Prediction model based on artificial neural network for pyrophyllite mechano-chemical activation as an integral step in production of cement binders

Abstract

The optimal outputs of pyrophyllite mechano-chemical activation in an ultra-centrifugal mill performing under different technological conditions were determined by analytical modeling and verified via Artificial Neural Network in order to be employed in the production of cement-based binders. Cluster Analysis and Principal Component Analysis were utilized in assessment of the effect of activation process parameters on the activated pyrophyllite quality. Artificial Neural Network which performed with high prediction accuracy, i.e. 0.914 during the training period, was sufficient for precise prediction of activated pyrophyllite quality in a wide range of processing parameters. The probability of utilization of observed activation products was estimated through interrelation of technological parameters (mesh size sieve, activation period, specific energy consumption) and acquired characteristics of pyrophyllite (grain diameter, specific surface area). The optimal products singled out from each activation sequence were used as mineral additives in the mix-designs of four cement binders (cement replacement portion was 30%). Influence of activated pyrophyllite additions on the cement chemistry, mineral phase compositions and microstructures of the cement binders were monitored by instrumental techniques (DTA/TGA, XRD, SEM). Activated pyrophyllite showed characteristics of pozzolana as it slightly accelerated early stages of hydration, decreased cement hydration energy and increased the quantity of cement mineral alite at later hydration stages. Micron-sized crystalline foila characteristic for mechanically activated pyrophyllite formed micro-reinforcement within cement binder microstructure.