Regenerated plastic quantitative detection method based on instrument analysis and machine learning and application thereof

Abstract

The invention belongs to the field of analysis and testing of high polymer materials, discloses a quantitative detection method for recycled plastics based on instrument analysis and machine learning and application of the quantitative detection method, and particularly relates to determination of the content of the recycled plastics (polyethylene glycol terephthalate (PET), polyethylene (PE) and polypropylene (PP)). The quantitative detection method comprises the following steps: acquiring organic small molecules in a plastic sample with known regenerated plastic content, and analyzing to obtain analysis result data; processing the data to obtain characteristic data, and constructing a regression model based on machine learning by using a database formed by different sample characteristic data; a sample to be detected is processed by the same processing method to obtain characteristic data, and the content of the regenerated plastic in the sample is detected by utilizing the constructed model. Based on instrument analysis and machine learning, the invention provides the quantitative detection method for the regenerated plastic, so that the regenerated plastic can be rapidly and accurately quantified, and the quantitative detection method can be applied to identification of the regenerated plastic and the raw plastic.

Description

Technical Field

[0001] The present invention belongs to the field of polymer material analysis and testing, and relates to a quantitative detection method for recycled plastics based on instrumental analysis and machine learning and an application thereof, and in particular to the determination of the content of recycled plastics (polyethylene terephthalate (PET), polyethylene (PE), and polypropylene (PP)). Background Art

[0002] The recycling and reuse of plastics is an important part of sustainable development. As the most widely used plastics, the recycling and reuse of PET, PE, and PP has received more and more attention. Some countries and regions have begun to put forward corresponding requirements for the content of recycled plastics in different products. For example, the EU "Packaging and Packaging Waste" regulations stipulate that the proportion of recycled plastics in disposable beverage containers with PET as the main component should be no less than 25% by 2025 and no less than 30% by 2030. At present, the content of recycled plastics in materials mainly depends on recycled certification, and there is no corresponding detection technology for confirmation. The development of a method for determining the content of recycled plastics is of great significance to promoting the development of the recycled plastics industry, ensuring food safety, and maintaining market order.

[0003] Post-consumer recycled plastics have gone through the process of consumer use, recycling and regeneration, which makes the types and contents of pollutants in recycled plastics different from those in virgin plastics, which can serve as a theoretical basis for the identification of virgin / recycled plastics. At present, there are some methods that combine instrumental detection and analysis technology with chemometrics or machine learning to identify pure virgin plastics and 100% recycled plastics. For example, the absorption spectra of the extracts of virgin and recycled plastic samples in the ultraviolet-visible light region are measured by ultraviolet-visible spectrophotometry (UV-vis); or the differences in volatile substances in virgin and recycled samples are analyzed by headspace-gas chromatography-mass spectrometry (HS-GC-MS), and then a machine learning classification model is established to achieve the purpose of identifying the gender of virgin and recycled plastics. However, there are no relevant reports on the determination method of the content of recycled plastics. Summary of the invention

[0004] In order to overcome the above-mentioned shortcomings and deficiencies of the prior art, the primary purpose of the present invention is to provide a quantitative detection method for recycled plastics based on instrumental analysis and machine learning.

[0005] Another object of the present invention is to provide an application of the above method in the identification of recycled plastics and virgin plastics. The method of the present invention can be applied to the identification of various recycled plastics, such as polyethylene terephthalate (PET), polyethylene (PE), polypropylene (PP), etc.

[0006] The purpose of the present invention is achieved through the following solutions:

[0007] In the first aspect, a quantitative detection method for recycled plastics based on instrumental analysis and machine learning includes: obtaining small organic molecules in a plastic sample with a known recycled plastic content and analyzing it to obtain analysis result data; processing the data to obtain characteristic data, and constructing a regression model based on machine learning using a database formed by characteristic data of different samples; using the same processing method for the sample to be tested to obtain characteristic data, and using the constructed model to detect the content of recycled plastic in the sample.

[0008] In the technical solution of the present invention, the organic small molecules in the sample can be obtained by conventional methods, such as solvent extraction, headspace method, dissolution precipitation method, etc.

[0009] Furthermore, it can be obtained by solvent extraction.

[0010] In the technical solution of the present invention, the method for analyzing organic small molecules may include spectroscopy, chromatography-mass spectrometry and the like.

[0011] Furthermore, the spectroscopy may include ultraviolet-visible spectrophotometry (UV-Vis), infrared spectroscopy (IR), and the like.

[0012] Further, spectroscopy may include ultraviolet-visible spectrophotometry (UV-Vis).

[0013] Furthermore, when ultraviolet-visible spectrophotometry (UV-Vis) is used for analysis, the analysis result data may be selected from the entire section, or preferably from a band with obvious differences.

[0014] In the technical solution of the present invention, processing data refers to preprocessing the analysis result data, such as smoothing, first-order derivative and baseline removal.

[0015] Furthermore, the analysis result data is further processed, such as using dimensionality reduction methods such as principal component analysis (PCA) or uniform manifold approximation and projection (UMAP) to process the data.

[0016] In the technical solution of the present invention, the establishment of machine learning regression models adopts a variety of different machine learning algorithms, such as random forest (RF), support vector machine (SVM), extreme gradient boosting (XGboost), etc. to establish multiple regression models. The model training is based on sample data with known recycled plastic content, and hyperparameters are optimized through 10-fold cross validation.

[0017] In the technical solution of the present invention, the optimal model can be selected from multiple regression models for use.

[0018] In the technical solution of the present invention, the database is obtained based on the characteristic data of plastic samples with different contents of known recycled plastics; including original plastic samples and recycled plastic samples.

[0019] Furthermore, the number of samples in the database is greater than or equal to 100.

[0020] Furthermore, the number of samples in the database is greater than or equal to 200.

[0021] Furthermore, the quantitative ratio of recycled plastic samples: (original plastic samples + recycled plastic samples) in the database may be 1:3-1:8.

[0022] In the technical solution of the present invention, the type of recycled plastic in the sample to be tested is consistent with the recycled plastic used to construct the database.

[0023] Furthermore, the plastic sample may be polyethylene terephthalate (PET), polyethylene (PE), polypropylene (PP), etc.

[0024] Secondly, the PCA model based on UV-visible spectroscopy of the present invention can achieve 100% accuracy in qualitative identification of recycled plastics, and can be applied to qualitative identification of recycled plastics and virgin plastics.

[0025] The present invention proposes for the first time a detection method for quantitatively detecting recycled plastics based on instrumental analysis and machine learning. Compared with the existing qualitative detection methods, the method of the present invention can not only achieve 100% accuracy in qualitative identification of recycled plastics, but also perform quantitative detection of recycled plastics, accurately determine the quantitative results, and more effectively perform quantitative analysis of recycled plastics. The method of the present invention is simple to operate, can quickly and accurately quantify recycled plastics, and can be applied to the identification of recycled plastics and virgin plastics.

[0026] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0027] (1) The method of the present invention uses multiple instrumental analyses and machine learning to quantify recycled plastics, which can accurately determine the quantitative results and more effectively perform quantitative analysis on recycled plastics.

[0028] (2) The instrument method used in the method of the present invention is simple and quick to operate and can quickly test samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0030] Figure 1The preprocessing results of the classification model for plastic samples: (A) original infrared spectrum data; (B) infrared spectrum data after baseline removal processing; (C) original ultraviolet spectrum data; (D) ultraviolet spectrum data after baselineremoval processing.

[0031] Figure 2 The classification model discrimination results of plastic samples: (A) Confusion matrix of infrared spectroscopy PCA-RF model training set results; (B) Confusion matrix of infrared spectroscopy PCA-RF model test set results; (C) Confusion matrix of ultraviolet spectroscopy PCA-RF model training set results; (D) Confusion matrix of ultraviolet spectroscopy PCA-RF model test set results.

[0032] Figure 3 This is the rmse of the three models after parameter adjustment: PCA-RF, PCA-SVM, and PCA-XGboost.

[0033] Figure 4-Figure 6 The quantitative model determination results for plastic samples. Figure 4 The quantitative model determination results for the 260-800nm ​​band; Figure 5 It is the quantitative model determination result of 400-800nm ​​band; Figure 6 This is the quantitative model determination result for the 500-700nm band. DETAILED DESCRIPTION

[0034] The present invention is further described in detail below in conjunction with examples, but the embodiments of the present invention are not limited thereto. The materials involved in the following examples can be obtained from commercial channels unless otherwise specified. The methods described are conventional methods unless otherwise specified.

[0035] In one embodiment, a quantitative detection method for recycled plastics based on instrumental analysis and machine learning comprises: obtaining small organic molecules in a plastic sample with a known recycled plastic content and analyzing the sample to obtain analysis result data; processing the data to obtain characteristic data, and constructing a regression model based on machine learning using a database formed by characteristic data of different samples; using the same processing method for the sample to be tested to obtain characteristic data, and using the constructed model to detect the content of recycled plastic in the sample.

[0036] In a specific embodiment of the present invention, the organic small molecules in the sample can be obtained by conventional methods, such as solvent extraction, headspace method, dissolution precipitation method, etc. In one embodiment, the small organic molecules are obtained by solvent extraction; in another embodiment, the small organic molecules are obtained by headspace method; in another embodiment, the small organic molecules are obtained by dissolution precipitation method.

[0037] In a specific embodiment of the present invention, the organic small molecules in the sample can be obtained by solvent extraction.

[0038] In a specific embodiment of the present invention, methods for analyzing organic small molecules may include spectroscopy, chromatography-mass spectrometry and the like.

[0039] In a specific embodiment of the present invention, spectroscopy may include ultraviolet-visible spectrophotometry (UV-Vis), infrared spectroscopy (IR), and the like.

[0040] In one embodiment, the method for analyzing small organic molecules is ultraviolet-visible spectrophotometry (UV-Vis); in another embodiment, the method for analyzing small organic molecules is infrared spectroscopy (IR); in yet another embodiment, the method for analyzing small organic molecules is chromatography-mass spectrometry.

[0041] In a specific embodiment of the present invention, when ultraviolet-visible spectrophotometry (UV-Vis) is used for analysis, the analysis result data is preferably selected from a band with obvious differences, such as 500-700 nm.

[0042] In a specific embodiment of the present invention, processing the data refers to preprocessing the analysis result data, such as smoothing, first-order derivative and baseline removal.

[0043] In a specific embodiment of the present invention, the analysis result data is further processed, such as using principal component analysis (PCA), uniform manifold approximation and projection (UMAP) and other dimensionality reduction methods to process the data.

[0044] In a specific embodiment of the present invention, the establishment of a machine learning regression model uses a variety of different machine learning algorithms, such as random forest (RF), support vector machine (SVM), extreme gradient boosting (XGboost), etc., to establish multiple regression models. The model training is based on sample data with known recycled plastic content, and hyperparameters are optimized through 10-fold cross validation.

[0045] In one embodiment, the machine learning regression model is established using the machine learning algorithm random forest (RF); in another embodiment, the machine learning regression model is established using the machine learning algorithm support vector machine (SVM); in yet another embodiment, the machine learning regression model is established using the machine learning algorithm extreme gradient boosting (XGboost).

[0046] In the technical solution of the present invention, the optimal model can be selected from multiple regression models for use. In one embodiment, the machine learning regression model is established using a machine learning algorithm random forest (RF).

[0047] In a specific embodiment of the present invention, the database is constructed based on characteristic data of plastic samples with different contents of known recycled plastics, including virgin plastic samples and recycled plastic samples.

[0048] In a specific embodiment of the present invention, the number of samples in the database is greater than or equal to 100, preferably greater than or equal to 200.

[0049] In a specific embodiment of the present invention, the quantitative ratio of recycled plastic samples: (original plastic samples + recycled plastic samples) in the database may be 1:3-1:8.

[0050] In a specific embodiment of the present invention, the type of recycled plastic in the sample to be tested is consistent with the recycled plastic used to construct the database.

[0051] In a specific embodiment of the present invention, the plastic sample can be polyethylene terephthalate (PET), polyethylene (PE), polypropylene (PP), etc. In one embodiment, the plastic sample is polyethylene terephthalate (PET); in another embodiment, the plastic sample is polyethylene (PE); in another embodiment, the plastic sample is polypropylene (PP).

[0052] The technical solution of the present invention is described in detail below by taking the determination of samples with different recycled PET contents by solvent extraction combined with UV-Vis method as an example.

[0053] The solvent extraction method is exemplified by dichloromethane extraction method. The PET sample is added with dichloromethane and subjected to ultrasonic heating treatment to obtain an extraction solution.

[0054] The heating treatment condition may be 45-55°C for 0.5-2h.

[0055] The mass volume ratio of PET sample and dichloromethane can be 1:3-1:10, g, mL.

[0056] PET samples can be frozen and crushed with liquid nitrogen before extraction to achieve better solid-liquid extraction effects.

[0057] The extract was analyzed and tested by UV-vis, and the obtained data was preprocessed, including smoothing, first-order derivative and baseline removal. The spectral interval was selected as the entire range or a band with obvious differences, such as 500-700nm.

[0058] The preprocessed data are further processed using dimensionality reduction methods such as principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) to obtain UV-vis characteristic data.

[0059] The model was established by using a variety of machine learning algorithms, such as random forest (RF), support vector machine (SVM), extreme gradient boosting (XGboost), etc., to establish multiple regression models with different recycled plastic contents. The model was trained based on the training set data, and the hyperparameters were optimized through 10-fold cross validation. The model accuracy was verified using the validation set data. Or after multiple models were established, the optimal model could be selected, and the final model could be established using the training set data. The model accuracy was verified using the validation set data.

[0060] When modeling, the number of PET samples with known recycled plastic content (0-100%) is preferably greater than or equal to 100, preferably 200; in the PET samples, the ratio of recycled plastic samples to (original plastic samples + recycled plastic samples) can be 1:3-1:8. A UV-vis characteristic database of PET samples with known recycled plastic content is obtained, and the database is divided into a training set and a validation set in a ratio of 4:1, which are used for model training and validation, respectively.

[0061] The PET sample with unknown recycled plastic content is extracted with the above-mentioned solvent and processed by UV-Vis method to obtain UV-vis characteristic data, and then the recycled plastic content in the PET sample can be determined by using the above-mentioned established model for detection.

[0062] Example 1: Determination of virgin and recycled PET samples by different analytical methods

[0063] 1.Analysis method: ATR-FTIR, UV-visible spectroscopy (UV-vis)

[0064] 2. Samples: There are 139 PET samples in total, including 66 batches of 100% recycled PET samples and 73 batches of virgin PET samples.

[0065] 3. Sample testing

[0066] The above samples were tested by ATR-FTIR method and UV-vis method respectively;

[0067] 3.1ATR-FTIR method detection: Use the Fourier transform infrared spectrometer supporting software FTIR-850 to control the operation of the instrument and test the spectrum of the particle samples in turn. Set the spectrum parameters as follows: Scan range 4000~600cm -1 The number of spectral scans was set to 16 and the resolution was 4 cm -1 , the Y-axis format is set to transmittance, and other parameters are set by default. Use air as the background, and collect the background before each sample collection. The ambient temperature is 18℃~21℃, the ambient humidity is 38%~50%, and the humidity difference between the collected background and the collected sample is controlled to be no more than 2%.

[0068] 3.2 UV-vis sample pretreatment: Use a high-throughput tissue grinder to grind the sample into powder to obtain sample powder; then add 6 mL of dichloromethane to each sample in a ratio of 1 g, seal it, and ultrasonically heat it at 50°C for 1 h. Use a 0.22 μm filter membrane to filter the extract, and the extract is used for subsequent UV-vis analysis.

[0069] UV-vis detection: The spectrum of the extract is measured by UV-visible spectrophotometer, and the paired control instrument is used to set the parameters in the acquisition and run, the spectrum absorption range is set to 260-800nm, the sampling interval is 1.0nm, the vertical axis of the spectrum is set to transmittance, and dichloromethane is used as a blank control. The above samples are tested separately, and finally all the spectral curves are derived and sorted, where the X axis is the wavelength and the Y axis is the transmittance.

[0070] 4. Raw data preprocessing: Some bands without peaks or with mixed peaks in the infrared and ultraviolet spectra will affect the results. The infrared and ultraviolet spectra of the samples are preprocessed by smoothing and baseline removal (see the results). Figure 1 ) to obtain the full-band data matrix.

[0071] 5. Modeling data preprocessing: Establish a PCA model for the data matrix to obtain the corresponding dimensionality reduction data, which has a lot of intersections.

[0072] 6. Quantitative model establishment - classification model: Divide the data set after PCA dimensionality reduction into training set and test set. Use random sampling method to select 20% samples from each type of samples to form the test set, and the rest as the training set. Use the training set data to build the random forest (RF) classification model, and use 10-fold cross validation to optimize the hyperparameters to obtain the accuracy of model training, and use all training data to retrain the model. Finally, use the test set data to examine the discrimination accuracy of the model. Figure 1 are the original data and preprocessed data results, Figure 2 The results are for the training set and the test set.

[0073] From the results, it can be seen that the PCA model based on UV-visible spectroscopy of the present invention can achieve 100% accuracy in qualitative identification of recycled plastics, and compared with infrared spectroscopy, the UV-visible spectroscopy selected in the present invention has a more significant identification effect and has greater potential for quantitative discrimination.

[0074] Example 2: Determination of the content of recycled PET samples based on different bands of UV-visible spectroscopy and machine learning

[0075] (1) Samples: There were 239 PET samples in total, including 66 batches of 100% recycled PET samples, 73 batches of virgin PET samples, and PET samples with different recycled contents (obtained by mixing recycled PET samples with virgin PET samples in proportion), including 20 batches of 10%, 30%, 50%, 70%, and 90% recycled samples.

[0076] (2) Sample pretreatment and UV-vis detection: the method is the same as in Example 1

[0077] (3) Raw data preprocessing: The UV-visible spectrum of the sample was smoothed and baseline removed, and the full band (260nm-800nm), 400-800nm ​​band, and 500-700nm were selected to form their respective corresponding matrices.

[0078] (4) Modeling data preprocessing: Establish a PCA model for the data matrix to obtain the corresponding dimensionality reduction data.

[0079] (5) Quantitative model establishment - regression model: The data set after PCA dimensionality reduction was divided into a training set and a test set. A random sampling method was used to select 20% of the samples from 100% recycled PET samples, original PET samples, and PET samples with different recycled contents to form the test set, and the rest were used as the training set. The training set data was used to build three regression models: support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGboost), and 10-fold cross validation was used to optimize the hyperparameters. The results are shown in Figure 3 Then, the optimal model (PCA-RF model) was selected based on the mean square error (MSE), and the model was retrained using all the training data. Finally, the mean square error of the model was examined using the test set data.

[0080] The content of the tested PET sample is obtained from the prediction results based on the regression model. Figure 4-Figure 6 Table 1 shows the evaluation index results of different bands of the PCA-RF model training set, which has the best effect among the three models. The root mean square error (RMSE) value of the whole band (260nm-800nm) is 4.84, R 2 The value is 0.987; the root mean square error (RMSE) value in the 400-800nm ​​band is 4.7, R 2 The value is 0.988; the root mean square error (RMSE) value in the 500-700nm band is 4.08, R 2 The value is 0.991. As shown in Table 1, the full band can provide a relatively complete range of predictions for machine learning, and selecting characteristic bands for modeling can obtain better model effects.

[0081] The test set results of the PCAA-RF model established in the 500-700 nm band are shown in Table 2, which proves that the content of recycled plastic in samples containing recycled plastic can be quantitatively detected based on the method of the present invention.

[0082] Table 1

[0083] Select Band Root mean square error (rmse) <![CDATA[R 2 ]]> 260-800nm 4.84 0.987 400-800nm 4.7 0.988 500-700nm 4.08 0.991

[0084] Table 2

[0085]

[0086]

[0087] As can be seen from the above, the model based on the instrumental testing method combined with machine learning of the present invention can quantitatively determine the content percentage of the recycled plastic sample. Compared with the infrared spectroscopy method, the content of recycled PET determined based on different bands of ultraviolet-visible spectroscopy and machine learning has a more significant effect. The method of the present invention can accurately, quickly and efficiently determine the content of recycled plastics.

[0088] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A quantitative detection method for recycled plastics based on instrumental analysis and machine learning, characterized in that include: Obtaining and analyzing organic small molecules in plastic samples with known recycled plastic content to obtain analysis result data; The data is processed to obtain characteristic data, and a regression model is constructed based on machine learning using a database formed by characteristic data of different samples; The samples to be tested were processed in the same way to obtain characteristic data, and the constructed model was used to detect the content of recycled plastic in the samples.

2. The method for quantitative detection of recycled plastics based on instrumental analysis and machine learning according to claim 1, characterized in that: The organic small molecules in the sample are obtained by solvent extraction, headspace method or dissolution precipitation method.

3. The method for quantitative detection of recycled plastics based on instrumental analysis and machine learning according to claim 1, characterized in that: The method for analyzing organic small molecules includes at least one of spectroscopy and chromatography-mass spectrometry.

4. The method for quantitative detection of recycled plastics based on instrumental analysis and machine learning according to claim 1, characterized in that: Processing the data refers to preprocessing the analysis result data, including at least one of smoothing, first-order derivative and baseline removal.

5. The method for quantitative detection of recycled plastics based on instrumental analysis and machine learning according to claim 1, characterized in that: The analysis result data is further processed by using principal component analysis, or uniform manifold approximation and projection dimensionality reduction methods.

6. The method for quantitative detection of recycled plastics based on instrumental analysis and machine learning according to claim 1, characterized in that: The establishment of machine learning regression models uses a variety of different machine learning algorithms, including at least one of random forest, support vector machine, and extreme gradient boosting, to establish multiple regression models.

7. The method for quantitative detection of recycled plastics based on instrumental analysis and machine learning according to claim 1, characterized in that: The database is obtained based on characteristic data of plastic samples with different contents of known recycled plastics; Includes virgin plastic samples and recycled plastic samples.

8. The method for quantitative detection of recycled plastics based on instrumental analysis and machine learning according to claim 1, characterized in that: The number of samples in the database is greater than or equal to 100.

9. The method for quantitative detection of recycled plastics based on instrumental analysis and machine learning according to claim 1, characterized in that: The quantitative ratio of recycled plastic samples: (original plastic samples + recycled plastic samples) in the database is 1:3-1:

8.

10. Application of the quantitative detection method for recycled plastics based on instrumental analysis and machine learning as described in any one of claims 1 to 9 in the identification of recycled plastics and virgin plastics.

Images