WO2024130805A1 - Identification method and system for polyethylene terephthalate recycled material - Google Patents

Abstract

An identification method and system for a polyethylene terephthalate (PET) recycled material. The method comprises: acquiring a PET sample; analyzing the PET sample to obtain feature data of the PET sample, the feature data comprising chain extender data and depolymerization product data; using the feature data to train a preset decision tree model until the preset decision tree model reaches a preset convergence condition, so as to obtain a PET recycled material identification model; and using the PET recycled material identification model to identify a PET sample to be identified, so as to determine the recycled material condition of said sample. Using chain extenders and depolymerization products to trace molecular structures and incorporating a model autonomous decision improve the accuracy of identification results of PET recycled materials.

Description

A method and system for identifying polyethylene terephthalate recycled materials Technical Field

The invention relates to the technical field of recycled material identification, and in particular to a method and system for identifying recycled polyethylene terephthalate materials.

Background technique

Recycled plastics (or recycled materials) have low carbon properties, and using recycled materials to improve the recycling rate of plastics has become an effective means of sustainable development. However, considering that the material reliability and quality of recycled materials have different degrees of deterioration, recycled materials are not suitable for all fields. Therefore, in order to trace the source of the product, it is necessary to identify whether the product contains recycled materials. Among them, the recycling volume of polyethylene terephthalate (PET) accounts for 32% of all plastic recycling volume, and is the main object of recycled material identification.

At present, the existing identification methods of recycled materials mainly rely on instrumental analysis combined with manual identification. For example, the oligomers in polyester are physically recovered in advance by dissolution-mixed precipitant precipitation-filtration method, dissolution-single precipitant precipitation filtration method or extraction method, and the oligomer peak information of polyester is detected by high performance liquid chromatography, and then the identification result is obtained manually based on the oligomer peak information; for another example, the detection signal of methyl benzoate para-substituent is obtained by high performance liquid chromatography, and then the identification result is obtained manually based on the detection signal; for another example, a sulfolane-dichloromethane dissolution precipitation extraction chemically recovers the first sequence of cyclic oligomers of recycled polyester, and then uses high performance liquid chromatography for identification. The above identification methods of recycled materials are mainly simple combinations of one or several different test methods, and the accuracy of the identification results is greatly affected by personal experience. Moreover, the material differences of recycled materials lead to different weights of the measurement results of different methods in the final result judgment, and it is difficult for the existing methods to reasonably quantify this, resulting in a certain subjectivity and inaccuracy in the identification results.

Summary of the invention

The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art and provide a method and system for identifying polyethylene terephthalate recycled materials, so as to reduce the subjective influence of the user's personal experience on the identification results of PET recycled materials and improve the accuracy of the identification results.

In order to solve the above technical problems, in a first aspect, the present invention provides a method for identifying polyethylene terephthalate recycled materials, comprising:

Obtaining PET samples, wherein the PET samples include PET recycled material samples and PET virgin material samples;

Performing sample analysis on the PET sample to obtain characteristic data of the PET sample, wherein the characteristic data includes but is not limited to chain extender data and depolymerization product data;

Using the characteristic data, training a preset decision tree model until the preset decision tree model reaches a preset convergence condition, thereby obtaining a PET recycled material identification model;

The PET recycled material identification model is used to identify the PET sample to be identified, and the recycled material situation of the PET sample to be identified is determined.

In some implementations, for the chain extender data and the depolymerization product data, sample analysis includes sample analysis based on thermal cracking and sample analysis based on alcohol depolymerization.

In some implementations, for the sample analysis based on thermal pyrolysis, performing sample analysis on the PET sample to obtain characteristic data of the PET sample includes:

The PET sample is detected by using a pyrolysis gas chromatography-mass spectrometer to obtain pyrolysis gas chromatography-mass spectrometer data, wherein the pyrolysis gas chromatography-mass spectrometer data includes chain extender data and depolymerization product data, wherein the thermal desorption temperature of the detection process is 250-300° C., maintained for 20 min, and the thermal cracking temperature of the detection process is 560-600° C., maintained for 0.2 min.

In some implementations, for sample analysis based on alcohol depolymerization, performing sample analysis on the PET sample to obtain characteristic data of the PET sample includes:

In the presence of an alcohol solvent and a catalyst, the PET sample is subjected to a depolymerization reaction using a stainless steel reactor or a microwave high-pressure synthesis reactor at a reaction temperature of 180-250° C. and a reaction time of 2-3 hours to obtain a PET sample solution;

After filtering and diluting the PET sample solution, the material components in the PET sample solution are detected by GCMS, LCMS or Py-GCMS to obtain spectrum data;

Using a preset chain extender GCMS database, LCMS database or Py-GCMS database, the spectrum data is searched for fragment ion peaks to obtain the chain extender data;

The spectral data are searched for fragment ion peaks using a preset depolymerization product GCMS database, LCMS database or Py-GCMS database to obtain the depolymerization product data.

In some implementations, the chain extender data includes chain extender data of epoxy resins, isocyanates, anhydrides, oxazolines, and amides related to the chain extender.

In some implementations, the feature data is used to train a preset decision tree model until the preset decision tree model reaches a preset convergence condition to obtain a PET recycled material identification model, including:

Performing dimension reduction on the feature data to generate a training sample set;

Using the training sample set, the preset decision tree model is trained, and the Gini coefficient of each feature in the training sample set at all possible split values is calculated;

Taking the feature data based on the minimum Gini coefficient as the optimal feature node, taking the segmentation value based on the minimum Gini coefficient as the optimal subspace, and establishing the left node data set and the right node data set when the optimal feature node is used as the root node;

Generate child nodes of the decision tree according to the left node data set and the right node data set until no more child nodes can be generated, thereby obtaining an original decision tree;

The original decision tree is pruned to obtain the PET recycled material identification model.

In some implementations, reducing the dimension of the feature data to generate a training sample set includes:

Based on the feature data, establish a feature data set;

Normalizing and zero-meaning the feature data set to obtain a target feature data set;

Based on the target feature data set, a covariance matrix is calculated, and eigenvalues and corresponding eigenvectors of the covariance matrix are determined;

Based on the sorting results of the eigenvalues, a number of corresponding eigenvectors are selected to form an eigenvector matrix;

The feature data is reduced in dimension based on the feature vector matrix to generate the training sample set.

In some implementations, the calculation function of the Gini coefficient is:

Among them, Gin(D,Ai) is the Gini coefficient of feature Ai in training sample set D, D1 is the data set composed of feature data when the cut value in training sample set D is not greater than ai, D2 is the data set composed of feature data when the cut value in training sample set D is not less than ai, Gini(D1) is the Gini coefficient of data set D1, and Gini(D2) is the Gini coefficient of data set D2.

In some implementations, generating child nodes of a decision tree according to the left node data set and the right node data set until no more child nodes can be generated to obtain an original decision tree includes:

Generate a new child node according to the left node data set and the right node data set as the next root node data;

The child node data set corresponding to the new child node is used as the next root node data set, and new child nodes are continuously generated until no more child nodes can be generated, thereby obtaining the original decision tree.

In some implementations, pruning the original decision tree to obtain the PET recycled material identification model includes:

Based on the preset loss function, the original decision tree is optimized until the original decision tree reaches the preset convergence condition, thereby obtaining the PET recycled material identification model, wherein the preset loss function is:

C _α (T _t ) = C (T _t ) + α |T _t |;

Wherein, T _t represents any node in the original decision tree, C(T _t ) represents the prediction error of node T _t , |T _t | represents the number of nodes, and α is a preset training parameter.

In some implementations, the step of using the PET recycled material identification model to identify the PET sample to be identified and determining the recycled material status of the PET sample to be identified includes:

Performing sample analysis on the PET sample to be identified to obtain target feature data of the PET sample to be identified;

Inputting the target feature data into the PET recycled material identification model and outputting a decision value;

If the decision value satisfies the preset decision value interval, it is determined that the PET sample to be identified contains recycled material.

In a second aspect, the present invention further provides a system for identifying recycled polyethylene terephthalate materials, comprising:

An acquisition module is used to acquire PET samples, wherein the PET samples include PET recycled material samples and PET virgin material samples;

An analysis module, used for performing sample analysis on the PET sample to obtain characteristic data of the PET sample, wherein the characteristic data includes but is not limited to chain extender data and depolymerization product data;

A modeling module, used to train a preset decision tree model using the feature data until the preset decision tree model reaches a preset convergence condition, thereby obtaining a PET recycled material identification model;

The identification module is used to identify the PET sample to be identified by using the PET recycled material identification model, and determine the recycled material situation of the PET sample to be identified.

Compared with the prior art, the present invention has at least the following beneficial effects:

The present invention uses the determination of chain extenders and depolymerization products to trace back to the molecular structure level of PET plastics, thereby improving the accuracy and efficiency of identification. A decision tree model is used as an identification model, and the model makes decisions independently, reducing the influence of subjective factors caused by human decision-making, thereby improving the accuracy of PET recycled material identification, especially more reliable identification of modified PET recycled material.

In addition, the present invention reduces the number of features of the model through dimensionality reduction to improve the model operation efficiency; through pruning optimization operations, it reduces the problem of model overfitting and enhances the generalization ability of the model.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic flow diagram of a method for identifying polyethylene terephthalate recycled materials according to an embodiment of the present invention;

FIG2 is a schematic diagram of a characteristic infrared spectrum shown in an embodiment of the present invention;

FIG3 is a schematic diagram of a thermal weight loss curve shown in an embodiment of the present invention;

FIG4 is a schematic diagram of an element distribution spectrum shown in an embodiment of the present invention;

FIG5 is a schematic diagram of a thermal desorption Py-GCMS test spectrum shown in an embodiment of the present invention;

FIG6 is a schematic diagram of a GCMS test spectrum of an alcoholysis product shown in an embodiment of the present invention;

FIG. 7 is a structural block diagram of a polyethylene terephthalate recycled material identification system according to an embodiment of the present invention.

Detailed ways

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Please refer to Figure 1. The present invention provides a method for identifying polyethylene terephthalate recycled materials, which can be applied to an identification system for polyethylene terephthalate recycled materials. The identification system can be integrated into a computer device, and the computer device includes but is not limited to tablet computers, laptop computers, desktop computers, physical servers, cloud servers and other devices. The computer device can be connected to a gas chromatograph-mass spectrometer GCMS, a liquid chromatograph-mass spectrometer LCMS, a pyrolysis gas chromatograph-mass spectrometer Py-GCMS and other instruments. The method includes steps S101 to S104, which are described in detail as follows:

Step S101, obtaining PET samples, wherein the PET samples include PET recycled material samples and PET virgin material samples;

Step S102, performing sample analysis on the PET sample to obtain characteristic data of the PET sample, wherein the characteristic data includes but is not limited to chain extender data and depolymerization product data;

Step S103, using the characteristic data, training a preset decision tree model until the preset decision tree model reaches a preset convergence condition, thereby obtaining a PET recycled material identification model;

Step S104, using the PET recycled material identification model to identify the PET sample to be identified, and determining the recycled material status of the PET sample to be identified.

In the present embodiment, since the PET recycled material is affected by light, oxygen, heat, humidity, stress, and environmental substances during the service stage, side reactions such as hydrolysis, pyrolysis, and oxidative degradation may occur during the recycling process, resulting in a certain degree of difference between its molecular chain, molecular structure, performance, and the characteristics of the pollutants contained therein and the new material. Under normal circumstances, the molecular weight of PET will decrease after recycling and cause performance degradation, so the PET recycled material is usually added with a bifunctional chain extender to increase the molecular weight or characteristic viscosity of the PET recycled material. Therefore, the present embodiment separates the chain extender and the residual pollutants, organic impurities, degradation products, oligomers, etc. remaining in the PET through a depolymerization reaction, and then establishes a PET recycled material identification model by detecting the subtle changes in the composition, structure, and performance of these materials in the PET recycled material, and identifies whether the unknown sample is a PET recycled material through the PET recycled material identification model.

Optionally, the PET samples include PET recycled materials and PET virgin materials from various manufacturers. Sample analysis includes, but is not limited to, spectral analysis, chromatographic analysis, mass spectrometry analysis, infrared spectrum analysis, and thermal analysis. Characteristic data also includes, but is not limited to, characteristic infrared spectrum data, thermogravimetric data, elemental data, melt flow rate data, morphology analysis data, fluorescent brightener data, and pyrolysis gas chromatography-mass spectrometry data.

Optionally, for the characteristic infrared spectrum data, an infrared spectrometer is used to measure the characteristic infrared spectrum of the PET sample. The infrared detection parameters of the infrared spectrometer include 32 scan times, the characteristic spectrum band is the average value of 32 scans, the scanning range is 4000-400 cm ^-1 , the interval of the data point is 0.5 cm ^-1 , the room temperature and humidity during scanning are controlled at 25°C/50RH respectively, and each sample is scanned once.

Optionally, for the thermogravimetric data, a TGA thermogravimetric curve of the PET sample is measured using a thermogravimetric analyzer. The test conditions of the thermogravimetric analyzer include: in a nitrogen atmosphere, after being kept at 30°C for 5 minutes, the temperature of the environment in which the PET sample is located is increased from 30°C to an aging temperature (330-380°C, preferably 340°C) at a heating rate of 20°C/min; or in an air atmosphere, the temperature is kept at the aging temperature for 60 minutes.

Optionally, for elemental data, an X-ray fluorescence spectrometer is used to scan the sample to obtain an element distribution spectrum of the sample, and the types and total number of detected elements such as antimony, bromine, titanium, etc. are analyzed and recorded based on the element distribution spectrum.

Optionally, for the melt flow rate data, a melt indexer is used to measure the melt flow rate data of the PET sample. The test conditions of the melt flow rate are a temperature of 265° C., a weight of 2.165 kg, a cutting time of 5 s, and drying at 120° C. for 15 h, and the melt flow rate MFR is obtained by testing.

Optionally, for the morphological analysis data, an optical microscope or a scanning electron microscope is used to record the number of noisy spots per 50 mm ² of the PET sample according to a five-point area method.

Optionally, for the fluorescent brightener data, ultraviolet spectrophotometry and fluorescence spectrophotometry are used to perform qualitative and quantitative analysis on the fluorescent substances.

Optionally, for pyrolysis gas chromatography-mass spectrometry data, the PET sample is detected by pyrolysis gas chromatography-mass spectrometry to obtain the pyrolysis gas chromatography-mass spectrometry data, wherein the thermal desorption temperature of the detection process is 250-300°C, maintained for 20 minutes, and the thermal cracking temperature of the detection process is 560-600°C, maintained for 0.2 minutes.

Furthermore, a mass spectrometry database of thermal desorption and thermal cracking substances of PET recycled materials was established to facilitate the analysis of thermal desorption and thermal cracking substances of the samples to be tested in the application stage. Among them, the characteristic fragment ion peaks include fragment ion peaks of substances such as benzene 3-(4-(tert-butyl)phenyl)-2-methylpropanal, biphenyl, straight-chain hydrocarbons (such as pentadiene, 1-hexene, 1-heptene and 3-eicosene), 3-phenyl-2H-chromene, 2-methoxyethanol, 1,2-dimethoxyethane, 1,2-ethylene glycol and diethylene glycol monoethyl ether.

Optionally, for the chain extender data and the depolymerization product data: in the presence of an alcohol solvent and a catalyst, a stainless steel reactor or a microwave high-pressure synthesis reactor is used to perform a depolymerization reaction on the PET sample, the reaction temperature is 180-250°C, and the reaction time is 2-3h to obtain a PET sample solution; after filtering and diluting the PET sample solution, GCMS, LCMS or Py-GCMS is used to detect the material components in the PET sample solution to obtain spectrum data; using a preset chain extender GCMS database, LCMS database or Py-GCMS database, a fragment ion peak search is performed on the spectrum data to obtain the chain extender data; using a preset depolymerization product GCMS database, LCMS database or Py-GCMS database, a fragment ion peak search is performed on the spectrum data to obtain the depolymerization product data.

The chain extender data include but are not limited to chain extender data of epoxy resins, isocyanates, anhydrides, oxazolines and amides.

Optionally, the alcohol solvent used in the depolymerization process can be one or more of methanol, ethanol, ethylene glycol, propylene glycol, 1,4-butanediol, etc., and the catalyst can be one or more of zinc acetate, magnesium acetate, manganese acetate, cobalt acetate, antimony acetate, titanium acetate, tin acetate, germanium acetate, lead acetate, ionic liquid and the like.

In some embodiments, the step S103 includes:

Performing dimension reduction on the feature data to generate a training sample set;

Using the training sample set, the preset decision tree model is trained, and the Gini coefficient of each feature in the training sample set at all possible split values is calculated;

Taking the feature data based on the minimum Gini coefficient as the optimal feature node, taking the segmentation value based on the minimum Gini coefficient as the optimal subspace, and establishing the left node data set and the right node data set when the optimal feature node is used as the root node;

Generate child nodes of the decision tree according to the left node data set and the right node data set until no more child nodes can be generated, thereby obtaining an original decision tree;

The original decision tree is pruned to obtain the PET recycled material identification model.

In this embodiment, PCA is used to reduce the number of features of the model and improve the model operation efficiency. At the same time, the established decision tree model is pruned and optimized to reduce the overfitting problem of the model and enhance the generalization ability of the model.

Optionally, reducing the dimension of the feature data to generate a training sample set includes:

Based on the feature data, establish a feature data set;

Normalizing and zero-meaning the feature data set to obtain a target feature data set;

Based on the target feature data set, a covariance matrix is calculated, and eigenvalues and corresponding eigenvectors of the covariance matrix are determined;

Based on the sorting results of the eigenvalues, a number of corresponding eigenvectors are selected to form an eigenvector matrix;

The feature data is reduced in dimension based on the feature vector matrix to generate the training sample set.

In this optional embodiment, principal component analysis (PCA) is used to reduce the dimension of the collected input data set: the feature data of the PET recycled material are combined into a feature data set, expressed as Y=[y ₁ ,y ₂ ,…,y _N ]∈RN ^×M , where N represents the number of data and M represents the number of features. Based on the z-score standardization method, the data is normalized; the normalized feature data is zero-meaned, that is, the feature mean is first calculated, then the mean is subtracted to obtain the target feature data set, and the covariance matrix is calculated. The eigenvalues and corresponding eigenvectors of the covariance matrix are calculated, the eigenvalues are sorted from large to small, the first k eigenvalues are selected, and then the k eigenvectors corresponding to the selected eigenvalues are used as column vectors to form an eigenvector matrix. The original feature data is reduced in dimension according to the eigenvector matrix, and new feature data are calculated as training samples.

Optionally, the calculation function of the Gini coefficient is:

Among them, Gini(D,Ai) is the Gini coefficient of the feature data Ai in the training sample set D, D1 is the data set composed of the feature data when the cut value in the training sample set D is not greater than ai, D2 is the data set composed of the feature data when the cut value in the training sample set D is not less than ai, Gini(D1) is the Gini coefficient of the data set D1, and Gini(D2) is the Gini coefficient of the data set D2.

In this optional embodiment, the decision tree model includes but is not limited to ID3 algorithm, C4.5 algorithm, classification and regression tree (CART) algorithm and random forest (RF) algorithm.

For example, the decision tree model takes the CART decision tree model as an example, selects the Gini coefficient as the feature standard, and constructs the model as follows: delete missing values from the training sample set D, and use the preprocessed training sample set D for training. Assuming that the feature data in the training sample set D has K categories, the probability of the Kth category is p _k , and the Gini coefficient of the feature data is calculated

Among them, for each feature data Ai, each classification value (i.e., target segmentation value) ai that may be obtained is obtained. According to ai, each feature data is divided into two parts, wherein all feature data with a segmentation value not greater than ai form a data set D1, and all feature data with a segmentation value not less than ai form a data set D2. The Gini coefficient corresponding to the feature data Ai is obtained by the above-mentioned Gini coefficient calculation function.

Optionally, generating child nodes of a decision tree according to the left node data set and the right node data set until no more child nodes can be generated, thereby obtaining an original decision tree, including:

Generate a new child node according to the left node data set and the right node data set as the next root node data;

The child node data set corresponding to the new child node is used as the next root node data set, and new child nodes are continuously generated until no more child nodes can be generated, thereby obtaining the original decision tree.

In this optional embodiment, each feature data Ai in the training sample set D is traversed, and the Gini coefficients of all possible split values ai under the feature data are calculated. The feature data A' and split value a' that minimize the Gini coefficient are selected as the optimal feature node and the optimal subspace, respectively. The feature data A' and split value a' divide the training sample set D into two parts, and establish the left and right nodes with the current feature node as the root node, where the left node data set D1' and the right node data set D2', and the root node is the data set containing all sample features. Starting from the root node, the generated child node data set (left node data set and right node data set) is used as the next root node data set, and new child nodes are continuously generated until no child nodes can be generated, and the original decision tree T ₀ constructed by multiple feature nodes is obtained.

Optionally, five-fold cross validation is used to evaluate the model performance of the original decision tree model.

Optionally, pruning the original decision tree to obtain the PET recycled material identification model includes:

Based on the preset loss function, the original decision tree is optimized until the original decision tree reaches the preset convergence condition, thereby obtaining the PET recycled material identification model, wherein the preset loss function is:

C _α (T _t ) = C (T _t ) + α |T _t |;

Wherein, T _t represents any node in the original decision tree, C(T _t ) represents the prediction error of node T _t , |T _t | represents the number of nodes, and α is a preset training parameter.

In this optional embodiment, the original decision tree is pruned and optimized based on the loss function C _α (T _t ) = C (T _t ) + α |T _t |. The pruning principle is that the loss function of the child node is less than the loss function of the root node, that is,

Then the minimum value of α is

In some embodiments, the method of using the PET recycled material identification model to identify the PET sample to be identified and determining the recycled material status of the PET sample to be identified includes:

Performing sample analysis on the PET sample to be identified to obtain target feature data of the PET sample to be identified;

Inputting the target feature data into the PET recycled material identification model and outputting a decision value;

If the decision value satisfies the preset decision value interval, it is determined that the PET sample to be identified contains recycled material.

In this embodiment, the sample to be identified is subjected to spectral analysis, chromatographic analysis, mass spectrometry analysis, infrared analysis and thermal analysis, and the spectrum obtained by the analysis is digitally converted to obtain target feature data reflecting the characteristics of the sample spectrum. PCA is used to reduce the dimension of the target feature data, and then the trained PET recycled material identification model is used for identification, and a decision value is output.

As an example and not a limitation, the analysis process of whether an unknown sample is PET recycled material is as follows:

step 1:

Infrared spectrum analysis of PET samples from different manufacturers: KBr salt tablet method was used, the number of scans was set to 32 times, the characteristic spectrum band was the average of 32 scans, the scanning range was 4000-400cm ^-1 , the interval of data points was 0.5cm ^-1 , the room temperature and humidity were controlled at 25℃/50RH respectively during collection, and the spectrum of each sample was collected once. The results are shown in Figure 2.

The TGA thermal gravimetric curves of PET samples from different manufacturers were measured: the TGA test conditions were as follows: in a nitrogen atmosphere, the temperature was kept constant at 30°C for 5 minutes, and then the temperature was increased from 30°C to an aging temperature of 340°C at a rate of 20°C/min; in an air atmosphere, the temperature was kept constant at the aging temperature for 60 minutes, and the TGA weight loss 5% data of the PET recycled material was obtained, and the data was normalized using the z-score standardization method. The results are shown in Figure 3.

XRF elemental analysis of PET samples from different manufacturers: XRF was used to scan the samples to obtain the element distribution spectrum of the samples, and the types and total number of elements detected, such as antimony, bromine, titanium, etc., were recorded. The data were normalized using the z-score standardization method. The results are shown in Figure 4.

The melt flow rate data of PET samples from different manufacturers were measured: the melt flow rate test conditions included a temperature of 265°C, a weight of 2.165 kg, a cutting time of 5 s, and drying at 120°C-15 h, and the data were normalized using the z-score standardization method.

Morphological analysis of PET samples from different manufacturers: Using an optical microscope or scanning electron microscope, the number of variegated spots per ⁵⁰ mm2 was recorded according to the five-point area method, and the data were normalized using the z-score standardization method.

Determination of fluorescent brighteners in PET samples from different manufacturers: UV spectrophotometry and fluorescence spectrophotometry were used to qualitatively and quantitatively analyze fluorescent substances, and the z-score standardization method was used to normalize the data.

The pyrolysis gas chromatography-mass spectrometry (Py-GCMS) data of PET samples from different manufacturers were measured: thermal desorption temperature 250-300°C, maintained for 20 min, thermal cracking temperature 560-600°C, maintained for 0.2 min. A mass spectrum database of thermal desorption substances and thermal cracking substances of PET recycled materials was established. The characteristic fragment ion peaks of PET recycled materials include benzene 3-(4-(tert-butyl)phenyl)-2-methylpropanal fragment ion peaks, biphenyl, straight-chain hydrocarbons (such as pentadiene, 1-hexene, 1-heptene and 3-eicosene), 3-phenyl-2H-chromene, 2-methoxyethanol, 1,2-dimethoxyethane, 1,2-ethylene glycol and diethylene glycol monoethyl ether, etc., and the data were normalized using the z-score standardization method, and the results are shown in Figure 5.

Determine the chain extender data and depolymerization product data of PET samples from different manufacturers: weigh 1g of sample, put it into a stainless steel reactor, add 50ml of 0.02-0.04mg/mL zinc acetate methanol solution, heat at 200-220℃ for 2-3h, cool to room temperature, open the reactor to filter the liquid, filter it through a 0.45μm organic filter membrane into a 2ml sample bottle, and use GCMS and Py-GCMS to obtain mixed mass spectrometry data of ion peaks related to depolymerization products and chain extenders. The ion peaks related to chain extenders include relevant ion peaks of epoxy resin, isocyanate, anhydride, oxazoline, and amide. The obtained mass spectrometry data are normalized by the z-score standardization method.

GCMS and Py-GCMS were used to detect mixed mass spectrometry data of ion peaks related to the depolymerization products and chain extenders. The related ion peaks related to the depolymerization products included linear oligomers, cyclic oligomers, p-methylbenzene end groups, isophthalic acid, diethylene glycol, and bisphenol A. The obtained mass spectrometry data were normalized by the z-score standardization method, and the results are shown in Figure 6.

Step 2:

The normalized data is reduced in dimension using the PCA algorithm and used as the input of the decision tree model. The PCA dimension reduction process is as follows:

First, the features are zero-meaned, that is, the feature mean is first calculated, and then the mean is subtracted;

Calculate the covariance matrix C, the formula is

Find the eigenvalues and corresponding eigenvectors of the covariance matrix, sort the eigenvalues from large to small, select the first k, and then use their corresponding k eigenvectors as column vectors to form an eigenvector matrix P∈R ^N×M .

According to P, the original data is reduced in dimension and the new observation matrix is calculated.

As a training sample, the calculation formula is

Based on the randomly selected features of the training samples, multiple sub-training sets are obtained, and a judgment rule model based on the decision tree algorithm is established, in which the judgment criteria are the model input factors, and the output is verified by the training labels of whether it is the result of recycled materials manually.

The decision tree model uses the CART decision tree algorithm. The model is constructed as follows: missing values are deleted from the data set D, and the preprocessed data set D is trained. Assuming that there are K categories, the probability of the Kth category is p _k , and the standard Gini coefficient of the feature data is calculated.

Among them, for each feature data Ai, for each target segmentation value ai that may be obtained, each feature data is divided into two parts, wherein all feature data with a segmentation value not greater than ai form a data set D1, and all feature data with a segmentation value not less than ai form a data set D2. The Gini coefficient corresponding to the feature data Ai is obtained by the above-mentioned Gini coefficient calculation function.

Traverse each feature data Ai in the data set D, calculate the Gini coefficient of all possible split values ai under the feature set, select the feature A' and split value a' that minimize the Gini coefficient as the optimal feature node and optimal subspace, feature data A' and split value a', divide the original data set into two parts, and establish the left and right nodes of the current feature node, where the left node is the data set D1' and the right node is the data set D2'. At this time, the root node is the data set containing all sample features.

Starting from the root node, the generated child node data set is used as the next root node data set to continue generating child nodes until no child nodes can be generated, thereby generating the original decision tree T ₀ .

Through the above steps, a PET recycled material identification model based on the decision tree algorithm was constructed, and the model performance was evaluated using a five-fold cross validation. In the five-fold cross validation, the training samples were divided into five parts, four of which were used as training sets each time, and the remaining one was used as a validation set. The five subsets were used as validation sets in turn, and the crossover was repeated five times to obtain five results, and the average of the five results was used as the performance indicator of the classifier or model.

The original decision tree T ₀ is used to calculate the sub-node loss function and root node loss function of all internal nodes t from bottom to top, and the regularization parameter α is obtained. The minimum α' below this value is selected for pruning, and the new tree obtained is pruned in the same way until the root node of the original decision tree. The sub-tree sequence obtained after pruning is cross-validated to obtain the best one, and the complexity of the tree model is reduced.

Step 3:

The PET samples to be identified were analyzed to obtain the spectrum, chromatography, mass spectrum, infrared, thermal analysis and other data of the samples to be tested. The spectra were digitally converted to obtain the recycled material data reflecting the characteristics of the sample spectra. PCA was then used to reduce the data dimension, and the trained PET recycled material identification model was used for identification. The number of features after dimensionality reduction was 30, the number of CART decision trees was 500, and the average accuracy was 88.29%. The accuracy of the decision tree model after pruning optimization was increased by 3%.

It should be noted that: (1) the present invention proposes for the first time a method for identifying PET recycled materials by comprehensively utilizing the detection of depolymerization products and chain extenders in PET recycled materials; (2) it combines multiple recycled material identification methods with artificial intelligence algorithms for the first time, not only comprehensively considering the impact of each test method on the results, but also using machine learning methods to consider the weight of each method; (3) a recycled material performance database is constructed, and on the basis of big data comparison, the constructed decision tree artificial intelligence algorithm model is used to determine the results, avoiding the influence of subjective factors; (4) with the expansion of the database and the increase in the number of training times, the accuracy of the identification results will be further improved; (5) the method is highly transferable, and professionals do not need to have rich technical accumulation in this field.

In order to perform the identification method of polyethylene terephthalate recycled materials corresponding to the above method embodiment, to achieve the corresponding functions and technical effects. Referring to FIG. 7, FIG. 7 shows a structural block diagram of a polyethylene terephthalate recycled material identification system provided by an embodiment of the present invention. For the sake of convenience, only the parts related to this embodiment are shown. The polyethylene terephthalate recycled material identification system provided by the embodiment of the present invention includes:

An acquisition module 701 is used to acquire PET samples, wherein the PET samples include PET recycled material samples and PET virgin material samples;

An analysis module 702 is used to perform sample analysis on the PET sample to obtain characteristic data of the PET sample, wherein the characteristic data includes but is not limited to chain extender data and depolymerization product data;

A modeling module 703 is used to train a preset decision tree model using the feature data until the preset decision tree model reaches a preset convergence condition, thereby obtaining a PET recycled material identification model;

The identification module 704 is used to identify the PET sample to be identified by using the PET recycled material identification model, and determine the recycled material status of the PET sample to be identified.

In some embodiments, the analysis module 702 is specifically used to:

In the presence of an alcohol solvent and a catalyst, the PET sample is subjected to a depolymerization reaction using a stainless steel reactor or a microwave high-pressure synthesis reactor at a reaction temperature of 180-250° C. and a reaction time of 2-3 hours to obtain a PET sample solution;

After filtering and diluting the PET sample solution, the material components in the PET sample solution are detected by GCMS, LCMS or Py-GCMS to obtain spectrum data;

Using a preset chain extender GCMS database, LCMS database or Py-GCMS database, the spectrum data is searched for fragment ion peaks to obtain the chain extender data;

The spectral data are searched for fragment ion peaks using a preset depolymerization product GCMS database, LCMS database or Py-GCMS database to obtain the depolymerization product data.

In some embodiments, the chain extender data includes chain extender data of epoxy resins, isocyanates, anhydrides, oxazolines, and amides related to the chain extender.

In some embodiments, the modeling module 703 includes:

A generating unit, used for reducing the dimension of the feature data to generate a training sample set;

A training unit, used to train the preset decision tree model using the training sample set, and calculate the Gini coefficient of each feature in the training sample set at all possible split values;

An establishing unit is used to use the feature data based on the minimum Gini coefficient as the optimal feature node, use the segmentation value based on the minimum Gini coefficient as the optimal subspace, and establish a left node data set and a right node data set when the optimal feature node is used as the root node;

An iteration unit, used to generate child nodes of a decision tree according to the left node data set and the right node data set, until no more child nodes can be generated, thereby obtaining an original decision tree;

A pruning unit is used to prune the original decision tree to obtain the PET recycled material identification model.

In some embodiments, the generating unit is specifically configured to:

Based on the feature data, establish a feature data set;

Normalizing and zero-meaning the feature data set to obtain a target feature data set;

Based on the target feature data set, a covariance matrix is calculated, and eigenvalues and corresponding eigenvectors of the covariance matrix are determined;

Based on the sorting results of the eigenvalues, a number of corresponding eigenvectors are selected to form an eigenvector matrix;

The feature data is reduced in dimension based on the feature vector matrix to generate the training sample set.

In some embodiments, the calculation function of the Gini coefficient is:

Among them, Gini(D,Ai) is the Gini coefficient of feature Ai in training sample set D, D1 is the data set composed of feature data when the cut value in training sample set D is not greater than ai, D2 is the data set composed of feature data when the cut value in training sample set D is not less than ai, Gini(D1) is the Gini coefficient of data set D1, and Gini(D2) is the Gini coefficient of data set D2.

In some embodiments, the iteration unit is specifically configured to:

Generate a new child node according to the left node data set and the right node data set as the next root node data;

The child node data set corresponding to the new child node is used as the next root node data set, and new child nodes are continuously generated until no more child nodes can be generated, thereby obtaining the original decision tree.

In some embodiments, the pruning unit is specifically used to:

Based on the preset loss function, the original decision tree is optimized until the original decision tree reaches the preset convergence condition to obtain the PET recycled material identification model, wherein the preset loss function is:

C _α (T _t ) = C (T _t ) + α |T _t |;

Wherein, T _t represents any node in the original decision tree, C(T _t ) represents the prediction error of node T _t , |T _t | represents the number of nodes, and α is a preset training parameter.

In some embodiments, the identification module 704 is specifically used to:

Performing sample analysis on the PET sample to be identified to obtain target feature data of the PET sample to be identified;

Inputting the target feature data into the PET recycled material identification model and outputting a decision value;

If the decision value satisfies the preset decision value interval, it is determined that the PET sample to be identified contains recycled material.

The above-mentioned polyethylene terephthalate recycled material identification system can implement the polyethylene terephthalate recycled material identification method of the above-mentioned method embodiment. The options in the above-mentioned method embodiment are also applicable to this embodiment and will not be described in detail here. The rest of the contents of the embodiment of the present invention can refer to the contents of the above-mentioned method embodiment and will not be described in detail in this embodiment.

The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims (10)

1. A method for identifying polyethylene terephthalate recycled materials, characterized by comprising:

   Obtaining PET samples, wherein the PET samples include PET recycled material samples and PET virgin material samples;

   Performing sample analysis on the PET sample to obtain characteristic data of the PET sample, wherein the characteristic data includes chain extender data and depolymerization product data;

   Using the characteristic data, training a preset decision tree model until the preset decision tree model reaches a preset convergence condition, thereby obtaining a PET recycled material identification model;

   The PET recycled material identification model is used to identify the PET sample to be identified, and the recycled material situation of the PET sample to be identified is determined.
2. The method for identifying polyethylene terephthalate recycled materials according to claim 1, characterized in that the sample analysis of the PET sample to obtain characteristic data of the PET sample includes:

   In the presence of an alcohol solvent and a catalyst, the PET sample is subjected to a depolymerization reaction using a stainless steel reactor or a microwave high-pressure synthesis reactor at a reaction temperature of 180-250° C. and a reaction time of 2-3 hours to obtain a PET sample solution;

   After filtering and diluting the PET sample solution, the material components in the PET sample solution are detected by GCMS, LCMS or Py-GCMS to obtain spectrum data;

   Using a preset chain extender GCMS database, LCMS database or Py-GCMS database, the spectrum data is searched for fragment ion peaks to obtain the chain extender data;

   The spectral data are searched for fragment ion peaks using a preset depolymerization product GCMS database, LCMS database or Py-GCMS database to obtain the depolymerization product data.
3. The method for identifying polyethylene terephthalate recycled materials as described in claim 2 is characterized in that the chain extender data includes chain extender data of epoxy resins, isocyanates, anhydrides, oxazolines and amides related to the chain extender.
4. The method for identifying polyethylene terephthalate recycled materials according to claim 1, characterized in that the preset decision tree model is trained using the characteristic data until the preset decision tree model reaches a preset convergence condition to obtain a PET recycled material identification model, comprising:

   Performing dimension reduction on the feature data to generate a training sample set;

   Using the training sample set, the preset decision tree model is trained, and the Gini coefficient of each feature data in the training sample set at all possible segmentation values is calculated;

   Taking the feature data based on the minimum Gini coefficient as the optimal feature node, taking the segmentation value based on the minimum Gini coefficient as the optimal subspace, and establishing the left node data set and the right node data set when the optimal feature node is used as the root node;

   Generate child nodes of the decision tree according to the left node data set and the right node data set until no more child nodes can be generated, thereby obtaining an original decision tree;

   The original decision tree is pruned to obtain the PET recycled material identification model.
5. The method for identifying polyethylene terephthalate recycled materials according to claim 4, characterized in that the dimension reduction of the feature data to generate a training sample set comprises:

   Based on the feature data, establish a feature data set;

   Normalizing and zero-meaning the feature data set to obtain a target feature data set;

   Based on the target feature data set, a covariance matrix is calculated, and eigenvalues and corresponding eigenvectors of the covariance matrix are determined;

   Based on the sorting results of the eigenvalues, a number of corresponding eigenvectors are selected to form an eigenvector matrix;

   The feature data is reduced in dimension based on the feature vector matrix to generate the training sample set.
6. The method for identifying polyethylene terephthalate recycled materials according to claim 4, characterized in that the calculation function of the Gini coefficient is:

   

   Among them, Gini(D,Ai) is the Gini coefficient of the feature data Ai in the training sample set D, D1 is the data set composed of the feature data when the cut value in the training sample set D is not greater than the target cut value, D2 is the data set composed of the feature data when the cut value in the training sample set D is not less than the target cut value, Gini(D1) is the Gini coefficient of the data set D1, and Gini(D2) is the Gini coefficient of the data set D2.
7. The method for identifying polyethylene terephthalate recycled materials according to claim 4 is characterized in that, according to the left node data set and the right node data set, the child nodes of the decision tree are generated until the child nodes can no longer be generated, and the original decision tree is obtained, comprising:

   Generate a new child node according to the left node data set and the right node data set as the next root node data;

   The child node data set corresponding to the new child node is used as the next root node data set, and new child nodes are continuously generated until no more child nodes can be generated, thereby obtaining the original decision tree.
8. The method for identifying polyethylene terephthalate recycled materials according to claim 4, characterized in that pruning the original decision tree to obtain the PET recycled material identification model comprises:

   Based on the preset loss function, the original decision tree is optimized until the original decision tree reaches the preset convergence condition, thereby obtaining the PET recycled material identification model, wherein the preset loss function is:

   C _α (T _t ) = C (T _t ) + α |T _t |;

   Wherein, T _t represents any node in the original decision tree, C(T _t ) represents the prediction error of node T _t , |T _t | represents the number of nodes, and α is a preset training parameter.
9. The method for identifying polyethylene terephthalate recycled materials according to claim 1, characterized in that the PET recycled material identification model is used to identify the PET sample to be identified, and the recycled material situation of the PET sample to be identified is determined, including:

   Performing sample analysis on the PET sample to be identified to obtain target feature data of the PET sample to be identified;

   Inputting the target feature data into the PET recycled material identification model and outputting a decision value;

   If the decision value satisfies the preset decision value interval, it is determined that the PET sample to be identified contains recycled material.
10. A polyethylene terephthalate recycled material identification system, characterized by comprising:

    An acquisition module is used to acquire PET samples, wherein the PET samples include PET recycled material samples and PET virgin material samples;

    An analysis module, used for performing sample analysis on the PET sample to obtain characteristic data of the PET sample, wherein the characteristic data includes but is not limited to chain extender data and depolymerization product data;

    A modeling module, used to train a preset decision tree model using the feature data until the preset decision tree model reaches a preset convergence condition, thereby obtaining a PET recycled material identification model;

    The identification module is used to identify the PET sample to be identified by using the PET recycled material identification model, and determine the recycled material situation of the PET sample to be identified.

Images