Corrosion Evaluation Model of Reinforcement in Concrete Based on ANN

This paper integrates two types of evaluation models based on the study of reinforced concrete structure durability and artificial neural network. Through learning various algorithms and structural forms in artificial neural network, we determine the most effective algorithm and choose the most appropriate network structure form, according to specific problems in this paper. Since convergence speed of traditional back propagation (BP) neural network is slow with large computational complexity, this paper introduces LM(Leverberg Marquart) algorithm to replace gradient descent method to correct weight and threshold in BP network. We establish artificial neural network model in corrosion degree of reinforcement under different conditions and implement practical engineering detection data for proof analysis. The results show that this model has fast training speed and high predicting accuracy so it is adapted in evaluating corrosive degree of reinforcement in concrete.