Preparation method of high-strength flame-retardant CNF / PVA composite film based on machine learning assistance, composite film and application

Abstract

The application discloses a preparation method of high-strength flame-retardant CNF / PVA composite film based on machine learning assistance, a composite film and application, and is based on an active learning framework. A material database containing material composition parameters, process parameters and performance indexes is constructed, a prediction model between material composition, process and performance is established, performance prediction and screening are carried out on different modification schemes and formulations, and thus a target composite film scheme meeting the collaborative requirements of flame-retardant performance and mechanical performance is obtained. Based on the screening result, cellulose nanofiber is subjected to multi-element collaborative modification, and is blended with PVA to form a film, so that the high-flame-retardant CNF / PVA composite film is prepared, and experiment verification is obtained. The method can reduce the trial and error cost in the material research and development process, improve the design efficiency of the novel green flame-retardant composite material, and realize the predictable design of the material performance.

Description

Technical Field

[0001] This invention relates to the field of polymer composite materials and smart material design technology, and more specifically to a method for preparing a high-strength flame-retardant CNF / PVA composite film based on machine learning assistance, as well as the composite film and its applications. Background Technology

[0002] Cellulose nanofibers, as a widely available, renewable, and high-strength bio-based nanomaterial, have received considerable attention in the field of polymer composites in recent years. Polyvinyl alcohol (PVA) possesses excellent film-forming properties, biocompatibility, and biodegradability; introducing cellulose nanofibers into a PVA matrix can significantly improve the mechanical properties and structural stability of the material. However, due to the inherent flammability of both cellulose and PVA, and the brittleness of cellulose itself, their composite films still face challenges in practical applications regarding insufficient flame retardancy and mechanical properties.

[0003] To improve the flame retardant properties of CNF / PVA composite films, existing technologies typically employ methods such as introducing phosphorus-, nitrogen-, or silicon-containing flame retardants, or chemically modifying cellulose nanofibers to impart flame-retardant functionality. However, these methods largely rely on experience to select modification paths and formulation ratios. With multiple modifying elements and process parameters present simultaneously, it is difficult to systematically assess the comprehensive impact of different variables on material properties, leading to long development cycles, high trial-and-error costs, and difficulty in achieving synergistic optimization between flame retardant and mechanical properties.

[0004] With the emergence of the Materials Genome Engineering (MGE) concept, machine learning-driven materials design methods offer new insights into addressing the aforementioned challenges. However, existing research largely focuses on the analysis and prediction of experimental results, lacking a systematic design approach that can proactively learn and screen existing formulations and processes before material preparation, guiding subsequent design through an "experiment-model-screening-validation-experiment" closed-loop process, and directly directing the materials preparation process. Therefore, a CNF / PVA composite film design and preparation scheme that integrates machine learning methods with the materials preparation process is urgently needed. Summary of the Invention

[0005] In view of this, the present invention addresses the problems of insufficient flame retardant performance of CNF / PVA composite films in the prior art, the reliance on experience in formulation design and the high cost of trial and error, by providing a method for preparing high-strength flame-retardant CNF / PVA composite films based on machine learning assistance, as well as the composite films and their applications.

[0006] This invention constructs a material database containing material composition parameters, preparation process parameters, and corresponding performance data, establishes a predictive model between material composition, process, and performance, predicts and screens candidate material schemes, thereby obtaining a target scheme that meets the synergistic requirements of flame retardant performance and mechanical performance before material preparation, and continuously optimizes the model through experimental verification and feedback.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for preparing high-strength flame-retardant CNF / PVA composite films based on machine learning assistance is disclosed. This method employs an active learning loop framework, organically combining material database construction, model prediction, high-throughput screening, and experimental verification. Specifically, it includes the following steps: 1) Construct a materials database. The materials database shall include at least: a. Modification methods and compositional parameters of cellulose nanofibers, including the introduction and content of B, N, P and Si elements; b. Preparation process parameters of CNF / PVA composite films, including CNF content, PVA aqueous solution concentration, dissolution time, film formation method and drying conditions; c. Performance data corresponding to the above material composition parameters and process parameters, including flame retardant performance indicators (limiting oxygen index, vertical burning rating) and mechanical performance indicators (tensile strength).

[0009] The database contains data from experimental data, literature data, and extended data generated by introducing Gaussian noise to improve the model's generalization ability.

[0010] 2) Based on this, a supervised learning method is used to establish a material performance prediction model to characterize the mapping relationship between material composition parameters and process parameters and flame retardant and mechanical properties. The prediction model can be a simple linear regression model, a Lasso regression model, an SVM model, a random forest model, an XGBoost model, or an LSTM model.

[0011] 3) Using the aforementioned prediction model, the performance of candidate CNF / PVA composite film design schemes is predicted. Based on the prediction results or model uncertainty information, a high-throughput screening strategy is employed to select target material schemes that meet the preset performance requirements. The weighting ratio of the target performance—limiting oxygen index and tensile strength—is 6:4. Simultaneously, characteristic contribution analysis is used to evaluate the impact of different compositional and process parameters on material properties to improve the model's interpretability.

[0012] 4) Based on the selected target schemes, CNF underwent multi-element synergistic modification and was blended with PVA aqueous solution to form a film, thus preparing a high-strength flame-retardant CNF / PVA composite film. The experimental results were further fed back into the materials database for model updates and optimization, thereby forming an active learning closed loop of "database-model-screening-experiment-feedback".

[0013] The multi-element synergistic modification process of the cellulose nanofibers includes the following steps, wherein the reaction conditions and parameters of each step can be adjusted according to actual needs.

[0014] (1) Boron is introduced into the cellulose nanofiber structure by borylation reaction. Preferably, sodium tetraborate decahydrate is used as the boron source, with a mass fraction of 0-5 wt% relative to the dry weight of the cellulose nanofiber. The pH of the reaction system is controlled at 10±0.2, the reaction solvent is deionized water, and the mass concentration of B / CNF in the resulting borylated cellulose nanofiber dispersion system is preferably about 2 wt%.

[0015] (2) Nitrogen-containing compounds are introduced into the boronized cellulose nanofibers via electrostatic adsorption, thereby achieving the introduction of nitrogen. Preferably, the nitrogen-containing compound is polyethyleneimine (PEI), with a molecular weight preferably of about 20,000 g / mol. The amount of PEI added is 0-5 wt% relative to the dry weight of B / CNF, the reaction solvent is deionized water, and the mass concentration of the resulting B / N-CNF dispersion system is preferably about 2 wt%. In this process, nitrogen is mainly bound to the surface of the cellulose nanofibers through electrostatic adsorption.

[0016] (3) Phosphorus is further introduced into the B / N-CNF obtained in step (2). Preferably, sodium phytate (PA-Na) is used as the phosphorus source, with a mass fraction of 0-10 wt% relative to the dry weight of B / N-CNF. The reaction solvent is a deionized aqueous solution, and the mass concentration of the obtained B / NP-CNF dispersion system is preferably about 2 wt%. The introduction of phosphorus can further enhance the char-forming ability and flame-retardant properties of cellulose nanofibers.

[0017] (4) Introducing silicon into the B / NP-CNF via a silane coupling reaction. Preferably, KH550 is used as the silicon source, with a mass ratio of KH550 to B / NP-CNF of 1:(10-20), and the reaction solvent is a deionized aqueous solution. The mass concentration of the resulting B / NP-Si-CNF dispersion is preferably about 2 wt%. This step introduces silicon onto the surface of cellulose nanofibers, thereby further improving the thermal stability and flame retardant properties of the material.

[0018] Through the above multi-step synergistic modification process, cellulose nanofibers modified with boron, nitrogen, phosphorus and silicon elements can be obtained, providing functional fillers for the subsequent preparation of high-strength flame-retardant composite films.

[0019] As can be seen from the above technical solution, compared with the prior art, the present invention has the following technical effects: 1) By using machine learning to screen material formulations and process parameters, the number of trial and error attempts in the material development process can be significantly reduced; 2) It can achieve a synergistic improvement in the flame retardant properties and mechanical properties of materials under the condition of multiple modifying elements and process parameters; 3) Improve the design efficiency and repeatability of new green composite materials.

[0020] 4) The composite film obtained exhibits a high limiting oxygen index in terms of flame retardancy, with the screening formula reaching over 33%, and can achieve V-0 rating through vertical burning test, while maintaining good mechanical properties. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 The active learning loop framework based on machine learning provided by this invention; Figure 2 This is a schematic diagram of the prediction results of the machine learning prediction model of the present invention on the test set; Figure 3 The thermogravimetric analysis test results are those of Examples 1-2 and Comparative Examples 1-4 of this invention; Figure 4 These are the UL-94 test results for Examples 1-2 and Comparative Examples 1-4 of the present invention; Figure 5 The limiting oxygen index test results are for Example 2 of the present invention and Comparative Examples 2-4. Figure 6 The mechanical performance test results are for Embodiment 2 of the present invention and Comparative Examples 2-4. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 In this embodiment, the machine learning-assisted design method according to the present invention includes the following: 1) Construct a materials database, which includes at least: a. Modification methods and compositional parameters of CNF, including boron (B) content, nitrogen (N) content, phosphorus (P) content, and silicon (Si) content. b. Preparation process parameters of CNF / PVA composite film, including CNF content, PVA concentration, dissolution time, and drying temperature; c. Flame retardant and mechanical property data corresponding to the material composition and process parameters, including limiting oxygen index (LOI) and tensile strength (Ts). 2) Based on the aforementioned material database, a material performance prediction model is established using machine learning methods to express and predict the relationship between material composition parameters, process parameters, flame retardant properties, and mechanical properties; 3) The predicted model is used to predict and screen the performance of candidate CNF / PVA composite film design schemes to obtain target material schemes that meet the preset performance requirements, and the interpretability of the model is further performed, including feature importance matrix and SHAP analysis. 4) Based on the target material scheme, conduct experimental verification, modify CNF, and prepare CNF / PVA composite films; 5) Perform performance tests on the obtained composite film and use the test results to update and iterate the material database and prediction model.

[0025] 6) The above steps, based on the active learning method, form an active learning loop framework. This active learning loop framework should include: a. Database construction; b. Model building and optimization; c. Explanatory output of the model and high-throughput filtering; d. Experimental verification and feedback.

[0026] Based on the above method, a material database containing different cellulose nanofiber modification pathways and their corresponding thermal stability data was constructed. Simple linear regression, Lasso regression, SVM, random forest, XGBoost, and LSTM were then established as material property prediction models based on this database. After comparison and verification, XGBoost was selected as the final prediction model. Figure 2 The model is used to predict and perform high-throughput screening of candidate modification schemes, with the limiting oxygen index (LOI) and tensile strength (Ts) as the screening targets, and their weight ratio is 6:4. Target cellulose nanofiber modification schemes are obtained.

[0027] Based on the screening results, cellulose nanofibers were subjected to multi-element synergistic modification with B, N, P, and Si, including the following steps to obtain the cellulose nanofibers (SCNF) screened by the method of this invention: (1) CNF dispersion was reacted with sodium tetraborate decahydrate at pH 10 and room temperature 20-30 ℃ at 400 rpm for 0.5-1 h, and then centrifuged, washed and lyophilized to obtain borated CNF (B / CNF). (2) Disperse B / CNF in water, add polyethyleneimine (PEI) solution, react at room temperature (20-30 °C) at 400 rpm for 0.5-1 h, and introduce nitrogen element through electrostatic adsorption to obtain B / N-CNF; (3) Disperse B / CNF in water, add polyethyleneimine (PEI) solution, react at room temperature (20-30 °C) at 400 rpm for 0.5-1 h, introduce nitrogen element through electrostatic adsorption to obtain B / N-CNF; (4) B / NP-CNF was reacted with KH550 silane coupling agent to introduce Si element. The reaction temperature was 20-30℃, the rotation speed was 400 rpm, and the time was 0.5-1 h to obtain the quaternary synergistic modified SCNF selected by the model screening. The above-mentioned modification methods specifically include: CNF elements introduce N and P elements through electrostatic adsorption.

[0028] In step (1), the sodium tetraborate decahydrate has a mass fraction of 2.6 wt% of CNF dry weight, a pH of 10, and is reacted with deionized water as the solvent. The B / CNF concentration is 2 wt%.

[0029] In step (2), the molecular weight of PEI is 20,000 g / mol, the mass fraction of PEI is 3.0 wt% of the dry weight of B / CNF, the reaction solvent is deionized water, and the concentration of B / N-CNF is 2 wt%.

[0030] In step (3), the mass fraction of PA-Na is 5.8 wt% of the dry weight of B / N-CNF, the reaction solvent is a deionized aqueous solution, and the concentration of B / NP-CNF is 2 wt%.

[0031] In step (4), the mass ratio of KH550 to B / NP-CNF is 1.8 wt%, the reaction solvent is a deionized aqueous solution, and the concentration of SCNF is 2 wt%.

[0032] Thermogravimetric analysis (TGA) was performed on the SCNF. The results showed that, compared with unmodified cellulose nanofibers, the SCNF exhibited higher thermal stability and higher char residue under high-temperature conditions (maximum decomposition temperature of 311.37℃, char residue of 52.64% at 800℃), indicating that the method of the present invention can effectively guide the selection of cellulose nanofiber modification schemes.

[0033] Example 2 Based on the SCNF obtained by the method in Example 1, the SCNF was loaded onto the PVA matrix. The specific process was as follows: SCNF was mixed with PVA solution, CNF content was 23 wt%, reaction temperature was 100 ℃, reaction time was 1.5 h, after the reaction was completed, the film was cast and dried at 50 ℃ for 17 h to obtain a high flame retardant SCNF23 / PVA composite film.

[0034] The SCNF23 / PVA composite film was subjected to tensile and flame retardant performance tests, including vertical burning tests and limiting oxygen index (LOI) tests. The test results show that the SCNF23 / PVA composite film maintains high tensile strength (83 MPa) while exhibiting excellent flame retardant properties, passing the vertical burning test (UL-94 V-0 rating) and possessing a high limiting oxygen index (LOI of 33.6%).

[0035] Comparative Example 1 Unmodified cellulose nanofibers (CNFs) were selected and subjected to thermogravimetric analysis (TGA) under the same conditions as in Example 1. The results showed that the CNFs exhibited low thermal stability during thermal decomposition (maximum degradation temperature of 299.76°C) and a significantly lower char residue (24.58%) than the SCNFs described in Example 1.

[0036] Comparative Example 2 In this comparative example, cellulose nanofibers were synergistically modified with boron, nitrogen, phosphorus, and silicon using the same method as in Example 1 to obtain B / NP-Si-CNF. Thermogravimetric analysis was performed on the B / NP-Si-CNF under the same conditions as in Example 1.

[0037] Test results show that, compared with the SCNF obtained by the method of the present invention in Example 1, the thermal stability and char residue of the B / NP-Si-CNF show a decreasing trend (maximum degradation temperature is 321.20 ℃, char residue at 800 ℃ is 44.45%), but are better than the initial CNF in Comparative Example 1, indicating that the method of the present invention has significant advantages in screening multi-element synergistic modification schemes.

[0038] Comparative Example 3 Pure PVA films were prepared using the same process as in Example 2, with a PVA solution concentration of 7 wt%, and the tensile and flame retardant properties of the resulting PVA films were tested.

[0039] Comparative Example 4 Using the same process as in Example 2, a CNF / PVA composite film was designed based on experience, wherein the CNF in polyvinyl alcohol had a mass fraction of 5 wt%. The tensile properties and flame retardant properties of the resulting composite film were tested.

[0040] Test results show that, compared with the SCNF23 / PVA composite film described in Example 2, the flame retardant performance of the SCNF5 / PVA composite film is significantly reduced (limiting oxygen index of only 22%), and it shows no flame retardant rating in the vertical burning test. The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0041] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for preparing a high-strength flame-retardant CNF / PVA composite film based on machine learning assistance, characterized in that, Includes the following steps: 1) Construct a materials database, which includes at least: a. The modification methods and compositional parameters of CNF, including boron content, nitrogen content, phosphorus content, and silicon content; b. Preparation process parameters of CNF / PVA composite film, including CNF content, PVA concentration, dissolution time, and drying temperature; c. Flame retardant and mechanical property data corresponding to the material composition and process parameters, including limiting oxygen index and tensile strength; 2) Based on the aforementioned material database, a material performance prediction model is established using machine learning methods to express and predict the relationship between material composition parameters, process parameters, flame retardant properties, and mechanical properties; 3) Use the prediction model to predict and screen the performance of candidate CNF / PVA composite film design schemes, obtain target material schemes that meet the preset performance requirements, and further perform model interpretability output, including feature importance matrix and SHAP analysis; 4) Based on the target material scheme, conduct experimental verification, modify CNF, and prepare CNF / PVA composite films; the modification of CNF involves sequentially subjecting CNF to borylation treatment, nitrogen introduction treatment, phosphorus introduction treatment, and silicon coupling treatment, specifically including the following steps: (1) CNF dispersion was reacted with sodium tetraborate decahydrate at pH 10 and room temperature 20-30℃ at 400 rpm for 0.5-1 h, and then centrifuged, washed and lyophilized to obtain B / CNF; (2) Disperse B / CNF in water, add polyethyleneimine solution, react at room temperature (20-30℃) at 400 rpm for 0.5-1 h, and introduce nitrogen element through electrostatic adsorption to obtain B / N-CNF; (3) Disperse B / CNF in water, add sodium phytate solution, react at room temperature (20-30℃) at 400 rpm for 0.5-1 h, introduce phosphorus element through electrostatic adsorption, and obtain B / NP-CNF; (4) B / NP-CNF is reacted with KH550 silane coupling agent to introduce silicon element. The reaction temperature is 20-30℃, the rotation speed is 400rpm, and the time is 0.5-1h to obtain quaternary synergistic modified B / NP-Si-CNF. (5) B / NP-Si-CNF is mixed with PVA solution, with B / NP-Si-CNF content of 0-30wt%, reaction temperature of 100℃, reaction time of 1h-2h, casting and film laying after reaction, and drying at 40-60℃ for 12-48h to obtain high flame retardant CNF / PVA composite film. 5) Perform performance tests on the obtained composite film and use the test results to update and iterate the material database and prediction model; The above steps, based on the active learning method, form an active learning loop framework, which includes: A. Database construction; B. Model establishment and optimization; C. Explanatory output of the model and high-throughput filtering; D. Experimental verification and feedback.

2. The method for preparing a high-strength flame-retardant CNF / PVA composite film based on machine learning assistance according to claim 1, characterized in that, The data in the material database mentioned in step 1) comes from experiments, literature searches, and data augmentation methods that introduce Gaussian noise; The material property prediction model mentioned in step 2) is a regression model based on supervised learning, including simple linear regression, Lasso regression, SVM, random forest, XGBoost and LSTM.

3. The method for preparing a high-strength flame-retardant CNF / PVA composite film based on machine learning assistance according to claim 2, characterized in that, The screening described in step 3) adopts high-throughput screening, and the screening parameters include limiting oxygen index and tensile strength, with a weight ratio of 6:4; candidate material schemes are selected for experimental verification based on the prediction results or uncertainty information of the prediction model; feature contribution analysis is performed on the prediction model to evaluate the influence of different material composition parameters and process parameters on material properties.

4. The method for preparing a high-strength flame-retardant CNF / PVA composite film based on machine learning assistance according to claim 1, characterized in that, In step (1), the sodium tetraborate decahydrate has a mass fraction of 0-5 wt% of CNF dry weight, a pH of 10 ± 0.2, is a deionized solvent, and has a B / CNF concentration of 2 wt%. In step (2), the molecular weight of polyethyleneimine is 20,000 g / mol, the mass fraction is 0-5 wt% of the dry weight of B / N-CNF, the reaction solvent is deionized water, and the concentration of B / N-CNF is 2 wt%.

5. The method for preparing a high-strength flame-retardant CNF / PVA composite film based on machine learning assistance according to claim 1, characterized in that, In step (3), the sodium phytate mass fraction is 0-10 wt% of the dry weight of B / N-CNF, the reaction solvent is a deionized aqueous solution, and the concentration of B / NP-CNF is 2 wt%. In step (4), the mass ratio of KH550 to B / NP-CNF is 1:(10-20), the reaction solvent is a deionized aqueous solution, and the concentration of B / NP-Si-CNF is 2wt%.

6. A high-strength flame-retardant CNF / PVA composite film, characterized in that, It is prepared by the preparation method according to any one of claims 1-5.

7. The application of the high-strength flame-retardant CNF / PVA composite film as described in claim 6 in flame retardancy of single-layer boards.

Images