Application Of Machine Learning To The Computational Simulation Of Natural Rubber Materials

With the continuous development of big data and artificial intelligence,the cross-fertilization between mathematical disciplines and various disciplines has become increasingly close,and the active exploration of the development and application of mathematical theories and machine learning algorithms in materials science,information and control engineering,and biomedicine has become a hot issue for research at home and abroad.Relying on the background of polymer material science,this study applies machine learning(ML)models and algorithms to the mechanical property prediction of natural rubber material computational simulations,and develops data enhancement and feature processing algorithms adapted to the data characteristics of polymer materials.In the face of the problem that molecular dynamics(MD)simulations are limited by low sample quality under high-throughput computation,we investigate data enhancement algorithms based on small sample size data to make the samples expand in the sample space as much as possible,which in turn enriches the sample space and solves the problem of insufficient data drive for small sample size data in traditional machine learning,giving some theoretical support and scientific guidance for practical applications in materials science.It gives some theoretical support and scientific guidance for practical applications in materials science.First,this paper applies a machine learning method trained using MD simulation data,and we adopt a new idea based on data augmentation by combining our previous experience in small sample prediction.The adopted machine learning methods include(1)a data augmentation algorithm based on nearest neighbor interpolation(NNI)with synthetic minority oversampling technique(SMOTE).In our experiments,the NNI algorithm achieves the effect of approximating the original sample distribution by interpolating in the adjacent areas of the original samples,while the SMOTE algorithm solves the problem of sample imbalance by interpolating at the edges of a few clusters;(2)extreme gradient boosting(XGBoost)model to predict the tensile stress of natural rubber.The debugged XGBoost model has high prediction accuracy with the guarantee of high performance values.Some traditional machine learning methods are also added to the comparison experiments,and the results show that XGBoost has a definite advantage in dealing with this problem.Secondly,this paper establishes a GAN-XGBoost-based machine learning framework for the prediction problem of computational simulation of natural rubber crystallinity,using a generative adversarial network(GAN)-based data augmentation algorithm to extend the pre-input dataset and a tree-structure algorithm-based XGBoost model to accelerate the prediction of crystallinity,and finally,a weight integration approach by feature importance analysis to analyze The details of the effects of phospholipid protein percentage(),hydrogen bonding strength()and non-hydrogen bonding strength()of natural rubber materials in relation to crystallinity prediction under dynamic conditions.The modeling framework mainly integrates the crystallinity algorithm of polymer materials based on molecular dynamics simulation with the data enhancement algorithm based on generative adversarial network,and finally predicts the crystallinity of natural rubber by XGBoost model.