Regenerated polyester identification method based on UV-Vis combined with chemometrics and machine learning algorithm

Abstract

The invention belongs to the technical field of polymer material analysis and detection, and particularly relates to a regenerated polyester identification method based on UV-Vis combined with chemometrics and a machine learning algorithm. The method comprises the following steps: respectively extracting original and regenerated polyester fiber samples by adopting an ultrasonic-assisted solvent, and carrying out UV-Vis determination on the extracted components to respectively obtain spectral characteristic data of the original and regenerated polyester fibers; sequentially carrying out data preprocessing and chemometrics analysis on the obtained spectral characteristic data to obtain a key data preprocessing mode for effectively distinguishing the native polyester fiber from the regenerated polyester fiber; a machine learning algorithm is adopted to establish an identification model according to spectral characteristic data obtained in a key data preprocessing mode, and the identification model is adopted to realize rapid and accurate identification of polyester fibers. A new solution is provided for identification of polyester fibers, the problem that raw polyester counterfeit regenerated polyester is solved, and healthy development of the polyester fiber material industry is promoted.

Description

Technical Field

[0001] The present invention belongs to the technical field of polymer material analysis and detection, and specifically relates to a recycled polyester identification method based on UV-Vis combined with chemometrics and machine learning algorithms. Background Art

[0002] Polyester fiber (hereinafter referred to as polyester) is a type of synthetic fiber obtained by the polymerization reaction of dibasic acid and diol. Polyester fiber occupies an important position in the global synthetic fiber. Polyester fiber is widely used in the textile industry due to its excellent properties such as wear resistance, strength, heat resistance, and hygroscopicity. With the continuous deepening of sustainable development and circular economy awareness, the development of the polyester textile industry has also ushered in new opportunities, and recycled polyester fiber (hereinafter referred to as recycled polyester) has gradually received more attention. The production of recycled polyester can reduce the consumption of petroleum-based raw materials, thereby bringing economic benefits of resource conservation, while also reducing the impact of waste synthetic plastic fibers on the environment and promoting ecological environmental protection. However, driven by this interest, the polyester textile industry may see the phenomenon of virgin polyester counterfeiting recycled polyester, affecting the sustainable and healthy development of the industry.

[0003] Therefore, the identification of virgin and recycled polyester has become a technical problem that urgently needs to be solved in the current textile industry. At present, numerous studies have shown that instrumental analysis combined with chemometric analysis or machine learning prediction models can effectively identify traditional plastics. Zhi-Feng Chen et al. established the identification of virgin and recycled polyethylene (PE) by combining ultraviolet-visible spectroscopy (UV-Vis) and ultra-high performance liquid chromatography quadrupole time-of-flight mass spectrometry (UPLC-Q-TOF-MS) with chemometric analysis; Hanke Li, Tian-Ying Hao et al. developed chemometric and machine learning prediction models for the identification of recycled polyethylene terephthalate (PET) based on the characteristics of volatile substances. These methods and models were established and developed based on PET plastic particles or bottle chip products. However, in the polyester textile industry, physically recycled polyester fibers are produced from waste bottles, waste polyester fibers, etc. as raw materials through more processes such as cleaning, drying, and melt spinning, resulting in obvious changes in the compound characteristics in the recycled fibers. The above UV-Vis-based method has a long analysis time and cannot be directly applied to the identification of polyester fibers. Preliminary studies have shown that the traditional headspace-gas chromatography-mass spectrometry method cannot effectively identify the characteristic differences between recycled and virgin polyester. The identification of recycled materials in textiles mainly relies on the certification of the polyester fiber recycling process. In addition, Fu Changfei et al. developed a pretreatment method based on methanol alcoholysis and swelling extraction, combined with liquid chromatography analysis to identify polyester fibers recycled by chemical and physical methods respectively, and subsequently transformed it into the recommended national standard GB / T 39026-2020 "Identification Method for Recycled Polyester (PET) Fibers". However, the analysis method proposed by them has cumbersome steps, high analysis costs, and there is a phenomenon of low discrimination accuracy in the industry, and it cannot quickly and accurately identify physically recycled polyester fibers. Summary of the Invention

[0004] Aiming at the above-mentioned shortcomings and deficiencies of the prior art, the purpose of the present invention is to provide a method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms. This method can efficiently and accurately identify and analyze colorless polyester fibers recycled by physical methods.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] A method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms, comprising the following steps:

[0007] (1) Ultrasonically assist solvent extraction of virgin polyester fiber and recycled polyester fiber samples respectively, and then perform UV-Vis scanning measurement on the extracted components to obtain the spectral characteristic data of extractable organic compounds in virgin polyester fibers and recycled polyester fibers respectively;

[0008] (2) Preprocess the spectral feature data obtained in step (1), and then perform chemometric analysis on the preprocessed data to obtain a key data preprocessing method for effectively distinguishing virgin polyester fibers from recycled polyester fibers.

[0009] (3) Using the spectral feature data obtained by the key data preprocessing method, adopt a machine learning algorithm to establish an identification model for recycled polyester, and then use the obtained identification model to achieve rapid and accurate identification of polyester fibers.

[0010] Further, the solvent used in the ultrasonic-assisted solvent extraction in step (1) is dichloromethane, ethyl acetate, methanol, n-hexane or tetrahydrofuran; preferably dichloromethane. Using dichloromethane as the extraction solvent can better achieve the extraction of differential compounds in polyester samples.

[0011] Further, the temperature of the ultrasonic-assisted solvent extraction in step (1) is 40 - 50 °C, and the time is 30 - 120 min.

[0012] Further, the data acquisition mode of the UV-Vis scanning determination in step (1) is transmittance, the wavelength scanning range is 260 - 800 nm, and the scanning interval is 1.0 nm.

[0013] Further, the data preprocessing method in step (2) is to perform at least one of smoothing processing, first derivative processing after smoothing, second derivative processing after smoothing, standardization processing after smoothing, multiplicative scatter correction processing after smoothing, and baseline removal processing after smoothing on the data using R language software.

[0014] Further, the chemometric analysis method in step (2) is: perform standardization processing on the spectral feature data after different data preprocessing in R language software, and then perform orthogonal partial least squares discriminant analysis (OPLS-DA), and screen and determine the key data preprocessing method for effectively distinguishing virgin polyester fibers from recycled polyester fibers according to the analysis result graph.

[0015] Further, the key data preprocessing method for effectively distinguishing virgin polyester fibers from recycled polyester fibers in step (2) is second derivative processing after smoothing.

[0016] Further, the method of establishing a discrimination model for recycled polyester by using a machine learning algorithm in step (3) is as follows: construct machine learning prediction models in R language software, including linear discriminant analysis (LDA), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost); use the stratified sampling method to divide the spectral feature data into a training set (such as 80%) and a test set (such as 20%) for training and validation of the discrimination model respectively; perform standardization processing and principal component analysis (PCA) dimensionality reduction processing on the training set and the test set; use the 10-fold cross-validation and grid search methods to optimize the parameters of the machine learning prediction models; evaluate the classification performance of the models according to the accuracy of multiple machine learning prediction models and the area under the curve of the receiver operating characteristic curve and screen the best machine learning prediction model; finally, use a confusion matrix diagram to visually evaluate the prediction ability of the selected machine learning prediction model.

[0017] Further preferably, the formula for calculating the accuracy is shown in the following formula (1):

[0018]

[0019] In the formula, TP is the true positive example, representing the positive sample predicted as the positive class by the model; TN is the true negative example, representing the negative sample predicted as the negative class by the model; FP is the false positive example, representing the negative sample predicted as the positive class by the model; FN is the false negative example, representing the positive sample predicted as the negative class by the model.

[0020] Further, the best machine learning prediction model is the support vector machine (SVM) model.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] The present invention uses ultraviolet-visible light (UV-Vis) to analyze different polyester samples; uses chemometrics analysis to screen out the best spectral feature data preprocessing method to reduce the baseline influence; uses four machine learning algorithms such as support vector machine (SVM) to establish a recycled polyester discrimination model. The results show that the UV-Vis spectral features of organic compounds can be better extracted from virgin and recycled polyester by ultrasonic-assisted extraction, and the spectral feature data based on smoothed second derivative processing can significantly distinguish the spectra of virgin and recycled polyester. Among the four machine learning algorithms, SVM has the best classification and prediction performance, and the prediction accuracies of the training set and the test set both reach 100%. The present invention develops an accurate and efficient recycled polyester discrimination model by combining UV-Vis with chemometrics and machine learning algorithms to solve the problem of confusion between recycled and virgin polyester fibers. Description of the Drawings

[0023] Figure 1Optimization results of extraction solvents based on representative polyester samples: (A) Dichloromethane (DCM); (B) Ethyl acetate (EAC); (C) Methanol (MeOH); (D) n-Hexane; (E) Tetrahydrofuran (THF).

[0024] Figure 2 Optimization results of extraction time based on representative polyester samples: (A) Virgin PET fiber; (B) Recycled PET fiber.

[0025] Figure 3 UV-Vis spectral characteristic diagrams of polyester based on different data pre-treatments: (A) Smoothing treatment; (B) First derivative treatment after smoothing; (C) Second derivative treatment after smoothing; (D) Normalization treatment after smoothing; (E) Multiplicative scatter correction treatment after smoothing; (F) Baseline removal treatment after smoothing.

[0026] Figure 4 OPLS-DA score diagrams of polyester UV-Vis spectral characteristic data based on different data pre-treatments: (A) Smoothing treatment; (B) First derivative treatment after smoothing; (C) Second derivative treatment after smoothing; (D) Normalization treatment after smoothing; (E) Multiplicative scatter correction treatment after smoothing; (F) Baseline removal treatment after smoothing.

[0027] Figure 5 Accuracy result diagram of the machine learning prediction model based on the polyester UV-Vis spectral data after second derivative treatment after smoothing.

[0028] Figure 6 Confusion matrix diagram based on the SVM machine learning prediction model. Detailed implementation manners

[0029] The present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings, but the implementation manners of the present invention are not limited thereto.

[0030] Embodiment 1

[0031] A method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms, comprising the following steps:

[0032] (1) Ultrasonic-assisted solvent extraction is respectively performed on virgin polyester fiber (RF, 44 kinds) and recycled polyester fiber (VF, 56 kinds) samples, and then UV-Vis scanning determination is performed on the extracted components to respectively obtain the spectral characteristic data of extractable organic compounds in virgin polyester fiber and recycled polyester fiber.

[0033] (2) Preprocess the spectral feature data obtained in step (1), and then perform chemometric analysis on the preprocessed data to obtain a key data preprocessing method that can effectively distinguish virgin polyester fibers from recycled polyester fibers.

[0034] (3) Using the spectral feature data obtained by the key data preprocessing method, adopt a machine learning algorithm to establish an identification model for recycled polyester, and then use the obtained identification model to achieve rapid and accurate identification of polyester fibers.

[0035] The optimization of the ultrasonic-assisted solvent extraction conditions in step (1) is as follows:

[0036] The difference in extractable organic compounds between virgin polyester fiber samples and recycled polyester fiber samples is related to the extraction solvent. Through comparative optimization of the extraction solvents, including: dichloromethane (DCM), ethyl acetate (EAC), methanol (MeOH), n-hexane (n-Hexane), and tetrahydrofuran (THF).

[0037] The sample extraction process is as follows:

[0038] 1) Accurately weigh 1.0 g (accurate to 0.01 g) of fiber sample into a 20 mL headspace vial for later use.

[0039] 2) Pipette and add 6 mL of extraction solvent into the headspace vial filled with the fiber sample. If necessary, appropriately cut the fiber sample to ensure that the test fiber can be completely immersed and then seal it. Ultrasonically extract for a certain time at 40 - 50 °C. After the ultrasonic extraction is completed, cool it to room temperature, and filter the extract through a 0.45 μm organic microporous membrane into a clean headspace vial for later use.

[0040] The process of UV-Vis scanning and determination of the sample is as follows:

[0041] After the instrument is turned on, preheat it for half an hour and set the parameters as follows: wavelength scanning range: 260 - 800 nm; sampling interval: 1.0 nm; data acquisition mode: transmittance.

[0042] Before the sample determination, pipette an appropriate amount of extraction solvent into a clean quartz cuvette as a blank reference sample to calibrate the instrument baseline. After the baseline calibration, pipette an appropriate amount of the filtered sample extract into a clean cuvette, and measure the test solution of the polyester fiber sample to be tested according to the above specified measurement conditions, and export and organize all the spectral feature data, including wavelength and transmittance.

[0043] The UV-Vis scanning and determination results of the extraction components extracted with different extraction solvents are as Figure 1 shown. Among them, dichloromethane (A) has a significant effect on the extraction of differential compounds in polyester samples among the five solvents.

[0044] Secondly, the differences in extraction times of 30 min, 60 min, and 120 min were compared, and the results are as Figure 2 shown. As Figure 2 can be seen, the differences between different classified polyester fibers did not change significantly under various extraction times.

[0045] The data preprocessing steps in step (2) are as follows:

[0046] Some miscellaneous peaks or non-peak-emitting bands that may be present in the spectral characteristics of the samples will interfere with the robustness of the discrimination model. Therefore, the characteristic spectral data is first imported into the R language software and data preprocessing is performed. In this embodiment, the following processing methods are preliminarily considered respectively to reduce the influence of baseline noise:

[0047] 1) Smoothing: The original spectral data is smoothed using a non-parametric regression box kernel function, and the bandwidth is set to 10. The spectrum is as Figure 3 shown in (A) therein.

[0048] 2) Derivative after smoothing: The first derivative and second derivative are respectively performed on the smoothed data. The spectra are as Figure 3 shown in (B) and (C) therein.

[0049] 3) Standardization after smoothing: The Z-score standard normal transformation is performed on the smoothed spectral data. The spectrum is as Figure 3 shown in (D) therein.

[0050] 4) Multiplicative scatter correction after smoothing: For the smoothed spectral data, the deviation of each spectrum from the sample mean reference spectrum is calculated, and the deviation is corrected using linear regression to eliminate the scattering effect. The spectrum is as Figure 3 shown in (E) therein.

[0051] 5) Baseline removal after smoothing: For the smoothed data, the baseline is calculated using polynomial fitting and subtracted. The spectrum is as Figure 3 shown in (F) therein.

[0052] The chemometric analysis process in step (2) is as follows:

[0053] In the R language software, pareto standardization is respectively performed on various preprocessed data, and orthogonal partial least squares discriminant analysis (OPLS-DA) is performed. According to the fitting situation of the established OPLS-DA model, the classification performance of different data preprocessing methods is evaluated using the OPLS-DA score plot ( Figure 4 ). As Figure 4It can be seen that the spectral feature data processed by smoothing and second derivative can significantly reflect the differences between the extracted compounds of virgin and recycled polyester, and effectively distinguish virgin and recycled polyester. Therefore, in this embodiment, the processing method of smoothing and second derivative is selected as the key data preprocessing method to process all polyester characteristic spectral data.

[0054] In step (3), the process of establishing an identification model for recycled polyester using a machine learning algorithm is as follows:

[0055] Four machine learning prediction models were constructed in the R language software, including Linear Discriminant Analysis (LDA), Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost). The processed polyester spectral data was divided into a training set (80%) and a test set (20%) according to the virgin and recycled categories by stratified sampling. The divided spectral data (training set and test set) were respectively subjected to standardization processing and principal component analysis (PCA) dimensionality reduction processing to further reduce the dataset dimension and remove redundant and useless information. Hyperparameter optimization of various machine learning prediction models was performed using 10-fold cross-validation and grid search. The classification performance of the model was evaluated by accuracy and the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. The calculation formula of accuracy is shown in Equation (1).

[0056]

[0057] In the formula, TP is the true positive (True Positive, TP), which represents the positive sample predicted as the positive class by the model; TN is the true negative (True Negative, TN), which represents the negative sample predicted as the negative class by the model; FP is the false positive (False Positive, FP), which represents the negative sample predicted as the positive class by the model; FN is the false negative (False Positive, FP), which represents the positive sample predicted as the negative class by the model.

[0058] As Figure 5 shown, sorted by the model accuracy (Accuracy), the Support Vector Machine (SVM) model has the highest model accuracy, followed by Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Linear Discriminant Analysis (LDA). To more intuitively display the model prediction results, the present invention uses a confusion matrix, as Figure 6As shown, the accuracy of both the training set and the test set predicted by the SVM model reached 100%, and all the native and recycled polyester samples were correctly predicted.

[0059] From the above, it can be seen that the present invention is based on spectral data, uses UV-Vis to measure the spectral characteristics of known native and recycled polyester fibers, screens out a data preprocessing method that can effectively distinguish between native and recycled categories through chemometric methods, and develops an accurate and efficient recycled polyester identification model based on the machine learning algorithm SVM, providing a new solution for the identification of recycled polyester fibers and promoting the healthy development of the polyester textile material industry.

[0060] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. A method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms, characterized in that: The steps include: (1) Virgin polyester fiber and regenerated polyester fiber samples were extracted by ultrasonic-assisted solvent, and then the extracted components were measured by UV-Vis scanning to obtain spectral characteristic data of extractable organic compounds in virgin polyester fiber and regenerated polyester fiber respectively; (2) performing data preprocessing on the spectral characteristic data obtained in step (1), and then performing chemometric analysis on the preprocessed data to obtain a key data preprocessing method for effectively distinguishing between virgin polyester fibers and regenerated polyester fibers; (3) Using the spectral feature data obtained by key data preprocessing and a machine learning algorithm, an identification model for recycled polyester is established. The obtained identification model is then used to achieve rapid and accurate identification of polyester fibers.

2. The method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms according to claim 1, characterized in that: The solvent used in the ultrasonic-assisted solvent extraction in step (1) is dichloromethane, ethyl acetate, methanol, n-hexane or tetrahydrofuran.

3. The method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms according to claim 1, characterized in that: The solvent used in the ultrasonic-assisted solvent extraction in step (1) is dichloromethane.

4. The method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms according to claim 1, characterized in that: The temperature of the ultrasonic-assisted solvent extraction in step (1) is 40-50° C. and the time is 30-120 min.

5. The method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms according to claim 1, characterized in that: The data acquisition mode of the UV-Vis scanning measurement in step (1) is transmittance, the wavelength scanning range is 260 to 800 nm, and the scanning interval is 1.0 nm.

6. The method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms according to claim 1, characterized in that: The data preprocessing method in step (2) is to use R language software to perform at least one of smoothing processing, first-order derivative processing after smoothing, second-order derivative processing after smoothing, standardization processing after smoothing, multivariate scattering correction processing after smoothing and baseline removal processing after smoothing on the data.

7. The method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms according to claim 1, characterized in that: The method of chemometric analysis described in step (2) is: the spectral feature data after different data preprocessing are standardized in R language software respectively, and then orthogonal partial least squares discriminant analysis is performed, and the key data preprocessing method that effectively distinguishes virgin polyester fiber and recycled polyester fiber is screened and determined according to the analysis result graph.

8. The method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms according to claim 1, characterized in that: The key data preprocessing method for effectively distinguishing virgin polyester fiber and regenerated polyester fiber in step (2) is smoothing followed by second-order derivative processing.

9. The method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms according to claim 1, characterized in that: The method of using a machine learning algorithm to establish an identification model for recycled polyester in step (3) is as follows: constructing a machine learning prediction model in R language software, including linear discriminant analysis, random forest, support vector machine and extreme gradient boosting; The spectral feature data were divided into a training set and a test set by stratified sampling method, which were used for training and verification of the identification model respectively. The training set and the test set were standardized and the principal component analysis was used for dimensionality reduction. The parameters of the machine learning prediction model were optimized by 10-fold cross validation and network search method. The classification performance of the model was evaluated based on the accuracy of multiple machine learning prediction models and the area under the curve of the receiver operating characteristic curve, and the best machine learning prediction model was selected. Finally, the confusion matrix diagram was used to intuitively evaluate the prediction ability of the selected machine learning prediction model. The calculation formula of the accuracy is shown in the following formula (1): Where TP is a true positive example, which indicates a positive sample predicted by the model as a positive class; TN is a true negative example, which indicates a negative sample predicted by the model as a negative class; FP is a false positive example, which indicates a negative sample predicted by the model as a positive class; FN is a false negative example, which indicates a positive sample predicted by the model as a negative class.

10. The method for identifying recycled polyester based on UV-Vis combined with chemometrics and machine learning algorithms according to claim 9, characterized in that: The optimal machine learning prediction model is a support vector machine model.

Images