7th International Conference on Managing Pavement Assets (2008)

Transcription

1 MEASURING SURFACE DISINTEGRATION (RAVELLING OR FRETTING) USING TRAFFIC SPEED CONDITION SURVEYS P. Scott Highways Agency, England Federated House, London Road Dorking, Surrey RH4 1SZ, England K. Radband and M. Zohrabi Stoneham Place, Stoneham Lane, Southampton, Hampshire SO5 9NW, England P. Sanders, S. McRobbie and A. Wright TRL, Crowthorne Berkshire. England ABSTRACT The two main deterioration parameters on the motorway and trunk road network in England are rutting in lane 1 and ravelling across the whole carriageway. Until recently, ravelling has not been measured at speed using lasers. Information on the condition of the pavement surface is used in the UK to determine the current state of the road and its deterioration rate, and hence determine what type treatment should be carried out, and when. For deterioration that appears as surface disintegration, known as ravelling or fretting, it has been found that both the measurement of the extent and degree of deterioration, and the implications of this on road safety, are not well understood. The Highways Agency commissioned research to develop an objective method for measuring the presence of fretting. Work has been undertaken to develop an algorithm to identify fretting on Hot Rolled Asphalt (HRA) surfaces using laser measured transverse profile. The algorithm development has been based on a methodology previously developed for application to texture profile data, but expanded to operate in three dimensions. This has produced a useable algorithm for detecting fretting, that has been shown capable of distinguishing between lengths containing high and low levels of fretting. Work has also been undertaken to assess the effect of fretting on skid resistance. It has been found that there is no evidence of local reduction in skid resistance associated with areas of fretting on HRA. The paper will report on the results and how this valuable additional tool can be employed in the maintenance of pavements, with intervention levels for safety and

2 Scott, Radband, Zohrabi, Sanders, McRobbie and Wright Page 2 BACKGROUND Many Hot Rolled Asphalt (HRA) surfaces have been in service for 25 years or more. Whilst these pavements remain structurally sound, the surfaces have, in some cases, started to show signs of progressive or sudden surface disintegration. This can affect the pavement performance and its ability to provide a smooth, safe and comfortable ride for users. As the pavement ages and the binder becomes oxidised this allows the chips to loosen and pull out of the surface, the asphalt then abrades rapidly. This phenomenon is known as fretting and surface disintegration is one of the main drivers for resurfacing, when levels of either safety or serviceability may be breached. To obtain best value in the operation and maintenance of the UK trunk road network it is imperative that road renewal schemes are accurately prioritised to ensure that the most beneficial schemes go ahead, and it is also important to ensure that such intervention is carried out at the optimum time (undertaking the scheme at the optimum point in the lifetime of the pavement to obtain the maximum use of the existing pavement, whilst ensuring no any expensive structural damage is incurred, and that safety is maintained). In order to do this the HA uses a Continuous Value Management approach to road renewal maintenance. This uses extensive data to determine the current state of the road and its deterioration rate, and then attempts to determine when treatment should be done. However, this approach relies on an accurate understanding of the failure mechanism and current condition of the pavements, so that the time between a pavement being identified as needing treatment soon to requiring intervention now can be judged, and the treatment applied before this latter stage is reached. This approach can be summarised in Figure 1 below. Some patching interventions Cat 1 defects Intervention period, with some possible unplanned small works interventions as pavement deteriorates Time FIGURE 1 Deterioration Curve and Maintenance Trigger Such an approach works well for deterioration where the monitoring is accurate and the deterioration processes are understood and relatively predictable, such as rutting. Defects such as fretting are currently hard to monitor accurately, and the processes which cause the defect are hard to predict. Lack of necessary knowledge of the onset of fretting and its progressive rate of failure into potholes and other surface defects provides a problem for those involved in the advance planning and funding of maintenance schemes. Temporary pothole repairs have a very limited life (perhaps as short as 24 hours in some conditions), therefore maintenance agents often prefer to use small (1m) patches to treat isolated defects. In most cases the spread of potholes caused by fretting over a large area of the carriageway results

3 Scott, Radband, Zohrabi, Sanders, McRobbie and Wright Page 3 in the need for a machine laid patch to provide longevity to the treatments. Hence, fretting becomes one of the main drivers of surface treatments. There is a clearly identified need to determine the best value intervention point to treat fretting. It is clear that money is being spent on the network to remove fretting failures and that untreated fretting may have safety implications. However, because the mechanism of fretting is not currently understood, and the defect itself is difficult to monitor accurately and objectively, it is unlikely that fretting sites are being targeted in a Best Value way. As the network ages and the proportion of HRA reaching the end of its life increases, it is becoming increasingly important to be able to identify and predict the progress of fretting in order to efficiently plan maintenance programmes in a proactive not reactive manner. Therefore a research project has been commissioned by the Highways Agency that aims to develop an objective method for measuring, reporting and monitoring the presence of fretting, with a view to improving the current knowledge of the mechanisms involved in producing fretting. This paper presents the early results of this project. CURRENT METHODS OF ASSESSING FRETTING CVI surveys by inspectors At present Coarse Visual Inspections (CVI) are performed each year, lane by lane, through the windscreen from a moving vehicle. They are carried out at traffic speed using an electronic recording device to record the presence and location of a range of defects. These surveys are not ideal, but give a general indication of the pavement condition. Defects are recorded as present or not present, with no judgement of the defect severity included. Once the survey has been completed the data is collated and presented as a percentage of each 1m length that contains each defect (fretting or patching). The results are subjective as they are dependant on the inspector s opinion and CVI surveys have variable levels of repeatability. Existing methods of automatic detection of fretting In the UK routine surveys of surface condition are also carried out under the TRAffic-Speed Condition Survey (TRACS) contract. Machines carrying out TRACS surveys record data on the surface of the pavement using a set of lasers mounted transversely on the front of the survey vehicle. Data is reported as the transverse and longitudinal profile of the pavement, and is interpreted using algorithms to estimate rutting, and ride quality. Further data is provided by a texture profile laser mounted in the nearside wheelpath, from which an estimate of the texture depth is obtained. An attempt has also been made to characterise the levels of fretting on the pavement using the texture profile data from this single laser, by processing and analysing the data using bespoke algorithms to identify the presence, or otherwise, of fretting. Some success has been found in applying the method to identify fretting on Hot Rolled Asphalt (HRA) surfaces. However, one of the main drawbacks of the current approach is that, because only a single laser is used, fretting present outside the wheelpath is not identified. SUMMARY OF THE RESEARCH Clearly, there would be benefit in obtaining a measure of fretting across the full width of the survey (traffic lane). Research has been undertaken to explore the potential of using the set of lasers mounted transversely on the front of the survey vehicle to obtain this measure. The Highways Agency s research survey vehicle, HARRIS1, which is operated on behalf of the Highways Agency by TRL, has been adapted to provide profile data from each of the 25 measurement lasers at a longitudinal spacing that, whilst less closely spaced than the texture profile data, is sufficient to provide a measure of the megatexture of the pavement, from

4 Scott, Radband, Zohrabi, Sanders, McRobbie and Wright Page 4 which fretting may be identified. Work has then been carried out to develop algorithms, based on the current approach taken with the single texture profile data, that exploit the three dimensional data provided by the 25 measurement lines to identify fretting across the width of the survey. This Enhanced Fretting Algorithm (EFA) should therefore provide a representative measurement of fretting regardless of where on the carriageway it occurs. In addition, the research has investigated the safety implications of fretting, by studying skid resistance values obtained from SCRIM on fretted sites and comparing them the results from unfretted sites. METHODOLOGY This work has concentrated on optimising the use of profile data to measure fretting on HRA. However, it has been conducted in the knowledge that it forms part of a larger process of Highways Agency research, having the objective of delivering a general measure of fretting based on data collected at traffic-speed. The work has used reference data, collected by trained inspectors on site, to assist in the development and assessment of the algorithms. Reference surveys Test sites exhibiting a range of fretting severities were selected and assessed to record the presence and severity of fretting. In order to provide more detail regarding the location of any fretting (to assist in the subsequent algorithm assessment) the test sites were separated into series of successive 5m subsections. These 5m subsections were also separated into five sections transversely to enable the recording of the fretting across the carriageway, in terms of a grid of data. Visual surveys performed onsite by trained inspectors, recording the severity of fretting on the grid using a scale from (no fretting) to 3 (no road surface left). In addition any cracks, joints, patches or features were recorded. Figure 2 shows some sample visual fretting survey results. This data was used as the reference source for the development and assessment of the fretting algorithms. Figure 2 Example of visual fretting survey output data. The final development and assessment of the algorithm made use of data from a total of 19 sites, which had a combined length of approximately 14km. All sites were surfaced with HRA, and located on the Highways Agency network. Constraints and development of the visual survey method. Weather conditions and light levels have a significant impact on the repeatability of this type of survey. Low light levels and shade, such as areas beneath bridges, make fretting much more apparent than bright light, particularly on wet pavements. On particularly sunny winter days facing into low sun fretting becomes very difficult to see at all. This may lead to severe inconsistencies between site visits. However, these conditions remain the same for any single visit to any single site; therefore the relative fretting within the site is still reliable. Profile data

5 Scott, Radband, Zohrabi, Sanders, McRobbie and Wright Page 5 Surveys were carried out with the HARRIS1 survey vehicle, collecting transverse profile data over 25 measurement lines spaced transversely at 15mm intervals and longitudinally at 6mm intervals. It can be considered that each of the 25 measurement lines contains a measurement of the megatexture of the pavement in the direction of travel of the survey vehicle. Figure 3 shows such an example of 5m of data from one of the measurement lines. The two red circles show locations where fretting may be present. Figure 4 shows all 25 measurement lines, displayed as a 3-D surface. FIGURE 3: Possible example of fretting as seen in a single longitudinal profile of surface. FIGURE 4 Example of 25 individual longitudinal profiles, used to produce a three dimensional representation of the pavement surface. Algorithm Development The development of the fretting algorithm has built on existing methods employed in the identification of fretting using single texture lasers, which has been applied elsewhere in Europe (Van Ooijen at al., 24) and implemented by the Highways Agency in the UK for use in TRACS surveys. The fretting algorithm uses the profile data to characterise the underlying texture of the pavement, and then identify local differences from this characteristic level. Locations which differ from the characteristic level by a sufficient depth and, over a significant length, are deemed to be fretted. The proportion of the road affected by fretting is reported. To summarise, for each of the 25 lines of profile data: The measured profile is filtered to remove long wavelengths, such as those representing the shape of the road, leaving only data which is representative of the surface texture of the pavement. A baseline is established, against which the measured, filtered values are assessed. This baseline value is calculated over relatively short lengths to enable the baseline value to react to changes in surface texture. In this case the Mean Profile Depth or MPD was used. This is an internationally defined method used in the assessment and measurement of surface texture depth. Each filtered measured profile point is compared against the baseline value. If the measured filtered profile value meets certain criteria in relation to the baseline then the presence of fretting is reported. These criteria are referred to as D (which is a unitless

6 Scott, Radband, Zohrabi, Sanders, McRobbie and Wright Page 6 parameter defining the required difference which must be observed between the baseline and the filtered profile before fretting can be reported) and L (which is the length of profile over which D must be exceeded before fretting can be reported). The percentage of filtered measurement data points over the preceding baseline length which have satisfied the criteria is reported as fretting. Algorithm development proceeded through the selection of optimum filter lengths, baseline lengths and parameter values. In particular, different values and combinations of D and L were explored and the results assessed, on localised lengths, against the reference. ASSESSMENT OF THE FRETTING ALGORITHM It was accepted that it would not be realistic to expect the fretting algorithm to deliver a direct like for like comparison with the reference visual survey data. A fundamental objective would be to obtain an algorithm that was able to highlight poor lengths of pavement (i.e. those containing fretting) that were also reported in the reference, whilst not falsely reporting fretting on lengths which are in fact sound. Therefore the level of detail in the results of the fretting algorithm data was reduced by calculating the level of fretting present on the pavement over grid squares of size.5m by.5m. Following a degree of experimentation it was concluded that each of the.5x.5m grid squares would contain a value representing the mean value of the fretting algorithm values reported within the grid square. This fretting grid data could then be directly compared with the reference data. However, because the reference data was only reported over four levels of intensity it was also necessary to develop threshold values for the fretting algorithm to enable direct comparison of the results. Figure compares the gridded, thresholded algorithm data with the reference, using four test sets of algorithm parameters. Here, white represents category (no fretting), green category 1, amber category 2 and red category 3 (severe) fretting. It can be seen that, by using the appropriate parameters, the fretting algorithm is able to identify and classify fretting reported in the visual survey. Figure 5 Grid representation of fretting showing reference data and the results of the fretting using different algorithm parameters. Although development of the algorithms is assisted by localised comparisons, as shown in Figure 5, examination of the entire test dataset (and comparison with the reference) requires further reduction of the data into longer lengths. Therefore the reference and fretting algorithm results were amalgamated into 1m subsection lengths, with one representative value for the entire carriageway width. This was done by calculating the 95th percentile of all the.5m grid values across the carriageway in a 1m length. These values were thresholded

7 Scott, Radband, Zohrabi, Sanders, McRobbie and Wright Page 7 using appropriate values to again produce four categorisations of fretting severity, at 1m resolution. FIGURE compares the thresholded reference and automated (fretting algorithm) data over approximately 3km of the test data set (with D=2.2, and L=12). It can be seen that there are areas of general agreement in the data. Where the reference reports the presence of fretting, it is usually reflected in the output of the algorithm, especially when the fretting is severe. However, Figure 6 also shows that there are lengths where discrepancies exist between the reference fretting and the results of the algorithm. Figure 7 compares the results of the fretting algorithm and reference data over 14km of the test dataset (without thresholding). Again, there is agreement between the two data sources, particularly in areas where the reference reported a lot of fretting, but there is a lot of noise in the data from the algorithm. This data is re-plotted in Figure 8, following thresholding. The two datasets in Figure 8 are offset for clarity. It can be seen that there is some agreement, with the general trends being followed, but there are discrepancies, notably between about 4 and 5m where the reference data shows there to be no fretting of note, but the algorithm consistently reports fretting at category 1. Work is ongoing to fully gauge the performance of the algorithm, identify any further areas for improvement, and assess the implications for the end user. FIGURE 6 1m aggregated plots showing sum of all primary reference fretting across carriageway in 1m subsection, and 95th percentile value of all EFA fretting across carriageway in 1m subsection.

8 Scott, Radband, Zohrabi, Sanders, McRobbie and Wright Page 8 Primary reference (Sum) Primary Reference EFA EFA output (95 th percentile) Chainage (m) FIGURE7 Primary reference and EFA output metrics for dataset. Primary reference categorisation Primary Reference EFA EFA output categorisation Chainage (m) FIGURE 8 Primary reference and EFA categorisation for dataset. SAFETY IMPLICATIONS OF FRETTING In parallel to the work being carried out to develop a way of objectively measuring fretting, a limited investigation has been carried out to determine whether skid resistance reduces on fretted surfaces. This has taken the form of SCRIM testing in the nearside wheelpath of pavements showing a range of fretting. Approach A detailed comparison of SCRIM measurements and ratings of fretting in five longitudinal strips, including the nearside wheelpath, was done for a total of 6.3km of HRA from the M4, the M3, the A34 and the A3. SCRIM testing was carried out in accordance with BSI (26) with three runs in the nearside wheelpath. SCRIM measurements were recorded at 5m intervals to enable direct comparison with visual measures of fretting that were also taken over 5m lengths. Results and observations Typical examples of the recorded SCRIM coefficients and visual fretting ratings are given in FIGURE and FIGURE. These show no obvious relationship between SCRIM and the degree of fretting.

9 Scott, Radband, Zohrabi, Sanders, McRobbie and Wright Page 9 SCRIM vs Fretting M4 MP Fretting rating 3 2 Patch Patch.3.2 SCRIM Coefficient Chainage (m) Fretting Rating SCRIM Run 1 SCRIM Run 2 SCRIM Run 3 FIGURE 9 Measurements of SCRIM and fretting at MP122 on the M4 Fretting vs SCRIM M4 MP Fretting rating Patches Patch SCRIM Coefficient Chainage (m) Fretting SCRIM Run 1 SCRIM Run 2 SCRIM Run 3 FIGURE 1 Measurements of SCRIM and fretting at MP38 on the M4 M4 MP122/3-MP122/9 SCRIM Coefficient Fretting FIGURE 11 Comparing SCRIM and fretting data over 6m of the M4 This was the case for each of the sites tested, although for pavements which were more significantly fretted (above the level of 1.5, say see Figure 11) it appeared that the skid resistance tends to increase as the severity of fretting increases. This increase in skid

10 Scott, Radband, Zohrabi, Sanders, McRobbie and Wright Page 1 resistance is not surprising as for a fretted pavement with a fretting rating of 2 or even 3; there are still a large proportion of chippings still remaining on the surface, and a proportion of exposed aggregate within the HRA matrix. It is noted that in some locations where skid resistance has fallen but little or no fretting has been recorded, substantial patching has been carried out. The rate of fretting is likely to be determined by a range of factors including the binder and aggregate types, exposure to de-icing salts, age and traffic loading. Individually and in combination, the influence of these factors on fretting is difficult to determine. Work is therefore needed to help define these influences and thus enable more accurate budgeting for maintenance programmes to be carried out. CONCLUSIONS The development of a quantitative, traffic-speed technique for the measurement of fretting is highly desirable, as it will enhance the identification and prioritisation of lengths of the network requiring maintenance. Work has been undertaken to develop an algorithm to identify fretting on HRA using transverse profile data. The algorithm development has been based on a methodology previously developed for TRACS, but expanded to operate in three dimensions. This has produced a useable algorithm for detecting fretting, that has been shown capable of distinguishing between lengths containing high and low levels of fretting. However, further development is required. This further development will form part of the Highways Agency s longer term goals of developing a method to identifying fretting on all surface types. Such further development may include advanced analysis techniques such as dynamic parameter selection, or include additional data sources such as images of the road surface. The SCRIM data did not show any evidence of local reductions in skid resistance associated with areas of fretting in the nearside wheelpath on HRA. There were, however, some localised lengths of lower skid resistance where patching had taken place. REFERENCES BSI (26). Methods for measuring the skid resistance of pavement surfaces. Sideways-force coefficient routine investigation machine. BS7941-1:26, British Standards Institution, London. Van Ooijen W, Van den Bol M and Bounmen F (24), 5th Symposium on Pavement Surface Characteristics (SURF24), June 24, Toronto, Canada