Exploring the experimental design and statistical modeling of cementitious composite systems using various sampling methods

Abstract

Advances in the development of concrete materials have led to increasingly complicated composite systems. Investigation of such complex systems by traditional sampling methods requires a large sample size to understand the influence and interactions of all important factors, and conducting full size experiments with a large number of samples could be onerous and expensive. To devise more efficient experimental design and statistical modeling, this study explores the use of alternative sampling methods, and two space filling techniques, Latin hypercube and space packing, are adopted to investigate a cement-fly ash-silica fume ternary paste system. The applicability of these space filling sampling methods was tested by modeling the material performance – evaluated here using microhardness – of the ternary paste system using Response Surface Methodology. It was found that the space filling design methods and microhardness test were non-ideal and provided poor results in their corresponding models due to relatively large noise caused by the proximity between sample points. However, Response Surface Methodology was confirmed to be not only a convenient tool for modeling the performance of composite systems but also a means for comparing sampling methods by their model accuracy.