Study On Prediction Of Common Concrete Strength By Artificial Neural Network

Compressive strength is one of the most important properties, which is the center content and important basis of structure design and construction. Code define strength of concrete structure member is gained after curing 28 days. This condition cannot meet the tempo of modern construction besides may leave incipient fault. Thus, developing concrete strength early stage measurement technique quickly and improving forecast accuracy have great significance. By borrowing ideas from the research findings on concrete strength forecast around the world and incorporating the principle of Artificial Neural Networks, the paper uses MATLAB Neural Networks Tool Box to research how to select input variable, network fabric, transfer function and the other parameter. A large quantity concrete standard samples are made under the same curing condition on three field substation in Inner Mongolia middle west region, which are become training samples and testing samples. This paper uses individually the basic BP algorithm, self-adaption learning ratio algorithm of additional momentum, L-M algorithm to train networks. After massive trial method and emulation results comparison, L-M algorithm is accepted finally and so normal concrete strength forecast model is established. The model can meet engineering needs, which also has others advantages like reasonable structure, quick convergence and high accuracy. Comparing to forecast results of the multivariant lineal regression model, BP network has higher accuracy because it can control forecast error within 3 percent.