7th International Conference on Managing Pavement Assets (2008) PROCEDURE TO PROCESS, HARMONIZE AND ANALYZE GRIP TESTER MEASUREMENTS Hernán de Solminihac1, Alondra Chamorro2, Tomás Echaveguren3 Pontificia Universidad Católica de Chile Departamento de Ingeniera y Gestión de la Construcción Vicuña Mackenna 4860, Macul Santiago Po Box 160, Correo 160, CP 6904411 Fax: (562)354-4806 Phone (562)354-4989 (1) Email: [email protected] (Corresponding Author) (2) Email: [email protected] (3) Email: [email protected] ABSTRACT Skid Resistance data is affected by several factors, such as temperature, climate seasonal variations, pavement surface conditions and operation of devices. Therefore, measures may differ considerably between each other and cannot be compared directly. This is especially critical when comparing skid resistance between different devices as they may also differ in the measuring principal being used. The study proposes a procedure to process and analyze skid resistance values measured with Grip Tester and, harmonize them to SCRIM (Sideway Coefficient Routine Inspection Machine) measures. The method considers four steps. First invalid data is detected and eliminated. Second, adjustment factors to correct temperature and speed variations are applied. Third, processed Grip Tester measurements are harmonized to SCRIM values. And finally, an analysis methodology to detect outlier data, singularities, data structural changes and segmentation is presented. To estimate adjustment factors measurements with Grip tester were performed at different speeds and in roads presenting important temperature oscillations. Finally, to harmonize Grip Tester measurements to SCRIM values, a particular method was developed based on the PIARC (World Road Association) experiment. From the study it was concluded that Grip Tester values have to be processed and corrected due to external factors affecting the measuring procedure. Once data is processed and harmonized, it can be analyzed and compared to measures obtained with other devices. INTRODUCTION Grip tester is a trailer that measures continuously and dynamically the longitudinal skid resistance coefficient of a pavement expressed as Grip Number (GN). Skid resistance is obtained in terms of the friction force observed between a partially locked wheel and the wet pavement. Measures may vary with temperature, climate seasonal variations, pavement surface conditions and the operation of the equipment. Measures performed with the same Grip Tester or between different Grip Testers can differ considerably because of these factors. Harmonization of Grip Tester with other devices is required when their measures need to be compared. This is because different devices measure skid resistance according to diverse principals. In the case of SCRIM (Sideway Coefficient Routine Inspection Machine), skid TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s). 7th International Conference on Managing Pavement Assets (2008) De Solminihac, Chamorro, and Echaveguren Page 2 coefficient is measured with a wheel at an angle of 20° to the direction of travel. Therefore, the transverse skid resistance of a pavement is obtained in terms of the SCRIM values (SidewayForce Coefficient: SFC). A research project lead by the Pontificia Universidad Católica de Chile in partnership with the public and private sector has been studying the skid resistance phenomena in Chile during the last four years. From this, a friction standard was developed for Chile. One of the scopes of the project was to define clear and precise procedures for the measurement of skid resistance with different devices. This paper presents the results of the processing and analysis procedure of Grip Tester measurements, considering also the harmonization of these measures to SCRIM values (SFC). Objective and Scope The main objective of the paper is to present a clear and precise procedure to process and analyze skid resistance values measured with Grip Tester, considering the harmonization of these values to SCRIM measures. The procedure considers four general steps: first, invalid GN data is detected and eliminated from the database; second, adjustment factors to correct temperature and speed variations are applied; third, processed GN data are harmonized to SFC and; finally, an analysis methodology to detect outlier data, singularities, data structural changes and segmentation is applied to the harmonized values. Research Methodology The research methodology followed for the development of the procedure considered three different experiments: temperature tests, speed tests and the harmonization experiment. Previous to these experiments a repeatability test was held with each device considered in the study. The measuring methodology and the repeatability tests followed a standard procedure described in detail in de Solminihac et al. (1; 2). This was done to ensure the precision and reliability of each device under constant testing conditions. From these, repeatability of Grip Tester devices, SCRIM and laser profiler was checked and approved. To estimate the temperature factor, measures with Grip Tester were performed in roads presenting different climates and important daily temperature oscillations. Measures showed a considerable difference when temperatures ranged between 15°C and 40°C. To obtain the speed factor, measures with three different Grip Testers were performed at 30 km/h, 50 km/h and 70 km/h. Tests were performed in roads presenting different geometrical and surface characteristics. Finally, to harmonize Grip Tester measurements to SFC, tests were performed with both devices at 30km/h, 50 km/h and 70 km/h in road sections presenting different skid resistance and macro texture levels. Macro texture of the selected sections was evaluated with a laser profiler. A particular method was developed to harmonize the devices based on the PIARC (World Road Association) experiment (3; 4) and the HERMES experiment (5; 6). Data obtained from the three main experiments was analyzed in detail. Correction factors, harmonization requirements and a methodology to detect outlier data, singularities, data structural changes and segmentation were defined from this analysis. As the final result of these, the procedure to process, harmonize and analyze grip tester measurements was designed and developed. TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s). 7th International Conference on Managing Pavement Assets (2008) De Solminihac, Chamorro, and Echaveguren Page 3 PROCEDURE TO PROCESS, HARMONIZE AND ANALYZE GRIP TESTER MEASUREMENTS General Framework The proposed procedure considers the four general steps presented in Figure 1 and described as follows: Field Data STEP 1 Data Pre-Processing Elimination of Invalid Data STEP 2 Data Processing Temperature Adjustment Speed Adjustment Adjusted Grip Number (GNA ) Need to Harmonize? NO YES STEP 3 Harmonization Equivalent Side Force Coefficient (SFCe) STEP 4 Data Analysis Outlier Data Singularities Structural Changes Segmented Data FIGURE 1 Procedure to Process, Harmonize and Analyze Grip Tester Measurements - - STEP 1: Grip Tester data obtained from the field is pre-processed by detecting and eliminating invalid data. STEP 2: Adjustment factors to correct temperature and speed variations are applied to the preprocessed data. The output of the data processing is an Adjusted Grip Number (GNA). STEP 3: The Adjusted Grip Number (GNA) values are harmonized to SFC. This step needs to be considered when skid resistance thresholds of a road network are defined in terms of SFC. The output of this step is an Equivalent Sideway-Force Coefficient value (SFCe). STEP 4: An analysis methodology to detect outlier data, singularities and data structural changes is applied to the harmonized values. As a result of this final step, the analyzed data is segmented and reported. TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s). 7th International Conference on Managing Pavement Assets (2008) De Solminihac, Chamorro, and Echaveguren Page 4 STEP 1: Data Pre-Processing Invalid data caused by singularities detected in the field that caused distortion to the measured data should be eliminated of the database for data processing and analysis, but should be identified in future reports. This is the case of local defects observed in the pavement (like potholes or shoving), surface contamination (dust, oil or rubber from tires) and physical elements (bridge metal joints, rumble strips or humps). These singularities should be detected and reported during data collection. STEP 2: Data Processing Temperature Adjustment Factor The objective of temperature adjustments is to correct Grip Tester measurements performed in periods during a day where the temperature gradient is high. GN values obtained in several test sections presenting different climatic and temperature regimens showed to be sensible under extreme conditions. Pavement temperature has logistic behavior during the day ranging between two high and low asymptotic temperature values, named thermal stability. Only when the pavement reaches its thermal stability, GN values are no longer affected by temperature oscillations. Outside the thermal stability range, GN values may be highly affected by temperature changes. The temperature adjustment factor should be applied when measures are made during moments of the day where the pavement temperature differs significantly to a reference condition. The principle of this factor is to take the GN values to an equivalent value. By convention, the reference values were defined as GN values measured at 20 °C. Equation 3 presents the recommended temperature adjustment factors for Grip Tester, where Ta stands for environmental temperature. ⎧0.10 − 0.005(Ta ) ; 10°C < Ta < 20°C ⎪ FT (20) = ⎨ 0.06 − 0.003(Ta ) ; 20 ≤ Ta ≤ 30°C ⎪0.04 − 0.002(T ) ; Ta > 30°C a ⎩ (3) This temperature adjustment factor is applied only during summer months, as during the rest of the year temperature oscillations do not affect measures significantly. It has to be noticed that the temperature factors were developed for Chilean temperature regimes and seasonal conditions. It is recommended that similar factors should be developed for local conditions, when not available. Speed Adjustment Factor The purpose of the speed factor is to take measures performed at different speeds to a reference speed of 50 Km/h. According to measuring recommendations for the Chilean road network (1), measures should be made at 50 ± 4 Km./h in secondary roads and at 70 Km./h ± 4 Km./h in primary roads and freeways. When traffic conditions cannot ensure this speed ranges, the driver may adjust the measuring speed to the traffic conditions to guarantee safety during surveys. For all cases when the reference speed of 50 ± 4 Km./h is not fulfilled, the correction factors presented in Equations 1 and 2 should be applied. The relationship between speed and Grip Number considered in these equations was estimated from the speed tests. TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s). 7th International Conference on Managing Pavement Assets (2008) De Solminihac, Chamorro, and Echaveguren − High Macro texture (> 1 mm) − Low Macro texture (< 1 mm) FS (50) = 0.08 − 0.0015( Se ) FS (50) = 0.30 − 0.006( Se ) Page 5 (1) (2) Calculation of the Adjusted Grip Number The Adjusted Grip Number (GNA) is estimated using Equation 4. The GNA is the corrected GN value obtained after eliminating invalid data and after applying the speed and temperature adjustment factors to the data set. This is the output value of the data processing step from the general procedure. GNA = GN + FS(50) + FT(20) (4) STEP 3: Harmonization of Grip Tester to SCRIM measures Harmonization can be defined as the action of making identical or minimizing the differences between related measures of similar scope but performed with different devices or procedures. Therefore, measurements of two or more equipments that evaluate the same phenomena but with different procedures can be compared between each other when harmonized. In the case of this study, the scope is to compare measurements of Grip Tester to SCRIM, by obtaining a GNA value equivalent to a SCRIM value, so called equivalent SFC (SFCe). Several skid resistance and friction harmonization methods have been developed, such as the International Friction Index (IFI), European Friction Index (EFI), International Runway Friction Index IRFI, ESDU method and direct correlations (3; 5; 6; 7). The proposed harmonization procedure considers concepts from the first three methodologies. Equations 6 and 7 were deducted from the theoretical approach of the IFI calculation procedure. SFCe = C ( Se2 , Sp2 , V ) { A1 + B1 [ (GNp) D( Se1 , Sp1 , V ) ]} (6) C ( Se2 , Sp2 , V ) = e V − Se2 Sp2 ; D ( Se1 , Sp1 , V ) = e Se1 −V Sp1 (7) Where, Sp1 and Sp2 are speed constants; V is the reference speed; Se1 and Se2 are the measuring speeds of Grip Tester and SCRIM respectively; and, A1 and B1 are calibration constants. With these two equations an equivalent SFC value can be estimated from a GN value obtained from a Grip Tester (GNp) that is representative of a Grip Tester fleet. The equations consider the correlation existing between macrotexture and skid resistance through the speed constants Sp1 and Sp2. The V constant is used to optimize the calibration between both equipments. In the PIARC experiment, from where IFI was derived, this speed was 60 Km/h, whereas in the EFI experiment this value was from 30 Km/h. An important issue is that the GNp value considered in the equations represents the GNA value of a particular device harmonized to a virtual Grip Tester representative of a Grip Tester fleet. When calibrating Equations 6 and 7 to local conditions, it is recommended that the virtual Grip Tester should be estimated in terms of the mean GN values of a fleet. TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s). 7th International Conference on Managing Pavement Assets (2008) De Solminihac, Chamorro, and Echaveguren Page 6 As an output of this step within the overall procedure, the SFCe value for a specific Grip Tester is obtained, which was already processed in the previous steps. This data is of particular relevance when comparing Grip Tester values to thresholds defined in terms of SFC. STEP 4: Data Analysis The objective of data analysis is to develop a skid resistance profile that can be easily compared to thresholds. The main steps are: (a) outlier detection, (b) preliminary segmentation, (c) change point detection, and (d) mean and standard deviation calculation. Outlier Detection Outliers are defined as data points observed beyond the GNA tendency. Outliers can be identified with the Leverage Test. The method measures the distance of an observed sample compared to a set of means (hi). This distance is the diagonal of a hat matrix in a linear regression. This distance is calculated with Equation 8. 1 (x i − x) 2 hi = + n (n − 1)S2x (8) Where n is the total data contained in the i-esim data set, x is the mean of the sample and 2 x S is the standard deviation. The limit value of h is calculated with Equation 9: i hi ≤ 6 n (9) This method was implemented in a digital worksheet, to simplify its application. Outlier detected in this step are not considered in the following steps, but have to be reported, and should not be eliminated from the final data report. Preliminary Segmentation Skid resistance threshold are defined as function of special sites category (such as straight lanes, horizontal curves, stopping site such as toll places or traffic signs). Then, to each site category are associates a specific threshold value. To compare on a common basis thresholds and GNA data, a first segmentation should be performed. The segmentation split GNA data into each site category existent on the lane measured. Change Point Detection To identify structural changes of the mean, a change point detection procedure that groups data statistically similar is applied. The method is applied independently on each site category and considers two steps: Segmentation and Optimization. The segmentation method finds the point where the decision function g(i) is maximized, by applying Equation 10. ⎤ n ⎡ i g(i) = ⎢ ∑ (x j − x) ⎥ i(n − i) ⎣ j=1 ⎦ 2 ; g k = max S(i) 1≤i≤ k (10) Where n is the total data contained in the analyzed section, xj is the value of the measurement in point j and x is the mean of the measurement in the analyzed section. The point TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s). 7th International Conference on Managing Pavement Assets (2008) De Solminihac, Chamorro, and Echaveguren Page 7 where g(i) is maximum in a section, is the point where the mean changes. It is also the first point of the second segment of the same section. The procedure is repeated in both segments, obtaining new maximum points g(i). The process ends when the function g(i) does not present more maximum points within. In some cases contiguous segments have similar means. Therefore, a statistical analysis has to be applied to identify whether it is possible to combine in one section means and standard deviations. With this, the number of segments obtained from the previous step is reduced. The following rules are applied for this: - If in two isolated consecutive segments the t-student test verifies the null hypothesis of equity of means, then the two segments are grouped together. The mean and standard deviation are recalculated for the new segment. - When three or more consecutive segments verify the null hypothesis of equity of means in each pair of consecutive segments, an analysis of variance for repeated measures is applied (ANOVA). Then, the Tuckey test of multiple comparisons is applied. If the null hypothesis of equity of means is verified for the group of segments, then the segments can be grouped together. Else, only consecutive segments that verify the Tuckey test can be grouped together. The outputs of this step are a series of segments with different length, where consecutive segments present means and standard deviations statistically different. Therefore, each segment represents the change point of the mean value of GNA. Estimation of Mean and Standard Deviation per Segment Once change points and outliers are identified, a systematic segmentation every 100 m is superposed to the segmentation procedure described in the last section. From this segmentation, the mean and standard deviation are estimated considering the following: 1. Case 1: No structural change exists. In this case the mean and standard deviation are calculated with all the data contained in the segment. 2. Case 2: A structural change exists. In this case the mean and standard deviation have to be calculated separately: o If the structural change is at the central third of the data set, the mean and standard deviation have to be calculated for each part of the segment. o If the structural change is at the first or final third of the segment, then the mean and standard deviation of the central third is representative of the segment. CONCLUSIONS This paper presented a general procedure for processing, harmonizing and analyzing skid resistance values measured with Grip Tester. With this, skid resistance data can be analyzed systematically and be consistently compared to threshold values defined in terms of SCRIM measures (SFC), or else in terms of other measures. From the study it was observed that skid resistance values measured in warm seasons are affected by daily temperature oscillations. An adjustment factor was therefore proposed to standardize Grip Tester values measured under different temperature conditions during warm seasons. Skid resistance values are also affected by measuring speed. Skid resistance measurements performed under heterogeneous traffic conditions may present considerable TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s). 7th International Conference on Managing Pavement Assets (2008) De Solminihac, Chamorro, and Echaveguren Page 8 alterations. The speed adjustment factor proposed in the present study standardizes measures performed at any speed to a reference speed of 50 Km/h. When skid resistance thresholds are referred to a specific device, measurements from other devices need to be harmonized to this equipment. This is the case of the Chilean standard, which is defined in terms of SCRIM values (SFC). The harmonization method presented in this study is applied to the case when Grip Tester measurements (GN) need to be harmonized to SFC. However, the proposed method can be generalized and applied with other devices. Skid resistance measurements generally present high variability. Therefore, an essential final step considered in the proposed procedure is the statistical analysis of data. This is especially important when comparing skid resistance data to threshold values. The analysis procedure proposed in the study considers the identification of outlier data, structural changes in the skid resistance measures and a consistent segmentation process. ACKNOWLEDGMENTS The authors are gratefully acknowledged to the partners and sponsors of the Research Project FONDEF D03I 1042: FONDEF-CONICYT Fondo de Fomento al Desarrollo Científico y Tecnológico, Dirección Nacional de Vialidad del Ministerio de Obras Públicas de Chile, Sociedad Concesionaria Autopista del Maipo, Sociedad Concesionaria Talca-Chillán, Sociedad Concesionaria Ruta de La Araucanía and Sociedad Concesionaria Ruta de los Ríos. REFERENCES 1. de Solminihac, H. Echaveguren, T. Chamorro, A. Vargas, S. y Bustos, M. Procedimiento de Medición y Procesamiento de Datos de Resistencia al Deslizamiento Medidos con Equipo Grip Tester. Technical Document TM-014-06-1. FONDEF D031 – 1042. Pontificia Universidad Católica de Chile. 2007a 2. De Solminihac, H. Echaveguren, T. Chamorro, A. Vargas, S. y Bustos, M. Procedimiento de Medición y Procesamiento Datos de Resistencia al Deslizamiento Medidos con Equipo SCRIM. Technical Document TM-014-06-2. FONDEF D031 – 1042. Pontificia Universidad Católica de Chile. 2007b. 3. Wambold, J. Antle, C. Henry J.J., and Rado, Z. International PIARC experiment to Compare and Harmonize Texture and Skid Resistance Measurments. Technical Commiteé C1. PIARC. 1995. 4. ASTM. Standard Practice for Calculating International Friction Index of Pavement Surface. ASTM E 1960 – 98. 1998. 5. Descornet, G. Proposal for a European Standard in Relation with the Skid Resistance of Road Surfacings. Final Report, Program of scientific support to standardization, part III. Belgian Road Research Centre. Brussels, Belgium. 1998. 6. Gothié, M. Harmonisation of European Routine and Research Measurement Equipment for Skid Resistance of Roads and Runways. The HERMES Project”. SURF 2004. 5th Symposium on Pavement Surface Characteristics. Toronto, Canada, June 6th - 10th. 2004. 7. ASTM Standard Practice for Calculating the International Runway Friction Index, ASTM E2100-04. 2004. TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).
