Estimation of Strength of Reactive Powder Concrete using Artificial Neural Network

Objective: Attempt is made to estimate the compressive strength of Reactive Powder Concrete (RPC) by developing Artificial Neural Network model. Methods/Analysis: RPC is a complex composite material made using a combination of constituent materials viz. fine quartz, silica fume and cement to form a cementitious matrix. Mix design for RPC is more complicated as it excludes the coarse aggregates and a low water/cement ratio is maintained. The mix proportioning methods applicable to normal concrete cannot be applied readily to RPC. The complexity in behavior of concrete with special constituents makes the mix design complicated. Artificial neural network (ANN) techniques can capture complex relations among input/output variables in a given system and this method has been implemented in Reactive Powder Concrete mix proportion design, for predicting its strength. Findings: This paper presents the development and application of ANN for predicting the strength of RPC using data related to 112 mix proportions and the results of the model are checked experimentally. Compressive strength of concrete is chosen as a function of the 'Water/Cement' ratio (w/c), 'Silica Fume/Cement' ratio (sf/c), and 'Quartz Sand/Cement' ratio (qs/c). The ANN model developed is tested for the 12 mix proportions randomly selected from experimental data for which the experimental compressive strengths were available. The compressive strength given by the model when compared with the experimental values showed an average error of only 3.99 %. Novelty/Improvement: The model developed has coefficient of determination R² equal to 0.9066 which indicates a significant enough correlation.