Authors : A.C.S. Silva , E.M. Bezerra , E.J.X. Costa and H. Savastano
Abstract: The backpropagation algorithm was utilized to implement a fiber cement composition simulator. Six predictors were used: Synthetic fiber supplier, content of synthetic fiber, supplier of the softwood cellulose pulp, refinement degree of softwood cellulose pulp, content of softwood cellulose pulp and refinement degree of hardwood cellulose pulp. The combination of the 6 predictors generated compositions that were used as the Artificial Neural Network (ANN) target in relation to the variables: modulus of rupture (y1), toughness (y2) and water absorption (y3) of the fiber cement composites at the total age of 28 days that were used as the neural network input. The ANN performance was 97.3 % of correct classification with kappa coefficients varying between 0.89 and 0.93. The results suggest that the ANN approach can be used to simulate the composite formulation based on mechanical and physical characteristics using historical data set from experimental results.
A.C.S. Silva , E.M. Bezerra , E.J.X. Costa and H. Savastano , 2007. Fiber Cement Composition Simulator Using Artificial Neural Networks . Journal of Engineering and Applied Sciences, 2: 1206-1212.