Hydration kinetics and strength estimation of fly ash blended cement based materials

Abstract

Fly ash, as an industrial by-product, has been widely adopted to replace part of cement. Addition of fly ash in cement based materials has changed the hydration mechanism and strength development characteristics. In this study, we specially focused on the hydration kinetics and strength estimation for fly ash blended cement based materials. Addition of fly ash generally impairs the reaction of silicate phase and promotes the reaction of aluminate phase. Meanwhile, from the perspective of isothermal heat flow, fly ash can retard the cement hydration heat. No generalized prediction model is available for accurate description of the early-age exothermic behaviour of fly ash blended cement based materials. In this study, we first try to adopt a regression ensemble model to predict the early-age exothermic behaviour of fly ash cement paste under different curing temperatures with different water to binder ratios and fly ash contents. The hyperparameters of the model are tuned by Bayesian Optimization with 5-fold cross validation. The results show that the optimized regression ensemble model can well predict the time-dependent heat flow rate. The cumulative heat can also be calculated by integrating the heat release rate over curing time. Meanwhile, the hydration kinetics of cement paste containing fly ash with or without mechanical activation is analyzed by nucleation and growth based models (Avrami equation and phase-boundary nucleation and growth model), three-parameter exponential function and silicate polymerization related reaction order model. The kinetic analysis revealed that the as-received fly ash has retards the early-age cement hydration, and mechanical activation of fly ash can offset to some extent the hindering effect. The deceleration stage can be well modelled by the reaction order model. The mathematically determined maximum heat release is also studied. The heat flow monitoring time, calculation approach and fly ash content have been found to affect the mathematically determined maximum heat release. Therefore, a much longer curing time is needed to obtain a reliable mathematically determined maximum heat release. The maturity-based strength function recommended by fib model code 2010 can be easily operated to estimate compressive strength of ordinary Portland cement concrete or high strength cement concrete. However, lack of calculation approaches and instruction for the determination of coefficient “s” and the apparent activation energy have limited its further application to fly ash cement concrete. In this study, we have proposed two methods to determine coefficient “s” for concrete under investigation. One is the equivalent mortar method, which considers the physical and chemical properties of the binder system in the concrete. This method has showed its superiority over the s-related mathematical formulas. The other is the machine learning algorithms. Three different kinds of machine learning algorithms including robust linear regression model, support vector machine and regression ensemble model are applied to determine coefficient “s”. The hyperparameters for the last two machine learning models are tuned by Bayesian optimization with 10- fold cross validation. The results show that the optimized regression ensemble model can well predict the s value and compressive strength. Moreover, the original strength function in fib model code is modified to incorporate the reaction rate constant, so that Arrhenius equation can be applied to determine the apparent activation energy Ea. The iterative searching method is also supplemented to find out the best-fit Ea. The Results show that the best-fit Ea or the one obtained by the linear Arrhenius plot is not always equal to the default Ea value (33.33 kJ/mol) given by fib model code, and can achieve relatively lower standard error for strength estimation of fly ash cement concrete.