Prediction of SIFCON compressive strength using neural networks and curve fitting model

Abstract

This paper presents the results of experimental investigation conducted to evaluate the possibilities of adopting Levenberg Marquardt (LM) based Artificial Neural Network (ANN) to predict the compressive strength of SIFCON (made with manufactured sand) with different percentage fibre fractions (8%, 10% and 12%) and different curing periods (7, 14, 21, 28, 35, 42, 49, 56, 63, 70, 77, 84, 91 and 98 days) as input vectors. The network has been trained with experimental data obtained from laboratory experimentation. The Artificial Neural Network learned the relationship for predicting the compressive strength of Slurry Infiltrated Fibrous concrete (SIFCON) in 400 training epochs. The input vector considered for the LM training phase includes curing periods of SIFCON concrete, fibre configuration, number of neurons, learning rate, momentum and activation functions. After successful learning, the LM based ANN models predicted the compressive strength satisfying all the constraints with an accuracy of about 95%. Results of LM algorithm are compared with the polynomial curve fitting method. Research results demonstrate that the proposed LM based ANN model is practical, predicts with high accuracy and beneficial. The various stages involved in the development of Levenberg Marquardt based Neural Network models are enumerated in brief in this paper.