In a significant advancement for the construction industry, researchers have developed a method to predict key performance metrics of asphalt pavements more efficiently, potentially transforming how materials are tested and utilized. The study, led by Ibrahim Asi from the Civil Engineering Department at Al-Ahliyya Amman University in Jordan, addresses the traditional challenges associated with determining Marshall stability (MS) and Marshall flow (MF) in hot mix asphalt (HMA) — processes that are often labor-intensive and costly.
Asi’s research leverages explainable machine-learning techniques, utilizing a comprehensive database of 721 data points that encompass critical variables such as aggregate percentage, asphalt content, and specific gravity. “Our goal was to streamline the testing process and provide a more sustainable approach to asphalt pavement design,” Asi stated. The results are promising; the CatBoost regression model, one of several tested, achieved impressive R² values of 0.835 for MS and 0.845 for MF, indicating a strong predictive capability.
The implications of this research extend beyond mere efficiency. By employing machine learning, the need for extensive laboratory testing can be significantly reduced, allowing engineers to preselect design variables with greater accuracy. This not only saves time and resources but also aligns with the growing emphasis on sustainable construction practices. “The ability to predict these parameters accurately means less waste and a more environmentally friendly approach to road construction,” Asi emphasized.
Moreover, the integration of Shapley values into the model provides insights into how different variables influence the predictions, offering a layer of transparency that is crucial for engineers and stakeholders alike. This feature enhances decision-making processes, enabling more informed choices in material selection and design.
As the construction sector increasingly seeks innovative solutions to meet both economic and environmental challenges, the findings from this study, published in ‘Transportation Engineering’, could pave the way for broader adoption of machine-learning applications in civil engineering. The potential for cost savings and improved project outcomes is immense, making this research a critical step forward in the quest for smarter, more efficient construction methodologies.
For more information on Ibrahim Asi’s work, you can visit his department’s webpage at Civil Engineering Department, Al-Ahliyya Amman University.