PET (Polyethylene Terephthalate) titanium sodium film for high-end wallboard and intelligent preparation method of PET titanium sodium film

Abstract

The invention discloses a PET titanium sodium film for a high-end wallboard and an intelligent preparation method of the PET titanium sodium film. The PET titanium sodium film for the high-end wallboard comprises the following components in parts by weight: 40-60 parts of PET powder, 15-20 parts of NaTi2 (PO4) 3, 0.8-1.6 parts of glycerol and 2-6 parts of a processing aid. Wherein the processing aid comprises 40%-60% of an anti-aging agent and 40%-60% of a flame retardant; the invention relates to the technical field of PET titanium sodium film preparation. According to the PET titanium sodium film for the high-end wallboard and the intelligent preparation method of the PET titanium sodium film, NaTi2 (PO4) 3 is added, so that the prepared PET film has a certain anti-ultraviolet function and relatively good mechanical properties, an ultraviolet shielding effect is achieved, the mechanical properties of the PET film can still be kept good after ultraviolet aging, and the service life of the PET film is prolonged.

Description

Technical Field

[0001] The invention relates to the technical field of PET sodium titanium film preparation, and in particular to a PET sodium titanium film for high-end wall panels and an intelligent preparation method thereof. Background Art

[0002] The Chinese patent with the publication number CN116462943B discloses a flame retardant and UV-proof PET film, which includes the following raw materials by weight: 50 parts of PET resin, 15-19 parts of flame retardant additives, 8-12 parts of kaolin, 0.8-1 parts of titanate coupling agents, 2-4 parts of toughening agents, and 0.8-1 parts of lubricants. The present invention adds flame retardant additives to the raw materials, which are polyester-type polymers, which can not only achieve uniform dispersion in the PET film, but also have higher stability, migration resistance and exudation resistance compared to ordinary organic flame retardants, and have a long-lasting flame retardant effect. The flame retardant additive has a PN-Si synergistic flame retardant mechanism and has a high-efficiency flame retardant effect; in addition, the flame retardant additive has high light aging resistance due to its high-density hindered amine structure, and synergizes with kaolin to greatly improve the UV protection performance of the PET film; the PET film is applied to the wallboard to provide safe and long-term protection for the wallboard. However, the PET film is prone to aging after long-term practical use, which makes it easy to break and affects its use effect. To this end, the present invention provides a PET sodium titanium film for high-end wall panels and an intelligent preparation method thereof to solve the above problems. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a PET sodium titanium film for high-end wall panels and an intelligent preparation method thereof, which solves the problems mentioned in the above background technology.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: a PET sodium titanium film for high-end wall panels and an intelligent preparation method thereof, the PET sodium titanium film for high-end wall panels comprises the following components in parts by weight: 40 to 60 parts of PET powder, 15 to 20 parts of NaTi2(PO4)3, 0.8 to 1.6 parts of propylene glycol, and 2 to 6 parts of processing aids;

[0005] Wherein, the processing aid comprises 40% to 60% of anti-aging agent and 40% to 60% of flame retardant.

[0006] Preferably, the anti-aging agent is one of UV-9 and UV-3853, and the flame retardant is selected from any one of aluminum hydroxide, magnesium hydroxide, antimony oxide and zinc borate.

[0007] A method for preparing a PET sodium titanium film for high-end wallboards, specifically comprising the following preparation steps:

[0008] Step 1: Place NaTi2(PO4)3 in distilled water and perform ultrasonic dispersion. After the NaTi2(PO4)3 is fully dispersed, transfer the mixture to an electric heating mantle and stir the mixture.

[0009] Step 2: dissolving glycerol, and pouring the dissolved glycerol into the solution in step 1, and continuing to stir;

[0010] Step 3, adding PET powder to the mixed solution in step 2, and letting it stand for a while after the addition is completed;

[0011] Step 4: Place in a constant temperature environment and continue stirring. After the PET powder is completely dissolved, add the processing aid, mix evenly, and transfer to a drying box for drying;

[0012] Step 5. Finally, put it into the pulling machine and pull it into film to obtain PET sodium titanium film.

[0013] Preferably, in step 1, the ultrasonic dispersion time is 20 to 40 minutes.

[0014] Preferably, in step 2, the dissolution method of glycerol is: 0.72 g of glycerol is placed in 62 mL of distilled water to dissolve it.

[0015] Preferably, in the step three, when adding the PET powder, it is added in small amounts and multiple times, and the standing time is 50 to 120 minutes.

[0016] Preferably, in step 4, the constant temperature is 65-75° C., the temperature of the drying oven is 65-75° C., and the drying time is 40-60 min.

[0017] Preferably, in step five, the thickness of the prepared PET sodium titanium film is 0.16 to 0.20 mm.

[0018] Furthermore, in steps 2 to 5, an intelligent learning algorithm is used to analyze the effects of different formulations during preparation, and the raw material ratio is optimized to improve the performance and quality stability of the wallboard; specifically:

[0019] Step 1: Collect a large amount of experimental data, including the performance indicators of wallboards with different raw material ratios: strength, wear resistance, weather resistance, and process parameters such as temperature, pressure, and mixing time during the preparation process; ensure the quality and integrity of the data;

[0020] Collect experimental data under different raw material ratios. These data are collected through systematic experimental design to ensure that various raw material ratio combinations are covered; each ratio combination contains corresponding wallboard performance index data; during the preparation process, record key process parameters, including temperature, pressure, and stirring time; these parameters are obtained through sensors or recording instruments of experimental equipment and need to be recorded together with the wallboard performance index data of each experimental sample;

[0021] Step 2: Extract features and preprocess the collected data, convert the raw material ratio and process parameters into numerical features that can be processed by the machine learning algorithm; feature scaling and standardization operations are required;

[0022] 2.1. Represent different raw material ratios as numerical features so that machine learning algorithms can process them; use the weight ratio of the raw materials as features, and use the weight ratio of each raw material such as PET powder, NaTi2(P04)3, glycerol, and processing aid as features;

[0023] Convert the process parameters in the preparation process into numerical features; including temperature, pressure, stirring time and other process parameters as features;

[0024] 2.2. Perform feature scaling on raw material ratios and process parameters and map them to an appropriate range; avoid instability in the model training process due to scale differences in feature values; use the minimum-maximum scaling method to perform feature scaling;

[0025] For categorical features, encoding is required to convert them into numerical features; use one-hot encoding or label encoding for processing;

[0026] Perform feature selection based on the correlation between features and wall panel performance indicators, and select features that have a greater impact on the model prediction performance; use statistical methods, correlation coefficient analysis or feature importance evaluation methods to perform feature selection;

[0027] Step 3: Select a suitable machine learning model to establish the relationship between raw material ratio and wallboard performance; considering the complex impact of raw material ratio on wallboard performance, select deep neural network DNN and gradient boosting tree GBT;

[0028] 3.3 Deep Neural Network DNN Implementation Process:

[0029] Design the structure of the neural network, including the input layer, hidden layer and output layer; according to the patent content, the number of nodes in the input layer should correspond to the number of characteristics of the raw material ratio and process parameters, and the number of nodes in the output layer should correspond to the number of wallboard performance indicators;

[0030] Use the training set data to train the neural network and adjust the network parameters to minimize the loss function; use the back propagation algorithm and optimizer for training;

[0031] 3.4DNN Improvements:

[0032] Introducing the attention mechanism: Introducing the attention mechanism before the input layer of the neural network to assign weights to the raw material ratio and process parameters; using the self-attention mechanism or the attention mechanism network to implement the importance assessment between the raw material ratio and the process parameters; specifically, setting the input layer to X, using the attention mechanism to obtain the weighted input feature X′=Attention(X);

[0033] Increase network depth and width: Increase the number of hidden layers and the number of neurons in each hidden layer, and use residual connections or batch normalization techniques to speed up network training and improve model stability;

[0034] Derivation of the attention mechanism formula: Let X∈R n ×m Represents the feature matrix of the input layer, where n represents the number of samples and m represents the number of features; the formula derivation process of the self-attention mechanism is as follows:

[0035] A=softmax(WXQ T )

[0036] X′=AX

[0037] Meaning of network structure parameters: n: number of samples; m: number of features of raw material ratio and process parameters; W, Q: weight matrix in the attention mechanism, used to calculate the importance of input features; A: attention score, indicating the importance of each feature; X′: weighted input feature, used to pass into the hidden layer of the neural network;

[0038] 3.3 Gradient Boosting Tree GBT Improvement:

[0039] Improve the gradient boosting tree model to better capture the complex effects of raw material ratio on wallboard performance;

[0040] Using domain knowledge of raw material ratios and process parameters, feature engineering is performed to extract more effective features to enhance the model’s ability to capture the relationship between raw material ratios and wallboard performance;

[0041] Enhanced regularization: During the model training process, regularization terms are added to limit the size of model parameters and prevent overfitting. For the gradient boosting tree model, regularization terms are added to the loss function, including L1 regularization terms or L2 regularization terms.

[0042] Add a regularization term to the loss function to limit the complexity of the model:

[0043]

[0044] Among them, L(y i ,F(x i )) represents the loss function of the model, y i Usually represents the prediction result or model function of the model. In the iterative process, y i Gradually approaching the true target value F(x i );λ represents the coefficient of the regularization term. This parameter controls the strength of regularization. The larger the parameter, the greater the influence of regularization. Ω(F) represents the complexity of the model;λ: the coefficient of the regularization term;

[0045] Model integration: Considering the complex influence of raw material ratio, model integration methods are used, including bagging and boosting; multiple gradient boosting tree models of different depths are integrated, and the prediction results of multiple models are synthesized by voting;

[0046] The above improvement scheme makes better use of the gradient boosting tree model to establish the association between the raw material ratio and the wallboard performance, and improves the prediction performance and generalization ability of the model; the association model between the raw material ratio and the wallboard performance is established by combining the two, and a suitable machine learning algorithm is selected to achieve performance optimization;

[0047] Step 4: Use the cross-validation method to evaluate and verify the trained model to ensure that the prediction performance of the model meets the requirements;

[0048] Divide the original data set into k subsets of equal size, select k-1 subsets as training sets each time, and use the remaining subset as validation set. Repeat this process k times, using a different validation set for validation each time.

[0049] The average value of the evaluation indicators obtained in each verification is used as the final evaluation result to evaluate the prediction performance of the model;

[0050] The data set contains N samples, and k-fold cross validation is performed. Each time, N / k samples are selected as the validation set, and the remaining samples are used as the training set; k: the number of cross-validation folds, indicating how many subsets the data set is divided into for cross-validation; N: the number of samples in the data set;

[0051] Calculate the evaluation indicators for each validation, and then take the average value as the final model evaluation result;

[0052] Step 5: Deploy the trained model to the production environment, monitor the raw material ratio and process parameters in real time, and make adjustments based on the model’s predicted results to optimize the wallboard performance;

[0053] 5.5 Deploy the trained model to the production environment to monitor and predict the wallboard performance in real time; deploy the model as a service, receive the real-time raw material ratio and process parameters as input, and output the predicted wallboard performance indicators;

[0054] 5.6 Real-time monitoring:

[0055] Set up sensors in the production environment to monitor changes in raw material ratios and process parameters in real time;

[0056] X: characteristic vector of raw material ratio and process parameters, including PET powder, NaTi2(P04)3, glycerol, and processing aids;

[0057] y: wall panel performance indicators, including strength, wear resistance, and weather resistance;

[0058] Transmit the monitored data to the model service for real-time prediction and performance evaluation;

[0059] ypre d : Wall panel performance index predicted by the model;

[0060] 5.7 Adjustment control:

[0061] After receiving the real-time data, the model service predicts the performance of the wall panel;

[0062] According to the results predicted by the model, determine whether the current raw material ratio and process parameters need to be adjusted;

[0063] If the wallboard performance index predicted by the model is different from the target value, the raw material ratio and process parameters are adjusted automatically or manually;

[0064] 5.8 Monitoring and Adjustment Strategy:

[0065] Threshold determination: Set a certain threshold. When the gap between the model prediction result and the actual performance index exceeds the threshold, the parameter adjustment is triggered. If the strength predicted by the model is set to be 5% lower than the target value, the raw material ratio and process parameters are automatically adjusted.

[0066] Feedback adjustment: Update and optimize the model based on feedback information from actual production; regularly collect actual production data and retrain the model to adapt to changes in the production environment and optimize model performance.

[0067] Beneficial Effects

[0068] The present invention provides a PET sodium titanium film for high-end wallboards and an intelligent preparation method thereof. Compared with the prior art, the PET sodium titanium film for high-end wallboards and the intelligent preparation method thereof have the following beneficial effects: the PET sodium titanium film for high-end wallboards and the intelligent preparation method thereof, by adding NaTi2(PO4)3, makes the prepared PET film have certain anti-ultraviolet function and good mechanical properties, plays a role of ultraviolet shielding, and its mechanical properties can still be maintained well after ultraviolet light aging, extending its service life; at the same time, a machine learning algorithm is used to analyze the effects of different formulas and optimize the raw material ratio to improve the performance and quality stability of the wallboard; such an algorithm finds the best raw material ratio by analyzing a large amount of experimental data, so that the performance of the wallboard is optimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 A tensile strength test provided for a PET sodium titanium film for high-end wallboard and its intelligent preparation method of the present invention;

[0070] Figure 2 The ultraviolet reflection spectrum provided for a PET sodium titanium film for high-end wallboards and its intelligent preparation method of the present invention;

[0071] Figure 3 The invention provides a test for the retention rate of elongation at break of a PET sodium titanium film for high-end wall panels and its intelligent preparation method.

[0072] Figure 4 The present invention is an algorithm flow chart of an intelligent control system for a PET sodium titanium film for high-end wall panels and its intelligent preparation method. DETAILED DESCRIPTION

[0073] 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.

[0074] Preparation method of sodium titanium phosphate: TiO(OH)2·0.6H2O, 85% H3PO4, and NaH2PO4·2H20 are placed in a grinder and ground evenly in a mass ratio of 2:2:1, and then the ground mixture is placed in an oven at 125°C and heated for 3.5 hours, and then cooled to room temperature.

[0075] The dissolution method of glycerol is: put 0.72g of glycerol into 62mL of distilled water to dissolve it.

[0076] Example 1

[0077] A method for preparing a PET sodium titanium film for high-end wallboards, specifically comprising the following preparation steps:

[0078] Step 1, weigh the raw materials: PET powder 40g, NaTi2(PO4) 320g, glycerol 1.6g, processing aid 6g, including UV-9 40%, aluminum hydroxide 60%, and set aside;

[0079] Step 2: Place NaTi2(PO4)3 in distilled water and perform ultrasonic dispersion for 20 minutes. After it is fully dispersed, transfer it to an electric heating mantle and stir it.

[0080] Step 3, dissolving the glycerol, and pouring the dissolved glycerol into the solution in step 2, and continuing to stir;

[0081] Step 4: Add PET powder to the mixed solution in step 3 in small amounts and several times, and let it stand for 50 minutes after the addition is completed;

[0082] Step 5: Place in a constant temperature environment of 65°C and continue stirring. After the PET powder is completely dissolved, add the processing aid, mix evenly, and transfer to a drying oven at 65°C for drying for 40 minutes.

[0083] Step 6. Finally, put it into the pulling machine and pull it into a film to obtain a PET sodium titanium film with a thickness of 0.16 mm.

[0084] Example 2

[0085] A method for preparing a PET sodium titanium film for high-end wallboards, specifically comprising the following preparation steps:

[0086] Step 1, weigh the raw materials: 50g PET powder, 18g NaTi2(PO4), 1.2g glycerol, 4g processing aid, including 50% UV-3853 and 50% magnesium hydroxide, and set aside;

[0087] Step 2: Place NaTi2(PO4)3 in distilled water and perform ultrasonic dispersion for 30 minutes. After it is fully dispersed, transfer it to an electric heating mantle and stir it.

[0088] Step 3, dissolving the glycerol, and pouring the dissolved glycerol into the solution in step 2, and continuing to stir;

[0089] Step 4: Add PET powder to the mixed solution in step 3 in small amounts and several times, and let it stand for 85 minutes after the addition is completed;

[0090] Step 5: Place in a constant temperature environment of 70°C and continue stirring. After the PET powder is completely dissolved, add the processing aid, mix evenly, and transfer to a drying oven at 70°C for drying. The drying time is 50 minutes.

[0091] Step 6. Finally, put it into the pulling machine and pull it into a film to obtain a PET sodium titanium film with a thickness of 0.18 mm.

[0092] Example 3

[0093] A method for preparing a PET sodium titanium film for high-end wallboards, specifically comprising the following preparation steps:

[0094] Step 1, weigh the raw materials: 60g PET powder, 15g NaTi2(PO4), 0.8g glycerol, 2g processing aid, including 60% UV-3853 and 40% zinc borate, and set aside;

[0095] Step 2: Place NaTi2(PO4)3 in distilled water and perform ultrasonic dispersion for 40 minutes. After it is fully dispersed, transfer it to an electric heating mantle and stir it.

[0096] Step 3, dissolving the glycerol, and pouring the dissolved glycerol into the solution in step 2, and continuing to stir;

[0097] Step 4: Add PET powder to the mixed solution in step 3 in small amounts and several times, and let it stand for 120 minutes after the addition is completed;

[0098] Step 5: Place in a constant temperature environment of 75°C and continue stirring. After the PET powder is completely dissolved, add the processing aid, mix evenly, and transfer to a drying oven at 75°C for drying for 60 minutes.

[0099] Step 6. Finally, put it into the pulling machine and pull it into a film to obtain a PET sodium titanium film with a thickness of 0.20 mm.

[0100] Example 4

[0101] A method for preparing PET sodium titanium film for high-end wallboards, in which the machine learning algorithm is used to analyze the effects of different formulas in the preparation of steps 2-5, and optimize the raw material ratio to improve the performance and quality stability of the wallboard. Such an algorithm finds the best raw material ratio by analyzing a large amount of experimental data, so that the performance of the wallboard is optimized. The specific steps are:

[0102] Machine learning algorithms are used to analyze the effects of different formulations and optimize the raw material ratio to improve the performance and quality stability of wallboards. Such algorithms analyze a large amount of experimental data to find the best raw material ratio to optimize the performance of wallboards.

[0103] Step 1: Collect a large amount of experimental data, including the performance indicators of wallboards with different raw material ratios: strength, wear resistance, weather resistance, and process parameters such as temperature, pressure, and mixing time during the preparation process; ensure the quality and integrity of the data;

[0104] 1.1 Data Collection:

[0105] Collect experimental data under different raw material ratios. These data are collected through systematic experimental design to ensure that various raw material ratio combinations are covered; each ratio combination contains corresponding wallboard performance index data;

[0106] During the preparation process, record key process parameters, including temperature, pressure, and mixing time; these parameters are obtained through sensors or recording instruments of the experimental equipment and need to be recorded together with the wallboard performance index data of each experimental sample;

[0107] 1.2 Data quality and integrity assurance:

[0108] Ensure the scientificity and reliability of the experimental design to avoid biased or misleading data results; use randomized experimental design or balanced design methods to ensure the representativeness of experimental data;

[0109] During the data recording process, ensure the accuracy and consistency of experimental data; use electronic recording systems and double verification methods to reduce the occurrence of data recording errors;

[0110] Ensure data integrity, including complete recording of all experimental sample data and corresponding process parameter data; avoid the impact of data loss or omission on subsequent analysis;

[0111] Step 2: Extract features and preprocess the collected data, convert the raw material ratio and process parameters into numerical features that can be processed by the machine learning algorithm; feature scaling and standardization operations are required;

[0112] 2.1 Data feature extraction:

[0113] Represent different raw material ratios as numerical features so that machine learning algorithms can process them; use the weight ratio of the raw materials as features, that is, the weight ratio of each raw material such as PET powder, NaTi2(P04)3, propylene glycol, and processing aid as features;

[0114] Convert the process parameters in the preparation process into numerical features; including temperature, pressure, stirring time and other process parameters as features;

[0115] 2.2 Feature processing flow:

[0116] Perform feature scaling on raw material ratios and process parameters to map them to an appropriate range; avoid instability in the model training process due to scale differences in feature values; use the minimum-maximum scaling method for feature scaling;

[0117] For categorical features, encoding is required to convert them into numerical features; One-Hot Encoding or Label Encoding is used for processing;

[0118] Perform feature selection based on the correlation between features and wall panel performance indicators, and select features that have a greater impact on the model prediction performance; use statistical methods, correlation coefficient analysis or feature importance evaluation methods to perform feature selection;

[0119] Step 3: Select an appropriate machine learning model to establish the relationship between the raw material ratio and the wallboard performance; considering the complex impact of the raw material ratio on the wallboard performance, select a complex model such as a deep neural network (DNN) or a gradient boosting tree;

[0120] 3.1 Deep Neural Network (DNN) Implementation Process:

[0121] Design the structure of the neural network, including the input layer, hidden layer and output layer; according to the patent content, the number of nodes in the input layer should correspond to the number of characteristics of the raw material ratio and process parameters, and the number of nodes in the output layer should correspond to the number of wallboard performance indicators;

[0122] Use the training set data to train the neural network and adjust the network parameters to minimize the loss function; use the backpropagation algorithm and optimizer (including Adam, SGD) for training;

[0123] 3.2DNN Improvements:

[0124] Based on the patent content, improve the deep neural network model to better capture the complex impact of raw material ratio on wallboard performance;

[0125] Considering the complexity of raw material ratio, more complex network structure and regularization technology are introduced to improve the model's fitting ability and generalization ability; the attention mechanism allows the model to pay more attention to the raw material ratio and process parameters that have an important impact on the prediction; this enhances the model's ability to capture the complex relationship between raw material ratios; by increasing the depth and width of the neural network, the model's fitting ability is improved, and the nonlinear relationship between raw material ratio and wallboard performance is better learned;

[0126] DNN specific implementation cases:

[0127] Introduce the attention mechanism: Introduce the attention mechanism before the input layer of the neural network to assign weights to the raw material ratio and process parameters; use the self-attention mechanism (Self-Attention) or the attention mechanism network (AttentionNetwork) to realize the importance evaluation between the raw material ratio and the process parameters; specifically, set the input layer to X, and use the attention mechanism to obtain the weighted input feature X′=Attention(X)

[0128] Increase network depth and width: Increase the number of hidden layers and the number of neurons in each hidden layer to increase the complexity of the model; use residual connection or batch normalization technology to speed up network training and improve model stability;

[0129] Derivation of the attention mechanism formula: Let X∈R n×m Represents the feature matrix of the input layer, where n represents the number of samples and m represents the number of features; the formula derivation process of the self-attention mechanism is as follows:

[0130] A=softmax(WXQ T )

[0131] X′=AX

[0132] Among them, W, Q, A represent weight matrices respectively, and X′ represents the weighted input features;

[0133] Meaning of network structure parameters: n: number of samples; m: number of features of raw material ratio and process parameters; W, Q: weight matrix in the attention mechanism, used to calculate the importance of input features; A: attention score, indicating the importance of each feature; X′: weighted input feature, used to pass into the hidden layer of the neural network;

[0134] Through the above improvement scheme and formula derivation process, the deep neural network model is better used to establish the relationship between the raw material ratio and the wallboard performance, and improve the prediction performance of the model;

[0135] 3.3 Gradient Boosting Tree (GBT) Implementation Process:

[0136] Build a gradient boosted tree model, including selecting a base learner (decision tree) and determining the number and depth of trees.

[0137] The training set data is used to train the gradient boosting tree model, and the gradient boosting algorithm is used to gradually fit the residuals and optimize the loss function;

[0138] 3.4GBT Improvements:

[0139] Specific improvement plan: Use domain knowledge of raw material ratio and process parameters to perform feature engineering and extract more effective features to enhance the model's ability to capture the relationship between raw material ratio and wallboard performance;

[0140] By adding regularization terms, including L1 regularization (Lasso) and L2 regularization (Ridge), the complexity of the model is controlled to prevent overfitting;

[0141] Considering the diverse effects of raw material ratio on wallboard performance, a model integration method is used, including integrating multiple gradient boosting tree models of different depths, to comprehensively consider the prediction results of multiple models; 3.5GPT specific implementation cases:

[0142] Secondary feature engineering: Perform secondary feature engineering to extract more effective features of raw material ratios and process parameters, including feature intersection and feature selection; extract the interaction features between the various components of the raw material ratio, and the interaction features between the process parameters;

[0143] Enhanced regularization: During the model training process, regularization terms are added to limit the size of model parameters and prevent overfitting. For the gradient boosting tree model, regularization terms are added to the loss function, including L1 regularization terms or L2 regularization terms.

[0144] Add regularization terms to the loss function to limit the complexity of the model

[0145]

[0146] Among them, L(y i ,F(x i )) represents the loss function of the model, λ represents the coefficient of the regularization term, this parameter controls the strength of regularization, the larger the parameter, the greater the influence of regularization, Ω(F) represents the complexity of the model; λ: the coefficient of the regularization term,;

[0147] Model integration: Considering the complex influence of raw material ratio, model integration methods are used, including bagging and boosting; multiple gradient boosting tree models of different depths are integrated, and the prediction results of multiple models are synthesized by voting;

[0148] Through the above improvement scheme, the gradient boosting tree model is better used to establish the association between the raw material ratio and the wallboard performance, and the prediction performance and generalization ability of the model are improved; the association model between the raw material ratio and the wallboard performance is established by combining the two, and the appropriate machine learning algorithm is selected to achieve performance optimization;

[0149] Step 4: Use the retention set or cross-validation method to evaluate and validate the trained model to ensure that the model's predictive performance meets the requirements;

[0150] 4.1 Holdout Set Method:

[0151] The original data set is divided into a training set and a test set. The training set is used for model training, and the test set is used for model evaluation.

[0152] During the training process, the cross-validation method is used to select the model's hyperparameters, including the number of hidden layer nodes and learning rate in the deep neural network, and the number of trees and learning rate in the gradient boosting tree;

[0153] The final trained model is evaluated on the reserved test set to assess the generalization ability and prediction performance of the model;

[0154] 4.2 Cross-validation method:

[0155] Divide the original data set into k subsets of equal size, select k-1 subsets as training sets each time, and use the remaining subset as validation set. Repeat this process k times, using a different validation set for validation each time.

[0156] The average value of the evaluation indicators obtained in each verification is used as the final evaluation result to evaluate the prediction performance of the model;

[0157] 4.3 Model Evaluation Metrics:

[0158] Model evaluation indicators include mean squared error (MSE), root mean square error (RMSE), and mean absolute error (MAE). MSE and RMSE measure the difference between the model prediction value and the true value, while MAE measures the absolute difference between the model prediction value and the true value.

[0159] The data set contains N samples, and k-fold cross validation is performed. Each time, N / k samples are selected as the validation set, and the remaining samples are used as the training set; k: the number of cross-validation folds, indicating how many subsets the data set is divided into for cross-validation; N: the number of samples in the data set;

[0160] Calculate the evaluation indicators (including MSE, RMSE, MAE) for each validation, and then take their average value as the final model evaluation result;

[0161] Holdout set method: Divide the dataset into training set and test set in a ratio of 80%-20%; train the model on the training set, and evaluate the prediction performance of the model based on the evaluation indicators (including MSE, RMSE, MAE) on the test set;

[0162] Cross-validation method: Divide the data set into k subsets and perform k-fold cross-validation;

[0163] Each time, a subset is selected as the validation set, and the remaining subsets are used as the training set to calculate the evaluation index; finally, the average value of the evaluation index obtained by k validations is calculated as the final model evaluation result;

[0164] Through the above specific implementation cases, the trained model is evaluated and verified to ensure that the prediction performance of the model meets the requirements, so as to ensure the reliability and effectiveness of the model in practical applications;

[0165] Step 5: Deploy the trained model to the production environment, monitor the raw material ratio and process parameters in real time, and make adjustments based on the model’s predicted results to optimize the wallboard performance;

[0166] 5.1 Deployment Model:

[0167] Deploy the trained model to a production environment to monitor and predict wallboard performance in real time; deploy the model as a service that receives real-time raw material ratios and process parameters as input and outputs predicted wallboard performance indicators;

[0168] 5.2 Real-time monitoring:

[0169] Set up sensors in the production environment to monitor changes in raw material ratios and process parameters in real time;

[0170] X: characteristic vector of raw material ratio and process parameters, including PET powder, NaTi2(PO4)3, glycerol, and processing aids;

[0171] y: wall panel performance indicators, including strength, wear resistance, and weather resistance;

[0172] Transmit the monitored data to the model service for real-time prediction and performance evaluation;

[0173] ypre d : Wall panel performance index predicted by the model;

[0174] 5.3 Adjustment control:

[0175] After receiving the real-time data, the model service predicts the performance of the wall panel;

[0176] According to the results predicted by the model, determine whether the current raw material ratio and process parameters need to be adjusted;

[0177] If the wallboard performance index predicted by the model is different from the target value, the raw material ratio and process parameters are adjusted automatically or manually;

[0178] 5.4 Monitoring and Adjustment Strategy:

[0179] Threshold determination: Set a certain threshold. When the gap between the model prediction result and the actual performance index exceeds the threshold, the parameter adjustment is triggered. If the strength predicted by the model is set to be 5% lower than the target value, the raw material ratio and process parameters are automatically adjusted.

[0180] Feedback adjustment: Update and optimize the model based on feedback information from actual production; regularly collect actual production data and retrain the model to adapt to changes in the production environment and optimize model performance; 5.5 Specific implementation cases:

[0181] Define the feature vector X to represent the raw material ratio and process parameters in the production process, including the following variables:

[0182] X=[x1,x2,x3,x4,x5,x6]

[0183] Where: x1: amount of PET powder; x2: amount of NaTi2(P04)3; x3: amount of propylene glycol; x4: amount of processing aids (UV-9 and aluminum hydroxide); x5: stirring time; x6: drying time and temperature

[0184] Define the wall panel performance index y, including the following variables:

[0185] y=[y1,y2,y3]

[0186] in:

[0187] y1: strength; y2: wear resistance; y3: weather resistance

[0188] Real-time monitoring of data transmission and prediction:

[0189] Real-time monitoring dataX t Transmitted to the model service, the model service predicts the input feature vector:

[0190] y pred =f(X t )

[0191] Where: X t is the real-time feature vector collected at time t; f represents the model function, which predicts the input features through the trained model; is the performance index of the wall panel predicted by the model.

[0192] Comparative Example 1

[0193] A method for preparing a PET sodium titanium film for high-end wallboards, specifically comprising the following preparation steps:

[0194] Step 1, weigh the raw materials: 40g PET powder, 1.6g glycerol, 6g processing aid, including 40% UV-9 and 60% aluminum hydroxide, and set aside;

[0195] Step 2, dissolving glycerol, and pouring the dissolved glycerol into the solution in step 2, and continuing to stir;

[0196] Step 3: Add PET powder to the mixed solution in step 3 in small amounts and several times, and let it stand for 50 minutes after the addition is completed;

[0197] Step 4: Place in a constant temperature environment of 65°C and continue stirring. After the PET powder is completely dissolved, add the processing aid, mix evenly, and transfer to a drying oven at 65°C for drying. The drying time is 40 minutes.

[0198] Step 5. Finally, put it into the pulling machine and pull it into a film to obtain a PET sodium titanium film with a thickness of 0.16 mm.

[0199] Tensile strength and tensile elongation test: carried out on a JSL-5000N electronic tensile testing machine. Test steps: clamp the sample between the upper and lower clamps of the testing machine, fix the upper clamp, and move the lower clamp up and down to align the sample with the center of the upper and lower clamps of the testing machine. Before the test, make the upper and lower elongation instruments and the two marking lines of the sample coincide. The elongation instrument cannot make the sample bear the load. The stretching rate is 75mm / min.

[0200] UV aging test: Place the sample in a UV aging box for 70 hours, and take samples every 10 hours. The distance between the UV lamp and the sample is 40mm.

[0201] Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0202] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0203] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A PET sodium titanium film for high-end wallboard, characterized in that: The PET sodium titanium film for high-end wallboards comprises the following components by weight: 40-60 parts of PET powder, 15-20 parts of NaTi2(PO4), 0.8-1.6 parts of glycerol, and 2-6 parts of processing aids; Wherein, the processing aid comprises 40% to 60% of anti-aging agent and 40% to 60% of flame retardant.

2. The PET sodium titanium film for high-end wallboard according to claim 1, characterized in that: The anti-aging agent is one of UV-9 and UV-3853, and the flame retardant is selected from any one of aluminum hydroxide, magnesium hydroxide, antimony oxide and zinc borate.

3. A method for preparing the PET sodium titanium film for high-end wallboard according to claim 1, characterized in that: The specific preparation steps include: Step 1: Place NaTi2(PO4)3 in distilled water and perform ultrasonic dispersion. After the NaTi2(PO4)3 is fully dispersed, transfer the mixture to an electric heating mantle and stir the mixture. Step 2: dissolving glycerol, and pouring the dissolved glycerol into the solution in step 1, and continuing to stir; Step 3, adding PET powder to the mixed solution in step 2, and letting it stand for a while after the addition is completed; Step 4: Place in a constant temperature environment and continue stirring. After the PET powder is completely dissolved, add the processing aid, mix evenly, and transfer to a drying box for drying; Step 5. Finally, put it into the pulling machine and pull it into film to obtain PET sodium titanium film.

4. The method for preparing a PET sodium titanium film for high-end wallboard according to claim 3, characterized in that: In the step 1, the ultrasonic dispersion time is 20 to 40 minutes.

5. The method for preparing a PET sodium titanium film for high-end wallboard according to claim 3, characterized in that: In steps 2 to 5, an intelligent learning algorithm is used to analyze the effects of different formulations during its preparation, and the raw material ratio is optimized to improve the performance and quality stability of the wallboard; Specifically: Step 1: Collect a large amount of experimental data, including the performance indicators of wallboards with different raw material ratios: strength, wear resistance, weather resistance, and process parameters such as temperature, pressure, and mixing time during the preparation process; ensure the quality and integrity of the data; Collect experimental data under different raw material ratios. These data are collected through systematic experimental design to ensure that various raw material ratio combinations are covered; each ratio combination contains corresponding wallboard performance index data; during the preparation process, record key process parameters, including temperature, pressure, and mixing time; These parameters are obtained through sensors or recording instruments of the experimental equipment and need to be recorded together with the wallboard performance index data of each experimental sample; Step 2: Extract features and preprocess the collected data to convert raw material ratios and process parameters into numerical features that can be processed by machine learning algorithms; Feature scaling and standardization operations are required; 2.

1. Represent different raw material ratios as numerical features so that machine learning algorithms can process them; use the weight ratio of the raw materials as features, and use the weight ratio of each raw material such as PET powder, NaTi2(PO4)3, glycerol, and processing aid as features; Convert the process parameters in the preparation process into numerical features; Including process parameters such as temperature, pressure, and stirring time as characteristics; 2.

2. Perform feature scaling on raw material ratios and process parameters and map them to an appropriate range; avoid instability in the model training process due to scale differences in feature values; use the minimum-maximum scaling method to perform feature scaling; For categorical features, encoding is required to convert them into numerical features; use one-hot encoding or label encoding for processing; Perform feature selection based on the correlation between features and wall panel performance indicators, and select features that have a greater impact on the model prediction performance; use statistical methods, correlation coefficient analysis or feature importance evaluation methods to perform feature selection; Step 3: Select a suitable machine learning model to establish the relationship between raw material ratio and wallboard performance; considering the complex impact of raw material ratio on wallboard performance, select deep neural network DNN and gradient boosting tree GBT; 3.1 Deep Neural Network DNN Implementation Process: Design the structure of the neural network, including the input layer, hidden layer and output layer; according to the patent content, the number of nodes in the input layer should correspond to the number of characteristics of raw material ratio and process parameters, and the number of nodes in the output layer should correspond to the number of wallboard performance indicators; Use the training set data to train the neural network and adjust the network parameters to minimize the loss function; use the back propagation algorithm and optimizer for training; 3.2DNN Improvements: Introducing the attention mechanism: Introducing the attention mechanism before the input layer of the neural network to assign weights to the raw material ratio and process parameters; using the self-attention mechanism or the attention mechanism network to implement the importance assessment between the raw material ratio and the process parameters; specifically, setting the input layer to X, using the attention mechanism to obtain the weighted input feature X′=Attention(X); Increase network depth and width: Increase the number of hidden layers and the number of neurons in each hidden layer, and use residual connections or batch normalization techniques to speed up network training and improve model stability; Derivation of the attention mechanism formula: Let X∈R n×m Represents the feature matrix of the input layer, where n represents the number of samples and m represents the number of features; the formula derivation process of the self-attention mechanism is as follows: A=softmax(WXQ T ) X′=AX Meaning of network structure parameters: n: number of samples; m: the number of features of raw material ratio and process parameters; W, Q: weight matrices in the attention mechanism, used to calculate the importance of input features; A: Attention score, indicating the importance of each feature; X′: weighted input features, used to pass into the hidden layer of the neural network; 3.3 Gradient Boosting Tree GBT Improvement: Improve the gradient boosting tree model to better capture the complex effects of raw material ratio on wallboard performance; Using domain knowledge of raw material ratios and process parameters, feature engineering is performed to extract more effective features to enhance the model’s ability to capture the relationship between raw material ratios and wallboard performance; Enhanced regularization: During the model training process, regularization terms are added to limit the size of model parameters and prevent overfitting. For the gradient boosting tree model, regularization terms are added to the loss function, including L1 regularization terms or L2 regularization terms. Add a regularization term to the loss function to limit the complexity of the model: Among them, L(y i ,F(x i )) represents the loss function of the model, y i Usually represents the prediction result or model function of the model. In the iterative process, y i Gradually approaching the true target value F(x i );λ represents the coefficient of the regularization term. This parameter controls the strength of regularization. The larger the parameter, the greater the influence of regularization. Ω(F) represents the complexity of the model;λ: the coefficient of the regularization term; Model integration: Considering the complex influence of raw material ratio, model integration methods are used, including bagging and boosting; multiple gradient boosting tree models of different depths are integrated, and the prediction results of multiple models are synthesized by voting; The above improvement scheme makes better use of the gradient boosting tree model to establish the association between the raw material ratio and the wallboard performance, and improves the prediction performance and generalization ability of the model; the association model between the raw material ratio and the wallboard performance is established by combining the two, and a suitable machine learning algorithm is selected to achieve performance optimization; Step 4: Use the cross-validation method to evaluate and verify the trained model to ensure that the prediction performance of the model meets the requirements; Divide the original data set into k subsets of equal size, select k-1 subsets as training sets each time, and use the remaining subset as validation set. Repeat this process k times, using a different validation set for validation each time. The average value of the evaluation indicators obtained in each verification is used as the final evaluation result to evaluate the prediction performance of the model; The data set contains N samples, and k-fold cross validation is performed. Each time, N / k samples are selected as the validation set, and the remaining samples are used as the training set; k: the number of cross-validation folds, indicating how many subsets the data set is divided into for cross-validation; N: the number of samples in the data set; Calculate the evaluation indicators for each validation, and then take the average value as the final model evaluation result; Step 5: Deploy the trained model to the production environment, monitor the raw material ratio and process parameters in real time, and make adjustments based on the model’s predicted results to optimize the wallboard performance; 5.1 Deploy the trained model to the production environment to monitor and predict the wallboard performance in real time; deploy the model as a service, receive real-time raw material ratios and process parameters as input, and output predicted wallboard performance indicators; 5.2 Real-time monitoring: Set up sensors in the production environment to monitor changes in raw material ratios and process parameters in real time; X: characteristic vector of raw material ratio and process parameters, including PET powder, NaTi2(P04)3, glycerol, and processing aids; y: wall panel performance indicators, including strength, wear resistance, and weather resistance; Transmit the monitored data to the model service for real-time prediction and performance evaluation; ypre d : Wall panel performance index predicted by the model; 5.3 Adjustment control: After receiving the real-time data, the model service predicts the performance of the wall panel; According to the results predicted by the model, determine whether the current raw material ratio and process parameters need to be adjusted; If the wallboard performance index predicted by the model is different from the target value, the raw material ratio and process parameters are adjusted automatically or manually; 5.4 Monitoring and Adjustment Strategy: Threshold determination: Set a certain threshold. When the gap between the model prediction result and the actual performance index exceeds the threshold, the parameter adjustment is triggered. If the strength predicted by the model is set to be 5% lower than the target value, the raw material ratio and process parameters are automatically adjusted. Feedback adjustment: Update and optimize the model based on feedback information from actual production; regularly collect actual production data and retrain the model to adapt to changes in the production environment and optimize model performance.

6. The method for preparing a PET sodium titanium film for high-end wallboard according to claim 3 or 5, characterized in that: In the step 2, the dissolution method of glycerol is: put 0.72g of glycerol into 62mL of distilled water to dissolve it.

7. The method for preparing a PET sodium titanium film for high-end wallboard according to claim 3 or 5, characterized in that: In the step 3, when adding PET powder, it is added in small amounts and multiple times, and the standing time is 50 to 120 minutes.

8. The method for preparing a PET sodium titanium film for high-end wallboard according to claim 3 or 5, characterized in that: In the step 4, the constant temperature is 65-75° C., the temperature of the drying oven is 65-75° C., and the drying time is 40-60 min.

9. The method for preparing a PET sodium titanium film for high-end wallboard according to claim 3 or 5, characterized in that: In the step 5, the thickness of the prepared PET sodium titanium film is 0.16-0.20 mm.

Images