Rheological Structure-activity Relationship Of Asphalt Based On Machine Learning

It is the key and difficult point to establish the correlation between composition structure and macro performance to realize the on demand design and numerical manufacturing of "Material Genome Project".The rheological structure-activity relationship of asphalt refers to the relationship between the macroscopic rheological properties of asphalt and the microscopic composition of asphalt.Its establishment is the core and key to accurately predict the rheological properties of asphalt and realize the ondemand design of asphalt.However,due to the extremely complex microstructure of asphalt,the existing relationship models mainly have the following problems: the prediction performance of ordinary models is single,the material,environment and other factors have a great impact,and the establishment of multi-scale corresponding relationship models is complex,which is difficult to popularize and apply.Machine learning breaks the establishment method of conventional relationship model,and establishes statistical relationship model based on a large amount of data to achieve the purpose of accurately predicting multiple performances at the same time.This kind of relationship model is not affected by external conditions and can avoid the difficult problem of complex connections among multiple scales.Therefore,this study adopts machine learning method to establish the rheostructure-activity relationship model of asphalt.Firstly,50 kinds of asphalt commonly used at present are collected.The dynamic rheological properties of asphalt were analyzed in detail.Based on the main curve of modulus and phase Angle,the rheological energy database of asphalt was established.By means of element analysis and INFRARED spectrum,the microstructure of different bitumen was measured.Based on element content and infrared spectrum curve,the microstructure database of bitumen was established.Secondly,study the theory of data of feature selection and feature extraction,determined the microscopic structure and rheological properties data of feature extraction methods,i.e.,the characteristics of pretreatment,through variance filter method based on principal component analysis(pca)feature dimension reduction,distance correlation coefficient method is used to search for mapping relationship with the rheological characteristic parameters of micro characteristic parameters;The feature selection and feature extraction of the data in the database are carried out by the deterministic method.Finally,the applicability of each learning algorithm in the machine learning theory to the establishment of bitumen rheological structure-activity relationship is studied.Three machine learning algorithms,namely support vector machine,correlation vector machine and random forest,are selected for comparative analysis.Based on the accuracy of the model established by different algorithms,the support vector machine is finally selected for the model establishment.The applicability of grid search method and cross validation method to model optimization was studied.A rheological structure-activity relationship model of asphalt based on optimized support vector machine was established and its accuracy was verified.In this paper,the rheological structure-activity relationship model of asphalt based on machine learning is established,which has important theoretical significance and application value for asphalt material design,production and even road new material research and development,and has certain guiding significance for the realization of other materials on demand design.