Modeling Pavement Roughness Using Simulated Vertical Accelerations Across Multiple Vehicles and Speeds

Abstract

Vibration data provided by vehicle fleets can offer a continuous measurement system for pavement roughness, complementing profilometer surveys for network-level management. However, its heterogeneous nature—varying suspensions and speeds—poses significant application challenges. This study examines whether vertical accelerations measured above the suspension can predict the International Roughness Index (IRI) without requiring hard-to-obtain vehicle parameters or complex techniques, thus facilitating practical adoption by road agencies. Quarter-car simulations were performed on 36 measured pavement profiles for 18 representative vehicles spanning realistic parameter ranges and travel speeds (30–120 km/h). The root mean square (RMS) of vertical acceleration was computed for each run and related to IRI through six linear regression forms, including logarithmic variants accounting for speed. Monte Carlo analysis showed that aggregating RMS across nine vehicles reduced data dispersion and yielded the lowest prediction error. Achieving the best trade-off in accuracy, the logarithm model using separate RMS acceleration and speed terms yielded a root mean square error of 0.21 m/km and an adjusted R² of 0.84 on the test phase. This work also shows that a smoothed curve of mean RMS acceleration and box plot graphs can highlight rough segments when no IRI data is available, providing a rapid screening tool for maintenance prioritization. These findings indicate that collecting data from roughly 8 to 10 vehicles per segment and applying a simple logarithm calibration can provide stable and interpretable IRI estimates. The approach can potentially scale with connected-vehicle or smartphone fleets and guide agencies in continuous condition monitoring and targeted inspections.