Prediction Of Aggregate Packing And Sand Concrete Performance Based On The Dense Packing Theory

Aggregate,as the largest component in concrete materials,its packing structure is an important indicator for evaluating the superiority of concrete performance.With the development of the field of concrete,it has become a research hotspot to improve the packing structure by using micro-aggregates as aggregate fillers.The particularity of particle size range of micro-aggregate is that there are a large number of fine particles with particle size range below 75μm,which is beyond the calculation range of fineness modulus.So the fineness modulus cannot be used as a criterion to measure the quality of its gradation.Due to the particularity of micro-aggregate particle characteristics,this study intends to seek a particle packing model that can accurately evaluate the actual packing rate of microaggregates,and to investigate the effect of the packing structure of micro-aggregates on the performance of sand concrete with the help of neural network technology.And to summarize and generalize it in order to evaluate the significant degree of its effect on the workability and mechanical properties of sand concrete.In this paper,the particle packing experiment of micro-aggregates was carried out and its data was plugged into common packing rate calculation models such as Furnas model,modified Toufar model,linear packing model,compressible packing model(CPM),and 3-parameter model.The wall effect,loosening effect and wedge effect model that jointly affect the packing rate under the interaction are analyzed.The Furnas Model in the binary packing state has a higher prediction accuracy with a correlation coefficient of 0.97;the CPM has the next highest prediction accuracy with a correlation coefficient of 0.97;the correlations of the predicted values of the remaining models are below 0.8 and the prediction accuracy fluctuates greatly widely.The correlation coefficient of the existing model under the multivariate packing system is less than-0.5,and the error is very large.However,the actual packing process of microaggregates is multivariate continuous gradation particle packing.So the current packing model is difficult to meet the calculation accuracy of the actual packing rate of micro-aggregates.In this paper,a multivariate continuous graded particle packing model based on coarse-grained soil is introduced,and its correlation coefficient is corrected to construct a nonlinear packing model in the field of micro-aggregates.Base on that the GA-BP neural network prediction model of sand concrete performance is constructed in this paper.In the model,four factors including aggregate void ratio,water-cement ratio,cement-sand ratio and admixture amount are taken as input variables,and fluidity,compressive strength and flexural strength are used as output variables to construct the neural network respectively.In this paper,the particle packing experiment of micro-aggregates is carried out to verify the effectiveness of the multivariate nonlinear particle packing model in the field of microaggregates.The relative error of the experimental measurements does not exceed 1% and the root mean square error is within 0.4%.The validity of the model is verified by comparing the prediction results of several aggregate particle packing models in the conventional concrete sector.The total correlation of the training set of GA-BP neural network model for the fluidity of sand concrete is 0.98.Twenty groups of data were predicted by this network model with a mean deviation of-1.49%,a mean absolute error of 2.31% and a root mean square error of3.01%.The total correlation of the training set of GA-BP neural network model for the compressive strength of sand concrete is 0.91.Twenty groups of data were predicted by this network model with an average deviation of 1.15%,an average absolute error of 8.14% and a root mean square error of 14.43%.The total correlation of the training set of GA-BP neural network model for the flexural strength of sand concrete is 0.92.Twenty groups of data were predicted by this network model with an average deviation of 2.39%,an average absolute error of 4.96%,and a root mean square error of 6.32%.It can be seen that GA-BP neural network prediction model of sand concrete performance has good feasibility and applicability.