By Thijs van Loon (the Netherlands)
The department of ‘Management and maintenance of public roads’ [BOW] of the province of Gelderland is entrusted with the task of ensuring road safety. To spend their budget well and to prevent nuisance, it is important to plan maintenance thoughtfully. This study focusses on predicting the end-of-life of DGD pavement on provincial roads. This porous type of asphalt covers a quarter of the provincial acreage and is very susceptible to ‘ravelling’; the loss of stones in the surface of asphalt pavements.
Ravelling is currently assessed by visual inspections. These estimations are less suitable for statistical analysis due to their subjectivity and bad repeatability. The province of Gelderland puts a lot of effort into collecting and archiving data. Since several years, special laser measurement vehicles are deployed. Which record 2D-profiles of the asphalt texture with millimeter precision at traffic speeds. In this research a correlation is made between visual inspection results and texture measurements.
This research ultimately tries to exploit these valuable datasets with the goal to predict the end of the civil lifetime due to ravelling. This is done by associating the growth of ravelling with environmental conditions where the asphalt is subjected to.
Deriving ravelling from texture measurements The goal of the first research question is to come up with a model to measure ravelling by texture measurements. This will be done via ‘the decisegment approach’; Visual inspections state the severity and extent of ravelling, and the most severe patch of ravelling is normative. The inspected area, a road segment, is divided into ten deci-segments. The roughness of the most severe decisegment can then be correlated with the inspected severity class. The extent class is given by the number of damaged decisegments.
Roughness can be derived from 2D texture measurements in multiple ways. 23 options have been assessed. The ‘Core roughness depth (Rk)’ shows the best ability to distinguish the severity of ravelling. The model was able to find 26% of the segments that reached end-of-life.
It is unknown whether this low accuracy is caused by the visual inspections or texture measurements. Since the patches with the highest roughness are correlated with the severity class the roughness thresholds are relatively high. Depending on the accuracy needed for the final use of this model it is advised to alter the probability thresholds. It is recommended to circumvent the decisegment approach by using more accurate data from detailed visual inspections.
Correlating environmental conditions The second research question focusses on 1) gathering possible environmental factors, 2) expressing these factors in geospatial data, and 3) correlate this data to ravelling.
First, a list of environmental factors which are expected to have influence is set up. The international literature on the ageing of porous pavements is limited and mostly aimed at highways. Interviews with professionals have been conducted. They noted location-dependent sources of damage such as agricultural traffic, tannic acids from leaf litter.
Secondly, geospatial information was gathered to represent these environmental factors. Ultimately, the following predictors are used; Age, tree cover, days of frost, hours of rain, heavy vehicles per day, light vehicles per day and the presence of levering forces.
Hereafter these environmental factors were correlated to both the visual inspections and the texture measurements. End-of-life according to texture measurements could not directly be predicted. The roughness per decisegment is predicted. Which is converted to an end-of-life diagnostic by means of ‘the decisegment approach’. The texture-based model performed slightly worse than the inspection-based model, but both showed significant results.
Environmental scenarios giving rules of thumb The influences of environmental factors are quantifiable by comparing different scenarios. The base scenario has no overhanging trees, no nearby crossings, an average amount of traffic, and a moderate climate. Nine other scenarios were set up where these factors are altered. The differences in the growth of ravelling are noticeable and followed expectations. The test error of the model is high relative to these differences. Rules of thumb have been set up, such as “When a segment of 5 years old is below a tree, it shows the same amount of ravelling as a segment which is 6 years old.”.
Use of the outcome Recommendations towards the province of Gelderland are made. Currently, warranties demanded for road longevity are expressed in CROW inspection classes. This research shows that the use of texture measurements to quantify ravelling has the potential to be more accurate than the subjectivity of inspections. Several recommendations on improving this quantification have been noted.
It is advised to use the environmental prediction model of this research to generate shortlists. This shortlist could then be inspected more regularly, in turn preventing emergency repairs and providing time to apply rejuvenation cures or search for integrative approaches.