Terahertz spectrum quantitative analysis method for rubber mixture
By combining terahertz spectroscopy with data processing algorithms, a quantitative analysis model for rubber mixtures was constructed, which solved the problems of complexity and high cost in traditional detection methods and enabled rapid and accurate detection of antioxidant content in rubber mixtures.
Patent Information
- Application Number
- CN202610053297.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-15
- Publication Date
- 2026-05-01
AI Technical Summary
Traditional testing methods are difficult to achieve highly sensitive and accurate quantitative analysis of antioxidants in rubber mixtures, and are complex, time-consuming, and costly, failing to meet the needs of rapid and non-destructive testing.
By employing terahertz time-domain spectroscopy combined with fast Fourier transform, principal component analysis (PCA), support vector regression (SVR) model and improved algorithm, and through preprocessing and dimensionality reduction of spectral data, a quantitative analysis model for rubber mixtures is constructed to achieve accurate detection of antioxidant content.
It enables rapid, accurate, safe, and environmentally friendly detection of antioxidant content in rubber mixtures, reducing detection complexity and cost while improving detection efficiency and accuracy.
Smart Images
Figure FT_1 
Figure FT_2 
Figure FT_3
Abstract
Description
Technical Field
[0001] This invention relates to the field of macromolecular material detection technology, and in particular to a terahertz spectroscopic quantitative analysis method for rubber mixtures. Background Technology
[0002] Rubber, as a key polymer material, plays an irreplaceable role in the national economy and modern industry due to its unique physical properties. To ensure the reliability and durability of rubber in various application environments, quality testing is crucial. Terahertz spectroscopy, with its excellent coherence, strong penetration, and non-destructive testing characteristics, demonstrates broad application potential in fields such as non-destructive testing of materials.
[0003] Antioxidants are indispensable additives in rubber products, playing a crucial role in delaying rubber aging and extending its service life. However, it is worth noting that some antioxidants, such as secondary amines and phenols, are toxic and polluting, potentially posing threats to the environment and human health. Furthermore, the amount of antioxidants added to rubber products is usually small, which brings significant challenges to detection. Traditional detection methods often struggle to achieve high sensitivity and accuracy in quantitative analysis when dealing with complex multi-component mixtures. Therefore, developing an efficient and accurate antioxidant detection technology, especially for the rapid determination of trace components in complex matrices, has become an urgent need in the current scientific research field.
[0004] Currently, traditional destructive detection methods include visual inspection, mass spectrometry, enzyme-linked immunosorbent assay (ELISA), and chromatography. While these methods can achieve high-precision analysis, they generally suffer from drawbacks such as being time-consuming, labor-intensive, complex to operate, highly dependent on chemical reagents, and requiring specialized operators and precision equipment. In recent years, non-destructive spectroscopic techniques, such as fluorescence spectroscopy, Raman spectroscopy, near-infrared spectroscopy, and hyperspectroscopy, have been widely used in material identification due to their rapid detection capabilities. However, these techniques also have limitations, such as fluorescence spectroscopy being limited by the self-luminescence properties of the sample, Raman spectroscopy being affected by fluorescence interference, hyperspectral technology having complex data processing, and near-infrared spectroscopy having limited sensitivity.
[0005] Therefore, while traditional testing methods can achieve the testing objectives, they are subject to numerous uncertainties, and most methods require significant time and tedious operations, demanding high levels of expertise from testing personnel and incurring high costs, making them unsuitable for rapid and non-destructive testing. Thus, the development of a reliable, accurate, rapid, safe, and environmentally friendly new method for testing rubber mixtures is imperative. Summary of the Invention
[0006] To address the aforementioned problems, this invention provides a quantitative analysis method for rubber mixtures based on terahertz time-domain spectroscopy. This method enables accurate detection of the content of two trace antioxidants (NBC and 44S) in a five-component rubber mixture. Furthermore, the pretreatment method is simple and convenient, solving the problems of low accuracy, high cost, cumbersome operation, long time consumption, and high professional requirements for testing personnel in traditional detection processes. This provides technical support and reference for the detection of low-content substances in other mixtures.
[0007] To achieve the above objectives, this invention proposes a quantitative analysis method for rubber mixtures based on terahertz time-domain spectroscopy, comprising the following steps.
[0008] S1: Prepare rubber mixture samples according to a certain proportion, crush the rubber mixtures of different proportions into powder using an electric pulverizer, and then dry, sieve, grind, weigh, and press into sheets. Finally, make smooth circular thin sheet experimental samples from the sieved rubber mixture powder in equal mass.
[0009] S2: Use a terahertz time-domain spectroscopy system to obtain time-domain spectral data information for reference and experimental samples.
[0010] S3: After converting the time-domain spectral data of rubber mixtures with different ratios obtained in S2 into corresponding frequency-domain spectral data through Fast Fourier Transform (FFT), the absorbance spectrum of the sample is obtained using parameter extraction methods.
[0011] S4: The SG, SNV and MSC algorithms are used to preprocess the spectral data respectively.
[0012] S5: Select the absorbance spectral data processed by the superior SG algorithm, and use principal component analysis (PCA) to reduce the dimensionality of the spectral data, eliminate redundant features and noise interference, and extract representative sample features.
[0013] S6: Use the SPXY algorithm to divide the data into training and prediction sets for model analysis.
[0014] S7: Using Support Vector Regression (SVR) as the data analysis model, GA-SVR, PSO-SVR, and FBO-SVR models were constructed to analyze the absorbance spectra of the mixture samples. In order to further improve the accuracy of the model prediction, the FBO algorithm was improved by introducing chaotic mapping, Cauchy variation, and reverse learning strategies. The final detection results of different models were compared.
[0015] The advantages of this invention are:
[0016] (1) Terahertz spectroscopy can provide both the amplitude information of substances that infrared spectroscopy can provide and the phase information of substances that infrared spectroscopy cannot provide. Therefore, for five-component rubber mixtures with different contents of antioxidants, they can be identified based on the "fingerprint" characteristics of the analyte. Applying this invention only requires establishing a quantitative analysis model for the five-component rubber mixture to be tested, which enables accurate detection of the antioxidant content in the rubber mixture.
[0017] (2) Before establishing the quantitative analysis model for rubber mixtures, the SG, SNV, and MSC algorithms were used to preprocess the spectral data. The absorbance spectral data processed by the SG algorithm, which performed better, was selected to effectively remove interference factors such as noise, baseline drift, and scattering. Furthermore, principal component analysis (PCA) was used to reduce the dimensionality of the spectral data, which significantly shortened the model training time, eliminated redundant features and noise interference, and improved the model's generalization ability and analysis efficiency.
[0018] (3) The quantitative detection method for five-component rubber mixtures with different contents of antioxidants provided by the present invention has advantages in terms of speed, accuracy, safety and environmental protection. Attached Figure Description
[0019] Figure 1 This is a schematic flowchart of a terahertz spectral quantitative analysis method for rubber mixtures according to the present invention.
[0020] Figure 2 This is a schematic diagram of the principle of the terahertz time-domain spectral transmission system of the present invention.
[0021] Figure 3 These are time-domain spectra of five-component rubber mixture samples with different proportions according to the present invention.
[0022] Figure 4 This is the time-domain spectrum of the five-component rubber mixture samples with different proportions of the present invention after averaging.
[0023] Figure 5 These are absorbance spectra of five-component rubber mixture samples with different proportions according to the present invention.
[0024] Figure 6 This is the absorbance spectrum of the five-component rubber mixture samples with different proportions according to the present invention.
[0025] Figure 7-1 , 7-2 Figures 7-3 and 7-4 show the prediction results of the support vector regression model after optimization using the genetic algorithm, particle swarm optimization algorithm, bitterling algorithm, and improved bitterling algorithm, respectively. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Please refer to the accompanying drawings for details. Figure 2 This invention provides a terahertz spectroscopic quantitative analysis method for rubber mixtures, specifically including the following steps:
[0027] S1: Prepare rubber mixture samples according to a certain proportion, crush the rubber mixtures of different proportions into powder using an electric pulverizer, and then dry, sieve, grind, weigh, and press into sheets. Finally, make smooth circular thin sheet experimental samples from the sieved rubber mixture powder in equal mass.
[0028] S2: Use a terahertz time-domain spectroscopy system to obtain time-domain spectral data information for reference and experimental samples.
[0029] S3: After converting the time-domain spectral data of rubber mixtures with different ratios obtained in S2 into corresponding frequency-domain spectral data through Fast Fourier Transform (FFT), the absorbance spectrum of the sample is obtained using parameter extraction methods.
[0030] S4: The SG, SNV and MSC algorithms are used to preprocess the spectral data respectively.
[0031] S5: Select the absorbance spectral data processed by the superior SG algorithm, and use principal component analysis (PCA) to reduce the dimensionality of the spectral data, eliminate redundant features and noise interference, and extract representative sample features.
[0032] S6: Use the SPXY algorithm to divide the data into training and prediction sets for model analysis.
[0033] S7: Using Support Vector Regression (SVR) as the data analysis model, GA-SVR, PSO-SVR, and FBO-SVR models were constructed to analyze the absorbance spectra of the mixture samples. In order to further improve the accuracy of the model prediction, the FBO algorithm was improved by introducing chaotic mapping, Cauchy variation, and reverse learning strategies. The final detection results of different models were compared.
[0034] Specifically, rubber mixture samples were prepared according to a certain proportion. Different proportions of rubber mixtures were pulverized into powder using an electric pulverizer, then dried, sieved, ground, weighed, and compressed into tablets. Finally, the sieved rubber mixture powder was used to make smooth, round, thin experimental samples of equal mass. Specifically, the experimental samples were mainly prepared using a tableting method. First, different proportions of rubber mixtures were pulverized into powder using a pulverizer. Then, the pulverized mixtures were sieved through a 200-mesh sieve. Next, the solid powder was placed in a drying oven at 50°C for 2 hours. Then, the weight of the rubber mixture powders of different proportions was weighed according to the proportions using an electronic analytical balance. The sample weight of each sample was 200 mg. Finally, the rubber mixture powder was poured into a tableting mold and compressed at a pressure of 8t for 60 seconds, resulting in round, thin tablets with a diameter of 13 mm, a thickness of approximately 1 mm, and a smooth surface.
[0035] Optionally, the terahertz spectrometer mentioned is the CCT-1800 terahertz spectrometer developed by Shenzhen Huaxun Fangzhou Technology Co., Ltd. This invention uses a terahertz time-domain spectrometer transmission module, which mainly includes a terahertz radiation generating element, a terahertz radiation detection element, a time delay device, and a femtosecond pulse laser, among other main components. For detailed principles, please refer to [link / reference needed]. Figure 2 .
[0036] Optionally, start the CCT-1800 terahertz spectrometer, open the laptop control software, select the projection module as the spectral mode, and set the parameters. Turn on the nitrogen regulating switch to fill the equipment with nitrogen, while ensuring that the gas flow regulating valve is at 15~20 NL / S to eliminate water vapor interference. After observing that the no-load time domain plot is normal, obtain the reference time domain spectral information. Then, place the experimental sample into the experimental sample slot of the instrument to start the measurement and obtain the time domain spectral data information of rubber mixtures with different proportions.
[0037] Optionally, the time-domain spectral data of the rubber mixtures with different ratios obtained in S2 are converted into corresponding frequency-domain spectral data using Fast Fourier Transform (FFT). Then, the absorbance spectrum of the sample is obtained using a parameter extraction method. Specifically, the time-domain spectra of the reference and sample obtained in S2 are used, and then a Fourier transform is performed on them. The transfer function related to parameters such as refractive index and extinction coefficient is calculated. Finally, the absorbance characterizing the material's absorption of terahertz waves is obtained through calculation. The calculation formula is as follows:
[0038]
[0039] Optionally, the absorbance spectral data obtained through calculation contains not only its own valid information but also other invalid information and noise, stray light, etc., which can negatively affect the modeling effect. Therefore, effective spectral preprocessing is required when building the model. In this invention, SG, SNV, and MSC algorithms are used to preprocess the spectral data respectively. Cross-validation results show that the SG algorithm exhibits superior performance. To further eliminate or reduce spectral intensity variations caused by physical factors (such as sample thickness, particle size, measurement conditions, etc.) and to make the data comparable on the same scale, the preprocessed spectral data is normalized to ensure the stability and reliability of subsequent analysis.
[0040] Optionally, the high dimensionality of the original absorbance spectral data and the strong correlation between points lead to a long model training time. However, not all data features are necessary; redundant features not only increase computational complexity but may also introduce noise, affecting model performance and efficiency. To address this issue, Principal Component Analysis (PCA) is used to reduce the dimensionality of the spectral data. PCA extracts principal components that explain the maximum variance of the data. Its advantages lie in effectively reducing data dimensionality, removing redundant information, retaining important features, and simplifying the data structure. Through PCA dimensionality reduction, not only is the model training time significantly shortened, but redundant features and noise interference are also eliminated, improving the model's generalization ability and analytical efficiency.
[0041] Optionally, the SPXY algorithm is used to divide the data into training and prediction sets for model analysis. By calculating the Euclidean distance matrix of samples in the independent and dependent variable spaces, a joint distance matrix is constructed after weighted combination. The samples with the largest distances are then progressively selected as the training set to ensure that the training and prediction sets are representative in terms of data distribution. In this invention, the spectral data is divided into a training set (144 spectral data points) and a prediction set (72 spectral data points) at a 2:1 ratio.
[0042] Optionally, Support Vector Regression (SVR) is used as the data analysis model. GA-SVR, PSO-SVR, and FBO-SVR models are constructed to analyze the absorbance spectra of the mixture samples. To further improve the accuracy of model predictions, the FBO algorithm is improved by introducing chaotic mapping, Cauchy mutation, and reverse learning strategies. The final detection results of different models are then compared. Specific improvements are as follows:
[0043] (1) Through systematic experimental design and spectral analysis, the differential characteristics of different antioxidant ratios in the time domain and absorbance spectrum were clarified. The signal-to-noise ratio of the spectral data and the generalization ability of the model were effectively improved by using SG preprocessing, PCA and SPXY dataset partitioning methods.
[0044] (2) In view of the characteristics of small sample and high-dimensional nonlinear spectral data, a support vector regression (SVR) model was introduced, and the significant advantage of IFBO in parameter optimization was verified by comparing the optimization effects of GA, PSO and FBO.
[0045] Optionally, model performance can be evaluated using accuracy on both the training and prediction sets. An accuracy value closer to 1 indicates a better correlation between the training and prediction sets.
[0046] This invention provides a specific embodiment; please refer to [link / reference]. Figure 3 , 4 5, 6, 7-1, 7-2, 7-3, 7-4.
[0047] The experimental subjects were multi-component mixtures consisting of nitrile rubber, silica, antioxidant N-phenyl-N'-(1,3-dimethylbutyl)-p-phenylenediamine, and two vulcanization accelerators (a toxic vulcanization accelerator, vinylthiourea, and a low-toxicity vulcanization accelerator, tetrabenzylthiuram disulfide). The content of the two vulcanization accelerators in the mixture was detected. All samples required for the experiment were provided by the National Rubber and Rubber Products Quality Supervision and Inspection Center (Guangxi). The experimental samples described in this invention are mainly prepared by the tableting method. Different proportions of rubber mixtures are crushed into powder using a pulverizer. The resulting pulverized mixtures are then sieved through a 200-mesh sieve. The solid powders are then placed in a drying oven and dried at 50°C for 2 hours. Next, the weights of the rubber mixture powders in different proportions are weighed using an electronic analytical balance, with each sample weighing 200 mg. Finally, the rubber mixture powders are poured into a tableting mold and pressed at a pressure of 8 t for 60 seconds to obtain a circular sheet with a diameter of 13 mm, a thickness of approximately 1 mm, and a smooth surface.
[0048] To ensure the quality of sample preparation, 5% polyethylene was added to all mixtures. All substances were weighed strictly according to the proportions in Table 1, and 36 samples were prepared for each proportion, for a total of 216 samples. One spectral data point was collected for each sample.
[0049] Table 1. Proportions of each substance in the five-component mixture of rubber and additives
[0050]
[0051] To reduce random error, the spectral data for each sample was obtained by measuring three times on both sides and dividing the result by six. During sample preparation, it is important to use a dehumidifier to maintain a relatively dry experimental environment to minimize interference from moisture.
[0052] Start the Z-3 terahertz spectrometer, open the laptop control software, select the projection module as the spectral mode, and set all parameters. Turn on the nitrogen regulator switch to fill the equipment with nitrogen, while ensuring the gas flow rate regulator valve is between 15 and 20 NL / S to eliminate water vapor interference. After observing that the no-load time domain plot is normal, obtain the reference time domain spectral information. Then, place the experimental sample into the experimental sample slot of the instrument to start the measurement and obtain the time domain spectral data information of rubber mixture samples with different proportions.
[0053] Using the time-domain spectral signals of the reference and sample obtained above, the time-domain spectral data of the rubber mixture sample is converted into corresponding frequency-domain spectral data through Fast Fourier Transform (FET). The transfer function related to parameters such as refractive index and extinction coefficient is calculated, and finally, the absorbance A, characterizing the material's absorption of terahertz waves, is obtained through calculation. The absorbance spectrum of the sample is then obtained. For details, please refer to [link to relevant documentation]. Figure 5 The formula for calculating absorbance is:
[0054]
[0055] ω is the terahertz angular frequency. Eref(ω) is the reference signal. Esam(ω) is the sample time-domain signal.
[0056] To clearly display the absorption peak characteristics of the spectrum, the absorbance spectrum of the sample is averaged. Please refer to [link / reference needed] for details. Figure 6 The spectral data were preprocessed using the SG, SNV, and MSC algorithms. The absorbance spectral data processed by the superior SG algorithm were selected, and Principal Component Analysis (PCA) was used to reduce the dimensionality of the spectral data, eliminating redundant features and noise interference to extract representative sample features. The SPXY algorithm was used to divide the data into training and prediction sets for model analysis, with the spectral data divided into a 2:1 ratio: a training set (144 spectral data points) and a prediction set (72 spectral data points).
[0057] The absorbance spectra of the mixture samples were analyzed by constructing GA-SVR, PSO-SVR, and FBO-SVM models respectively. Please refer to the model prediction results for details. Figure 7-1 , 7-27-3. Based on the information in the figure, it can be seen that the GA-SVR model performs the worst in predicting mixtures with 0% and 5% NBC (Non-Nutritional Preservative) content. In comparison, the prediction results of the PSO-SVR and FBO-SVM models are closer to the zero error line. Furthermore, upon closer observation, the FBO-SVM model performs better when the NBC content is 2%. To further improve the accuracy of the model predictions, the FBO algorithm is improved by introducing chaotic mapping, Cauchy mutation, and back-learning strategies. Please refer to Figure 7-4 for the specific model prediction results. As shown in the figure, the performance of the IFBO-SVR model is significantly improved.
[0058] To more intuitively analyze the prediction effects of different models, the parameters involved in the model analysis process have been summarized, as shown in Table 2.
[0059] Table 2. Parameters related to model analysis after applying the optimized algorithm.
[0060]
[0061] Interpreting the relevant parameters of the model analysis yields the following conclusions: First, the original bitterling optimization algorithm, after optimizing the SVR model, achieved a correlation coefficient Rc as high as 0.9575 and a root mean square error of only 0.0059 on the prediction set, both superior to the results achieved by the other two algorithms. Second, the improved bitterling optimization algorithm further improved the SVR model's performance on both the training and prediction sets. Specifically, the correlation coefficient Rc on the prediction set increased by 0.0304 compared to the original algorithm, and the root mean square error decreased by 0.0035, demonstrating excellent parameter optimization capabilities. In conclusion, employing appropriate optimization algorithms can effectively improve the model's predictive performance, thereby enabling accurate quantitative analysis of low-content antioxidants in mixtures.
[0062] It should be noted that the above specific embodiments are only for detailed explanation of the technical solution of the present invention, and should not be used to limit the scope of the present invention. Those skilled in the art should understand all and part of the process of the above embodiments, and equivalent substitutions made in accordance with the claims of the present invention still fall within the scope of the present invention.
Claims
1. A terahertz spectroscopic quantitative analysis method for rubber mixtures, characterized in that, Includes the following steps S1: Prepare rubber mixture samples according to a certain proportion, crush the rubber mixtures of different proportions into powder using an electric pulverizer, and then dry, sieve, grind, weigh, and press into sheets. Finally, make smooth circular thin sheet experimental samples from the sieved rubber mixture powder in equal mass. S2: Use a terahertz time-domain spectroscopy system to obtain time-domain spectral data information for reference and experimental samples; S3: After converting the time-domain spectral data of rubber mixtures with different ratios obtained in S2 into the corresponding frequency-domain spectral data through Fast Fourier Transform (FFT), the absorbance spectrum of the sample is obtained by using parameter extraction method. S4: The SG, SNV and MSC algorithms are used to preprocess the spectral data respectively; S5: Select the absorbance spectral data processed by the superior SG algorithm, and use principal component analysis (PCA) to reduce the dimensionality of the spectral data, eliminate redundant features and noise interference, and extract representative sample features. S6: Use the SPXY algorithm to divide the training set and prediction set data for model analysis; S7: Using Support Vector Regression (SVR) as the data analysis model, GA-SVR, PSO-SVR, and FBO-SVR models were constructed to analyze the absorbance spectra of the mixture samples. In order to further improve the accuracy of the model prediction, the FBO algorithm was improved by introducing chaotic mapping, Cauchy variation, and reverse learning strategies. The final detection results of different models were compared.
2. The terahertz spectroscopic quantitative analysis method for rubber mixtures according to claim 1, characterized in that, The SG, SNV, and MSC algorithms were used to preprocess the spectral data, and the cross-validation results showed that the SG algorithm performed better.
3. The terahertz spectroscopic quantitative analysis method for rubber mixtures according to claim 1, characterized in that, Principal component analysis (PCA) was used to reduce the dimensionality of the spectral data, eliminating redundant features and noise interference, in order to extract representative sample features.
4. The terahertz spectroscopic quantitative analysis method for rubber mixtures according to claim 1, characterized in that, The SPXY algorithm was used to divide the data into training and prediction sets for model analysis. The spectral data were divided into a training set (144 spectral data points) and a prediction set (72 spectral data points) in a 2:1 ratio.
5. The terahertz spectroscopic quantitative analysis method for rubber mixtures according to claim 1, characterized in that, By combining an improved bitterling optimization algorithm (IFBO), the accurate detection of two trace antioxidants (NBC and 44S) in a five-component rubber mixture was achieved. The optimization effects of GA, PSO, and IFBO were compared, validating the significant advantages of IFBO in parameter optimization. On the prediction set, the correlation coefficient (Rp) reached 0.9879, and the root mean square error (RMSEP) was reduced to 0.0024. Specific improvements are as follows: (1) Through systematic experimental design and spectral analysis, the differential characteristics of different antioxidant ratios in the time domain and absorbance spectrum were clarified. The signal-to-noise ratio and model generalization ability of the spectral data were effectively improved by using SG preprocessing, PCA and SPXY dataset partitioning methods. (2) In view of the characteristics of small sample and high-dimensional nonlinear spectral data, a support vector regression (SVR) model was introduced, and the significant advantage of IFBO in parameter optimization was verified by comparing the optimization effects of GA, PSO and FBO.