Spatial-Temporal Data Lakes for Predicting Porous Asphalt Lifetime

Abstract

Porous Asphalt (PA) is widely implemented on Dutch highways due to its ability to reduce noise. However, PA typically has a shorter lifetime than dense asphalt. Therefore, planning maintenance strategies is essential by employing Artificial Intelligence (AI), which depends on data quality and availability. Recent advancements in monitoring tools have increased the quantity and characteristics of data in pavement engineering. However, it introduces a new challenge in storing and managing the datasets.
A data lake architecture is proposed to manage and process a heterogeneous pavement dataset while ensuring data quality to support AI-based predictive modeling. Therefore, the objective of the research is to develop a data lake architecture that supports pavement lifetime prediction.
The methodology consists of three phases. First, system architecting requires understanding the dataset characteristics, designing the data lake framework, and designing the metadata structure. Next, the implementation includes a multi-layered architecture, consisting of data ingestion, storage, processing, consumption, and governance layers. A case study on predicting pavement lifetime in terms of raveling is performed by incorporating multiple variables, such as traffic, climatic conditions, and road maintenance history. Finally, the developed framework is evaluated using quantitative performance metrics and qualitative assessment of user experience and data management practices.
The research demonstrates that applying data lake architecture in pavement engineering supports pavement data management and efficient data processing. Moreover, integrating AI-based models with a structured data management system can support data-driven maintenance planning, which can extend pavement lifetime.