Development Of Software For High-throughput Computation Of Composite Physical Property By Effective Medium Theory

The Materials Genome Initiative proposes a new material development mode driven by high-throughput computation,high-throughput experiments,and material databases,bringing an innovative development direction of "data+artificial intelligence" in the future.Grasping these historical opportunities,overcoming key material technology barriers,and changing China’s long-standing backward situation in new material technology are of strategic importance.This paper is one part of the project "Cross-scale high-throughput automatic integrated calculation algorithms and software of functional material".The goal of this paper is the cross-scale highthroughput computation software and data sharing platform.The main research contents and conclusions are listed as follows:(1)Based on the modified effective medium theory,using mixed programming of C++and Qt,two cross-platform software named "Composite Studio physical property calculation software" and the "Composite lithium ion dielectric conductivity" were developed,which included three physical property calculation modules of elastic modulus,dielectric constant and ionic conductivity.A working mode for constructing high-throughput microstructure parameter combinations at one time was designed.Three computing cores of physical properties were written using the C++language with high calculation efficiency.To edit microstructure parameters such as volume fraction,aspect ratio of reinforce particles,orientation distribution,macroscopic orientation and layer thickness ratio of the interface,Qt was used to implement a GUI with convenient operation and friendly interaction.Real-time plotting analysis of the calculation results was realized by QCustomPlot.The speed can reach more than 10,000 combinations in each computation,which is of high-throughput computation.(2)A "material genetic engineering—composite material physical property calculation platform" at microstructure scale was designed and developed.Its computing cores include effective medium theory and phase field method,which have 5 modules of elastic modulus,dielectric constant,ionic conductivity,local electric field distribution and dielectric breakdown.A web service framework and the online highthroughput computation process were designed.To exchange the data between different calculation modules,standardized formats of input scripts and output files were defined clearly.The website system was developed using Python language,and computing cores were written by C++and Fortran language.This website has been integrated into the overall platform of the Materials Genome Initiative of China.Through the public data pool,the data from other calculation modules such as the first-principle method at atomic scale were imported,and the cross-scale high-throughput computation of physical properties of composite materials was realized in addition.(3)A research were performed to study the ability of machine learning methods to simplify the effective medium theoretical model and predict the physical properties of materials.The data obtained by high-throughput computation based on effective medium theory was used as a training set,and regression learning is performed using three machine learning methods,support vector machine,decision tree and random forest.Performance of the three machine learning methods under different parameters was studied.The results show that under appropriate parameter settings,good regression can be achieved.However,when predicting new test set,the prediction results of random forest show stepped curves,which means there is problem of overfitting and the generalization ability of random forest regression is poor.Moreover,the prediction result of support vector regression on the test set is smoother and more realistic,showing a better generalization ability.In conclusion,it is feasible to use machine learning to imitate a complex theoretical calculation model,and is expected to obtain a faster method for predicting material properties.In this paper,based on the improved effective medium theory,high-throughput calculation software for the physical properties of composite materials was developed.Stand-alone software containing multiple physical properties computing cores and a website platform that can be combined with other computing tools for cross-scale computation were design and realized.The simplification of effective medium theoretical models and the prediction of physical properties of materials using machine learning methods were explored.This research will benefit for the high-throughput computation and data-driven researches on new materials,and is expected to play an important role in the innovation of the Material Genetic Engineering in China.