Neural network model for preformed-foam cellular concrete

Cellular concrete is a lightweight material consisting of portland cement p aste or mortar with a homogeneous void or cell structure created by introdu cing air or gas in the form of small bubbles (usually 0.1 to 1.0 mm in diam eter) during the mixing process. This material has traditionally been used in heat insulation and sound attenuation, nonload bearing walls, roof decks , and is gaining wider acceptance in tunneling and geotechnical application s. A major concern with the production of cellular concrete is achieving pr oduct consistency and predictability of performance. Producers of the mater ial have generated extensive experimental data over the years, but the anal ysis of such data using traditional statistical tools has not produced reli able predictive models. This research investigates the use of artificial ne ural networks (ANN) to predict the performance of cellular concrete mixture s. The ANN method can capture complex interactions among input/output varia bles in a system without any prior knowledge of the nature of these interac tions and without having to explicitly assume a model form. Indeed, such a model form is generated by the data points themselves. This paper describes the database assembled, the selection and training process of the ANN mode l, and its validation. Results show that production yield, foamed density, unfoamed density, and compressive strength of cellular concrete mixtures ca n be predicted much more accurately using the ANN method compared to existi ng parametric methods.