Research On Prediction Technology Of Cement Microstructure Elements Distribution Based On Conditional Probability Density Estimation

As one of the most common building materials,cement plays a pivotal role in the economic development in out country.However,as the world’s largest cement producer,China is currently in a situation of "large but not strong".Therefore,the development of high-performance and high-quality cement has become an urgent problem to be solved.From a microscopic perspective,the phase information in cement is one of the important determining factors for the macroscopic properties of cement.Therefore,the phase information that can reflect the characteristics of the cement microcosm plays an important role in exploring and simulating the cement hydration process,and is an important link in exploring the principle of cement hydration and then developing high-performance cement.At present,it is difficult to directly obtain the phase distribution information in cement with the existing technology.In order to achieve the acquisition and analysis of cement phase distribution information,researchers usually use a scanning electron microscope(SEM)equipped with an energy dispersive spectrometer(EDS)to capture cement element distribution images,which can intuitive display the distribution of elements in the cement,that is,the position and relative content.The phase distribution can be obtained by analyzing the distribution images of various elements.However,the acquisition of element distribution images through SEM and EDS still has certain disadvantages.Many researchers encountering the problem of high time and economic costs for using EDS,and even do not have the conditions for use.At the same time,in the process of obtaining element distribution images,due to the limitation of the scanning area of SEM,EDS cannot obtain both high-resolution and large-scale element distribution images,and the obtained image data is relatively small in the actual corresponding area of the cement sample.Since large-scale modeling has important practical significance in the phase of exploring and simulating cement hydration,which brings great obstacles for researchers to carry out large-scale phase analysis.In order to solve the above problems,this thesis mainly conducts the following research:(1)Element distribution prediction of cement microstructure based on conditional probability density estimation.To solve the problem of high economic and time costs even lack of using conditions when using EDS to obtain element distribution images,this thesis proposes a method to predict the corresponding element distribution by cement microstructure images.This method builds the conditional probability density function of the cement microstructure image and the relative content of elements at each pixel to modeling the relationship between them.On this basis,if it is sampled,it can get the element distribution image.This method can achieve more accurate element distribution prediction in the experiment.(2)The synthesis of large-scale cement microstructure image.Due to the SEM scanning area is limited,it is impossible to obtain both high-resolution and large-scale cement microstructure images and element distribution images,which influences the prediction of large-scale element distribution.To solve the above problem,this thesis proposed a large-scale cement microstructure image synthesis method based on convolutional neural network,which uses CNN to model image texture information and upsampling.This method can synthesize large-scale cement microstructure images based on local cement microstructure images under the premise of ensuring high resolution.The experimental results show that this method can achieve relatively accurate large-scale cement microstructure image synthesis,which lays the foundation for obtaining large-scale element distribution.(3)The design and development of on-line prediction system for the distribution of largescale cement microstructural elements.To apply the model more conveniently and quickly,this thesis designs and implements an online prediction system for the distribution of large-scale cement microstructure elements,which integrates the above two models.Users can apply the large-scale cement microstructure image synthesis model to synthesize cement microstructure images,and apply the cement microstructure element distribution prediction model to the synthesized large-scale microstructure images,and finally realize the large-scale cement microstructure element distribution prediction.