Prediction method for leaching resistance of perfume glass bottle
By combining chemical kinetic models and nanosensors with machine learning methods, the problems of long cycle, high cost and low precision in predicting the anti-leaching performance of perfume glass bottles have been solved. Accurate and rapid predictions of different glass materials and perfume formulas have been achieved, adapting to stability under long-term storage conditions and meeting the perfume industry's demand for packaging safety.
Patent Information
- Application Number
- CN202510971538.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-15
- Publication Date
- 2025-09-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing technologies for evaluating the anti-leaching performance of perfume glass bottles have the disadvantages of long cycles, high costs, limited parameter coverage, and poor model generalization ability. They are difficult to adapt to the combination of different glass materials and perfume formulas, and cannot ensure the reliability of prediction results under long-term storage conditions.
By integrating chemical kinetics, nanosensing and machine learning, a reaction kinetics model was established by preparing standard simulation liquid, and nanosensing detection was performed using modified gold nanoparticles. The prediction model was trained using the random forest algorithm, combined with multiple sets of parallel experiments for verification, and the model parameters were optimized to improve the prediction accuracy.
It achieves accurate and rapid prediction of different glass formulas and perfume ingredients, reduces interference from external factors, ensures the stability of the model under long-term storage conditions, significantly reduces experimental materials and time costs, and meets the needs of rapid iteration and efficient R&D.
Smart Images
Figure CN120703013A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of glass product quality detection, and in particular to a method for predicting the anti-leaching performance of a perfume glass bottle. Background Art
[0002] The leaching resistance of perfume glass bottles directly impacts the stability and safety of perfumes and is a key technical indicator in the cosmetics packaging industry. Existing technologies primarily rely on long-term immersion testing to assess glass leaching resistance. This involves placing glass bottle samples in a simulated perfume environment and continuously monitoring ion leaching concentrations for months or even years. This method has significant drawbacks, including long cycle times, high costs, and limited parameter coverage.
[0003] Traditional prediction methods are often based on empirical formulas or single kinetic models, which only reflect the effects of temperature and time on leaching, ignoring the synergistic effects of glass composition, surface treatment processes, and the complex components of perfumes. For example, perfumes containing aldehydes easily react with sodium ions in glass, accelerating the leaching process. Existing models fail to incorporate these chemical interaction parameters, resulting in prediction errors often exceeding 20%.
[0004] The application of nano-detection technology has limitations: existing ion detection mostly relies on inductively coupled plasma mass spectrometry, which has high accuracy but requires offline sampling and cannot monitor interfacial reactions in real time; although gold nanoparticle sensors have real-time performance, they lack surface modification optimization, have insufficient selective recognition ability for sodium ions, are easily interfered by organic acids and metal ions in perfumes, and have large detection deviations.
[0005] The application of machine learning in material property prediction is still immature. Existing models often use a single algorithm, with input features consisting of only 3-5 parameters and less than 300 training data sets. This results in poor model generalization and makes it difficult to adapt to different combinations of glass materials and fragrance formulas. Furthermore, model validation lacks systematic multi-parameter control experiments, making it impossible to ensure the reliability of predictions under long-term storage conditions.
[0006] As the perfume industry places increasing demands on packaging safety, there is an urgent need to establish a multidimensional prediction method that integrates chemical kinetics, nanosensing, and machine learning to accurately and rapidly predict the anti-leaching performance under different glass formulations, surface processes, perfume ingredients, and storage conditions, providing a scientific basis for glass bottle material optimization and perfume storage solution design. Summary of the Invention
[0007] (1) Technical problems solved
[0008] In view of the shortcomings of the prior art, the present invention provides a method for predicting the anti-leaching performance of perfume glass bottles.
[0009] (2) Technical solution
[0010] A method for predicting the anti-leaching performance of a perfume glass bottle comprises the following steps:
[0011] S1: Prepare a standard simulating solution by mixing 5-15% by mass of ethanol, 0.1-0.5% by mass of phenoxyethanol, 0.01-0.1% by mass of citronellal, 0.05-0.2% by mass of benzoic acid, and the remainder by mass of deionized water, and adjust the pH to 3.0-5.5 to form a simulated perfume system;
[0012] S2: Build a reaction kinetics model: Construct an ion migration model at the glass bottle-perfume interface based on the following reaction equation:
[0013]
[0014] Among them, the reaction rate constant is:
[0015]
[0016] A1: Pre-exponential factor of sodium ion leaching, value is 1.2×10 -4 s -1 ;
[0017] A2: Pre-exponential factor of calcium ion leaching, value is 8.5×10 -5 s -1 ;
[0018] E a1 : Sodium ion leaching activation energy, 55 kJ / mol;
[0019] E a2 : Activation energy of calcium ion leaching, 62 kJ / mol;
[0020] R: gas constant, 8.314 J / (mol·K);
[0021] T: absolute temperature, unit K;
[0022] S3 nanosensor detection: Gold nanoparticles modified with 3-aminopropyltriethoxysilane were dispersed in a simulated liquid, and the absorbance change at a wavelength of 520-540 nm was monitored by UV-visible spectroscopy. A linear relationship between absorbance and sodium ion concentration was established. ΔA = 0.025 [Na + ]+0.003, where △A is the absorbance change value, [Na + ] is the sodium ion concentration, unit: mg / L;
[0023] S4 machine learning prediction: Input glass composition, surface treatment process parameters, and simulated liquid condition characteristic variables, train the prediction model through the random forest algorithm, and output the predicted value of sodium ion leaching concentration for 50-365 days.
[0024] Preferably, the simulation liquid in S1 further contains 0.01-0.05% of 2-bromo-2-nitropropane-1,3-diol and 0.005-0.02% of ethyl salicylate to enhance the simulation of preservatives and fragrance ingredients.
[0025] 3. The method for predicting the anti-leaching performance of perfume glass bottles according to claim 1, wherein the reaction kinetics model in S2 further includes a diffusion term of the glass surface coating:
[0026]
[0027] C: concentration of sodium ions in the coating, unit mol / m 3 ;
[0028] t: time, unit s;
[0029] D: Diffusion coefficient, value is 2.5×10 -12 m 2 / s;
[0030] x: Coordinate in the coating thickness direction, unit: m;
[0031] C H+ (x=0): hydrogen ion concentration at the interface between the coating and the solution, in mol / m 3 ;
[0032] L: coating thickness, ranging from 100-200nm.
[0033] Preferably, the nanosensing detection in S3 specifically comprises: mixing a gold nanoparticle solution with a concentration of 0.1-0.5 mg / mL and a simulation solution at a volume ratio of 1:5, incubating at 25-60° C. for 1-4 hours, and then detecting the absorbance.
[0034] Preferably, the machine learning model training in S4 uses more than 500 sets of experimental data, and the characteristic variables include: glass composition, physical properties, surface parameters, and simulated liquid conditions.
[0035] Preferably, the method further includes an S5 verification step: selecting 100 groups of samples with different parameter combinations for actual immersion experiments, and using inductively coupled plasma mass spectrometry (ICP-MS) to detect the actual leaching concentration; comparing the predicted value with the measured value, and calculating the mean absolute error (MAE); when the MAE is greater than 10%, adding 200 groups of training data and adjusting the model hyperparameters, with the number of decision trees being 200-300 and the maximum depth being 10-12 layers, and retraining until the MAE is ≤8%.
[0036] Preferably, the ultraviolet-visible spectrum detection in S3 uses a fiber optic probe with a dual-beam optical path design, a detection wavelength range of 500-600nm, a scanning rate of 100nm / min, a detection sensitivity of 0.001absorbance unit, a data collection interval of 10s, and three parallel detections are performed to take the average value to reduce the error.
[0037] Preferably, the random forest algorithm parameter settings in S4 are: the number of decision trees is 100-500, the maximum depth is 8-15 layers, the minimum number of samples for node splitting is 5-10, the minimum number of samples for leaf nodes is 1-3, the feature sampling ratio is 0.6-0.8, the Gini coefficient is used as the splitting criterion, the parameters are optimized through 5-fold cross validation, and the model training iteration number is 500-1000 times until the loss function converges.
[0038] Preferably, the preparation of the simulated liquid in S1 is carried out in a nitrogen-protected glove box, and the dissolved oxygen content is monitored online by fluorescence and controlled at ≤0.5 mg / L; the mixing order is: first, deionized water is heated to 40°C, benzoic acid, phenoxyethanol, and citronellal are added in sequence, stirred until completely dissolved, and then ethanol is added. Finally, the pH is adjusted to 3.0-5.5 with 0.1 mol / L hydrochloric acid or sodium hydroxide, and the mixture is allowed to stand for 24 hours before use.
[0039] Preferably, the prediction method is suitable for predicting the anti-leaching performance of perfume glass bottles made of different materials such as borosilicate glass, soda-lime silicate glass, and aluminosilicate glass. The prediction results can be used to guide the optimization of glass bottle formula and the adjustment of surface treatment process.
[0040] (3) Beneficial technical effects
[0041] Compared with the existing technology, the beneficial effects of the present invention are:
[0042] 1. This method incorporates multiple parameters such as glass composition, surface treatment process, perfume formula ingredients, and storage environment conditions into the analysis system. In particular, the introduction of detailed parameters such as trace components in glass and coating porosity enables the prediction model to truly reflect the complex scenarios of multiple factors in actual applications, avoiding prediction bias caused by the lack of key parameters.
[0043] 2. By modifying gold nanoparticles with silane, the selective recognition ability of target ions is enhanced, effectively reducing the interference of organic acids, fragrances and other ingredients in perfumes; combined with ultraviolet-visible spectroscopy detection with a dual-beam optical path design, the influence of external factors such as ambient light and instrument fluctuations is further reduced, making the real-time monitoring data closer to the actual leaching state, and solving the error problem caused by the sampling process in traditional offline detection.
[0044] 3. By expanding the size of the training dataset and refining the classification of feature variables, the model can adapt to the combination scenarios of different types of glass and different formulas of perfumes. The comprehensive verification process ensures the predictive stability of the model under long-term storage conditions through multiple sets of parallel experiments and parameter adjustment mechanisms. It can be directly used to guide formula optimization and surface process improvement in glass bottle production, providing a reliable reference for actual production without relying on long-term immersion experiments.
[0045] 4. By combining kinetic models with machine learning, this method can complete the prediction of long-term performance in a short period of time, significantly reducing experimental material consumption and time costs, while avoiding product development delays caused by long waiting cycles, and meeting the industry's demand for rapid iteration and efficient R&D. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 This is a flow chart of a method for predicting the anti-leaching performance of a perfume glass bottle proposed by the present invention;
[0047] Figure 2 It is a histogram comparison of the errors of different prediction methods;
[0048] Figure 3 1. It is a broken line comparison chart of the predicted period of each embodiment and the comparative example;
[0049] Figure 4 It is a comparison chart of the mean absolute error between the embodiments and the comparative examples. DETAILED DESCRIPTION
[0050] according to Figures 1 to 4 , the specific implementation methods of the present invention are as follows:
[0051] Example 1: Prediction of Soda-Lime Silicate Glass Leaching Resistance
[0052] Glass sample parameters
[0053] Soda-lime silicate glass is selected, and its composition is 70% silicon dioxide, 15% sodium oxide, 8% calcium oxide, 2% aluminum oxide, and 5% magnesium oxide, all by mass. The glass density is 2.5g / cm 3 , the thermal expansion coefficient is 10×10 -6 / ℃. The glass surface is treated with a silica coating with a coating thickness of 150nm, a porosity of 0.5%, and a roughness Ra of 0.3 microns.
[0054] Simulation fluid formula
[0055] A simulant solution was prepared, by weight, containing 10% ethanol, 0.3% phenoxyethanol, 0.05% citronellal, 0.1% benzoic acid, 0.03% 2-bromo-2-nitropropane-1,3-diol, and 0.01% ethyl salicylate, with the balance being deionized water. The preparation was performed in a nitrogen glove box with an oxygen concentration of 0.08%. After stirring and dissolving the components at 40°C, the pH was adjusted to 4.5 with 0.1 mol / L hydrochloric acid, maintaining a dissolved oxygen content of 0.4 mg / L.
[0056] Prediction Step
[0057] S1: After the simulated liquid is prepared, let it stand for 24 hours. Take 50 ml and seal it in a brown bottle for later use.
[0058] S2: Establish a kinetic model and substitute the parameters for calculation. Set the temperature T = 313 Kelvin, i.e. 40°C, and calculate the reaction rate constant k1 = 1.2×10 -4 ×exp(-55000 / (8.314×313))=3.2×10 -6 / s; the diffusion coefficient D in the diffusion term is 2.5×10 -12 square meters / s, coating thickness L=150nm.
[0059] S3: Gold nanoparticles with a particle size of 20 nm were modified with 3-aminopropyltriethoxysilane to prepare a 0.3 mg / mL solution. The solution was mixed with the simulated liquid at a volume ratio of 1:5 and incubated at 50°C for 2 hours. The absorbance change was detected using a fiber optic probe with a probe spot size of 0.8 mm at a wavelength of 530 nm. The absorbance change formula is ΔA = 0.025 [Na + ]+0.003.
[0060] S4: 12 feature variables were input, including glass composition, coating parameters, and simulation fluid conditions. The random forest algorithm was used to set 200 decision trees, with a maximum depth of 10 layers and a feature sampling ratio of 0.7. 5-fold cross-validation was used to train the model.
[0061] S5: Three groups of parallel samples were taken for a 180-day actual immersion experiment. The actual leaching concentration was detected by ICP-MS and compared with the predicted value. The average absolute error (MAE) was 7.2%, which was less than or equal to 8%, indicating that the model was effective.
[0062] Example 2: Prediction of anti-leaching performance of borosilicate glass
[0063] Glass sample parameters
[0064] Borosilicate glass is selected, and its composition is 75% silicon dioxide, 12% sodium oxide, 8% boron oxide, 3% aluminum oxide, and 2% calcium oxide, all by mass. The density of the glass is 2.45 grams per cubic centimeter, and the thermal expansion coefficient is 6×10 -6 / ℃. The glass surface is uncoated and has a roughness Ra of 0.2 microns.
[0065] Simulation fluid formula
[0066] A simulant solution was prepared by weight containing 15% ethanol, 0.5% phenoxyethanol, 0.1% citronellal, 0.2% benzoic acid, 0.05% 2-bromo-2-nitropropane-1,3-diol, and 0.02% ethyl salicylate, with the balance being deionized water. The preparation was performed in a nitrogen glove box with an oxygen concentration of 0.06%. After dissolving the components by stirring at 45°C for 30 minutes, the pH was adjusted to 3.5 with 0.1 mol / L hydrochloric acid. High-purity nitrogen was then passed through the solution for 30 minutes to deoxygenate it, maintaining the dissolved oxygen content at 0.3 mg / L. The solution was then sealed and stored in an inert gas-protected storage bottle.
[0067] Prediction Step
[0068] S1: After the preparation of the simulated liquid, let it stand for 24 hours at a temperature of 25℃±1℃. Transfer 50 ml to a brown glass sample bottle, seal it and store it in a refrigerator at 4℃ as the backup solution for testing.
[0069] S2: Establish a kinetic model and substitute the parameters for calculation. Set the temperature T = 333 Kelvin (60°C). Based on the kinetic characteristics of borosilicate glass leaching, the reaction activation energy Ea = 48000 J / mol, and calculate the reaction rate constant k1 = 1.5×10 -4 ×exp(-48000 / (8.314×333))=4.8×10 -6 / s; because there is no coating on the surface, the diffusion resistance can be ignored, so the diffusion term is omitted, and the model is the first-order reaction kinetic equation of the change of leaching concentration with time: C(t)=C0×(1-exp(-k1t)).
[0070] S3: Gold nanoparticles with a particle size of 20 nm were modified with 3-aminopropyltriethoxysilane at a concentration of 5%. After reacting in an ethanol / water mixture at 30°C for 2 hours, a 0.5 mg / mL solution was prepared. The solution was mixed with the simulated liquid at a volume ratio of 1:5, with a total system of 3 mL. The solution was incubated in a 50°C water bath for 2 hours at an oscillation rate of 150 rpm. A fiber optic probe was used for detection, with a probe spot size of 0.8 mm. The absorbance change was measured at a wavelength of 525 nm. The absorbance change formula is ΔA = 0.022 [Na + ]+0.005[B 3+ ]+0.001.
[0071] S4: Input 10 feature variables, including glass composition (5 items such as silicon dioxide, sodium oxide, and boron oxide), simulated liquid conditions (3 items such as pH, dissolved oxygen, and ethanol content), temperature, and time. The random forest algorithm sets 300 decision trees, a maximum depth of 12 layers, a minimum number of sample splits of 10, a minimum number of sample leaf nodes of 4, a feature sampling ratio of 0.75, and uses 5-fold cross-validation to train the model. The training set:test set = 8:2, and the model fitting accuracy R 2 =0.96.
[0072] S5: Three parallel groups of glass samples, each containing five identical specimens (2 cm × 2 cm × 0.3 cm), were ultrasonically cleaned for 15 minutes and then subjected to 50-day and 365-day immersion tests in a 60°C constant-temperature oscillating chamber at 60 rpm. One milliliter of sample was removed at each time point and replenished with an equal amount of fresh simulant to maintain a constant volume. Actual leached concentrations were determined using ICP-MS with an internal standard of 115In and a detection limit of 0.01 μg / L. Compared to the predicted values, the 50-day predictions deviated by 5.8%, and the 365-day predictions deviated by 7.6%. The resulting mean absolute error (MAE) was 6.8%, which is less than or equal to 8%, indicating that the model is suitable for uncoated glass.
[0073] Example 3: Prediction of the anti-leaching performance of high-alumina silicate glass
[0074] Glass sample parameters
[0075] High-alumina silicate glass is selected, and its composition is 68% silicon dioxide, 14% sodium oxide, 5% aluminum oxide, 5% calcium oxide, and 8% magnesium oxide, all by mass. The density of the glass is 2.55 grams per cubic centimeter, measured by the water displacement method with an accuracy of ±0.01g / cm 3 ;The thermal expansion coefficient is 9×10 -6 / °C, measured by thermomechanical analyzer in the range of 30-300°C. A silica coating was deposited on the glass surface by plasma-enhanced chemical vapor deposition. The coating thickness was 200 nm, measured by ellipsometer, averaging 5 points. The porosity was 0.3%, measured by mercury intrusion porosimetry. The roughness Ra was 0.25 microns, measured by white light interferometry, scanning an area of 10 μm × 10 μm. The adhesion between the coating and the substrate was 12 N / cm, tested by the cross-hatch method.
[0076] Simulation fluid formula
[0077] The simulated liquid was prepared by mass fraction, containing 5% ethanol, 0.1% phenoxyethanol, 0.01% citronellal, 0.05% benzoic acid, and the balance being deionized water, with a resistivity of 18.2 MΩ·cm. The preparation process was carried out in a nitrogen glove box with an oxygen content of 0.07%. Benzoic acid was first dissolved in deionized water and stirred at 40°C for 15 minutes. Then, ethanol and other ingredients were added and stirred for 20 minutes until completely dissolved. The pH was adjusted to 5.5 with 0.1 mol / L nitric acid, and nitrogen was passed through for 20 minutes to control the dissolved oxygen content in the simulated liquid to 0.5 mg / L. The solution was stored at 4°C in the dark until ready for use.
[0078] Prediction Step
[0079] S1: After the preparation of the simulated liquid, let it stand for 24 hours at a temperature of 25°C. Take 50 ml of it and put it into a brown bottle. Seal it with a polytetrafluoroethylene cap. Label it and store it in a refrigerator at 4°C to prevent the volatilization of the fragrance components.
[0080] S2: Establish a kinetic model and substitute the parameters for calculation. Set the temperature T = 313 Kelvin (40°C), the activation energy of high aluminosilicate glass leaching Ea = 52000 J / mol, and calculate the reaction rate constant k1 = 1.3×10 -4 ×exp(-52000 / (8.314×313))=3.5×10 -6 / s; diffusion coefficient D in the diffusion term = 2.2×10 -12 square meters / s, the presence of Al2O3 reduces the ion diffusion rate, the coating thickness L = 200nm = 2×10 -7 m, diffusion resistance coefficient F = L / D = 9.1×10 4 s / m.
[0081] S3: Gold nanoparticles with a particle size of 20 nm were modified with 3-aminopropyltriethoxysilane to prepare a 0.3 mg / mL solution. The dispersion medium was a simulated liquid blank, mixed with the simulated liquid at a volume ratio of 1:5, and the total system was 3 mL. The solution was incubated at 50°C for 4 hours with an oscillation rate of 150 rpm to ensure sufficient reaction at low ethanol concentrations. A fiber optic probe was used for detection, with a probe spot size of 0.8 mm. The absorbance change was measured at a wavelength of 530 nm. The absorbance change formula is ΔA = 0.023 [Na + ]+0.018[Al3 + ]+0.002.
[0082] S4: 12 feature variables were input, including glass composition (silicon dioxide, sodium oxide, aluminum oxide, calcium oxide, magnesium oxide content), coating parameters (thickness 200nm, porosity 0.3%, roughness Ra 0.25μm), simulated liquid conditions (pH 5.5, dissolved oxygen 0.5mg / L, ethanol 5%), temperature, and time. The random forest algorithm set 100 decision trees, a maximum depth of 10 layers, a minimum number of sample splits of 10, a minimum number of sample leaf nodes of 4, and a feature sampling ratio of 0.6. The Al2O3 content weight was specifically included, with a weight coefficient of 1.2, which is higher than other components. The model was trained using 5-fold cross-validation, and the model training set R 2 =0.95.
[0083] S5: Three groups of parallel glass samples, each with four samples of the same specifications, were taken for actual immersion experiments. Under constant temperature conditions of 40°C and an oscillation rate of 120 r / min, samples were taken at 90 days and 180 days, respectively. The actual leaching concentration was detected by ICP-MS. The matrix effect was eliminated through internal standard correction. Compared with the predicted value, the mean absolute error (MAE) was 7.5%, which is less than or equal to 8%, indicating that the model can capture the inhibitory effect of high aluminum components on leaching.
[0084] Example 4: Prediction of Anti-leaching Performance of Glass by Optimizing Surface Coating
[0085] Glass sample parameters
[0086] Soda-lime silicate glass is selected, and its composition is 70% silicon dioxide, 15% sodium oxide, 8% calcium oxide, 2% aluminum oxide, and 5% magnesium oxide, all by mass. The density of the glass is 2.5 grams per cubic centimeter, and the thermal expansion coefficient is 10×10 -6 / °C. A silica coating was deposited on the glass surface using plasma-enhanced chemical vapor deposition. The coating thickness was 100 nm, produced by magnetron sputtering, with a uniformity deviation of ≤5%. The porosity was 0.1%, optimized by controlling the sputtering power. The roughness Ra was 0.1 micron, as measured by atomic force microscopy. The coating hardness was 8 GPa, as measured by nanometer indentation, indicating improved wear resistance.
[0087] Simulation fluid formula
[0088] A simulant solution was prepared, by weight, containing 10% ethanol, 0.3% phenoxyethanol, 0.05% citronellal, 0.1% benzoic acid, 0.03% 2-bromo-2-nitropropane-1,3-diol, and 0.01% ethyl salicylate, with the balance being deionized water. The preparation was performed in a nitrogen glove box with an oxygen concentration of 0.08%. After stirring and dissolving the components at 40°C, the pH was adjusted to 4.0 with 0.1 mol / L hydrochloric acid, maintaining a dissolved oxygen content of 0.4 mg / L.
[0089] Prediction Step
[0090] S1: After the preparation of the simulated liquid, let it stand for 24 hours at a temperature of 25℃±1℃. Take 50 ml and store it in a sealed brown bottle to ensure that the pH is stable at 4.0±0.1. This is used as the standby liquid for testing.
[0091] S2: Establish a kinetic model and substitute the parameters for calculation. Set the temperature T = 313 Kelvin, i.e. 40°C, and calculate the reaction rate constant k1 = 1.2×10 -4 ×exp(-55000 / (8.314×313))=3.2×10 -6 / s; diffusion coefficient D in the diffusion term = 2.2×10 -12 m2 / s, the ion diffusion rate is reduced due to coating optimization, and the coating thickness L = 100nm = 1×10 -7 m, diffusion resistance coefficient F = L / D = 9.1×10 4 s / m, ion penetration resistance R = L / (D×(1-porosity)) = 1×10 -7 / (2.0×10 -12 × 0.999) = 5.0 × 10 4 s / m.
[0092] S3: Gold nanoparticles with a particle size of 20 nm were modified with 3-aminopropyltriethoxysilane to prepare a 0.3 mg / mL solution, mixed with the simulated solution at a volume ratio of 1:5, and incubated at 50°C for 2 hours. Since the leached ion concentration was lower after the coating optimization, the absorbance change formula was corrected to ΔA = 0.028 [Na + ]+0.002, using a fiber optic probe with a probe spot size of 0.8mm and absorbance changes at a wavelength of 530nm.
[0093] S4: Input 12 feature variables, including glass composition (5 items such as silicon dioxide, sodium oxide, and calcium oxide), coating parameters (thickness 100nm, porosity 0.1%, roughness Ra0.1μm), simulated liquid conditions (pH 4.0, dissolved oxygen 0.4mg / L, ethanol 10%, etc.), temperature, and time. In the random forest algorithm, the weight coefficients of coating thickness, porosity, and roughness are 1.5, 1.8, and 1.3 respectively. Set 500 decision trees, the maximum depth is 12 layers, the minimum number of sample splits is 10, the minimum number of sample leaf nodes is 4, the feature sampling ratio is 0.75, and a 5-fold cross-validation is used to train the model. The model test set R 2 =0.97.
[0094] S5: Three groups of parallel glass samples were taken, with five samples of the same specifications in each group, measuring 2cm×2cm×0.3cm. A 180-day actual immersion experiment was conducted. At 40°C and pH=4.0, ICP-MS was used to detect the actual leaching concentration. Compared with the predicted value, the maximum deviation was 8.1%, and the mean absolute error (MAE) was 6.5%, which is less than or equal to 8%, indicating that the model can accurately predict the inhibitory effect of coating optimization on leaching.
[0095] Comparative example: Traditional empirical formula prediction method
[0096] Method Parameters
[0097] Use the industry's common empirical formula: Leaching concentration = k × t 0 · 5 ×exp(-Ea / (RT)), only three parameters are input: sodium oxide content 15%, temperature 40°C, and time 180 days.
[0098] Prediction results
[0099] This method does not take into account other glass components, coating parameters, simulated liquid pH, and fragrance components, resulting in a 22% deviation between the predicted value and the actual leaching concentration. It is also unable to distinguish between different glass types (such as borosilicate and soda-lime glass), and the prediction accuracy is far lower than the method of the present invention.
[0100] The prediction errors of the examples and comparative examples are shown in the following table:
[0101] Table 1
[0102]
[0103] The MAE of the methods of the present invention are all between 6.5% and 7.5%, significantly lower than the 22% of the comparative example. Furthermore, the prediction cycle is only 4-5.5 hours, much shorter than the 30 days (720 hours) of the comparative example. This fully demonstrates the significant advantages of the present invention in prediction accuracy and efficiency, and solves the problems of long cycles and large errors in traditional methods. The errors of different prediction methods are shown in the following table:
[0104] Table 2
[0105] project Example 1 Example 2 Example 3 Example 4 Comparative Example 50-day forecast error (%) 6.8 6.5 7.3 6.2 21.5 180-day forecast error (%) 7.2 6.8 7.5 6.5 22.0 365-day forecast error (%) 7.5 7.0 7.8 6.9 23.2
[0106] The prediction error of the method of the present invention is stable at 6.2%-7.8% at various time lengths, while the error of the comparative example increases slightly with time, ranging from 21.5% to 23.2%; at the same time, a single prediction of the present invention takes only 4-5.5 hours, far less than the 720 hours of the comparative example, indicating that the present invention can not only accurately capture the leaching laws in long-term storage, but also efficiently adapt to different scenarios, which is superior to traditional methods.
[0107] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.
Claims
1. A method for predicting the anti-leaching performance of perfume glass bottles, characterized in that: The following steps are involved: S1: Prepare a standard simulating solution by mixing 5-15% by mass of ethanol, 0.1-0.5% by mass of phenoxyethanol, 0.01-0.1% by mass of citronellal, 0.05-0.2% by mass of benzoic acid, and the remainder by mass of deionized water, and adjust the pH to 3.0-5.5 to form a simulated perfume system; S2: Build a reaction kinetics model: Construct an ion migration model at the glass bottle-perfume interface based on the following reaction equation: Among them, the reaction rate constant is: A1: Pre-exponential factor of sodium ion leaching, value is 1.2×10 -4 s -1 ; A2: Pre-exponential factor of calcium ion leaching, value is 8.5×10 -5 s -1 ; E a1 : Sodium ion leaching activation energy, 55 kJ / mol; E a2 : Activation energy of calcium ion leaching, 62 kJ / mol; R: gas constant, 8.314 J / (mol·K); T: absolute temperature, unit K; S3 nanosensor detection: Gold nanoparticles modified with 3-aminopropyltriethoxysilane were dispersed in a simulated liquid, and the absorbance change at a wavelength of 520-540 nm was monitored by UV-visible spectroscopy. A linear relationship between absorbance and sodium ion concentration was established. ΔA = 0.025 [Na + ]+0.003, where △A is the absorbance change value, [Na + ] is the sodium ion concentration, unit: mg / L; S4 machine learning prediction: Input glass composition, surface treatment process parameters, and simulated liquid condition characteristic variables, train the prediction model through the random forest algorithm, and output the predicted value of sodium ion leaching concentration for 50-365 days.
2. The method for predicting the anti-leaching performance of a perfume glass bottle according to claim 1, characterized in that: The simulated liquid in S1 further contains 0.01-0.05% of 2-bromo-2-nitropropane-1,3-diol and 0.005-0.02% of ethyl salicylate to enhance the simulation of preservatives and fragrance ingredients.
3. The method for predicting the anti-leaching performance of a perfume glass bottle according to claim 1, characterized in that: The reaction kinetics model in S2 also includes the diffusion term of the glass surface coating: C: concentration of sodium ions in the coating, unit mol / m 3 ; t: time, unit s; D: Diffusion coefficient, value is 2.5×10 -12 m 2 / s; x: Coordinate in the coating thickness direction, unit: m; C H+ (x=0): hydrogen ion concentration at the interface between the coating and the solution, in mol / m 3 ; L: coating thickness, ranging from 100-200nm.
4. The method for predicting the anti-leaching performance of a perfume glass bottle according to claim 1, characterized in that: The nanosensing detection in S3 specifically includes: mixing a gold nanoparticle solution with a concentration of 0.1-0.5 mg / mL and a simulation solution at a volume ratio of 1:5, incubating at 25-60° C. for 1-4 hours, and then detecting the absorbance.
5. The method for predicting the anti-leaching performance of a perfume glass bottle according to claim 1, characterized in that: The machine learning model training in S4 uses more than 500 sets of experimental data, and the characteristic variables include: glass composition, physical properties, surface parameters, and simulated liquid conditions.
6. The method for predicting the anti-leaching performance of a perfume glass bottle according to claim 1, characterized in that: It also includes the S5 verification step: 100 groups of samples with different parameter combinations are selected for actual immersion experiments, and the actual leaching concentration is detected by inductively coupled plasma mass spectrometry (ICP-MS); the predicted values are compared with the measured values, and the mean absolute error (MAE) is calculated. When MAE is greater than 10%, 200 groups of training data are added and the model hyperparameters are adjusted. The number of decision trees is 200-300, the maximum depth is 10-12 layers, and retraining is carried out until MAE ≤ 8%.
7. The method for predicting the anti-leaching performance of a perfume glass bottle according to claim 1, characterized in that: The UV-visible spectrum detection in the S3 uses a fiber optic probe with a dual-beam optical path design, a detection wavelength range of 500-600 nm, a scanning rate of 100 nm / min, a detection sensitivity of 0.001 absorbance unit, a data collection interval of 10 s, and an average value is taken by three parallel detections to reduce errors.
8. The method for predicting the anti-leaching performance of a perfume glass bottle according to claim 1, characterized in that: The random forest algorithm parameter settings in S4 are as follows: the number of decision trees is 100-500, the maximum depth is 8-15 layers, the minimum number of samples for node splitting is 5-10, the minimum number of samples for leaf nodes is 1-3, the feature sampling ratio is 0.6-0.8, the Gini coefficient is used as the splitting criterion, the parameters are optimized through 5-fold cross-validation, and the model training iterations are 500-1000 times until the loss function converges.
9. The method for predicting the anti-leaching performance of a perfume glass bottle according to claim 1, characterized in that: The simulation liquid in S1 was prepared in a nitrogen-protected glove box, and the dissolved oxygen content was monitored online by fluorescence and controlled at ≤0.5 mg / L. The mixing order was as follows: first, deionized water was heated to 40° C., benzoic acid, phenoxyethanol, and citronellal were added in sequence, and after stirring until completely dissolved, ethanol was added. Finally, the pH was adjusted to 3.0-5.5 with 0.1 mol / L hydrochloric acid or sodium hydroxide, and the mixture was allowed to stand for 24 hours before use.
10. The method for predicting the anti-leaching performance of a perfume glass bottle according to claim 1, characterized in that: The prediction method is applicable to predicting the anti-leaching performance of perfume glass bottles made of different materials, such as borosilicate glass, soda-lime silicate glass, and aluminosilicate glass. The prediction results can be used to guide the optimization of glass bottle formula and the adjustment of surface treatment process.