Neural Network Analysis of Chloride Penetration in Stressed Concrete

Two neural networks (NN), back-propagation neural network (BPNN) and the radial basis function neural network (RBFNN), are proposed to analyze the chloride penetration in stressed concrete. In order to generate the training and testing data for the NNs, an accelerated chloride aggressive experiment was carried out, and the influence of stress level of concrete on chloride transport process was taken into account especially. Then, based on the experimental results, the BPNN and RBFNN models which all take the stress level of concrete, water-cement ratio, cement-fine aggregate, cement-coarse aggregate ratio and testing age as input parameters were built and all the training and testing work was performed in MATLAB. It can be found that the two NN models seem to have a high prediction and generalization capability in evaluation of chloride penetration, and the mean absolute percentage errors of test data sets are 9.68% and 8.45% for BPNN and RBFNN, respectively. NN technology is a feasible way to analyze and predict chloride penetration in concrete.