Soft computing to predict the porosity and permeability of pervious concrete based on mix design and ultrasonic pulse velocity

Abstract

The present study explores the potential of machine learning to predict the porosity and permeability of pervious concrete constructed on mix parameters (compaction energy, aggregate-to-cement ratio and aggregate size) and ultrasonic velocity. The prediction models use non-destructive measurements and mixed design variables, which can help the construction sector apply the models without any theoretical expertise. The study uses 225 data samples from an experimental study. This study used six machine learning algorithms, namely, linear regression, artificial neural networks, boosted decision tree regression, random forest regression, K-nearest neighbour and support vector regression, to determine the best predictive model. The results show that the ANN model is the best technique for predicting the porosity of pervious concrete (R2 = 0.9502 for training datasets and R2 = 0.8958 for testing datasets) and boosted decision tress for permeability of pervious concrete (R2 = 0.9323 for training datasets and R2 = 0.7574 for testing datasets). The sensitivity analysis of the random forest regression model reveals that ultrasonic pulse velocity is the most influential parameter for the prediction of porosity and permeability of pervious concrete. The proposed models provide a more accurate method for estimating the porosity and permeability of pervious concrete.