The Prediction Of Chloride Ion Diffusion Coefficient About Fly Ash Concrete Based On Artificial Neural Network

Chloride ion penetration is one of the main factors affecting concrete durabilityof the coastal areas. Chloride ion into the concrete can cause steel take off blunt andrust. Along with the increase of corrosion, corrosion product volume swell resulting inspalling or flaking of the protective layer of concrete, resulting in reinforced concretestructure damage which shorten the life of concrete structures.Domestic and foreign scholars, experts carried out a large number of theoreticaland experimental researches about fly ash which can be used to improve the durabilityof concrete as an effective mineral admixture. Fly ash is widely used in concreteengineering with the deepening of the study.Referred to the design theory of fly-ash content concrete, this paper formulateddifferent water-cement ratio and different content of fly ash concrete on the basis ofprevious research. It concluded that different water-cement ratio and fly ash concretehad different workability, strength, chloride ion diffusion of the concrete throughexperimental analysis and comparisons and its durability was evaluated.This paper used neural network toolbox functions of MATLAB7.0to study thetransfer function (activation function), the learning function, learning algorithm andthe choice of other parameters through a lot of engineering experiments and test dataand set up multi-layer forward neural network (BP) prediction model about how topredict the chloride ion diffusion coefficient. Through the comparison of a lot oftraining and simulation, the results found that the average error of about8.3%,asmaller error, but the accuracy needed to be further improved.Based on fuzzy theory which is applied to artificial network, this paper set upadaptive neuro-fuzzy inference systems (ANFIS) which combined their advantages.Through the training and verification about the fitting and predictive ability of themodel, the results show that the average error was7.04%which was high predictionaccuracy and meet the needs of the project. The achievement would provide referenceon the durability design of concrete and concrete testing.