The mixture was stirred thoroughly at room temperature until a uniform and transparent titanium dioxide / europium oxide heterojunction dispersion was formed.

[0066] The above dispersion was subjected to dispersibility tests, particle size distribution tests, solvent compatibility tests, photocatalytic activity tests, UV shielding ability tests, and crystallinity tests.

[0067] In this embodiment, a specific data processing and model building method will be given for steps 2) to 4), as described below:

[0068] After conducting multiple sets of experiments under different conditions according to the above procedure, a database of product properties corresponding to reaction variables is formed.

[0069] Non-numerical parameters such as dispersibility, particle size distribution, and solvent compatibility in product performance are quantified: dispersibility is defined as {0,1}, where 0 represents that it cannot be monodispersed and 1 represents that it can be monodispersed; particle size distribution is broken down into the lower limit of particle size and the particle size distribution interval. For example, the particle size distribution of 8-35nm is broken down into a lower limit of 8nm and a particle size distribution interval of 27nm; for solvent compatibility types of 3 or fewer, the original data is retained, and those of more than 3 are uniformly regarded as being compatible with 4 solvents.

[0070] The numerical parameters of the product performance, such as photocatalytic activity, ultraviolet shielding ability, and crystallinity, are explained. The photocatalytic activity is quantified based on the reaction rate constant in the dye degradation experiment, with untreated titanium dioxide as the baseline and 1. The ultraviolet shielding ability is quantified based on the ultraviolet cutoff rate of the mixed film of titanium dioxide dispersion and fluorocarbon coating. The crystallinity is quantified based on the X-ray diffraction analysis results, with untreated TiO2 as the baseline and 1.

[0071] Unithermal coding is used for variables such as the type of metal salt and the type of surface modifier in the reaction to distinguish between different metal salts (those without metal salts are defined as NO) and surface modifiers.

[0072] Based on the above data processing, by normalization, the maximum and minimum values ​​of the variable data are used to scale data of different orders of magnitude to the [0,1] interval.

[0073] Calculate the correlation coefficients between the experimental variables and different variables in the product performance, and determine the linear correlation between the experimental variables and the product performance variables. The correlation coefficient between experimental variables x1 and x2 is as follows:

[0074]

[0075] Correlation coefficient between product properties y1 and y2:

[0076]

[0077] The correlation coefficient range is [-1, 1]. The closer the correlation coefficient is to 1 (or -1), the more significant the linear positive correlation (or negative correlation) between the variables. This indicates that one variable can be obtained by linear transformation of another variable. Therefore, selecting only one variable for modeling can reduce the complexity of the model and increase its computational efficiency.

[0078] Heatmaps are used to visualize the calculation results, providing a clear visual representation of the correlation coefficients between various variables. Heatmaps for experimental variables and product properties are shown below. Figure 2 , Figure 3 As shown, variables with a correlation coefficient greater than 0.9 are filtered out, such as... Figure 3 As shown, the correlation coefficients between the lower limit of particle size and monodispersity, particle size distribution spacing, and solvent compatibility type in the product properties are -0.93, 0.96, and 0.91, respectively. This indicates that there is a significant linear correlation between these three product properties and the lower limit of particle size. Therefore, the lower limit of particle size is used to represent these three product properties in order to reduce the dimensionality of database parameters.

[0079] The dimensionality-reduced data is randomly divided into a training set and a test set, with the training set accounting for 20% of the total data.

[0080] A random forest model is constructed and trained using a training set. Model parameters are adjusted, such as the optimization criterion, the number of decision trees, and the maximum depth of decision trees. In this embodiment, the optimized number of decision trees is 150, the maximum depth of decision trees is 50, and the variance of the model's prediction results is used as the optimization criterion.

[0081]

[0082] The model's prediction results are evaluated and validated using a test set. Evaluation metrics include mean squared error (MSE).

[0083]

[0084] Coefficient of determination (R) 2 ).

[0085]

[0086] The model's predictions fit the experimental measurements as follows: Figure 4 As shown, the model has a coefficient of determination of over 0.9 for the lower limit of particle size, photocatalytic activity, UV shielding ability and crystallinity, indicating that the model can accurately predict the performance of the product.

[0087] The range of experimental variables was selected, and the experimental variables corresponding to the product performance indicators were screened through iterative screening using a prediction model. In this example, the experimental variables screened after 10 iterations were: the volume ratio of titanium source to water (17.25), reaction temperature (185℃), reaction time (18h), type of metal salt (cerium), molar ratio of metal salt to titanium source (1.8%), and type of surface modifier (RS710). The corresponding predicted product performances were: lower limit of particle size (15nm), photocatalytic activity (0.04), ultraviolet shielding ability (0.93), and crystallinity (1.00).

[0088] Based on the experimental variables in the above prediction results, the dimensionality of the experimental variables in the prediction results was increased according to the correlation coefficients between the variables shown in the thermogram, resulting in all experimental variables for synthesizing nano-TiO2-based particles with the corresponding product properties. Nano-TiO2-based particles were prepared according to the experimental variables for experiments; the products were characterized by morphology, dispersibility testing, particle size testing, photocatalytic activity testing, and UV shielding ability testing. The morphology of the prepared products was characterized by transmission electron microscopy, such as... Figure 5 The particles are well dispersed, exhibiting a monodisperse state, and have a uniform particle size. The dispersion of the product can be visually represented by a photograph of the dispersion, such as... Figure 6 A 5% (w / w) dispersion was transparent and exhibited a pronounced Tyndall effect, maintaining its transparent dispersion state even after six months. The particle size distribution of the product was measured using a nanoparticle size analyzer. Figure 7 The average particle size of the product is around 15 nm, with a narrow particle size distribution. The photocatalytic activity of the product is evaluated by the reaction rate constant of dye degradation under ultraviolet light irradiation, such as... Figure 8 In degradation experiments with three dyes—Rhodamine B, Methylene Blue, and Reactive Red 120—the photocatalytic activity of the products under predicted conditions was substantially lower than that of untreated titanium dioxide. The UV shielding capability of the products was evaluated by the UV cutoff rate of the composite coating of the dispersion and fluorocarbon paint, such as... Figure 9 In the visible light region, the coating with the product added under the predicted conditions showed only a slight decrease in transmittance compared to the base coating, indicating that the prepared titanium dioxide-based UV-shielding filler was uniformly dispersed in the organic matrix and maintained high transparency. Furthermore, in the UV region, the coating with the product added under the predicted conditions showed a significant improvement in UV cutoff compared to the base coating, with UV cutoffs reaching 85.2% and 99.9% in the UVA and UVB regions, respectively, demonstrating the excellent UV shielding capability of the prepared titanium dioxide-based UV-shielding filler. The crystallinity of the product was illustrated by X-ray diffraction patterns, such as... Figure 10The crystallinity of the product under the predicted conditions did not change compared to untreated titanium dioxide, indicating that it maintained good crystallinity. This demonstrates that using machine learning to optimize monodisperse titanium dioxide-based UV-shielding fillers with low photocatalytic activity is a reliable method.

[0089] Example 2

[0090] To illustrate the structure of the prepared titanium dioxide-based UV-shielding filler, a higher molar ratio of cerium salt / titanium source was used, with a cerium nitrate / tetrabutyl titanate molar ratio of 10%. The remaining reaction conditions remained consistent with steps 1-3 in Example 1. The microstructure of the product was characterized by transmission electron microscopy, as shown below. Figure 11 The diagram shows that titanium dioxide forms a heterojunction with amorphous cerium oxide, indicating that a titanium dioxide-based cerium oxide heterojunction has been successfully prepared. By adjusting the electronic structure of the two through the heterojunction interface, the recombination of charge carriers is promoted to reduce photocatalytic activity.

[0091] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. Those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for preparing multi-objective TiO2-based nanoparticles optimized by machine learning, characterized in that, Includes the following steps: 1) Prepare a precursor solution for preparing nano-TiO2-based particles, synthesize wet aggregate precipitates of nano-TiO2-based particles by solvothermal method, centrifuge and wash the wet aggregate precipitates, add a surface modifier and disperse them in an organic solvent, and obtain nano-TiO2-based particles after post-treatment; then, prepare multiple groups of nano-TiO2-based particles with different properties by changing experimental variables; the post-treatment includes anti-solvent precipitation, centrifugation and freeze drying; 2) Evaluate the product performance of multiple groups of nano-TiO2-based particles prepared in step 1). The experimental variables and corresponding product performance of multiple groups of nano-TiO2-based particles are combined into a dataset. Then, the experimental variables and product performance in the dataset are normalized and scaled to obtain the scaled dataset. 3) Based on the scaled dataset, calculate the correlation coefficients between experimental variables and between product performances. Represent multiple experimental variables with correlation coefficients greater than 0.9 using one of the experimental variables to obtain dimensionality-reduced experimental variables; at the same time, represent multiple product performances with correlation coefficients greater than 0.9 using one of the product performances to obtain dimensionality-reduced product performances; thus, the dataset is dimensionality-reduced to obtain the dimensionality-reduced dataset. 4) Construct an initialization model based on machine learning methods; train and optimize the initialization model using the dimensionality-reduced dataset to obtain a performance prediction model for nano-TiO2-based particles; 5) Input the product performance of the required nano-TiO2-based particles into the nano-TiO2-based particle performance prediction model; the nano-TiO2-based particle performance prediction model outputs the corresponding dimension-reduced experimental variables; then, based on the dimension-reduced experimental variables output by the prediction model and the correlation coefficients of the experimental variables, perform dimension-upgrading to obtain all the experimental variables required for preparing nano-TiO2-based particles, and prepare nano-TiO2-based particles according to all the experimental variables.

2. The preparation method according to claim 1, characterized in that: The precursor solution mentioned in step 1) includes a titanium source, a metal salt, water, a hydrolysis inhibitor, and an organic solvent; The titanium source includes one or more of tetrabutyl titanate, tetraethyl titanate, isopropyl titanate, titanium tetrachloride, titanium oxysulfate, and titanium acetylacetone. The metal salts include sodium chloride, copper chloride, manganese chloride, tin chloride, lanthanum chloride, ferric nitrate, cerium nitrate, europium nitrate, aluminum nitrate, zirconium nitrate, lanthanum nitrate, gallium nitrate, terbium nitrate, and samarium nitrate; The hydrolysis inhibitors include one or more of nitric acid, hydrochloric acid, sulfuric acid, phosphoric acid, formic acid, acetic acid, salicylic acid, lactic acid, tartaric acid, and citric acid; The organic solvent includes one or more of methanol, ethanol, isopropanol, n-butanol, tert-butanol, octanol, isocetyl alcohol, propylene glycol, benzyl alcohol, acetone, butanone, propylene glycol methyl ether acetate, tetrahydrofuran, dimethyl sulfoxide, chloroform, ethyl acetate, butyl acetate, benzene, toluene, xylene, heptane, n-hexane, cyclohexane, and petroleum ether.

3. The preparation method according to claim 2, characterized in that, The experimental variables include the volume ratio of titanium source to water, reaction temperature, reaction time, type of metal salt, molar ratio of metal salt to titanium source, and type of surface modifier.

4. The preparation method according to claim 1, characterized in that: The precursor solution in step 1) is homogenized by multi-stream feeding and rapid stirring, and the homogenization time is controlled between 0.05 and 60 min; the nano-TiO2-based particles are a composite of titanium dioxide and other metal oxides; the solvothermal reaction temperature is 80-250℃, the solvothermal reaction pressure is 0.01-10 MPa, and the solvothermal reaction time is 1-72 h.

5. The preparation method according to claim 1, characterized in that... The surface modifier mentioned in step 1) includes one or more of long-chain organic acids, phosphate esters, titanates, and silane coupling agents; The long-chain organic acid is a carboxylic acid having 12-24 carbon atoms, including lauric acid, myristic acid, palmitic acid, stearic acid, arachidic acid, oleic acid, and linoleic acid. The phosphate esters include one or more of the following: 4-hydroxybutyl phosphate dihydrogen ester, 2-ethylhexyl phosphate dihydrogen ester, 9-octadecene-1-ol phosphate ester, isooctyl phosphate ester, dodecyl phosphate ester, alkyl diphenyl phosphate ester, tetradecyl phosphate ester, lauryl polyoxyethylene ether phosphate ester, and tridecyl polyoxyethylene ether phosphate ester. The titanate comprises one or more of the following: isopropyl tris(dodecylbenzenesulfonyl) titanate, isopropyl triisostearate titanate, bis(acetylacetonyl) diisopropyl titanate, isopropyl tris(dioctyl pyrophosphate) titanate, isopropyl tris(dioctyl phosphate) titanate, isopropyl dioleoyl phosphate (dioctyl phosphate) titanate, di(octylphenol polyoxyethylene ether) phosphate, and tetraisopropyl di(dioctyl phosphite) titanate. The silane coupling agent includes one or more of the following: vinyltrimethoxysilane, vinyltriethoxysilane, dodecyltrimethoxysilane, dodecyltriethoxysilane, hexadecyltrimethoxysilane, hexadecyltriethoxysilane, phenyltrimethoxysilane, γ-aminopropyltrimethoxysilane, 3-glycidyl etheroxypropyltrimethoxysilane, γ-methacryloyloxypropyltrimethoxysilane, N-β-aminoethyl-γ-aminopropyltriethoxysilane, methoxytriethylene glycol etherylpropyltrimethoxysilane, 11-mercaptoundecyltrimethoxysilane, 3-mercaptopropyltrimethoxysilane, and diethylphosphorylethyltriethoxysilane. The organic solvent includes one or more of methanol, ethanol, isopropanol, n-butanol, tert-butanol, octanol, isocetyl alcohol, propylene glycol, benzyl alcohol, acetone, butanone, propylene glycol methyl ether acetate, tetrahydrofuran, dimethyl sulfoxide, dichloromethane, chloroform, ethyl acetate, butyl acetate, benzene, toluene, xylene, heptane, n-hexane, cyclohexane, and petroleum ether.

6. The preparation method according to claim 1, characterized in that, In step 2), the product performance of nano-TiO2-based particles is evaluated through dispersibility testing, particle size distribution testing, solvent compatibility testing, photocatalytic activity testing, UV shielding capability testing, and crystallinity testing. The experimental variables and product performance in the dataset are normalized by using the maximum and minimum values ​​in the dataset to scale data of different orders of magnitude to the [0,1] interval.

7. The preparation method according to claim 1, characterized in that; In step 3), the correlation coefficients between experimental variables for: Correlation coefficient between product properties Where x1 and x2 are two different experimental variables; z1 and z2 are two different product properties; and N is the amount of data.

8. The preparation method according to claim 1, characterized in that, In step 4), the initialization model based on the machine learning method is a random forest model. The dimensionality-reduced dataset is randomly divided into a training set and a test set. The random forest model is trained and optimized using the training set, and the performance prediction model of the trained nano-TiO2-based particles is evaluated and verified using the test set.

9. The preparation method according to claim 8, characterized in that, In step 4), the evaluation indices for the performance prediction model of nano-TiO2-based particles include mean square error (MSE) and coefficient of determination (R²). 2 : Where N is the amount of data. For the prediction result, y i This represents the actual results for the test set.

10. The application of nano-TiO2-based particles prepared by the method described in claim 1 as a UV-absorbing filler in a transparent UV-protective coating, characterized in that, The nano-TiO2-based particles dispersed in toluene are thoroughly mixed with the fluorocarbon coating main agent, and then the fluorocarbon coating curing agent is added and mixed evenly again. This mixture is then used as a transparent UV-protective coating for scraping, spin coating, and spraying.

Images