Prediction of crack spacing of bending reinforced concrete by strain compliance approach and neural network

The present study presents an investigation of the performance of Neural Networks applied to predict the primary crack spacing of reinforced concrete members subjected to flexural loading. The investigation is carried out on the same data set as was used by the authors to develop and validate the strain compliance approach for cracking analysis of reinforced concrete elements. The aforementioned approach is an alternative technique to analyze concrete cracking in an accurate and highly flexible way. In order to explore the bending specimen data set and further substantiate the strain compliance technique, neural networks were employed due to their ability to find and adapt to patterns in the data, provided they exist. Due to scarcity of available published experimental data of bending experiments, a multiple run approach was adopted together with surrogate data based comparison of artificial neural networks. The results revealed the performance of the neural networks to be slightly superior to the strain compliance technique, particularly in the control of the scatter of predictions. Moreover, the findings lead to the conclusion that the reinforced concrete flexural specimen data set is sufficiently consistent and relatively noise free to be used for the development of new cracking analysis methods.