Prediction of characteristics of pervious concrete by machine learning technique using mix parameters and non-destructive test measurements

Ensuring quality in pervious concrete poses challenges, limiting its use. This work investigates the potential of machine learning to forecast its properties, offering a novel and explicable approach. Five machine learning techniques were employed on 300 experimental data points, considering mix parameters (aggregate size, aggregate-to-cement ratio and compaction) and non-destructive measurements (ultrasonic velocity, electrical resistivity). Artificial Neural Networks (ANNs) excelled, achieving the highest accuracy (R2 > 0.97) in predicting porosity and compressive strength, with root mean squared errors (RMSEs) of 0.02 and 0.5 MPa, respectively. Sensitivity analysis revealed a dominant influence of compaction energy, aggregate-to-cement ratio and ultrasonic velocity, while aggregate size and resistivity had compromised impact. Findings suggest that machine learning models, particularly ANNs, can be reliably implemented in construction quality control, facilitating real-time assessments and ensuring optimal performance of pervious concrete. This predictive capability can improve quality control processes, ultimately encouraging broader adoption in the construction sector and contributing to more sustainable infrastructure.