Experimental Materials Science of Concrete

    As I turn now to the experimental materials science of concrete, I will only mention a few, key items and some general, philosophical comments on what I think is needed in order for experiments on cement-based materials to work better with computational theory.

    The first item is material characterization, and it is a very important one. Let’s first look at cements and mineral admixtures. We all know that there are differences in how cements react with various admixtures, differences in fly ashes, etc. But there is no way that we can sort out why these differences occur if we don’t characterize the materials thoroughly. One might argue that there already is material characterization, e.g. oxide analysis for cements and Bogue calculations. This is true, but obviously this is not good enough, since the reasons for differences among cements cannot be sorted out using only this information. Also, the Bogue calculations are notoriously inaccurate, especially for the minor clinker phases, but fortunately will be one day replaced by quantitative X-ray diffraction techniques based on the Rietveld method [11]. We also need to go to particle-level characterization of cements [12], since how the different clinker phases are arranged, on average, in the cement particles affects reactivity. Fly ashes and other multi-phase mineral admixtures need to be characterized in this way as well [13]. And if we really want to understand how different mineral admixtures react together with cement over time, then detailed characterization of their mutual hydration is very important to obtain [14].

    Aggregate shape and chemistry significantly affect concrete properties. In the short term, aggregate shape plays a large role in fresh concrete rheology and early age mechanical properties [15,16], and in the long term, aggregate chemistry plays a large role in durability in things like alkali-silica reaction. Both shape and chemistry need to be characterized, and the current characterization tests in use are inadequate in that they often cannot distinguish between “good” performers and “bad” performers.

     One might complain that the models proposed need too much input effort, since this kind of detailed materials characterization is necessary for their proper use. Models can only operate on what has been givens them. If the appropriate chemical/physical details are not given to the models, then they cannot distinguish differences in behavior that are based on these details – good models are powerful, but not magical. If the models do have this kind of detailed characterization, then they have the possibility of working very well to predict properties and results of many tests.

    Quantitative models are designed to take real material information and solve for real properties based on valid mathematical descriptions of material behavior. With care, they can then sometimes predict the results of empirical tests – sometimes. Empirical tests can be helpful when you don’t know anything about the relevant mechanisms. However, empirical test results can also be very misleading. For example, in concrete workability, sometimes two different concrete mixtures will have the same slump cone value and entirely different workability performances. So our end goal is not computational materials science models of the slump test, though in the short run this may be useful for current practice, but we need to be able to predict fundamental rheological quantities that a rheometer measures. This will allow us to predict concrete workability parameters like pumping performance. As another example, in durability tests that are empirically accelerated without knowledge of the degradation mechanisms and of how these mechanisms depend on temperature, the wrong mechanism could be accelerated and so the results would have no meaning for real service life behavior. Unleashing the power of computational materials science models goes hand in glove with a better experimental materials science understanding of what material behaviors actually need to be known.

Next: Computational Materials Science of Concrete Previous: A tribute to Dr. Geoffrey J.C. Frohnsdorff