A framework of integrating machine learning model and pavement life cycle assessment to optimize asphalt mixture design

Accurate International Roughness Index (IRI) deterioration patterns are key to quantifying carbon emissions (CEs) in the use and maintenance and rehabilitation (M&R) phases of pavement life cycle assessment (LCA). Despite evidence that initial IRI has a significant impact on IRI deterioration, most LCA studies still used an ideal fixed value for initial IRI. Therefore, enhancing initial IRI data quality is essential for improving LCA model. To develop an initial IRI prediction model, a dataset containing data on surface layer material, pavement structure, and construction conditions was collected from the Shanxi Department of Transportation. The best of four machine learning models used in this study has been able to achieve an excellent performance (R2 = 0.831). Factors such as compaction number and mixture ratio significantly affect initial IRI. Based on three real cases, the integration of best initial IRI prediction model with LCA and heuristic optimisation algorithm was used to obtain mixture ratios corresponding to the optimal life cycle CEs. For three cases, the total CEs of the optimized results were reduced by 22.7 % (314,942 Kg CO2-eq), 16.7 % (242,565 Kg CO2-eq) and 17.1 % (251,931 Kg CO2-eq) compared to the real results, with the use and M&R phases contributing over 90 % of the total reduction. This illustrates the need to consider the impact of materials on the use and M&R phases when evaluating material benefits using pavement LCA. These processes can also build a framework for designing surface layer mixture to minimize CEs, which can provide a reference for other material production.