A Materials Genome-Based Toughness Classification and Prediction Method for Polyester Powder Coatings

Abstract

This invention discloses a method for classifying and predicting the toughness of polyester powder coatings based on materials genomics. This method converts the chemical structure of coating raw materials into SMILES expressions, and uses the RDKit program to convert these SMILES structural expressions into binary string molecular fingerprints, forming a materials genome. A database corresponding to the materials formulation, process, and coating performance is then constructed. Simultaneously, the toughness of polyester powder coatings is classified into three categories for subsequent performance prediction of novel powder coatings. This invention uses data such as materials genome structure, formulation, and process as input sets, and the corresponding coating toughness as the output parameter. After dividing the data into training and validation sets, Python software is used to perform machine learning on the relationship between materials genome structure, formulation, process, and performance changes, constructing a toughness prediction model for polyester coatings for the design and formulation development of novel powder coating products. This method has the advantages of convenience, speed, and high accuracy, effectively shortening the development cycle of novel polyester coating products and can be used to guide the improvement of material toughness and product functional design.

Description

Technical Field

[0001] This invention belongs to the field of polymer material performance prediction technology, specifically involving a method for predicting the toughness of polyester powder coatings based on materials genome. Background Technology

[0002] Traditionally, large-scale experiments and analytical testing are essential processes in the development of new materials. However, the development of new materials requires significant resources and equipment, necessitating long-term and large-scale repetition and adjustments of a certain quantity of materials. This makes the development process extremely slow. The emergence of computer simulation methods, particularly density functional theory (DFT), Monte Carlo simulations, and molecular dynamics, has driven the first computational revolution in materials science, enabling researchers to more effectively explore the relationship between atomic structure, aggregate structure, and material properties using computers. The combination of experiments and computer simulations has significantly reduced the time and cost of materials design. However, computer simulations still cannot establish correlations between monomer structure, proportions, and preparation processes and many macroscopic properties of materials (such as toughness, tensile strength, and thermal stability), thus hindering the prediction of these material properties. With the accumulation of materials data, the improvement of machine learning algorithms, and the exponential growth of computing power, materials genome-based methodologies for predicting materials properties have gradually become a research hotspot for scientists. Machine learning algorithms are the core of materials genome research and development, and have been widely applied in biopharmaceuticals and chemical synthesis, beginning to emerge in the field of materials science.

[0003] Polyester powder coatings, due to their excellent "4E" (high efficiency, economy, environmental protection, and resource conservation) characteristics, have been widely popular since their emergence, with applications spanning home appliances, automobiles, construction, engineering machinery, 3C products, and highway guardrails. Currently, research on polyester powder coatings mainly focuses on material modification, formulation development, and process optimization. Research on establishing databases of existing resin structures, formulations, and performance data, as well as on predicting the performance of novel resin materials, is relatively limited. The development of polyester powder coating products based on materials genomics is essentially nonexistent. Therefore, this invention develops a powder coating performance prediction technology based on materials genomics. It establishes a database of the correspondence between material genome structure, formulation, ratio, processing conditions, and coating performance, and develops machine learning-based coating performance prediction models (such as toughness, color difference, and gloss retention) to predict the performance of novel polyester powder coatings. This aims to accelerate the R&D cycle of resin products, enabling their application in home appliances, automobiles, and construction. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the toughness classification of polyester powder coatings based on materials genome. This method is based on historical experimental data, integrates materials genome structure and performance to form a corresponding database, and uses artificial intelligence algorithms to construct a powder coating toughness classification prediction model. The prediction results are highly accurate and can accelerate the research and development cycle of resin products.

[0005] The above-mentioned objective of this invention can be achieved through the following technical solution: a method for predicting the toughness classification of polyester powder coatings based on materials genome, comprising the following steps:

[0006] (1) Organize historical data on polyester powder coatings, including raw material chemical structure, ratio, molecular characteristic parameters, process and coating toughness data, to form a database;

[0007] (2) Convert the chemical structure of the raw materials in step (1) into a SMILES expression, convert the SMILES structure expression into a binary string molecular fingerprint, form material gene data, and fill it into the database accordingly;

[0008] (3) The database in step (2) is divided into groups, with one part used as the training set and the remaining part as the test set. The raw material chemical structure, ratio, molecular characteristic parameters and process data of the training set are used as input parameters, and the coating toughness data of the training set are used as output parameters. The model is trained using machine learning algorithms to construct a coating toughness classification prediction model. The raw material chemical structure, ratio, molecular characteristic parameters and process data of the test set are used as input parameters to predict the toughness category of the coating and verify the accuracy of the coating toughness classification prediction model.

[0009] (4) Extract the material gene data obtained in step (2) as feature parameters, and use the principal component analysis algorithm to analyze the feature parameters to determine the key gene structure of the toughness of polyester powder coating.

[0010] (5) Design a new polyester powder coating formulation containing the key gene structure in step (4). Use the raw material chemical structure, ratio, molecular characteristic parameters and process data of the new polyester powder coating formulation as input parameters and input them into the coating toughness classification prediction model established in step (3) to predict its toughness classification. Perform synthesis verification on the formulation with better toughness in the prediction results. Determine the coating toughness of the new polyester powder coating formulation as the predicted category through detection and analysis, so as to accelerate the development of the new polyester powder coating.

[0011] Therefore, the method of the present invention can obtain the toughness classification of the coating based on the structural data and the designed formulation data, which can be used to guide the design and performance improvement of new powder coating products.

[0012] In the above-mentioned powder coating toughness classification and prediction method based on materials genome:

[0013] In fact, the data types of the polyester powder coating database mentioned in step (1) include, but are not limited to, the raw material chemical structure, proportion, molecular characteristic parameters, process and coating toughness of the coating.

[0014] Preferably, in step (1), the coating toughness is classified into three categories: no cracks on both sides are defined as category 0, no cracks on the front side and cracks on the back side are defined as category 1, and cracks on both sides are defined as category 2.

[0015] Preferably, in step (2), the RDKit program is used to convert the SMILES structure expression into a binary string molecular fingerprint.

[0016] Preferably, the binary string molecular fingerprint in step (2) includes molecular fingerprints such as Morgan, Dayligt, Pubchem, MACCS, and FP2.

[0017] Preferably, the database in step (2) includes the chemical structure of powder coating raw materials, binary string molecular fingerprints, proportions, molecular characteristic parameters, process and coating toughness data.

[0018] Preferably, the molecular characteristic parameters include viscosity, molecular weight, and acid value, and the process includes pressure and glass transition temperature.

[0019] Preferably, in step (3), Python software is used to train the model using machine learning algorithms.

[0020] Preferably, the machine learning algorithm in step (3) is one or more of the following: improved neural network, support vector machine, random forest, regression analysis, deep learning and XGboost.

[0021] Preferably, in step (3), the raw material chemical structure, ratio, molecular characteristic parameters and process data of the training set are used as input parameters, and the coating toughness data of the training set are used as output parameters. The mapping relationship between the input parameters and the toughness category is constructed through machine learning algorithm, thereby predicting the toughness category of the coating, calculating the ROC curve and accuracy AUC of the model. When AUC≥0.95, the prediction result is reliable and the coating toughness classification prediction model has high accuracy. The raw material chemical structure, ratio, molecular characteristic parameters and process data of the test set are used as input parameters to predict the toughness category of the coating, calculate the ROC curve and accuracy AUC of the model. When AUC≥0.95, the prediction result is reliable and the accuracy of the coating toughness classification prediction model is verified.

[0022] Preferably, the key gene structure described in step (4) includes one or more of the following: benzene ring structure, tertiary carbon structure, tetramethylene, hexamethylene, cyclohexyl, tert-butyl, isopropyl, ether group and carboxyl group.

[0023] Compared with the prior art, the present invention has the following advantages:

[0024] (1) Traditional powder coating development methods are basically trial and error methods. The method of this invention is based on historical powder coating formulation data. It constructs a database through raw material structure, ratio, process, performance, etc., and uses machine learning algorithms to mine the key genes of powder coating toughness and predict its toughness classification, which can be used to guide the design and performance improvement of new powder coating products.

[0025] (2) This invention is the first to use materials genomics for performance prediction of powder coatings, fully exploring the key genes that affect the performance of materials structure, and using them to guide the development of new powder coating products;

[0026] (3) The method of the present invention has the advantages of being convenient, fast and accurate, which can effectively reduce the test time and product development cycle, and can be used to guide the design and performance improvement of new powder coating products. Attached Figure Description

[0027] Figure 1 This is the database of powder coating resins used in Examples 1-2;

[0028] Figure 2 This refers to the process of converting the molecular structure into binary digits in Examples 1-2;

[0029] Figure 3 The ROC curve and accuracy of the powder coating toughness three-class classification training set results in Example 1;

[0030] Figure 4 These are the ROC curves and accuracy rates of the powder coating toughness prediction set in Example 1.

[0031] Figure 5 The ROC curve and accuracy of the powder coating toughness three-class classification training set results in Example 2;

[0032] Figure 6 The ROC curve and accuracy of the powder coating toughness prediction set in Example 2 are shown. Detailed Implementation

[0033] Example 1

[0034] The powder coating toughness classification and prediction method based on materials genome provided in this embodiment includes the following steps:

[0035] (1) Organize historical formulation data for polyester powder coatings, including data on the chemical structure, proportions, molecular characteristic parameters, processes, and coating toughness of the raw materials, to form a database, such as... Figure 1As shown; among them, the coating toughness is classified into three categories, including no cracks on both sides of impact (Class 0), no cracks on the front side but cracks on the back side (Class 1), and cracks on both sides (Class 2), etc.

[0036] (2) The chemical structure of the raw materials is converted into SMILES expressions and populated into the database accordingly; the RDKit program is used to convert the SMILES structural expressions into binary string molecular Morgan fingerprints, such as... Figure 2 As shown, material genome data is generated and populated into the database;

[0037] (3) Different formulations and coating toughness categories were grouped, with 80% of the formulations used as the training set and the remaining 20% ​​as the test set. A hybrid learning algorithm combining SVM and random forest was constructed using Python software to train the coating toughness classification prediction model, forming a toughness classification model. The ROC curve of the model is shown below. Figure 3 As shown, the AUC is 0.9627;

[0038] (4) Using the gene structure, ratio, molecular parameters, and process of the test set as input parameters, predict the toughness category of powder coatings for different formulations, obtain the toughness data of the formulations, and calculate the ROC curve and AUC, such as... Figure 4 As shown, the calculated AUC is 0.9576, indicating that the prediction result is reliable.

[0039] (5) Extract the gene feature data of powder coating obtained in step (2) as feature parameters, and use the principal component analysis algorithm to analyze the feature parameter data to determine the key gene structure of powder coating toughness, namely para-benzene ring structure, meta-benzene ring structure, tertiary carbon structure, cyclohexane structure, and hexamethylene structure.

[0040] (6) By designing a novel powder coating formulation containing key genes, namely neopentyl glycol 41.9% (mass percentage, the same below), terephthalic acid 43.4%, isophthalic acid 9.5%, cyclohexanediol 4.6%, hexanediol 0.6%, viscosity 6320 (mPa·s), molecular weight 4519 (g / mol), acid value 30.4 (mgKOH / g), pressure 98 (Pa), glass transition temperature 63.3 (unit: °C), its coating toughness category is predicted, and the toughness category is 0, which is used to guide the optimization of the formulation and product development;

[0041] (7) Based on the predicted toughness category of the formulation, a synthesis verification was conducted through melt polycondensation. The synthesized polyester raw materials were 41.9% neopentyl glycol, 43.4% terephthalic acid, 9.5% isophthalic acid, 4.6% cyclohexanediol, and 0.6% hexanediol. The resulting polyester had a viscosity of 6320 mPa·s, a molecular weight of 4519 g / mol, an acid value of 30.4 mgKOH / g, a synthesis pressure of 98 Pa, and a glass transition temperature of 63.3 °C. After crosslinking and thermosetting with triglycidyl isocyanurate (TGIC), a powder coating was prepared. The falling ball impact toughness test of the powder coating was conducted according to the standard method of GB / T 14485-1993. It was found that the toughness of the coating prepared by the new powder coating formulation was the predicted category 0, thus accelerating the development of the new polyester powder coating.

[0042] Example 2

[0043] (1) Organize historical formulation data of polyester resin powder coatings, including data on the chemical structure, proportions, processes, and toughness of the coating raw materials, to form a database, such as... Figure 1 As shown; among them, the coating toughness is classified into three categories, including no cracks on both sides of impact (Class 0), no cracks on the front side but cracks on the back side (Class 1), and cracks on both sides (Class 2), etc.

[0044] (2) The chemical structure of the raw materials is converted into SMILES expressions and populated into the database accordingly; the RDKit program is used to convert the SMILES structural expressions into binary string molecular Daylight fingerprints, such as... Figure 2 As shown, material genome data is generated and populated into the database;

[0045] (3) Different formulation-coating toughness categories were grouped, with 80% of the formulations used as the training set and the remaining 20% ​​as the test set. Using Python as the programming language, a Transformer deep learning network with a recursive structure was built on the PyTorch architecture based on the sequence prediction mechanism of the LSTM network. Simultaneously, the self-attention module in the traditional Transformer was replaced with a parallel module of sliding self-attention and cross-attention, effectively improving computational speed while maintaining high prediction accuracy. The network was trained using the aforementioned training set to obtain the weight model used for testing. The ROC curve of the model is shown below. Figure 5 As shown, the model's AUC is 0.9710;

[0046] (4) Using the gene structure, ratio, molecular characteristics, and process of the test set as input parameters, predict the toughness category of powder coatings for different formulations, obtain the toughness data of the formulations, and calculate the ROC curve and AUC of the prediction set, such as... Figure 6 As shown, the calculated AUC is 0.9598, indicating that the prediction result is reliable.

[0047] (5) Extract the gene feature data of powder coating obtained in step (2) as feature parameters, and use the principal component analysis algorithm to analyze the feature parameter data to determine the key gene structure of powder coating toughness, namely para-benzene ring structure, meta-benzene ring structure, tertiary carbon structure, isobutyl structure, and tetramethylene structure.

[0048] (6) By designing a novel powder coating formulation containing key genes, namely neopentyl glycol 44.2%, terephthalic acid 42.4%, isophthalic acid 8.5%, cyclohexanediol 5.2%, hexanediol 1.1%, viscosity 6410 (mPa·s), molecular weight 5521 (g / mol), acid value 32.9 (mgKOH / g), pressure 98 (Pa), glass transition temperature 64.0 (°C), its coating toughness category was predicted, and the toughness category was 0, which was used to guide the optimization of the formulation and product development.

[0049] (7) Based on the predicted toughness category of the formulation, a synthesis verification was conducted through melt polycondensation. The synthesized polyester raw materials were 44.2% neopentyl glycol, 42.4% terephthalic acid, 8.5% isophthalic acid, 5.2% cyclohexanediol, and 1.1% hexanediol. The synthesized polyester had a viscosity of 6410 mPa·s, a molecular weight of 5521 g / mol, an acid value of 32.9 mgKOH / g, a synthesis pressure of 98 Pa, and a glass transition temperature of 64.0 °C. After crosslinking and thermosetting with triglycidyl isocyanurate (TGIC), a powder coating was prepared. The falling ball impact toughness test of the powder coating was conducted according to the standard method of GB / T 14485-1993. It was found that the toughness of the coating prepared by the new powder coating formulation was the predicted category 0, thus accelerating the development of the new polyester powder coating.

[0050] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for classifying and predicting the toughness of polyester powder coatings based on materials genome, characterized by: Includes the following steps: (1) Organize historical data on polyester powder coatings, including raw material chemical structure, ratio, molecular characteristic parameters, process and coating toughness data, to form a database; (2) Convert the chemical structure of the raw materials in step (1) into a SMILES expression, convert the SMILES structure expression into a binary string molecular fingerprint, form material gene data, and fill it into the database accordingly; (3) The database in step (2) is divided into groups, with one part used as the training set and the remaining part as the test set. The raw material chemical structure, ratio, molecular characteristic parameters and process data of the training set are used as input parameters, and the coating toughness data of the training set are used as output parameters. The model is trained using machine learning algorithms to construct a coating toughness classification prediction model. The raw material chemical structure, ratio, molecular characteristic parameters and process data of the test set are used as input parameters to predict the toughness category of the coating and verify the accuracy of the coating toughness classification prediction model. (4) Extract the material gene data obtained in step (2) as feature parameters, and use the principal component analysis algorithm to analyze the feature parameters to determine the key gene structure of the toughness of polyester powder coating. (5) Design a new polyester powder coating formulation containing the key gene structure in step (4). Use the raw material chemical structure, ratio, molecular characteristic parameters and process data of the new polyester powder coating formulation as input parameters and input them into the coating toughness classification prediction model established in step (3) to predict its toughness classification. Perform synthesis verification on the formulation with better toughness in the prediction results. Determine the coating toughness of the new polyester powder coating formulation as the predicted category through detection and analysis, so as to accelerate the development of new polyester powder coating. In step (3), the raw material chemical structure, ratio, molecular characteristic parameters and process data of the training set are used as input parameters, and the coating toughness data of the training set are used as output parameters. The mapping relationship between the input parameters and the toughness category is constructed through machine learning algorithm, so as to predict the toughness category of the coating. The ROC curve and accuracy AUC of the model are calculated. When AUC≥0.95, the prediction result is reliable and the coating toughness classification prediction model has high accuracy. The raw material chemical structure, ratio, molecular characteristic parameters and process data of the test set are used as input parameters to predict the toughness category of the coating. The ROC curve and accuracy AUC of the model are calculated. When AUC≥0.95, the prediction result is reliable and the accuracy of the coating toughness classification prediction model is verified. The key gene structure mentioned in step (4) includes one or more of the following: benzene ring structure, tertiary carbon structure, tetramethylene, hexamethylene, cyclohexyl, isobutyl, isopropyl, ether group and carboxyl group.

2. The method for classifying and predicting the toughness of polyester powder coatings based on materials genomes according to claim 1, characterized in that: In step (1), the coating toughness is classified into three categories: no cracks on both sides are defined as category 0, no cracks on the front side and cracks on the back side are defined as category 1, and cracks on both sides are defined as category 2.

3. The method for predicting the toughness classification of polyester powder coatings based on materials genomes according to claim 1, characterized in that: In step (2), the RDKit program is used to convert the SMILES structure expression into a binary string molecular fingerprint.

4. The method for predicting the toughness classification of polyester powder coatings based on materials genome according to claim 1, characterized in that: The binary string molecular fingerprint mentioned in step (2) includes Morgan, Dayligt, Pubchem, MACCS, or FP2 molecular fingerprints.

5. The method for predicting the toughness classification of polyester powder coatings based on materials genomes according to claim 1, characterized in that: The database mentioned in step (2) includes the chemical structure of polyester powder coating raw materials, binary string molecular fingerprints, proportions, molecular characteristic parameters, process and coating toughness data.

6. The method for predicting the toughness classification of polyester powder coatings based on materials genomes according to claim 5, characterized in that: The molecular property parameters include viscosity, molecular weight, and acid value, and the process includes pressure and glass transition temperature.

7. The method for predicting the toughness classification of polyester powder coatings based on materials genomes according to claim 1, characterized in that: In step (3), Python software is used to train the model using machine learning algorithms.

8. The method for predicting the toughness classification of polyester powder coatings based on materials genomes according to claim 1, characterized in that: The machine learning algorithms mentioned in step (3) include one or more of the following: improved neural networks, support vector machines, random forests, regression analysis, deep learning, and XGboost.

Images