Study Of Prediction Of Mix Proportion Of Self-compacting Concrete With Manufactured Sand Based On Artificial Neural Network

Due to its excellent fresh and hardened properties,self-compacting concrete(SCC)is widely used in many construction projects and the amount is increasing year by year.As one of the main raw materials for concrete,river sand is a short-term non-renewable resource with limited reserves,and its application is increasingly restricted by various conditions.In recent years,manufactured sand as a suitable substitute for river sand is used in self-compacting concrete.Meanwhile,with the development of concrete materials and the demand for concrete performance and function in the construction industry,a large amount of admixtures and mineral admixtures are required in the production process of SCC.It is particularly important to determine the mix proportion of concrete with specific properties.But the current proportion design method of selfcompacting concrete is mainly based on specifications and empirical formulas to determine the raw material ratio,and then continuously adjusted and optimized through experiments to obtain a better-performing concrete.This method not only has complicated theoretical calculations,tedious work,but also consumes a lot of raw materials,which cannot meet the requirements of green,environmental and economic for the construction industry.Without detailed understanding of the basic physical mechanism of the artificial neural network(ANN),it is possible for researchers to effectively establish the model between the output vectors and the input vectors according to the possible complex relationship.The artificial neural network is used to design the mix proportion of manufactured sand self-compacting concrete,which can not only effectively establish the model between parameters of physical performance of materials,workability,compressive strength and the quality of concrete components,but also reduce the workload of calculation of concrete mix ratio,reduce the number of test trials and time and money costs.First of all,according to the "CECS 203-2006 Self-Compacting Concrete Application Technical Regulations" and "JGJT 283-2012 Self-Compacting Concrete Application Technical Regulations",the mix proportion of self-compacting concrete is designed.The effect of grading of manufactured sand on rheological properties,workability and compressive strength of self-compacting concrete is studied.The experimental results show that the effect of grading of manufactured sand on the yield stress of mixture is greater than that on plastic viscosity.This is because the content of particle size and water absorption of manufactured sand is different from natural sand,resulting in the difference of paste content and contact force of the particles.When the water-binder ratio is 0.30 and the sand rate is 46%,the compressive strength of self-compacting concrete made of fine sand,medium sand and coarse sand has all reached the design standard of 50 MPa,but the workability of them is different.In this experiment,the medium sand used in the study with uniform particle size distribution is most suitable for preparing manufactured sand self-compacting concrete with excellent workability and compressive strength.The fine sand has a high content of 0.15 ~ 0.3mm particle size(about 43%).The specific surface area of the fine particles is larger,which results in an increase in the water absorption and water demand of the concrete mixture and poor fluidity.Coarse sand has a high content of 2.36 to 4.75mm(about 49%),which reduces the cohesiveness of the concrete mixture and causes the phenomenon of segregation and bleeding.Next,the effect of water-binder ratio,the content of fly ash and sand rate on the workability and compressive strength of manufactured sand self-compacting concrete is studied and the key points of design of mix proportion of manufactured sand selfcompacting concrete are discussed.The result shows that when the water-binder ratio is in the range of 0.36-0.40,the content of fly ash is less than 15%,and the sand rate is in the range of 48%-52%,the self-compacting concrete has good workability and high compressive strength,which provides guidance for the practical application of selfcompacting concrete in engineering.Thirdly,based on Adam’s adaptive learning rate algorithm of ANN,BP neural network models are constructed to predict the proportion of self-compacting concrete.The training data of network is from experiment.The most optimal prediction network is determined by increasing the complexity of the structure of ANN.The results show that with the increasing of the number of hidden neuron nodes,the dimensions of the output layer,and the number of hidden layer layers,the prediction accuracy of the network for components of self-compacting concrete is getting higher and higher.At this time,the prediction accuracy of the network for components of the manufactured sand self-compacting concrete has reached the expected standard.The correlation coefficient(R2)of the actual and the predicted is above 0.9 and approaches 1,and the average absolute percentage error(MAPE%)is less than 5%.Fourthly,this paper analyzes the sensitivity of the BP neural network using the grey correlation analysis method.Sensitivity analysis shows that the slump and slump flow of manufactured sand self-compacting concrete have the greatest influence on the network output,while the particle size of the manufactured sand has the smallest effect on the network output.Finally,based on the above research,this paper develops a manufactured sand selfcompacting concrete mix ratio prediction software using the Py Qt5(designer)framework.The software interface is simple to operate and easy to use,greatly improving the mixture proportion prediction efficiency.