Prediction of fracture parameters of concrete using an artificial neural network approach

Abstract In this study, detailed experimental results of a series of concrete fracture tests on published literatures were collected into a database named as CFT-DB. Based on large amount of experimental data, an artificial neural network (ANN) approach was employed to build connections between some influential factors and certain fracture parameters, and therefore to predict the corresponding fracture parameters. The fracture parameters investigated mainly were the initial cracking fracture toughness K I c i n i , the unstable fracture parameters K I c u n and the fracture energy GF. They correspond to crack initiation, critical unstable crack propagation and average energy release during the complete fracture process. Results showed that ANN models have better performance than existing empirical formulas when predicting concrete fracture parameters. By adding more variables as ANN input, the accuracy of ANN models was further improved. Parametric study was then performed based the constructed ANN models to evaluate the impact of each input parameters on concrete fracture parameters. Specially, size effect of these concrete facture parameters was discussed. Research of the present paper can be seen as a summary of past researches on fracture parameters. However, it constructs prediction model with higher accuracy and deepens the understanding of influential factors on fracture parameters.