Artificial-intelligence based prediction of flexible pavement condition deterioration using multi-temporal...

In recent years, Egypt has witnessed a remarkable development in infrastructure projects as part of its Egypt Vision 2030, which aims to achieve sustainable development. Among the most prominent of these projects is the construction and development of efficient highway networks, which are a cornerstone for supporting transportation and connecting different governorates and cities. Given the vital role these roads play in economic and social development, monitoring their condition assessment and evaluating their efficiency regularly is crucial to improve traffic flow, enhancing road safety and reducing operating and maintenance costs. This paper investigates using Terrestrial Laser Scanning (TLS) as a high resolution, repeatable alternative for managing flexible pavement surface defects and develops a data-driven framework to predict pavement deterioration depending on several observations epochs. The objective is to quantify pavement surface distresses, track their temporal evolution, identify the distress characteristics most strongly associated with pavement condition loss, and forecast future pavement performance to support proactive maintenance planning. A 20.1 km roadway of the Kafr El-Sheikh city-Tanta city direction in Egypt was scanned with TLS instrument at eight positions and surveyed over four observations epochs (November 2022, July 2023, April 2024 and March 2025). Point clouds were processed to extract several distress types and the Pavement Condition Index (PCI) was computed for every roadway section and for all observations epochs. Multiple Linear Regression, Decision Tree, Random Forest, and XGBoost models were used in conjunction with correlation and regression analysis to assess distress-PCI relationships and forecast pavement condition. Across the studied roadway, PCI declined in every section with losses ranging from 5.4 to 45% over 28 months; pothole area was the fastest growing distress (up to + 45% per year). Correlation and regression analysis identified rutting, lane-to-shoulder drop-off and linear cracking as the distresses most strongly associated with reduced serviceability and a Random Forest model ranked raveling, linear cracking and rutting as the dominant PCI predictors. Per-section deterioration trend models (mean R² = 0.91) forecast that sections 2 and 5 will reach structural failure (PCI < 25) before 2030. The paper's novelty is that it integrates repeated multi-temporal TLS observations, objective three-dimensional distress quantification, PCI-based assessment, statistical analysis, explainable machine-learning prediction, and long-term deterioration forecasting within a unified pavement-management framework. The results show that combining multi-temporal TLS observations with advanced statistical and machine-learning techniques can significantly improve pavement asset management by allowing for accurate distress quantification, reliable deterioration forecasting and proactive maintenance planning, ultimately improving roadway performance, safety and lifecycle sustainability.