Towards a mix design model for the prediction of permeability of hot-mix asphalt

• Published in: Construction & building materials 2019-10, Vol.221, p.637-642
• Main Authors: Blaauw, Sheldon A., Maina, James W., Horak, Emile
• Format: Article
• Language: English
• Subjects: Bailey; DASR; Design; Durability; HMA; Machine-learning; Mix; Packing; Permeability
• ISSN: 0950-0618; 1879-0526
• DOI: 10.1016/j.conbuildmat.2019.06.082

[Display omitted] •Use of Probability Density Function to delineate sections based of propensity for permeability.•Use of regression analysis of permeability control as a function of the micro level aggregate subset.•Use of machine learning with the Gaussian Process Regression to predict porosity.•The Bailey ratios and porosity correspond to permeability potential and field performances.•Use of the binary aggregate ratio to predict the permeability of an asphalt mixture. Hot-Mix Asphalt (HMA) is a designed pre-mix of graded aggregates and bituminous-based binder, hot applied on road pavements to provide a smooth and safe riding surface and protect the underlying layers. Current HMA mix design methods are focusing, mostly, on the strength of the mixture. However, the mixture permeability, which is an equally important performance parameter of an HMA layer – strongly influencing the durability of the HMA and underlying layers – is not always pertinently used as a mix design parameter. Exclusion of permeability is attributed to uncertainty on the accuracy of measured permeability, as well as poor correlation among predicted, field, and laboratory results. In this paper, it is hypothesised that practical and reliable methods are available to predict and measure the permeability of HMA in the field and laboratory as well as to successfully relate permeability parameters to long-term performance of HMA. The objective of the study is to establish these relationships and validate an HMA design model to predict durability. Numerous theoretical models were evaluated and the selected model is based on regression analysis of field data. Initially, predominant variables influencing layer permeability were found to be void proportion, binder content as well as aggregate grading and packing. Subsequently, regression analyses of data showed that within reasonable variation of the other variables, layer permeability is strongly correlated to aggregate grading and packing, as described by rational Bailey ratios, hence the focus of refining the model on the latter parameters. The research confirmed the hypothesis by showing an acceptable level of confidence in the relationships. The study objective was achieved to the extent that the predicted levels of permeability correlated well with field observations of durability performance. This was enhanced with machine learning modelling of the data sets available from recent as-built road sections with similar design and specifica