Abstract:
Compressive strength is often predicted from porosity or mix-design parameters. The inability to
convert laboratory scale-compaction-energy to industrial-scale limits the application of performance
prediction models. This study analyses compactability as a predictor which translates easily across
studies and applications. Pervious concrete specimens were cast using eight different aggregate-tocement
ratios (2.5–7.0), compaction-levels (0–75 blows from standard-Proctor rammer) and 2-types
of compaction energy distribution (single-layer and three-layer distributions). 672 specimens
were tested for wet-density, theoretical-porosity, measured-porosity, compressive-strength and
compactability. The Analysis-of-Variance and classification techniques, quadratic discriminant
analysis and the Boosted Forest algorithm classified two groups of specimens based on five
performance parameters with 94% accuracy. This indicated a significant difference imparted by
compaction energy distribution on pervious concrete performance. Gaussian Process Regression
predicted measured the porosity and compressive strength of samples with compactability and
aggregate-to-cement ratio with 94% accuracy for both groups of specimens separately. The
combined sample matrix model yielded high accuracy (86%) but failed on marginal observations
significantly.
