Neural network modelling of the effects of inorganic impurities on calcium aluminate cement setting

Neural network analysis was used to construct models of setting time as a function of mix composition and addition concentration using existing data for calcium aluminate cements. The models were able to represent the non-linear temperature dependency of setting times and generalised from the literature data to find exponential relationships between setting time and addition concentration. An induction time estimated from the results of different types of setting measurements was used in modelling to determine the following ranking of induction times: Ba > Ca ≥ Sr ≥ Mg ≥ K ≈ Na ≥ control ≫ Li and citrate ≥ Br⁻≫ [Formula: see text] > C1⁻> control ≈ [Formula: see text] ≥ OH⁻> [Formula: see text] ≈ [Formula: see text] > [Formula: see text] High variability of the results reported in the literature resulted in a root-mean-square error of 174% for the best model, which may provide a method-independent estimate of the interlaboratory reproducibility for setting time measurement.