Easy infiltration of grouting material in Porous Asphalt Mixture (PAM) is one of the crucial phases in the construction of semi-flexible pavement (SFP). The gradation and compaction primarily influence the pores and their arrangement in PAM, which affects the air voids and permeability of porous mixes and effectiveness of slurry infiltration. Complexity of gradation and compaction makes it very difficult to predict air voids and permeability in PAM. Statistical analysis of these parameters based on experimental data requires very high volume of samples which is missing in literature. Compaction effort has been kept constant in most of them which otherwise is important for stability. Also, machine learning models to predict mix design of PAM has not been explored. This study investigates the PAM using systematic experimentation and modelling approaches such as Response Surface Methodology (RSM) and Artificial Neural Networks (ANN) to establish new and low trial-based guidelines for establishing a relationship between air voids and permeability greater than 0.12 cm/sec, while ensuring the stability of the mix. Using a Simplex Lattice design approach, five distinct types of aggregates were utilized as input variables and twenty-six PAM mixes were generated, which were subjected to 5 compaction levels. A total of 260 mixes were tested and then models were prepared based on RSM and ANN, which considered two output variables: air voids and permeability. RSM and ANN models demonstrated strong agreement between predicted values and actual experiment outcomes. It is observed that optimal permeability in PAM is achieved by adjusting aggregate proportions based on compaction levels. Proportion limits for low compaction suggested 60–70 % coarse aggregates and up to 20 % fines. For higher compaction (25–75 blows), it is recommended to increase the coarse aggregates up to70–90 % and limit fines up to 10 %. All proposed limits satisfied minimum permeability criteria (> 0.12 cm/sec) while ensuring the stability of the mixes. The proposed modelling approach offers greater efficiency and accuracy and reduces trial batch time, minimizing the need for destructive experiments and material wastage.
Read full abstract