An Assessment of the Models to Predict Pavement Performance

Project Details
STATE

NE

SOURCE

NTL

END DATE

03/23/18

RESEARCHERS

Duckworth, William; Nath, Ravi; Ekpoke, Victor

SPONSORS

Creighton University; Midwest Transportation Center; US Department of Transportation

LINKS

Products

Project description

Data collected by the Iowa Department of Transportation (DOT) regarding road conditions across the state of Iowa were used to model pavement condition index (PCI). The data were for calendar year 2013, with the exception of updated PCI values from 2014 and 2015 and indicators of the resurfacing of road segments in 2014 and 2015. The data file provided by the Iowa DOT consisted of nearly 4,000 observations. Eighteen different road conditions and measures were considered as possible model inputs. Of the 18 measures, 11 were used in the final prediction of PCI in 2014 and 2015 for portland cement, composite, and asphalt cement pavement types. These measures included International Roughness Index (IRI), friction value, age, average daily traffic, PCI value in 2013, number of lanes, daily temperature change, surface type, pavement thickness, speed limit, and reconstructed kips. Series of multiple regression models were developed for the different pavement types, including aggregated pavement types with combined data. The results reveal that all 11 variables except age have a statistically significant relationship with PCI. The efficacies of the derived models, as measured by R2 values, range from 61% to 83%. Additional analyses also show that the efficacies of the derived models, as measured by root mean square error (RMSE) values, range from 6.29 to 9.52. The authors can interpret the RMSE values as indicating that approximately 95% of all prediction values should fall within 12.58 and 19.04 of the PCI values predicted by the models. Therefore, it is concluded that linear predictive models, which involve distress and descriptive characteristics of road conditions, provide a reasonable basis for estimating PCI. However, these models can be further improved by examining nonlinear effects.
TOP