Modeling Simulation And Some Applications Of Machine Learning And Multiphysics Coupling In Material Field

The world is the superposition of matter and time,and materials are the material basis of human survival.Research into the structural design and performance of materials is the mission of people to improve society and the research topic that remains unchanged forever.Modern materials have evolved into an era of intelligence,polymorphism and high performance.The numerical methodology is also constantly updated and improved to meet the needs of society.At present,there are many urgent problems to be solved in the performance research of new materials,pattern recognition and the internal structure of materials in the multi-physical field environment.Traditional methods have faced a difficult situation.Machine learning and multi-physical field coupling have become the general trend.Therefore,it is necessary to study the internal structure of the material.The purpose of this paper is to use machine learning and multi-physics field theory to model and solve some problems that have not been solved or need to be improved in the material field.Machine learning and multi-physical field theory are two numerical tools for material modeling and simulation.The extension and contribution of this paper in the field of material mechanics are mainly four points,which will be shown as follows:Firstly,the ARIMA-FEM method is proposed to solve the variable force prediction problem of porous elastic thin plates.The variable load prediction with time series is combined with elastic mechanics,which has the visualization function of load prediction and Stress-Strain cloud image.CST,LST and Q4 elements are used to discrete and solve rectangular,square and circular porous elastic plates.The advantages of this method are highlighted by comparing ARIMA with various prediction methods.In addition,the consistency of numerical convergence and theoretical convergence is verified in this paper.The correctness of the numerical method is illustrated.Secondly,the deformation theory of single-field and multi-field thin plates is established,and a hybrid element numerical method is proposed in the force field deformation,which realizes the free collocation of sub-domain grids and basis functions.In the single temperature field,it is found that the temperature change of the boundary of the graphene film under the condition of thermal convection is faster than that under the adiabatic condition.In the thermal-mechanical coupling field,the laser temperature distribution formula of periodic motion is also given,and it is found that the strain energy density of the monocrystalline silicon thin plate increases with the increase of temperature.Thirdly,in the mechanical structure analysis of nuclear pressure vessel(RPV),several models are established in this paper,from 1D to 3D,and from single field to multiple fields.In the 2D mechanical model,this paper proposes a numerical method(CMMDTCFEM)of continuum damage dynamics combined with truncated plane FEM,which can well describe the damage caused by damage factor on the Young’s modulus and Poisson’s ratio of the vessel.In addition,in order to make up for the deficiency of the truncated plane method of RPV,this paper gives the solving process of axisymmetric FEM.The experimental results show that the Young’s modulus decreases with the increase of temperature.The Poisson’s ratio increases with the increase of temperature.Then,the axial and hoop stress changes are analyzed in the thermal-mechanical coupling physical field.It is found that the axial stress of RPV is slightly larger than the hoop stress near the pipe mouth.Finally,the error estimation of the h-p method is also given.Fourth,this paper proposes an evaluation method for automobile shell design based on semi-supervised machine learning mode.The numerical results of the two-dimensional fluid-solid coupling model are given in the numerical part,and the specific evaluation methods and strategies are also given.The establishment of this method provides a basis for future large-scale production and evaluation.Promote the scientific design and efficient audit of the production process.In short,machine learning and the field of multi-physics are two knowledge systems,and they are very different in terms of principles.The former is data-driven cognition,and the latter is a combination of traditional physical models.However,they are both the two cornerstones for the scientific community to explore the unknown world.There are still many mathematical principles to be explained and deeply excavated in machine learning.For example,the collapse of the model caused by the discontinuous mapping of deep learning is still an unsolved problem.In addition,there are also some key problems in the multi-physics field,such as complex model interface,slow convergence,and optimization of computational memory.The research in this paper provides a new idea for the multi-physics field modeling,performance research and comprehensive application of materials.Machine learning can make the identification,classification and prediction of materials with defects more concise and efficient.The Multiphysical field is a comprehensive model to study the performance and structural changes of materials,which can reflect the inherent nature of things under the combined action of multiple physical quantities.Therefore,the method and simulation model proposed in this paper have a certain industrial application background and also reflect the scientific research concept of interdisciplinary application,diversification,intelligence and systematization.It provides a reliable reference and theoretical basis for machine learning and multiphysical field application in the field of materials.