|
The concrete structure which contact with water frequently in cold region, damaged by the freeze-thaw effects, where corrosion cracking and surface erosion is very widespread, which make the structure prematurely loss its durability. In generally consideration, freeze-thaw damage of concrete is a role of a physical process, and its damage mechanism is very complex. Currently, it’s more feasible to establish empirical models using experimental and engineering test data. And it is difficult to gives a determined mathematical model by regression analysis, as there are many affecting factors of frost resistance and they have very complex relationships.Artificial neural network is a theory model of human brain, and a information processing system based on imitate the structures and functions of brain networks, which will achieve complex logic operations and non-linear relationship and be very suitable for dealing with the difficulty which is hard to establish accurate mathematical model and ease to be collected samples. In view of this, the article uses the artificial neural network method to study the frost durability of concrete.This article draws on the research results of the theory of concrete freeze-thaw damage at home and abroad as well as the application of artificial neural network, based on the analysis of freeze-thaw damage of concrete mechanisms and factors, set up artificial neural network prediction models of three indicators of concrete frost resistance of the ordinary concrete and air-entraining concrete, which include the relative dynamic elastic modulus, of concrete, mass loss, the cube compressive strength. In addition, preliminarily establish the equivalent relational model of neural environment and laboratory freeze-thaw cycles, which based on collecting data of the works of actual freeze-thaw cycles. These models provide guidance for testing, designing, construction, assessment, repairing and reinforcement of the frost resistance of concrete, and establish a basis for further study on freeze-thaw performance.This article write codes using neural network toolbox of MATLAB,and establish the models respectively using BP neural network and RBF neural network as well as study how to select their input vector, training algorithms, network infrastructure and other related parameters. After a lot of trial and comparing with their simulation results, the article determines the best structures and performance parameters of BP neural network and RBF neural network models. And the predict results show that the model is accurate, whose error is less that is able to meet the actual needs, which established by artificial neural network. Generally speaking, it’s better to use the BP network model to predict the frost resistance target of plain concrete, but regarding the air-entrained concrete it is more accurate to predict the frost resistance target forecast using RBF network. Because RBF codes is simple, the curve of error is smoother, and easy to apply , therefore for the equivalent relational model of indoor and outside thaw-freeze cycle number of times, uses the RBF network to be more ideal.
|