Study On Relationship Between Pore Structure And Strength Of Concrete Based On Artificial Neural Networks

It's set a higher request to researching in performances of the most popular architectural material—concrete with the development of capital construction, and the strength has been one of the most important mechanical property. In order to master the rule of changes of the concrete strength well, we must establish connections between the microcosmic or mesoscopic structure within the concrete material and the strength. Concrete is a kind of porous material with heterogeneous and nonuniform nature, therefore, it expresses quite complicated nonlinear relationship between the mesoscopic structure and the strength. The quantitative study before always put forward some simplified relations by creating mathematic models, however, it lacked of the strict theoretical derivation and those semi-empirical relations just satisfied with practical demand of projects in a certain extent. Artificial neural networks(ANNs), which are developing at very fast speed in these ten years, are bionic techniques by imitating principle of brain operation. ANNs are especially accomplished in solving complex nonlinear problems, and none of definite mathematic expressions should be needed. This text makes use of ANNs to build the bridge connecting the mesoscopic structure of concrete and the main aspect of mechanical property—strength, and the author wishes to make certain contribution to the study in this direction.This text divides five chapters altogether. Chapter one firstly introduces the current situations of the pore structure in domestic and international research, including the pore structure's test and measurement study of the pore structure model and current research situations of pore structure and strength relation;then, the development history and the general application situations of artificial neural networks are reviewed. Chapter two analyses the development course of the pore structure and strength relation model, and the differences of the models existing nowadays. Emphatically, it clarifies that the model considering pore size distribution is more reasonable than that just considering porosity. Chapter three emphasizes onintroducing the concept n characteristics and basic principle of ANNs, especially presents the algorithm deduction ., shortage and improved methods of backpropagation(BP) neural network which is used in this text;in addition, wide-ranging applications of ANNs in civil engineering are generalized. In chapter four, BP models are established. The relationship between macroscopic and mesoscopic level of concrete is proposed from two sides: the strength predictive model predicts the compression strength of concrete with mean distribution radius and four different porosities;and porosity predictive model predicts porosity with compression strength and components of concrete. Chapter five is a conclusion, it concludes primary research outcomes of this text and makes corresponding prospect to the future study.The main achievement of this text is the creation of two BP neural network predictive models. Compared to traditional regression analysis, strength predictive model presents more accurately;on the other hand, porosity predictive model is a new research field, so it has an important meaning to improve the model.