U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-HRT-05-054
Date: September 2005
Quantification of Smoothness Index Differences Related To Long-Term Pavement Performance Equipment Type
Chapter 8: Conclusions and Recommendations
High-quality longitudinal profile data have been collected with all three inertial profilers that have been used in the LTPP program-the DNC 690, T-6600, and ICC profilers. The data collected by these profilers provide a valuable resource for researchers. Good agreement in IRI values between the DNC 690 and T-6600 profilers was observed. Also, good agreement in IRI values between the T-6600 and ICC profilers was observed. This indicates that the IRI values in the LTPP database can be used for analysis of roughness progression at test sections without any adjustments being made to the IRI values obtained by the different profilers. The research also indicated that using IRI to evaluate profiler repeatability, accuracy, and reproducibility is not sufficient. The following are conclusions and recommendations based on the analyses that were conducted for this project.
Conclusions: Data collected by inertial profilers do not accurately portray very narrow features such as cracks in either AC or PCC pavements, or joints in PCC pavements. This is not an error, but rather an effect of the low-pass filtering that is performed on the profile data. Evaluation of 25-mm (1-inch) data collected by both the T-6600 and ICC profilers over a joint in a PCC pavement showed that the joint appeared in the profile as a feature that was spread over a distance of 75 mm (3 inches), when the width of the joint was actually closer to 10 mm (0.4 inches). This happens because of the low-pass filter that is applied to the profile data. The low-pass filter applied to the profile data will distort narrow downward features by attenuating the depth of the feature, and will also spread the feature over a distance that is more than the actual width of the feature.
Conclusions: In the LTPP program, the 25-mm (1-inch) data obtained from the T-6600 and ICC profilers are processed using the ProQual software. ProQual applies a 300-mm (11.8 inch) moving average onto the 25-mm (1-inch) data and extracts data at 150-mm (5.9-inch) intervals. These data are used to compute IRI values, and this averaged data are uploaded to the LTPP database. The application of the moving average onto the 25-mm (1-inch) data attenuates features with wavelengths less than 1 m (3 ft). Detailed profile features cannot be observed in the ProQual-processed data because of this effect. The moving average also can cause distortion of the profile data, and the averaged data can show features that are not actually present in the pavement while eliminating features that are present.
Recommendations: Although the profile data that are currently in the LTPP database can be used for many research purposes, these data are not useful for researchers who are interested in short-wavelength data or for those who are interested in examining minute details in the profile (e.g., pavement distresses and joints in PCC pavements). Therefore, it is recommended that a procedure be put in place where the 25-mm (1-inch) interval data are easily available to researchers.
Conclusions: Since the DNC 690 profiler recorded profile data at 152.4-mm (6-inch) intervals, when comparing data from the T-6600 profiler with the DNC 690 profiler, only the 150-mm (5.9-inch) interval ProQual-processed data from the T-6600 profiler can be compared. Comparison of profile data for the two profilers showed good agreement, although there were some differences in the profiles for sections that had significant long-wavelength content. These differences are attributed to the different long-wavelength cutoff filter values used with the two profilers (91 m (300 ft) for the DNC 690 profiler, and 100 m (328 ft) for the T-6600 profiler). An evaluation of the profile data indicated that the long-wavelength cutoff filtering technique used in both of these profilers appeared to be similar. Good agreement in IRI values for the DNC 690 and T-6600 profilers was observed.
Conclusions: Since the 25-mm (1-inch) interval data were available for both the T-6600 and ICC profilers, a comparison of the 25-mm (1-inch) interval data for the two profilers could be performed. Comparison of profile data for the two profilers using PSD plots indicated that there were differences in both the short and long wavelengths. There was good agreement in the profile data for the two profilers for wavelengths between 1 and 40 m (3 and 131 ft). For wavelengths less than 1 m (3 ft), the ICC profiler usually showed a higher wavelength content than the T-6600 profiler. This is attributed to the smaller footprint of the ICC profiler, which most likely caused texture effects and the higher magnitude of narrow features to be recorded. For wavelengths greater than 40 m (131 ft), the T-6600 profiler recorded more wavelength content than the ICC profiler. This is attributed to differences in the long-wavelength filtering techniques that are used by the two profilers, although both profilers apply an upper-wavelength cutoff filter of 100 m (328 ft). When the ProQual-processed data for the two profilers are compared using PSD plots, only the differences at the higher wavelengths will be seen, and good agreement between the two profilers will be seen for wavelengths less than 1 m (3 ft) since the application of the moving average attenuates the short-wavelength features. Good agreement in IRI, which is primarily influenced by wavelengths between 1 and 30 m (3 and 100 ft), was obtained for data collected by these two profilers.
Conclusions:In the LTPP program, IRI values are computed using ProQual, which uses the IRI algorithm documented in World Bank Technical Report 46.(9) The IRI algorithm applies a 250 mm (9.8-inch) moving average onto the profile data before computing IRI if the data recording interval is less than 250 mm (9.8 inches). Literature published after the World Bank report have indicated that the moving average applied by the IRI algorithm should be omitted if the profile data have already been subjected to a moving average.(2,30) Before computing IRI, ProQual applies a 300-mm (11.8-inch) moving average onto the 25-mm (1-inch) profile data collected by the T-6600 and ICC profilers; however, this moving average is not applied to the 152.4-mm (6-inch) interval profile data collected by the DNC 690 profiler, which already has been subjected to a 304.8-mm (12-inch) moving average. When ProQual computes IRI from such data, the moving average in the IRI algorithm is applied to the data. Thus, during computation of IRI, the profile data are subjected to two moving averages, one applied by ProQual and the other applied by the IRI algorithm. This will cause a slight downward bias in the IRI values. A limited analysis using a set of sections with IRI ranging from 0.9 to 2.8 m/km (57 to 178 inches/mi) showed that the current LTPP procedures for computing IRI result in a downward bias in IRI ranging from 0.7 to 2.3 percent.
The IRI is influenced by the sampling interval. Thus, if a researcher obtains 25-mm (1-inch) interval LTPP profile data and computes IRI, the resulting IRI values will be slightly higher than the corresponding IRI values that are stored in the LTPP database. This is because IRI in the LTPP database have been computed after the 25-mm (1-inch) profile data have been processed using ProQual, which applies a 300-mm (11.8-inch) moving average onto the data, then extracts data points at 150-mm (5.9-inch) intervals and uses these data to compute IRI.
Recommendations: The current procedure used by ProQual to compute IRI slightly underestimates the IRI value. The procedure described in the World Bank report on which the IRI computation procedure used in ProQual is based does not specifically indicate that the moving average applied by the IRI algorithm should be omitted if the profile data have already been subjected to a moving average.(9) Since all time-sequence IRI values for a test section have this bias, the bias in the IRI values will not affect roughness progression studies performed using these data. Reprocessing profile data to compute IRI values by omitting the moving average applied by the IRI algorithm will be a major undertaking that will require a vast amount of resources. The very slight bias in IRI is not a major error that justifies reprocessing the IRI values. Thus, it is recommended that no changes be made to the current procedure for computing IRI values.
Conclusions: A variety of factors can cause the IRI obtained from the Dipstick data to differ from the IRI obtained from the profiler data. The factors that contribute to the differences between the Dipstick IRI and the profiler IRI are:
Despite these limitations, data from past LTPP comparisons have shown that good agreement between profiler IRI and Dipstick IRI, typically within ±0.16 m/km (±10 inches/mi), can usually be obtained at sections that do not have significant distress. Current procedures for LTPP profiler comparisons use the average profiler IRI obtained from five error-free runs for comparison with the Dipstick IRI. This procedure helps smooth out some of the variability in the profiler runs.
Recommendations: One of the tasks in this project was to provide recommendations on recalculating IRI because of differences between profiler IRI and Dipstick IRI. The current IRI values in the LTPP database that were computed from profiler data are considered to be accurate, and no recalculation of IRI is necessary.
Conclusions: The profile data for repeat runs that were collected by the T-6600 profiler before June 2000 may show poor repeatability at a few sections where an extremely rough localized feature is present on the pavement. This is because such a feature can cause the accelerometer(s) in the profiler to exceed the range and contaminate the long wavelengths in the profile data collected after such an event. However, IRI computed from such profiles still will be accurate, since the contaminated wavelengths are outside of the wavelength range influencing the IRI. The range of the accelerometers in LTPP's T-6600 profilers was increased during May-June 2000, and data collected after that were not expected to show such behavior.
Conclusions: In some instances, PSD plots of data collected by T-6600 profilers showed either one or two spikes for wave numbers greater than 5 cycles/m (1.52 cycles/ft), which corresponds to wavelengths less than 0.2 m (0.7 ft). This indicates that there is some contamination in the profile data. This phenomenon will not be seen in the profile data that are in the LTPP database, since wavelengths of less than 1 m (3 ft) have been attenuated in these profiles by applying the moving average. The cause of this contamination is not known. This contamination does not affect the IRI or RN because it occurs at wavelengths outside of the wavelength range influencing both of these indices. The presence of this contamination will not affect many of the analyses that can be performed using the 25-mm (1-inch) profile data.
Recommendations: Researchers who are using the 25-mm (1-inch) data and performing research that make use of the extremely short wavelengths should be aware of this issue. They should evaluate the data and determine whether any contamination is present in the short wavelengths and, if it is present, determine whether it will impact the research that they plan to do with that data.
Conclusions:Data obtained from the 1998 LTPP profiler comparison indicated that some contamination was present in the data collected by the right sensor of the Western profiler for wave numbers greater than 1 cycle/m (0.3 cycle/ft), which corresponds to wavelengths less than 1 m (3 ft). This phenomenon will not be seen for ProQual-processed data that are uploaded to the LTPP database, since short wavelengths are attenuated by the application of the moving average. The contamination does not affect IRI because it occurs at wavelengths that are outside of the range of wavelengths influencing the IRI. However, the contamination is affecting the RN computed from 25-mm (1-inch) data. The presence of this contamination does not mean that the 25-mm (1 inch) data obtained from this sensor are not usable. However, 25-mm (1-inch) data with this contamination cannot be used for any analyses that make use of wavelengths of less than 1 m (3 ft). This phenomenon was not noted in the data collected by the Western profiler during the 1996 verification study nor in the data collected during the 2000 profiler comparison. Therefore, it appears that this phenomenon occurred sometime between 1996 and 1998, and was fixed sometime between 1998 and 2000, or perhaps this is an intermittent problem.
Recommendations: It is recommended that a set of 25-mm (1-inch) interval data collected by the Western profiler at regular time periods (e.g., 6-month intervals) be obtained and reviewed to pinpoint when the problem with the sensor began, when the problem was fixed, and to identify the cause of the problem. Overlaid PSD plots of left- and right-sensor data for the set of profiles can be used to investigate this issue. Researchers who use 25-mm (1-inch) data and conduct research involving short wavelengths should be aware of this issue, since this contamination can be interpreted as a pavement effect.
Topics: research, infrastructure, pavements and materials
Keywords: research, infrastructure, pavements and materials,IRI, inertial profilers, Dipstick, pavement data collection, pavement profile, profile measurement, profiler, LTPP.
TRT Terms: Pavements--Smoothness, Roads--Riding qualities, Pavements--Evaluation, Pavements--Inspection, Profilometers, Ride quality, Data collection