Design method of high-flame-retardant tin-based polymer-containing composite material based on machine learning

Abstract

The invention belongs to the technical field of flame-retardant material design, and discloses a design method of a high-flame-retardant tin-based polymer-containing composite material based on machine learning, and the method comprises the following steps: S1, model construction; s2, acquiring an experimental data set; s3, dividing a data set; s4, performing model training; and S5, model application: applying the trained machine learning model, and outputting a predicted design scheme of the high-flame-retardant tin-based polymer-containing composite material. Based on the genetic algorithm and the ridge regression model, the flame-retardant mechanism of the tin-based polymer composite material is deduced by calculating the reaction degree of each substance according to the reaction degree of the substance and the deduction mechanism, the design of the flame-retardant polymer composite material with high flame-retardant performance can be guided, and compared with a trial-and-error design method, the method is low in cost and easy to implement. The method has high practical application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flame-retardant composite material design, and particularly relates to a design method of high flame-retardant tin-containing polymer composite material based on machine learning. BACKGROUND

[0002] Polymer materials have been widely used in many fields such as electronics, automobiles, packaging and medical treatment due to their light weight, high strength, chemical corrosion resistance, convenient processing and functional designability. However, polymer materials release a large amount of heat energy (the total heat release amount THR is generally high) when burning, and have poor thermal stability. The high THR is usually accompanied by sustained combustion and flame spread, which may cause a chain fire and pose a serious safety hazard. In addition, due to the poor thermal stability, the structure cannot remain intact during the combustion process, which poses a threat to people's life and property, which to some extent limits its more extensive application.

[0003] The flame-retardant mechanism has certain guiding significance for designing composite materials with excellent flame-retardant performance. Given the huge virtual space and huge time and cost consumption, it is challenging for traditional “experimental trial and error method” to explore the flame-retardant mechanism for designing polymer composite materials with high flame-retardant performance. The data-driven machine learning method reveals the hidden correlation in the data, predicts the potential rules in the unknown field and identifies potential new materials according to the data, providing an efficient and accurate way to accelerate the development of new materials. However, the machine learning model is usually a black box model, and the relationship between the features and the performance cannot be clearly explained. In addition, not every feature has a strong correlation with the performance, and feature engineering is needed to explore the most relevant feature set.

[0004] Linear regression algorithm is an interpretable algorithm that can reveal the relationship between features and performance through a linear formula. Ridge regression algorithm (Ridge) is one of the linear regression algorithms, which is an improved least square estimation method and an improved linear regression algorithm for solving the problem of multicollinearity. By introducing an L2 regularization term in the loss function to constrain the model complexity, overfitting is effectively alleviated.

[0005] Feature engineering can be understood as an optimization process, and genetic algorithm is a bionic optimization algorithm that simulates the natural selection and genetic variation mechanism in natural evolution to solve complex search and optimization problems. It first generates an initial population by randomly generating a set of candidate solutions, then evaluates the fitness according to the fitness function, then selects excellent individuals according to the fitness evaluation, and then generates new individuals through crossover and mutation to get a new generation of population and repeat the above steps.

[0006] However, since the above-mentioned technologies have their own application methods and objects, there are still many technical difficulties in combining them to design flame-retardant polymer composites, and no related reports have been seen. The conventional exploration of the flame-retardant mechanism of polymer composites has the defects of high cost and long time consumption, which must be improved. SUMMARY

[0007] In view of the deficiencies in the above background art, the present application aims to provide a design method for high-flame-retardant tin-containing polymer composites based on machine learning, which is based on a machine learning model combining genetic algorithm and ridge regression. The design method for high-flame-retardant tin-containing polymer composites is simple, fast, low-cost and reliable in results, so as to solve the problems of high cost and long time consumption in exploring the flame-retardant mechanism of polymer composites in the prior art.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical solutions: A design method for high-flame-retardant tin-containing polymer composites based on machine learning, comprising the following steps: S1, model construction Construct a machine learning model combining genetic algorithm and ridge regression machine learning method for predicting the thermal stability and flame-retardant reaction mechanism of high-flame-retardant tin-containing polymer composites; S2, obtaining experimental data set Collect the total heat release of flame-retardant polymer composites, the mass fraction of fillers and matrix and related performance from experimental data, and perform thermal stability data analysis to obtain an experimental data set; S3, dividing data set Randomly divide the experimental data set into a training set and a test set in proportion, use the machine learning model of step S1, and based on the genetic algorithm, heuristically select the optimal feature subset of the ridge regression model in the feature space to determine the coefficient R 2 As an index for selecting the optimal feature subset, the optimal feature subset and the random seed for dividing the training and test sets are selected; S4, model training Train the interpretable ridge regression model using the training set and the test set to obtain the trained machine learning model and the regression equation; S5, model application Deploy the model, use the trained machine learning model to calculate the reaction degree of each component in the known tin-containing polymer composite, and according to the reaction degree of each component, obtain the tin-containing flame-retardant mechanism and parameter data, and then apply them to the design of new tin-containing polymer composites to output the predicted design scheme of high-flame-retardant tin-containing polymer composites.

[0009] Wherein, step S1 model construction, based on genetic algorithm and ridge regression combined feature selection, specifically includes the following steps: S1-1, first randomly generate random seeds, divide the data set into training data set and test data set according to 4:1, set the genetic iteration number of genetic algorithm to 20 to 50 times, the crossover probability is 0.5 to 0.9, the mutation probability is 0.7 to 1, the population number is 35 to 50, the method of 5-fold cross validation is used to evaluate the feature subset, assuming that the sample number is n, randomly select to The number of features as the initial length of the feature, the ridge regression model as the model of screening features, the model parameters remain the initial model parameters, the fitness function of genetic algorithm is the root mean square error of the model, the optimization goal of genetic algorithm is to minimize the RMSE of the model, when selecting, if the number of features is less than two, return the maximum value of RMSE, in this way, the invalid subset is punished; In the process of crossover and mutation, two-point crossover and bit flip mutation are adopted; S1-2, then select the optimal feature subset according to the model performance obtained by leave-one-out cross validation method, R 2 and RMSE as the index of selecting the optimal subset, the higher R 2 , the lower RMSE, the better the feature subset.

[0010] The high flame-retardant tin-containing polymer composite material in step S1 is a tin-containing flame retardant, including polymer flame-retardant composite materials with polyamide (PA), polycarbonate / acrylonitrile-butadiene-styrene (PC / ABS), polyformaldehyde (POM), polybutylene terephthalate (PBT), polypropylene (PP), polyvinyl chloride (PVC) as the matrix.

[0011] Step S2 obtains the experimental data set, including the following steps: S2-1, the initial data set is obtained by experimental data, and the following data of a plurality of polypropylene flame-retardant polymer composite materials is collected from the experimental data: total heat release, mass fraction of formula, total oxygen consumption, limiting oxygen index (LOI), peak heat release rate (PHRR), total smoke release, maximum average heat release rate (MARHE), specific extinction area, maximum effective combustion heat and mass loss rate; S2-2, then reveal the thermal stability of the material through data analysis.

[0012] Step S3 divides the data set, including the following steps: S3-1, use the genetic algorithm combined with ridge regression machine learning model constructed in step S1 to select the optimal feature subset and the optimal random seed for dividing the training and test set; S3-2, after determining the optimal feature set, a ridge regression algorithm is used to establish a model, and the experimental THR data of part of the flame-retardant polymer composite in the training set randomly divided from the experimental data set and the mass fraction of PP and MPP and the total oxygen consumption are used as features to establish the leave-one-out cross-validation result of the ridge regression model; S3-3, the leave-one-out cross-validation Pearson correlation coefficient (R) of the training set is determined, the coefficient of determination R 2 , the root mean square error RMSE, and the robustness of the model is determined; S3-4, the THR data of part of the flame-retardant polymer composite in the test set randomly divided from the experimental data set and the mass fraction of PP and MPP and the total oxygen consumption data are used as features to establish the test set of the ridge regression model, and the generalization ability of the model is verified; S3-5, the linear equation is calculated based on the established ridge regression model.

[0013] The method for inferring the flame-retardant mechanism in step S5 is to infer the reaction mechanism in the flame-retardant process according to the reaction degree of the substance, comprising the following steps: S5-1, the reaction degree of each substance is calculated according to the actual oxygen consumption and the oxygen consumption when each substance is completely burned. The actual oxygen consumption of each substance is calculated according to the following formula: ; The theoretical oxygen consumption of each substance is calculated according to the element content of each substance and the corresponding oxygen consumption. Taking polypropylene as an example, first, the mass fraction of carbon is calculated according to the molecular formula C3H6 of polypropylene: ; Among them, is the relative atomic mass of carbon, is the relative molecular mass of polypropylene, according to the mass fraction and the mass of polypropylene, the amount of substance of carbon element can be calculated: ; Among them, is the mass of polypropylene, after obtaining the amount of substance of carbon element, according to the proportion coefficient in the chemical equation, the corresponding oxygen consumption is obtained, and the oxygen consumption of all elements reacting with oxygen in each substance is added, then the theoretical oxygen consumption is calculated; S5-2, the reaction degree of each substance is calculated according to the theoretical oxygen consumption and the actual oxygen consumption: ; S5-3, for each data, the calculation is performed once, and finally, the average value represents the reaction degree of each substance, and the flame-retardant mechanism is analyzed according to the reaction degree of each substance.

[0014] Compared with the prior art, the present application has at least the following advantages and remarkable features: 1、The present application designs a machine learning framework for revealing the thermal stability and flame retardant mechanism of tin-based polymer composite materials by combining existing experimental data and various algorithm frameworks, and enables simple analysis of the flame retardant mechanism of polymer flame-retardant polymer composite materials.

[0015] 2、The present application uses a framework based on genetic algorithm and ridge regression algorithm to quickly analyze the flame retardant mechanism of polymer flame-retardant polymer composite materials, and the flame retardant mechanism of flame-retardant polymers can be obtained through calculation only, which is faster, more convenient, lower in cost and better in result reliability than traditional experimental methods.

[0016] 3、The design method provided by the present application uses a model combining genetic algorithm and ridge regression to analyze the flame retardant mechanism of flame-retardant polymer composite materials, which has practical value for designing flame-retardant polymer composite materials with excellent flame retardant performance and thermal stability, and also provides a new tool for the field of machine learning model assisted mechanism exploration of flame-retardant materials. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description only illustrate the embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to these drawings without creative labor.

[0018] Fig. 1 Comparison diagram of test set experimental value and predicted value of the ridge regression model of the embodiment of the present application.

[0019] Fig. 2 Comparison diagram of test set experimental value and predicted value of the ridge regression model of the embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the technical problems, technical solutions and beneficial effects of the present application more clear, the present application will be further described in detail below with reference to the drawings.

[0021] Basic embodiment A design method of high flame-retardant tin-based polymer composite material based on machine learning, comprising the following steps: S1、Model construction Construct a genetic algorithm for predicting the thermal stability and flame retardant reaction mechanism of high flame-retardant tin-based polymer composite materials, and a machine learning model combining ridge regression machine learning method; In the model construction process, feature selection based on genetic algorithm and ridge regression combination includes the following steps: S1-1, first randomly generate random seeds, divide the data set into training data set and test data set according to 4:1, set the genetic iteration number of genetic algorithm to 20 to 50 times, the crossover probability to 0.5 to 0.9, the mutation probability to 0.7 to 1, the population number to 35 to 50, and use 5-fold cross validation method to evaluate the feature subset, assuming that the sample number is n, randomly select to the number of features as the initial length of the features, the ridge regression model as the feature selection model, the model parameters remain the initial model parameters, the fitness function of genetic algorithm is the root mean square error of the model, and the optimization goal of genetic algorithm is to minimize the RMSE of the model, when selecting, if the number of features is less than two, return the maximum value of RMSE, and punish the invalid subset in this way; In the process of crossover and mutation, two-point crossover and bit flip mutation are adopted; S1-2, then select the optimal feature subset according to the model performance obtained by leave-one-out cross validation method, R 2 and RMSE as the index for selecting the optimal subset, the higher R 2 , the lower RMSE, and the better feature subset.

[0022] The high-flame-retardant tin-containing polymer composite is a polymer flame-retardant composite material containing a tin-containing flame retardant, including a polymer flame-retardant composite material taking polyamide (PA), polycarbonate / acrylonitrile-butadiene-styrene (PC / ABS), polyformaldehyde (POM), polybutylene terephthalate (PBT), polypropylene (PP), and polyvinyl chloride (PVC) as the matrix.

[0023] S2, obtaining experimental data set Collecting the total heat release amount, mass fraction of filler and matrix, and related performance of the flame-retardant polymer composite material from the experimental data, and performing thermal stability data analysis to obtain the experimental data set, including the following steps: S2-1, the initial data set is obtained by sorting the experimental data, and the following data of the polypropylene flame-retardant polymer composite material is collected from the experimental data: total heat release amount, mass fraction of formula, total oxygen consumption, limiting oxygen index (LOI), peak heat release rate (PHRR), total smoke release amount, maximum average heat release rate (MARHE), specific extinction area, maximum effective combustion heat, and mass loss rate; S2-2, then reveal the thermal stability of the material through data analysis.

[0024] S3, dividing the data set The experimental data set is randomly divided into a training set and a test set in proportion, a machine learning model of step S1 is adopted, and an optimal feature subset of the ridge regression model is heuristically screened in the feature space based on a genetic algorithm to determine the coefficient R 2 As an index for screening the optimal feature subset, the optimal feature subset and the random seed for optimally dividing the training and test sets are screened; Among them, the data set is divided, including the following steps: S3-1, the genetic algorithm combined with the ridge regression machine learning model constructed in step S1 is used to screen the optimal feature subset and the random seed for optimally dividing the training and test sets; S3-2, after determining the optimal feature set, the ridge regression algorithm is used to establish a model, and the part of the experimental THR data of the flame-retardant polymer composite material in the training set randomly divided from the experimental data set and the mass fraction of PP and MPP and the total oxygen consumption are used as features to establish the leave-one-out cross-validation result of the ridge regression model; S3-3, the leave-one-out cross-validation Pearson correlation coefficient (R) of the training set, the coefficient of determination R2, and the root mean square error RMSE are determined to judge the robustness of the model; S3-4, the THR data of part of the flame-retardant polymer composite material in the test set randomly divided from the experimental data set and the mass fraction of PP and MPP and the total oxygen consumption data are used as features to establish the test set of the ridge regression model, and the generalization ability of the model is verified; S3-5, the linear equation is calculated based on the established ridge regression model.

[0025] S4, model training An interpretable ridge regression model is trained using the training set and the test set to obtain a trained machine learning model, and a regression equation is obtained; S5, model application The model is deployed, the trained machine learning model is used to calculate the reaction degree of each component in the known tin-based polymer composite material, and according to the reaction degree of each component, the tin-based flame-retardant mechanism and parameter data are obtained, which are applied to the design of new tin-based polymer composite materials, and the predicted design scheme of high-flame-retardant tin-based polymer composite materials is output.

[0026] The method of inferring the flame-retardant mechanism is to infer the reaction mechanism in the flame-retardant process according to the reaction degree of the substance, including the following steps: S5-1, the reaction degree of each substance is calculated according to the actual oxygen consumption of each substance and the oxygen consumption when it is completely burned. The actual oxygen consumption of each substance is calculated according to the following formula: ; The theoretical oxygen consumption of each substance is calculated according to the element content of each substance and the corresponding oxygen consumption. Taking polypropylene as an example, first, the mass fraction of carbon is calculated according to the molecular formula C3H6 of polypropylene: ; wherein, is the relative atomic mass of carbon, is the relative molecular mass of polypropylene, according to this mass fraction and the mass of polypropylene, the amount of substance of carbon element can be calculated: ; wherein, is the mass of polypropylene, after obtaining the amount of substance of carbon element, according to the proportion coefficient in the chemical equation, the corresponding oxygen consumption is obtained, and the oxygen consumption of all elements in each substance that react with oxygen is added to calculate the theoretical oxygen consumption; S5-2, calculate the reaction degree of each substance according to the theoretical oxygen consumption and the actual oxygen consumption: ; S5-3, for each data, the calculation is performed once, and finally, the average value represents the reaction degree of each substance, and according to the reaction degree of each substance, the flame retardant mechanism is analyzed.

[0027] The following will be described in detail in combination with the drawings and specific embodiments.

[0028] Embodiment 1 Referring to the accompanying Figs. 1-2 , the design method of high flame-retardant tin-containing polymer composite material based on machine learning provided in this embodiment is a specific application of the basic embodiment, which specifically uses the machine learning framework for revealing the flame-retardant mechanism of tin-containing polymer composite material based on genetic algorithm and ridge regression, quickly analyzes the flame-retardant mechanism of flame-retardant polymer composite material, and has certain reference significance for subsequent design of flame-retardant polymer composite material. The framework can be applied to polymer flame-retardant composite materials with polyamide (PA), polycarbonate / acrylonitrile-butadiene-styrene (PC / ABS), polyformaldehyde (POM), polybutylene terephthalate (PBT), polypropylene (PP), and polyvinyl chloride (PVC) as the matrix.

[0029] This embodiment takes tin-containing polypropylene flame-retardant composite material as an example, and its design method includes the following steps: (1) Construct a genetic algorithm-based ridge regression model framework. First, randomly generate a random seed, divide the data set into training and test sets in a 4:1 ratio, set the genetic algorithm iteration number to 20-50, the crossover probability to 0.5-0.9, the mutation probability to 0.7-1, the population size to 35-50, and use the 5-fold cross-validation method to evaluate the feature subset. Assuming the sample size is n, randomly select n / 10 to n / 3 features as the initial length of the features. The ridge regression model is used as the feature selection model, and the model parameters remain the same as the initial model parameters. The fitness function of the genetic algorithm is the root mean square error (RMSE) of the model, and the optimization goal of the genetic algorithm is to minimize the RMSE of the model. When selecting, if the number of features is less than two, return the maximum value of the RMSE. In this way, the invalid subset is penalized. In the cross-variation process, we use two-point crossover and bit flip mutation. Next, we select the optimal feature subset based on the model performance obtained by the leave-one-out cross-validation method, R 2 and RMSE as indicators for selecting the optimal subset. The higher R 2 , the lower the RMSE, and the better the feature subset.

[0030] (2) The initial data set is obtained by analyzing experimental data. From the experimental data, we collected 54 total heat release amounts of polypropylene flame-retardant polymer composites, the mass fraction of the formula, the total oxygen consumption, the limiting oxygen index (LOI), the peak heat release rate (PHRR), the total smoke release amount, the maximum average heat release rate (MARHE), the specific extinction area, the maximum effective combustion heat, and the mass loss rate. Data analysis reveals the thermal stability of the materials. According to the data analysis, the mass loss rates of about 57% of the flame-retardant polymer composites are less than 30%, indicating that these flame-retardant polymer composites have good stability. In addition, among the 54 collected flame-retardant polymer composites, 8 have both low total heat release and low mass loss rate. This indicates that flame-retardant polymer composites composed of PP, PAPP, ZS, and anti-dripping agents have good stability and good flame-retardant performance.

[0031] (3) The mass fraction of the formula, the total oxygen consumption, the limiting oxygen index (LOI), the peak heat release rate (PHRR), the total smoke release amount, the maximum average heat release rate (MARHE), the specific extinction area, and the maximum effective combustion heat of the polypropylene flame-retardant polymer composite are used as the input of the modeling features. To reduce the impact of dimensional differences between features on the model, we use the standardization method in scikit-learn to standardize the features.

[0032] (4) The genetic algorithm ridge regression machine learning framework established in step (1) is used to screen the optimal feature subset and the optimal random seed for dividing the training and test sets.

[0033] (5) After determining the optimal feature set, a model is established using the ridge regression algorithm, and the results of leave-one-out cross-validation of the ridge regression model established by the 43 experimental THR data of the flame-retardant polymer composite materials in the training set randomly divided from the experimental data set and the mass fraction and total oxygen consumption of PP and MPP as features are shown in Fig. 1 The Pearson correlation coefficient (R) of the training set leave-one-out cross-validation is 0.999, the determination coefficient R 2 is 0.999, and the root mean square error RMSE is 0.067, indicating that the robustness of the model is good. The test set results of the ridge regression model established by the THR data of 11 flame-retardant polymer composite materials in the test set randomly divided from the experimental data set and the mass fraction and total oxygen consumption of PP and MPP as features are shown in Fig. 2 The results of the test set are R of 0.997, R 2 of 0.994, and RMSE of 0.447. This result can indicate that the model has good generalization ability. The linear equation is calculated based on the established ridge regression model.

[0034] (6) The reaction degree of each substance is calculated according to the actual oxygen consumption and the oxygen consumption when completely combusted. The actual oxygen consumption of each substance is calculated according to the following formula: ; The theoretical oxygen consumption of each substance is calculated according to the element content and the corresponding oxygen consumption. Taking polypropylene as an example, first, the mass fraction of carbon is calculated according to the molecular formula C3H6 of polypropylene: ; wherein, is the relative atomic mass of carbon, is the relative molecular mass of polypropylene, according to this mass fraction and the mass of polypropylene, the amount of substance of carbon element can be calculated: ; wherein, is the mass of polypropylene, after obtaining the amount of substance of carbon element, according to the proportion coefficient in the chemical equation, the corresponding oxygen consumption can be obtained, and the oxygen consumption of all elements reacting with oxygen in each substance is added to calculate the theoretical oxygen consumption. According to the theoretical oxygen consumption and the actual oxygen consumption, the reaction degree of each substance is calculated: ; For each piece of data, the calculation is performed once, and finally, the average value represents the reaction degree of each substance. According to the reaction degree of each substance, its flame retardant mechanism can be analyzed, and it can be applied to the design of new materials next time.

[0035] Embodiment 2 The design method of high-flame-retardant tin-containing polymer composite based on machine learning provided in this embodiment further calculates the covariance between the independent variable matrix and the dependent variable matrix according to the least squares method, estimates the regression coefficient, and adds a penalty term to control the regression coefficient in the least squares method through the ridge regression method, thereby reducing the risk of model overfitting. The regression coefficient is optimized through gradient descent until the loss function is minimized. According to the variable and the regression coefficient, the formula is obtained. The linear regression expression obtained based on the established ridge regression model is as follows: ; If the coefficient of total oxygen consumption is the largest, it indicates that there is a very strong positive correlation between the total heat release and the total oxygen consumption, and the total oxygen consumption is the most important characteristic affecting the total heat release.

[0036] Embodiment 3 The design method of high-flame-retardant tin-containing polymer composite based on machine learning provided in this embodiment is further based on the most important total oxygen consumption characteristic analyzed in Embodiment 1 and Embodiment 2, and calculates the reaction degree of each substance according to the physical and chemical method, as shown in Table 1.

[0037] Based on the calculated reaction degree of each substance, the invention can analyze the reaction mechanism of the entire combustion process. For polypropylene, its reaction degree is 44.72%, indicating that after adding the flame retardant, the combustion degree of polypropylene in the combustion process is reduced, indicating that the flame retardant plays a role in flame retardation. The remaining three flame retardants and additives are piperazine pyrophosphate (PAPP), melamine polyphosphate (MPP), and anti-dripping agent (ADP). Among them, PAPP has the highest reaction degree, followed by MPP. It is indicated that in the condensed phase, PAPP and MPP generate a large amount of phosphate to form a dense carbon layer, which plays a role in flame retardation. In the gas phase, PAPP and MPP generate non-combustible gases such as phosphoric acid, which also prevent the next step of combustion. For ADP, its reaction degree is 41.21%, indicating that it plays a certain role in preventing polypropylene from dripping. The calculation of the reaction degree of each substance provides data support for revealing the flame retardant mechanism of the flame-retardant polymer composite. The reaction degree of each substance in the combustion process is shown in the following Table 1.

[0038] Table 1

[0039] In summary, the machine learning-based design methods for highly flame-retardant tin-containing polymer composites provided by the above-mentioned embodiments of the present invention focus on using genetic algorithms and ridge regression models. By calculating the reaction degree of each substance and then inferring the flame retardant mechanism of the tin-containing polymer composite based on the reaction degree of the substance, it can guide the design of flame-retardant polymer composites with high flame retardant properties. It is less expensive than the trial-and-error design method and has high practical application value. The framework model constructed by this design method can assist in the analysis of flame retardant mechanisms and also has certain guiding significance for the design of various other tin-containing polymer composites with high-performance flame retardant properties.

[0040] The above description is only an embodiment of the present invention and is not intended to limit the scope of protection of the present invention; the scope of protection of the present invention is defined by the claims in the appended claims, and all equivalent changes and modifications made based on the invention are within the scope of protection of the patent of the present invention.

Claims

1. A method for designing highly flame-retardant tin-based polymer composite materials based on machine learning, characterized in that: It includes the following steps: S1. Model construction Construct a genetic algorithm to predict the thermal stability and flame retardant reaction mechanism of highly flame-retardant tin-based polymer composites, combined with a machine learning model using ridge regression machine learning methods; S2. Obtaining experimental dataset The total heat release, mass fraction of filler and matrix, and related properties of flame-retardant polymer composites were collected from experimental data, and thermal stability data were analyzed to obtain an experimental data set. S3. Divide the dataset The experimental data set is randomly divided into a training set and a test set in proportion. The machine learning model in step S1 is used to heuristically screen the optimal feature subset of the ridge regression model in the feature space based on the genetic algorithm to determine the coefficient R 2 As an indicator for screening the optimal feature subset, screening the optimal feature subset and the random seed for the optimal division of the training and test sets; S4. Model training The training set and test set are used to train the interpretable ridge regression model to obtain the trained machine learning model and the regression equation; S5. Model Application Deploy the model and use the trained machine learning model to calculate the reaction degree of each component in the known tin-containing polymer composite material. Based on the reaction degree of each component, obtain the tin-containing flame retardant mechanism and parameter data, and then apply it to the design of new tin-containing polymer composite materials to output the predicted design scheme of highly flame-retardant tin-containing polymer composite materials.

2. The method for designing a highly flame-retardant tin-containing polymer composite material based on machine learning according to claim 1, characterized in that: Step S1: Model construction, based on feature screening combining genetic algorithm and ridge regression, specifically includes the following steps: S1-1. First, randomly generate random seeds and divide the data set into training data set and test data set at a ratio of 4:

1. Set the number of genetic algorithm iterations to 20 to 50, the crossover probability to 0.5 to 0.9, the mutation probability to 0.7 to 1, the population size to 35 to 50, and use the 5-fold cross validation method to evaluate the feature subset. Assume that the number of samples is n and randomly select arrive The number of features is used as the initial length of the features, the ridge regression model is used as the model for screening features, the model parameters are kept as the initial model parameters, the fitness function of the genetic algorithm is the root mean square error of the model, and the optimization goal of the genetic algorithm is to minimize the RMSE of the model. During selection, if the number of features is less than two, the maximum value of the RMSE is returned, and invalid subsets are penalized in this way; in the crossover mutation process, two-point crossover and bit flip mutation are used; S1-2, then select the optimal feature subset based on the model performance obtained by the leave-one-out cross-validation method, R 2 and RMSE as the indicators for selecting the optimal subset, when R 2 The higher it is, the lower the RMSE is and the better the feature subset is.

3. The method for designing a highly flame-retardant tin-containing polymer composite material based on machine learning according to claim 1, characterized in that: The highly flame-retardant tin-based polymer composite material in step S1 is a tin-based flame retardant, including a polymer flame-retardant composite material with polyamide PA, polycarbonate / acrylonitrile-butadiene-styrene PC / ABS, polyoxymethylene POM, polybutylene terephthalate PBT, polypropylene PP, and polyvinyl chloride PVC as the matrix.

4. The method for designing a highly flame-retardant tin-containing polymer composite material based on machine learning according to claim 1, characterized in that: Step S2 obtains the experimental data set, including the following steps: S2-1. The initial data set was obtained by collating experimental data. The following data of multiple polypropylene flame retardant polymer composites were collected from the experimental data: total heat release, mass fraction of the formula, total oxygen consumption, limiting oxygen index (LOI), peak heat release rate (PHRR), total smoke release, maximum average heat release rate (MARHE), specific extinction area, maximum effective heat of combustion, and mass loss rate; S2-2. Then reveal the thermal stability of the material through data analysis.

5. The method for designing a highly flame-retardant tin-containing polymer composite material based on machine learning according to claim 1, characterized in that: Step S3 divides the data set, including the following steps: S3-1, using the genetic algorithm constructed in step S1 combined with the ridge regression machine learning model to select the optimal feature subset and the random seed for optimally partitioning the training and test sets; S3-2. After determining the optimal feature set, a ridge regression algorithm was used to establish a model. The results of the leave-one-out cross-validation method were used to establish the ridge regression model using some flame-retardant polymer composite experimental THR data and the mass fraction and total oxygen consumption of PP and MPP in the training set randomly divided from the experimental data set as features; S3-3. Leave-one-out cross-validation of the training set Pearson correlation coefficient R and determination coefficient R 2 , root mean square error RMSE, to judge the robustness of the model; S3-4. The THR data of some flame-retardant polymer composites and the mass fraction and total oxygen consumption data of PP and MPP in the test set randomly divided from the experimental data set are used as the test set for the ridge regression model established by features to verify the generalization ability of the model; S3-5. Calculate the linear equation based on the established ridge regression model.

6. The method for designing a highly flame-retardant tin-containing polymer composite material based on machine learning according to claim 1, characterized in that: The method for inferring the flame retardant mechanism in step S5 is to infer the reaction mechanism in the flame retardant process based on the reaction degree of the substance, and includes the following steps: S5-1. Calculate the degree of reactivity of each substance based on its actual oxygen consumption and the oxygen consumption during complete combustion. The actual oxygen consumption of each substance is calculated using the following formula: Actual oxygen consumption of each substance (g) = total oxygen consumption (g) x mass fraction of each substance (%) The theoretical oxygen consumption of each substance is calculated based on the element content of each substance and the corresponding oxygen consumption. Taking polypropylene as an example, the mass fraction of carbon is first calculated based on the molecular formula of polypropylene, C3H6: ； in, is the relative atomic mass of carbon, is the relative molecular mass of polypropylene. Based on this mass fraction and the mass of polypropylene, the amount of carbon can be calculated: ； in, is the mass of polypropylene. After obtaining the amount of carbon, the corresponding oxygen consumption is obtained according to the proportional coefficient in the chemical equation. The oxygen consumption of all elements that react with oxygen in each substance is added together to calculate the theoretical oxygen consumption. S5-2. Calculate the reaction degree of each substance based on the theoretical oxygen consumption and the actual oxygen consumption: ； S5-3. This calculation is performed for each piece of data. Finally, the average value is used to represent the reaction degree of each substance. Based on the reaction degree of each substance, its flame retardant mechanism is analyzed.

Images