Cellular Automata Evolution Modeling And Performance Analysis Of Cement Microstructure Based On Deep Learning

Cement is the basic material of the construction industry.It is widely used in roads,bridges,buildings and other building facilities,and is considered as “food for the construction industry”.Recent years,the total cement production of China has ranked first in the world.However,at present,cement industry still has problems.The production cost is relatively high and the labor force is low.The production of high-performance cement is not high and the development of new materials is insufficient.These greatly restrict the advanced development of cement industry.Therefore,there is an appetence for a deeper understanding of cement and deeper research to improve the performance of cement.Cement is not only complex in composition,but its hydration process is also quite complicated.The physical changes and chemical reactions have not been completely penetrated,which increases the difficulty for understanding its mechanism.The simulation and modeling of cement microstructure evolution plays an important role in understanding cement hydration,the development of microstructures and guiding the development of new materials.In order to better understand the hydration process and the microstructure evolution of cement,this study apply computer to model and simulate the microstructure evolution of cement.Since computer technology is applied to the material field,the modeling of cement microstructure evolution has been well developed.At present,the research on simulation of cement microstructure is mainly divided into two genres.One is based on a single particle hydration model and the other is a hydration model based on digital images.However,neither model has a more realistic initial microstructure of cement.Their evolutionary mechanism is artificially derived and contains a number of simplifications and assumptions.Therefore,the wide variations exist between real and simulated microstructures.In view of the existing problems in the modeling of cement microstructure evolution,this research is devoted to producing realistic microstructures and building more realistic simulation model of microstructure evolution,further predicting and analyzing cement performance.The main research content in this paper covers the following aspects:(1)Generating initial microstructure of cementThe texture synthesis in this paper focuses on the two-dimensional cross-sectional images of the cement microstructure image.To solve the problem of time-consumption in the texture synthesis based on sample image,this study is the first time to apply gene expression programming in the rapid synthesis of cement texture.Experiments show that this method does improve the speed of microstructure generation.In view of the shortcomings of the first synthetic texture being too trivial,gray remapping was performed and statistical features of the texture were added in the second study.The improved method is capable of forming phases in cement microstructures such as small cement particles,pores and air voids.In addition,in order to generate a more realistic initial microstructure,this paper is the first attempt to generate cement microstructure using generative adversarial networks.As can be seen from the generated images that be saved from the optimization process,the model does continuously optimize the generation effect.Compared with the real cement image,results prove that generative adversarial networks can generate cement microstructures that are confused with real images.(2)Modeling of cement microstructure evolutionIn this study,the cellular automata is used as the evolution model.The deep learning is the first time to extract evolution rules from the real cement microstructure data,promoting the evolution of cellular automata.This study solves the unreality caused by simplification and hypothesis in previous models,and avoids the influence of artificial derivation mechanism on simulation results.The experimental results show that the process can veritably reflect the changes of cement microstructure and display good generalization ability in cement samples with different components.The evolution process of three phases in four different cement is analyzed.In addition,this study also proposes a variable capacity dynamic stratified sampling method to improve the representativeness of the extracted sample.Many experiments in various standard data and cement data indicate that this method can improve the representativeness of the sample.(3)Prediction and analysis of cement performanceIn this study,a setting time prediction model of cement is constructed using the broad learning system(BLS).This model represents the potential relationship between the chemical compositions,physical properties and the initial setting time or final setting time of cement.Compared with other methods,this method is proved to be efficient.The generalization ability of the method on the different sample is proved according to the comparison with the test data.The validity of the model is proved by comprising with previous studies.The study enabled a series of simulations and analysis from the initial microstructure generation of the cement to the modeling of microstructure evolution,further to the prediction of cement performance.This research can save a lot of manpower resources and has important guiding value for the production of high-performance cement and the development of new materials.