The Power of Statistical Learning Applied to the Proportioning of Fiber-Reinforced Concrete Mixes

Fiber-reinforced concrete (FRC) presents flexural load-bearing capacity in the cracked state, and residual flexural strength parameters are the basis of the material’s characterization and specification, together with compressive strength. However, the incorporation of fibers also affects the workability of the fresh mix. The correlations between these parameters and the dosage, size, and type of fibers as well as the relative amounts of the other constituents in a FRC mix are usually described separately for different specific mixes, with limited general validity. Furthermore, residual flexural strength parameters are mutually interdependent, and therefore conventional approaches that regard them as independent variables fail to make the most of the information which extracted from characterization tests results. The project “Optimization of Fiber-Reinforced Concrete using Data Mining” (abbreviated as OptiFRC), funded by the Concrete Research Council / ACI Foundation, has undertaken the first meta-analysis of FRC mixes and their main properties. An exhaustive database comprising nearly 2,000 cases of FRC mixes and their properties has been compiled from papers published in indexed journals. All this information has been analyzed from a data analytics perspective in order to develop statistical models for the multi-objective optimization of FRC mix designs. Semi-empirical equations have been obtained to relate residual flexural strength, compressive strength, and slump to the FRC mix proportioning, not only in terms of average values but also to account for their variability and sensitivity to changes.