The key project objective was to assess and evaluate the feasibility and accuracy of

custom software used in smartphones to measure road roughness from the

accelerometer data collected from smartphones and compare results with PASER

(Pavement Surface and Evaluation Rating System) and IRI (International Roughness

Index) measurement values collected from the same roadway segments. This

project is MDOT’s first large implementation of a customized Android smartphone

to collect road roughness data using a methodology developed from previous

research work performed by UMTRI. Accelerometer data collection was

performed via Android-based smartphones using a customized software application

called DataProbe. During the project’s initial phase smartphones were installed in

each of nine Michigan Department of Transportation (MDOT) vehicles driven by

MDOT employees. These same vehicles also were used during 2012 and 2013 to collect data on road distress using PASER Ratings for comparison.

The DataProbe software application was used to collect data and transmit it to a

University of Michigan Transportation Research server, where it was sorted, stored,

and analyzed. All MDOT regions are represented in this analysis that compares

road roughness ratings for nearly 6000 one tenth of a mile road segments. For the

second phase of the project, road distress (PASER Rating) data was collected in

2014 simultaneously with an MDOT vehicle equipped with an IRI device and two

DataProbe smartphones and two UMTRI vehicles equipped with five DataProbe

smartphones.

The analysis of the 2012 and 2013 data found that there were a number of

significant predictors of IRI road roughness including: the phone and the vehicle

used to collect the data, the speed of the vehicle collecting the data, the type of road

surface, date of data collection, and accelerometer variance. By including quadratic

terms to adjust for non-linear relationships and interactions among the predictors

studied in this project, the multiple regression model predicted nearly 45 percent and

43 percent of the variance in IRI values, respectively. An analysis of commonly

used IRI categories (3 level/5 level) using ordinal logistic regression found that

DataProbe accurately predicted these categories 68/71 percent of the time (2012

data), 77/76 percent of the time (2013 data).

Analysis of the data collected in 2014 showed multiple regression models with

variance among accelerometer measurements and speed accounting for 37 percent

of the variance, while the ordinal logistic regression accurately predicted the IRI (3

level/5 level) categories 86/83 percent of the time. These results are promising

when considering the near term application of the DataProbe technology for smaller

locales that drive over their local roads more often, generating web-based road

roughness visuals of each of the roads in their jurisdiction. In the longer term, statewide

road roughness measurement may be performed through the crowd-sourcing

model available through Connected Vehicle initiatives, where all vehicles will be

equipped with devices that support safety applications as well as other applications

such as those that measure road roughness.