Pervious concrete: Improved hydraulic conductivity models, measurement methods, and mix design optimization

Abstract

Evaluating the flow in pervious concrete (PC) correctly is one of the most important issues to be addressed for modeling infiltration, designing pavement, and preparing mixes. A multilayer neural network (MNN) model for the hydraulic conductivity K of PC is developed to use parameters that can be measured easily. 1140 data points representing PC specimens from 60 studies were used to train, test, and validate the model. This data set was used for statistical analysis which revealed the significant factors affecting the hydraulic conductivity and allowed the evaluation, calibration, and validation of 23 collected K models, and the development of multiple linear regression (MLR-K) K model. Fine aggregate content, the usage of fillers, and aggregate type are more significant than water-to-cement (w/c) and aggregate-to-cement ratios (a/c) for predicting K, but total porosity and aggregate size are better than all mix design variables for predicting hydraulic conductivity. The MNN model for K passed the limits of acceptability defined by this study (number of times observations variability is greater than the mean error > 0.70 and Nash-Sutcliffe efficiency > 0.65), while the other models did not pass these limits. The impracticality of using Darcy’s hydraulic conductivity KD to describe the flow through single sized pervious concrete (SSPC) was experimentally verified, as the Reynolds number ranged between 3 and 623 for all mixes at all hydraulic gradients. In addition, we evaluated the non-Darcian effect in SSPC, which ranged between 29-85% over the range of hydraulic gradients. However, when evaluating the relative error for estimating specific discharge using Darcy’s law, we found that invalid KD can explain the hydraulic behavior in all SSPC mixes up to frequent flooding depths (≤ 1.25 gradients) if and only if KD was measured at hydraulic gradient of at least 4.5 using falling head permeameter. Two methods were proposed to measure Izbash’s hydraulic conductivity: a two-point method (non-Darcian approach), which is valid for all PC species and a one-point method (Darcian Approach), which is limited to SSPC mixes made within the limits of mix design variables used in this study. Additionally, in attempt to create a balance between PC hydraulic and mechanical characteristics, the effects of selected mix design parameters on physical, hydraulic, and mechanical characteristics of SSPC were studied. The shape of aggregates had a statistically significant effect (p < 0.001) on the permeability of SSPC, which is defined in terms of non-Darcian hydraulic coefficients, the Izbash hydraulic conductivity KP and the Forchheimer hydraulic conductivity KF. Aggregate to cement ratio and aggregate size contributed the most to non-Darcian permeability (70%), while aggregate shape contributed less than 5%. Moreover, the connectivity of pores predicted non-Darcian permeability metrics better than total or effective porosities. Empirical models were generated to relate KP, KF, and the inertial resistance coefficient ω to mix design variables— w/c, a/c, aggregate size and shape—through general linear regression and pore space parameters—connectivity and median pore size—through non-linear regression. The inertial resistance coefficient ω is more sensitive to mix design parameters than the Izbash exponent n is