Method for predicting toughness of organic modified mineral polymer composite material based on machine learning

Abstract

The invention relates to the technical field of composite materials, in particular to a method for predicting toughness of an organic modified mineral polymer composite material based on machine learning. According to the method, a reliable prediction model for the interface binding energy between organic modified inorganic matters and polymer molecules in a composite material system is successfully constructed. The machine learning model developed by the invention can quickly analyze and learn mass material data, accurately capture a complex nonlinear relationship between an organic modifier structure and the performance of the composite material, have efficient prediction speed for interface bonding energy, and quickly screen and accurately evaluate the mechanical performance of the composite material. According to the method, the most potential candidate can be screened from numerous organic modifiers in a short time, the research and development period is greatly shortened, and the research and development cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of composite materials technology, and in particular to a method for predicting the toughness of organic modified mineral / polymer composite materials based on machine learning. Background Technology

[0002] Composite materials, due to their unique performance advantages, have become key materials driving the development of numerous industries. High-toughness composite materials, in particular, play an irreplaceable role in fields with extremely stringent material performance requirements, such as aerospace, automotive manufacturing, and defense. High-toughness composite materials can effectively resist impact loads and prevent brittle fracture under complex stress environments, thereby significantly improving product safety and reliability.

[0003] Currently, improving the performance of composite materials through organic modification is a common strategy. However, traditional methods face many challenges in the organic modification process. On the one hand, traditional experimental trial-and-error methods require the synthesis and testing of a large number of composite material samples with different organic modifier formulations, a process that is not only time-consuming and labor-intensive but also costly. On the other hand, from a theoretical calculation perspective, accurately simulating the complex physicochemical interactions between organic modifiers and the composite matrix is ​​extremely difficult. Due to the numerous interatomic interactions and various complex chemical reaction mechanisms involved, existing theoretical models struggle to accurately describe these interactions, leading to significant discrepancies between predicted results and actual conditions. Summary of the Invention

[0004] The purpose of this invention is to address the aforementioned shortcomings of existing technologies by proposing a method for predicting the toughness of organic modified mineral / polymer composite materials based on machine learning.

[0005] The present invention provides a method for predicting the toughness of organically modified mineral / polymer composite materials based on machine learning, comprising the following specific steps: Step S1: Collect structural feature data of organic modifiers from the PubChem database as input dataset samples for machine learning training; Step S2: Calculate the interfacial bonding energy between the (102) crystal plane of gypsum or by-product gypsum after modification with the above-mentioned organic modifier and polymer molecules using molecular dynamics. E int (102), which serves as the target dataset sample for machine learning training, together with step S1, constitutes the training data for machine learning. Step S3: Use the Magpie machine learning descriptor to generate 153 features for each material after conversion in step S2, find a pair of features whose absolute value of the Pearson correlation coefficient between features is greater than 0.9, and delete any one of the features. Step S4: The feature-processed dataset from step S3 is randomly divided into a training set and a test set according to the proportions. Step S5: Use the training set and test set from step S4 to train the deep learning algorithm and optimize the model using Bayesian hyperparameter tuning to obtain the trained deep learning model. Step S6: Integrate the SHAP model interpretation package with the optimal deep learning model obtained in step 5 to perform feature weight analysis. The top 6 features with the highest weight values ​​are retained as input data. Step S7: Randomly sample organic modifiers from the database. The numerical features derived from the structural data of the organic modifiers are then used to generate 153 features using the Magpie machine learning descriptor. Six features obtained from the analysis in Step S6 are selected as input data and fed into the deep learning model obtained in Step S5 to obtain the interfacial bonding energy between the (102) crystal plane of gypsum and polymer molecules after modification with the aforementioned organic modifier. E int (102); Step S8, through interface integration E int (102) Predict the toughness and interfacial bonding energy of composite materials. E int The larger the absolute value of (102), the greater the toughness of the composite material; Where, ∆ E = E int = E 聚合物-有机改性剂-石膏 -( E 聚合物 + E 有机改性剂-石膏 ) in, E 聚合物-有机改性剂-石膏 It is the total energy of the polymer-organic modifier-gypsum composite system. E 聚合物 and E 有机改性剂-石膏 These refer to the energy of the polymer system and the organic modifier-gypsum system, respectively. Steps S1 and S2 have no specific order.

[0006] Furthermore, in step S1, the molecular weight range of the collected organic modifiers is 0-1000 g / mol. Furthermore, the feature-enhanced dataset is randomly divided into a training set and a test set in a 4:1 ratio.

[0007] Furthermore, in step S7, the molecular weight of the organic modifier ranges from 10 to 600 g / mol.

[0008] Furthermore, in step S7, the organic modifier contains one or more of -COOH, -OH, -SH, and N(CH3)3.

[0009] Furthermore, in step S7, 88 organic modifiers are randomly sampled from the database each time.

[0010] Further, in step S5, the number of iterations n_estimators of the deep learning model is 10-500, and the initial learning rate learning_rate_init is e. -5 -e -1 .

[0011] Furthermore, the polymer is any one of polypropylene, polyvinyl chloride, polyethylene, and polycarbonate. Furthermore, by-product gypsum includes one of phosphogypsum, desulfurized stone, fluorogypsum, and citric acid gypsum.

[0012] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention successfully constructed a reliable predictive model for the interfacial binding energy between organic modified inorganic materials and polymer molecules in composite material systems. Compared with traditional technical approaches, this method can effectively evaluate the mechanical properties of composite materials without preparing actual samples and conducting mechanical property tests. Its significant advantage lies in that by avoiding cumbersome sample preparation and time-consuming performance testing procedures, it greatly shortens the research and development cycle, reduces resource consumption, and effectively reduces preparation and time costs. This efficient and convenient characteristic provides researchers and developers with a powerful tool for rapidly screening and accurately evaluating the mechanical properties of composite materials.

[0013] (2) The machine learning model developed in this invention can rapidly analyze and learn from massive amounts of material data, accurately capture the complex nonlinear relationship between the structure of organic modifiers and the properties of composite materials, and has a high-efficiency prediction speed for interfacial bonding energy, enabling rapid screening and accurate evaluation of the mechanical properties of composite materials. It can quickly screen the most promising candidates from numerous organic modifiers, greatly shortening the R&D cycle and reducing R&D costs. This efficient screening method helps to overcome the bottlenecks of traditional organic modification. Compared to traditional experiments, it can more quickly predict the toughness of composite materials. This helps to accelerate the screening of composite materials applicable to high-toughness fields, greatly speeding up the R&D cycle and promoting innovative development in the mechanical application field of composite materials. Attached Figure Description

[0014] Figure 1 MAE and RMSE data for MLPR and DL machine learning algorithms; Figure 2R for MLPR and DL machine learning algorithms 2 data; Figure 3 The Pearson correlation coefficient among the six retained features; Figure 4 To load the trained DL model onto the interfacial binding energies of the 88 randomly selected organic modifiers mentioned above. E int (102) The numerical distribution for prediction; Figure 5 Interfacial binding energy for molecular dynamics calculations E int (102) and the interface binding energy predicted by the model E int (102) comparison; Figure 6 A comparison of the toughness of the composite material before and after modification. Detailed Implementation

[0015] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0016] Example 1 Step S1: Calculate the interfacial bonding energy between phosphogypsum modified with 31 organic modifiers and polypropylene molecules using molecular dynamics. E int (102) Data; Interface integration energy E int The stability index is an indicator used to measure the interactions between components and the stability of the system. This index can be calculated using the formula below. E int The larger the negative value, the higher the stability of the system, and the greater the possibility of adsorption or binding. This means that when E int When the value is high and negative, the additive is more likely to act on the crystal surface. Conversely, when the value is low and negative, the additive is more likely to act on the crystal surface. E int When the value is 0 or positive, the additive is difficult to adsorb onto the crystal.

[0017] ∆ E = E int = E 聚丙烯-有机改性剂-磷石膏 -( E 聚丙烯 + E 有机改性剂-磷石膏 ) in, E 聚丙烯-有机改性剂-石膏It is the total energy of the polypropylene-organic modifier-phosphogypsum composite system. E 聚丙烯 and E 有机改性剂-磷石膏 These refer to the energy of the polypropylene system and the organic modifier-gypsum system, respectively.

[0018] Step S2: Obtain the structural data of the above 31 organic modifiers with molecular weights in the range of 20-200 g / mol from the PubChem database.

[0019] Step S3: Use feature hashing to transform the text features in the organic modifier structure data, and finally convert all features into numerical types.

[0020] Step S4: The transformed features from step S3 are used to generate 153 features using the Magpie machine learning descriptor.

[0021] Step S5: The feature-processed dataset from step S4 is randomly divided into a training set (80%) and a test set (20%).

[0022] Step S6: The training and test sets from step S5 are used to train two machine learning algorithms, MLPR and DL. Bayesian hyperparameter tuning is used to optimize the model. A 10-fold cross-validation method is employed to evaluate R... 2 MAE and RMSE yielded the best trained model for DL, such as Figure 1 and Figure 2 As shown.

[0023] Step S7: Use the SHAP model interpretation package to perform feature weight analysis by integrating it with the optimal ML obtained in step S6.

[0024] Step S8: Retain the top 6 features with the highest weights from step S7 and calculate the Pearson correlation coefficients among these 6 features, such as... Figure 3 As shown in Table 1, the meanings of numbers 1-6 are given in Table 1. Additionally, structural information of 88 organic modifiers containing functional groups such as -COOH, -OH, and -SH, with molecular weights ranging from 10 to 600 g / mol, was obtained from the PubChem database using a random sampling method. Using the above six main features as input data, a pre-trained deep learning model was loaded to analyze the interfacial binding energies of the 88 randomly obtained organic modifiers. E int (102) Make predictions. For example... Figure 4 As shown.

[0025] Step S9: Select the maximum value obtained in step S8. E int(102) data, corresponding to n-decanoic acid. Molecular dynamics was used to calculate the interfacial bonding energy between the (102) crystal plane of gypsum and polypropylene molecules after n-decanoic acid modification. E int (102) compares the two, such as Figure 5 As shown, the error is within 0.5%, indicating that the DL algorithm model has accurate predictive capabilities.

[0026] Step S10: Phosphogypsum was modified with n-decanoic acid and used as a filler in polypropylene to prepare a composite material. Its toughness was tested. Figure 6 As shown, the fracture tensile strain of the modified composite material is increased by about 16.6 times compared with the unmodified composite material, exhibiting excellent toughness, and further demonstrating the speed and accuracy of the prediction method.

[0027] The specific preparation method of the composite material is as follows: after grinding phosphogypsum, it is calcined at 600℃ for 4 hours to obtain calcined phosphogypsum. Then, decanoic acid and calcined phosphogypsum are mixed with anhydrous ethanol at a mass ratio of 1:10:180 for a period of time, dried, and then ground to obtain modified phosphogypsum filler.

[0028] Modified phosphogypsum filler obtained using a twin-screw extruder and low-density polyethylene particles were mixed at a mass ratio of 1:9 at 215 °C and then dried at 60 °C to obtain a high-toughness phosphogypsum-reinforced composite material.

[0029] Table 1. First 6 Feature Descriptors and Their Meanings

[0030] This embodiment employs an efficient machine learning method to study the interfacial bonding energy of composite materials. E int A highly efficient prediction model for composite materials was successfully constructed through modeling. The model achieved a determination coefficient of 0.601, a MAE of 7.151, and an RMSE of 8.793, demonstrating high prediction accuracy. This embodiment, by combining database analysis and theoretical calculations, established an accurate and efficient prediction model with advantages such as low cost and environmental friendliness. This machine learning-based method achieves high-precision prediction of interfacial bonding energy, providing strong technical support for the development of high-toughness polypropylene composite materials suitable for mechanical applications, and accelerating their design and application.

[0031] For any points not covered above, existing technologies shall apply.

[0032] Although specific embodiments of the present invention have been described in detail by way of examples, those skilled in the art should understand that the above examples are for illustrative purposes only and are not intended to limit the scope of the invention. Those skilled in the art can make various modifications or additions to the described specific embodiments or use similar methods to replace them, without departing from the direction of the invention or exceeding the scope defined by the appended claims. Those skilled in the art should understand that any modifications, equivalent substitutions, improvements, etc., made to the above embodiments based on the technical essence of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the toughness of organically modified mineral / polymer composite materials based on machine learning, characterized in that, The specific steps include the following: Step S1: Collect structural feature data of organic modifiers from the PubChem database as input dataset samples for machine learning training; Step S2: Calculate the interfacial bonding energy between the (102) crystal plane of gypsum or by-product gypsum after modification with the above-mentioned organic modifier and polymer molecules using molecular dynamics. E int (102), which serves as the target dataset sample for machine learning training, together with step S1, constitutes the training data for machine learning. Step S3: Use the Magpie machine learning descriptor to generate 153 features for each material after conversion in step S2, find a pair of features whose absolute value of the Pearson correlation coefficient between features is greater than 0.9, and delete any one of the features. Step S4: The feature-processed dataset from step S3 is randomly divided into a training set and a test set according to the proportions. Step S5: Use the training set and test set from step S4 to train the deep learning algorithm and optimize the model using Bayesian hyperparameter tuning to obtain the trained deep learning model. Step S6: Integrate the SHAP model interpretation package with the optimal deep learning model obtained in step 5 to perform feature weight analysis. The top 6 features with the highest weight values ​​are retained as input data. Step S7: Randomly sample organic modifiers from the database, and use Magpie machine learning descriptors to generate 153 features from the numerical features of the structural data of the organic modifiers. Select 6 features obtained from the analysis in step S6 as input data and input them into the deep learning model obtained in step S5 to obtain the interfacial bonding energy between the (102) crystal plane of gypsum or by-product gypsum and polymer molecules after modification with the above organic modifier. E int (102); Step S8, through interface integration E int (102) Predict the toughness and interfacial bonding energy of composite materials. E int The larger the absolute value of (102), the greater the toughness of the composite material; Where, ∆ E = E int = E 聚合物-有机改性剂-石膏 -( E 聚合物 + E 有机改性剂-石膏 ) in, E 聚合物-有机改性剂-石膏 It is the total energy of the polymer-organic modifier-gypsum composite system. E 聚合物 and E 有机改性剂-石膏 These refer to the energy of the polymer system and the organic modifier-gypsum system, respectively. Steps S1 and S2 have no specific order.

2. The method for predicting the toughness of organically modified mineral / polymer composite materials based on machine learning as described in claim 1, characterized in that, In step S1, the molecular weight range of the collected organic modifiers is 0-1000 g / mol.

3. The method for predicting the toughness of organically modified mineral / polymer composite materials based on machine learning as described in claim 1, characterized in that, The feature-enhanced dataset is randomly divided into training and test sets in a 4:1 ratio.

4. The method for predicting the toughness of organically modified mineral / polymer composite materials based on machine learning as described in claim 1, characterized in that, In step S7, the molecular weight range of the organic modifier is 10-600 g / mol.

5. The method for predicting the toughness of organically modified mineral / polymer composite materials based on machine learning as described in claim 1, characterized in that, In step S7, the organic modifier contains one or more of -COOH, -OH, -SH, and N(CH3)3.

6. The method for predicting the toughness of organically modified mineral / polymer composite materials based on machine learning as described in claim 1, characterized in that, In step S7, 88 organic modifiers are randomly sampled from the database each time.

7. The method for predicting the toughness of organically modified mineral / polymer composite materials based on machine learning as described in claim 1, characterized in that, In step S5, the number of iterations n_estimators of the deep learning model is 10-500, and the initial learning rate learning_rate_init is e. -5 -e -1 .

8. The method for predicting the toughness of organically modified mineral / polymer composite materials based on machine learning as described in claim 1, characterized in that, The polymer is any one of polypropylene, polyvinyl chloride, polyethylene, and polycarbonate.

9. The method for predicting the toughness of organically modified mineral / polymer composite materials based on machine learning as described in claim 1, characterized in that, By-product gypsum includes one of the following: phosphogypsum, desulfurized stone, fluorogypsum, and citric acid gypsum.

Images