A study on explicit formulation of sorptivity of concretes containing mineral admixtures

In this thesis, mathematical models derived from gene expression programming (GEP) and artificial neural network (ANN) were used for prediction of sorptivity of concretes. For this, 151 data samples were collected from the previous studies.The common predicttion parameters were selected as water-to-binder ratio (w/b), total binder content (B), compressive strength of 150 mm cube speciment at 28 days (fc,28), aggregate-to-binder ratio (Agg./B) and age of concrete (A). Additionally, in order to evaluate the performance of the proposed models an experimental study was also conducted. The study was carried out on water cured and air cured concretes produced by w/b ratio of 0.45 with total binder content of 400 kg/m3. Moreover, silica fume (SF) and fly ash (FA) were used in different replacement levels. Total 9 different concrete mixtures with binary and ternary blends of SF and FA were produced. Both of the proposed models were proved to be effective enough for prediction of sorptivity of concretes. However, NN models was more accurate than GEP model. Moreover, validation study also indicated that the proposed mathematical models can be utilized as reliable prediction tools for estimation of sorptivity of concretes. Keywords: Sorptivity of concrete, Mineral admixtures, Artificial Neural Network, Gene Expression Programming