Predicting of Load Carrying Capacity of Reactive Powder Concrete and Normal Strength Concrete Column Specimens using Artificial Neural Network

Abstract

In present work, prediction of the maximum loading carrying capacity of reactive powder concrete and normal strength concrete column specimens using artificial neural networks (ANNs), were studied experimentally. Twenty-three column specimens were cast and tested in the experimental work, five of which were cast using normal strength concrete (NSC). The other eighteen columns were cast using Reactive Powder Concrete (RPC) with different steel fibers volume fraction (1 and 2%) and four cases was taken from previous studies were used as final trained. The cross-sections of all column test specimens were (150x150 mm), (100x100 mm), (75x75 mm) and (50x50 mm) with a length of 900 mm. The column specimens were tested under concentric axial compression load up to failure. The result of experimental work demonstrated that the results obtained by ANN model are reasonably agree with the experimental results with higher coefficient of determination value (R^2 = 0.993). The distribution of the errors is well, around the zero axis and the uniform distribution of the extra four case study. The results showed that the residual values (experimental and predicted) for all column specimens are within acceptable range. Based on the reported results using ANN model, it can be revealed that the parameters (cross-section and compressive strength) of the column specimens are the greatest important factor affecting on the output parameter of the model. However, the other parameters such as percentage of steel fibers and the spacing between the lateral reinforcement, is unimportant with respect to the importance of the other parameters.