Prediction Of Rubber Mechanical Properties Based On Long And Short-term Neural Network

At present,with the continuous development and maturity of computer technology and machine learning algorithms,interdisciplinary research has become a hot topic of discussion in recent years.In the field of materials science,the selection of new materials with good properties and the modeling of quantitative structure-activity relationships have become hot topics in materials science computing research.However,computational simulation and traditional experimental measurement often consume a lot of time and resources,and are limited by some experimental conditions and theoretical foundations.Hence,there is a crying need to develop a novel method to accelerate the discovery and design process of new materials.In recent years,the use of machine learning methods to discover and design materials with specific properties has received more and more attention,and great progress has been made in time efficiency and prediction accuracy.While performance research is a key point in the design of new materials,it is very important to predict the performance of materials.The research object of this subject is in the process of molecular dynamics simulation,the stress-strain relationship in the rubber stretching process under the real experimental tensile strain rate,so that some important mechanical properties of the material can be quickly calculated.The research purpose of this subject is to design a mechanical property prediction algorithm based on artificial neural network technology and the qualitative theoretical relationship of the rheological properties of elastomer materials to minimize the gap between the mechanical properties of molecular dynamics simulation and the real experimental values.To better guide the molecular structure design of rubber with specific properties.In this paper,with a preliminary solution to the stress-strain relationship theory of the rubber tensile process in elastomer materials,different solutions are proposed from two perspectives:First,from a data-driven perspective,a rubber mechanical performance prediction model based on the Convolutional Long Short Term Memory(CNN-LSTM)network is proposed.It can better fit the complex nonlinear relationship between time series and stress data,so as to analyze the hidden laws in the series data.Experimental results show that this method not only improves the prediction accuracy of stress prediction at low strain rates,but also greatly reduces time consumption.Second,considering the characteristics of the material itself,this paper proposes a machine learning method based on the fusion of data and knowledge to solve this problem.Since both viscoelastic materials and fractional calculus have certain memory characteristics,this paper performs fractional differential transformation on the variable data so that the fractional derivative can better describe the dynamic change process of stress.Therefore,this paper constructs a fractional-order-long short-term memory neural network model(FO-LSTM)to predict the stress-strain relationship of rubber.The experimental results prove that the predicted value of the FO-LSTM model is closer to the real experimental value,and the prediction error is the smallest,which proves the availability of the proposed method.In summary,the prediction model of rubber mechanical properties proposed in this paper opens up a new idea for material properties calculation,and can even effectively guide the design of molecular structures.