Establishment of Statewide Axle Load Spectra Data using Cluster Analysis

Transcription

1 KSCE Journal of Civil Engineering (2015) 19(7): Copyright c2015 Korean Society of Civil Engineers DOI /s TECHNICAL NOTE Highway Engineering pissn , eissn Establishment of Statewide Axle Load Spectra Data using Cluster Analysis Jeongho Oh*, Lubinda F. Walubita**, and Joe Leidy*** Received July 9, 2014/Revised October 6, 2014/Accepted October 21, 2014/Published Online December 19, 2014 Abstract The role of traffic data in the Mechanistic-Empirical (M-E) pavement design is significantly crucial in assessment of pavement performance throughout the pavement design life. Providing a well-qualified traffic data for the M-E pavement design requires a huge effort with respect to collecting and analyzing the data. The Texas Department of Transportation (TxDOT) manages 30 Weighin-Motion (WIM) stations and provided classification and weight data used for this study. Evaluating the classification data first, vehicle classification distribution along with the percentage of trucks, average annual daily truck traffic, and monthly adjustment factor were established. Processing axle load data provided the number of axles per truck and axle load spectra for each of the traditional 13 vehicle classes. To develop statewide axle load spectra data, the cluster analysis was conducted using vehicle classification distribution and the Class 9 tandem axle load spectra data to provide traffic input data for the Texas M-E flexible pavement design program where the load spectra data are not available due to the absence of WIM station. As a result, six clusters for both variables were identified and a guideline was successfully established to use those clusters to generate axle load spectra inputs for a given set of truck traffic classification data. Keywords: mechanistic-empirical pavement design, axle load spectra data, weigh-in-motion, and cluster analysis 1. Introduction The Texas Department of Transportation (Texas DOT or TxDOT) continues to use the Texas Flexible Pavement System (FPS) program for design, rehabilitation, and flexible pavement evaluation, which follows a similar approach adopted in the 1993 edition of the AASHTO Guide for Design of Pavement Structures (AASHTO, 1993). In recognition of the potential benefits of the Mechanistic- Empirical (M-E) pavement design guide process developed under NCHRP project 1-37A (ARA, 2004), TxDOT has made an effort to develop a new M-E flexible pavement design program, the TxM-E. Similar to the MEPDG, TxM-E will have load-spectra traffic inputs for estimation of the magnitude, configuration, and frequency of the traffic loads that are applied throughout the pavement design life. Several states have conducted research on characterizing traffic data in accordance with the M-E pavement design concept. Arkansas DOT sponsored a research project to develop statewide volume adjustment factors (Tran and Hall, 2007). They found that a considerable variability in Truck Traffic Classification (TTC) employed in the MEPDG seemed to better define roadway groups. In addition, the influence of the statewide class distribution factors on predicted rutting and fatigue cracking was found to be significant while the impact of statewide monthly and hourly distribution factors were not substantial compared to class distribution factors. Characterization of truck traffic in California for M-E design was conducted using Weigh in Motion (WIM) data and cluster analysis (Lu and Harvey, 2006). Cluster analysis was found to be useful in providing insight into the structure of the WIM site data and revealing certain special traffic patterns. It was also observed that truck traffic composition changes significantly day by day and among sites, but has little seasonal variation. Indiana DOT assessed technologies in processing WIM and Automated Vehicle Classification (AVC) data to meet the requirements of traffic inputs for their M-E design implementation (Li et al., 2005). They developed special programs to process the WIM and AVC data for their purposes. Despite the small number of WIM stations available, a couple of states employed a rational cluster approach along with regression analysis in order to identify individual clusters with respect to monthly distribution factors, hourly distribution factors, vehicle classification distribution factors, and gross vehicle weight load spectra for Class 9 trucks (Lu and Harvey, 2009; Wang et al., 2011). *Member, Assistant Professor, Dept. of Railroad Facility Engineering, Korea National University of Transportation, Uiwang , Korea (Corresponding Author, j-oh@ut.ac.kr) **Research Scientist, Texas A&M Transportation Institute, College Station 77843, TX, USA ( l-walubita@tamu.edu) ***Flexible Pavement Design Engineer, Texas Department of Transportation, Austin TX, USA ( Joe.Leidy@txdot.gov) 2083

2 Jeongho Oh, Lubinda F. Walubita, and Joe Leidy Fig. 1. Distribution of Texas WIM Stations TxDOT has collected volume and weigh data from 30 Weighin-Motion (WIM) sites as illustrated in Fig. 1. WIM station locations cover highway functional classes from Interstate Highways (IH), to National Highways (US), State Highways (SH), and Farm to Market (FM) roadway systems. Among the 30 WIM stations, three WIM stations are included in the Long Term Pavement Performance (LTPP) program. The TxM-E flexible pavement design program aims to characterize traffic data based on Level 1 traffic input as follows: Truck traffic volume (base year information) Monthly adjustment factors, Axle load distribution factors, Axles per truck, and General traffic inputs Although extensive efforts are being made by TxDOT to collect traffic data as aforementioned, it is still necessary to systematically characterize traffic data statewide to implement the Texas M-E flexible pavement design program since it is not feasible to collect the requisite traffic data for every single roadway to allow calibration of flexible pavement M-E performance models in the long term. This paper presents the results of this study and recommended guidance for using the developed traffic and load spectra inputs. 2. State of the WIM System in Texas Several factors have increased the importance of collecting truck loading information to assess pavement damage and plan maintenance and rehabilitation activities. These include: legislative approval given on November 27, 2002 allowing Mexican trucks to travel on U.S. roads beyond border commercial zones (Texas DOT, 2003), and increases in energy exploration and production (wind, oil, gas) in various parts of the State. In line with this, TxDOT has made an effort to collect adequate truck loading information as funding allows. As of August 2002, TxDOT had 17 WIM data sites statewide that covered 68 lanes of travel for use in federal reporting and pavement and bridge design. Currently, a total of 30 WIM sites are active in Texas (Texas DOT, 2003). 3. Traffic Data for Development Statewide Axle Load Spectra TxDOT provided traffic data collected at 29 WIM stations (i.e., Stations numerically numbered 1 through to 29 in Fig. 1) collected from 2010 through 2012 for this study. The file format used is the vehicle Classification record (C-record), with extension CLA. Each vehicle classification file contains a 1-month record of hourly traffic volume by vehicle class. The other file format that is used to process axle load spectra is the vehicle Weight record (W-record), with extension WGT. This file contains a 1- month hourly record based GVW, number of axles, axle spacing, and axle weight. Researchers processed the data set for individual WIM sites to generate the following inputs: C-Record Vehicle Class Distribution (VCD): from Class 4 to 13 Percent trucks: dividing the sum of total vehicles counted (from Class 1 to 13) by the sum of heavy vehicles counted (from Class 4 to 13, which is illustrated in Fig. 2.) Average annual daily truck traffic W-Record Axle load spectra for each vehicle class Number of axles per truck It should be noted that the single axle category incorporates both the steering axle and single axles with dual wheels in the assessment of the axle load spectra and the number of axle per truck KSCE Journal of Civil Engineering

3 ESTABLISHMENT OF STATEWIDE AXLE LOAD SPECTRA DATA USING CLUSTER ANALYSIS Fig. 2. Commercial Vehicle Classification Schema (Hajek et al., 2005) 4. Traffic Volume Data Analysis and Results Prior to conducting the cluster analysis, the volume data was processed for each WIM site. Fig. 3 shows an example of VCD for different WIM sites. The distribution of Class 5 and 9 trucks was found to be predominant among vehicle types, which is analogous to the other State s experiences (Tran and Hall, 2007; Lu and Harvey, 2006; Li et al., 2005; Lu and Harvey, 2006; 2009). The trend of distribution seems to differ per highway functional class. As functional class goes down (from IH to FM), the proportion of Class 9 tends to decrease. This distinct trend in the VCD data motivated these researchers to conduct cluster analysis for all the WIM sites. With respect to seasonal variation of the VCD, the standard deviation (Stdev) and Coefficient of Variation (COV) were assessed as presented in Table 1. The averages of the COV for Class 5 and 9 were 10.6 and 7.4 percent, respectively. FM 2670 (Site# LTPP-0808) exhibited the Fig. 3. Example of Vehicle Class Distribution Table 1. Coefficient of Variation of TTC of Class 5 and 9 Trucks along with Percent of Truck and AADTT WIM ID Highway COV (%) ID C5 C9 % of Truck AADTT 142 US IH US IH IH US US IH US IH SH US US US IH SH IH US SH SH SH IH IH FM IH US IH FM FM Min Avg Max Vol. 19, No. 7 / November

4 Jeongho Oh, Lubinda F. Walubita, and Joe Leidy highest COV; which could be attributed to specific commercial activity predominantly associated with seasonal farm activities. However, these authors are of the opinion that the seasonal variation of the VCD is not significant enough to take into account for the cluster analysis. Table 1 also includes the percentage of trucks and Average Annual Daily Truck Traffic (AADTT), which are primary traffic inputs to the TxM-E flexible pavement design program. As theoretically expected, the highest AADTT comes from Interstate highways (e.g., IH 10) while the lowest AADTT is from Farm-market roads (e.g., FM 2670); see Table Traffic Weight Data Analysis and Results The authors processed traffic weight data to obtain axle load distribution factors for individual vehicle classes and the number of axle per truck based on data from 29 WIM stations. Axle load spectra consist of the histograms of axle load distributions for each of five axle types: steering, other single, tandem, tridem, and quad. For the quality check in processing the traffic weight data, the GVW of each individual vehicle was compared to the sum of identified axle weights. Fig. 4 shows an example of the axle load distribution of Class 9 trucks analyzed. It is interesting to note that the distribution of the steering and single axles is somewhat different. This difference is expected to have an influence in the assessment of pavement performance. When comparing against legal axle weight restrictions, a small portion of overweight axles were detected. According to the current U.S. and State traffic code, 89-kN single and 160-kN tandem axles are legally permitted; less than 5 percent overweight axles were counted for the data analyzed. Table 2 shows an example of the comparison of the number of axles per truck. Since the steering axle is regarded independently in this study unlike the MEPDG, it is necessary to combine two numbers (steering and single) to compare with the MEPDG default values. For the WIM station considered in this example (i.e., Station# 531), the numbers seems to be comparable except for Class 13. This discrepancy (i.e., Class 13) is one of the reasons and justifications why this study attempts to generate load spectra data for the TxM-E instead of just employing the MEPDG default values that are nationally based and may not be the case to certain specific roads in Texas. Fig. 4. Axle Load Distribution of Class 9 Truck at WIM Station 536: (a) Steering and Single Axle, (b) Tandem Axle Table 2. Comparison of the Number of Axles per Truck MEPDG Default WIM Station #531 on IH-35 Single Tandem Tridem Quad Steering Single Tandem Tridem Quad Class Class Class Class Class Class Class Class Class Class KSCE Journal of Civil Engineering

5 ESTABLISHMENT OF STATEWIDE AXLE LOAD SPECTRA DATA USING CLUSTER ANALYSIS 6. Cluster Analysis In reality, it is not feasible to collect WIM data to characterize axle load spectra data for every highway or project. Several studies have attempted to establish regional and statewide axle load spectra data using statistical methods such as cluster analysis (Tran and Hall, 2007; Lu and Harvey, 2006; Li et al., 2005). In this study, K-Means cluster analysis was applied to group the multivariate response into informative clusters. The K- Means clustering predefines the number of clusters (Hardle and Simar, 2003). Given a predefined cluster, K clusters are created by associating every observation with the nearest mean. The centroid of each of the K clusters then becomes the new mean, and the above steps are repeated until convergence has been reached (Hardle and Simar, 2003). These researchers first conducted cluster analysis on the VCD. A number of clusters from 2 to 7 was tested, and the acceded number of clusters was determined when the Mean Square Error (MSE) was minimized. For this analysis, the MSE of Class 5 and 9 trucks was the primary focus since these two truck classes are the most representative regardless of the functional class of highway in Texas. As shown in Fig. 5, the number of clusters in terms of VCD was determined to be six. When six clusters are considered, increasing the number of clusters to seven does not contribute significantly to the reduction of the MSE. Fig. 6 shows the six VCD clusters, which are mainly affected by Class 5 and 9 truck distributions. The second cluster analysis was then carried out on the axle load distribution. For this clustering, only the Class 9 tandem axle load spectrum was considered in this study since it is the most predominant truck class that influences pavement performance in terms of weight and frequency. Previous studies also used Class 9 tandem axle load spectrum for their cluster analysis due to its effectiveness (Tran and Hall, 2007; Lu and Harvey, 2006). Fig. 7 shows the reduction in MSE with increasing the number of clusters; the optimum number of clusters was determined as six. Six clustered Class 9 tandem axle load spectra are shown in Fig. 8. To characterize the axle load spectra for any given highway from the clustered variables, it requires combining two clusters (VCD and Class 9 tandem axle load spectrum). Fig. 9 illustrates the interaction between two clusters. The interaction seems to have an irregular pattern between two clusters except Cluster 2 of VCD, which are mostly interacted with Cluster 2 of Class 9 tandem axle load spectrum. Based on this interaction, a guideline was proposed to establish statewide axle load spectra as presented in Table 3. Although there are two cluster groups (V4- L5 and V5-L6) which have only one site to represent the corresponding cluster, since the LTPP sites exhibit unique Fig. 5. MSE versus Number of VCD Clusters Fig. 7. MSE versus Number of Class 9 Tandem Axle Load Spectrum Clusters Fig. 6. Comparison of VCD Clusters Fig. 8. Class 9 Tandem Axle Load Spectrum Clusters Established Vol. 19, No. 7 / November

6 Jeongho Oh, Lubinda F. Walubita, and Joe Leidy formulation was established based on the weighted average concept that accounts for the number of available WIM stations per each group identified: LS9=(L1+6*L2+L3)/8 (1) 7. Application and Verification Fig. 9. Interaction between Two Clustered Groups characteristics, it is of the opinion that the appropriateness of clusters can be permitted for the use at the current stage of study. Further analysis will be conducted along with additional WIM data to verify these clusters obtained. To use this proposed guideline, representative axle load spectra for truck Class 4 to 13 for the individual six clusters (L1 to L6) were established herein. Here is the example of application of this guideline. If a site is classified with respect to the VCD as cluster 2, the axle load spectra data of this site can then be generated based on the weighted average formula given in Table 3 (second row). The Based on the preceding cluster analysis discussion, the following steps can be used to generate the axle load spectrum of a given highway section. 1) Process traffic volume data to establish VCD. These researchers are making an effort to collect traffic volume data using a traffic tube system for 100 highway test sections that are to be used for TxM-E flexible pavement design calibration among others. 2) Determine the VCD cluster by taking the least absolute difference error between the measured and established VCDs. 3) Generate the axle load spectra based on the guideline given in Table 3. The proposed guideline and procedures were incorporated into a Microsoft Excel spreadsheet for implementation purpose. Using the spreadsheet, the verifications were made on two highway test sections where WIM data are available as shown in Figs. 10 and 11. The results indicated that the generated axle load spectra seem to provide a comparable trend against the WIM data. In addition, it should be noted that the measured VDC from the traffic tube system appears to provide reliable traffic counts compared to the WIM data, which is promising for the characterization of axle load spectra. Figure 12 exhibits the comparison of prediction of rutting and fatigue cracking using the ME PDG program. In this example, a Table 3. Recommendations to Characterize Statewide Axle Load Spectra *The first denotes the VCD cluster and the last one denotes the Class 9 tandem axle load spectrum cluster KSCE Journal of Civil Engineering

7 ESTABLISHMENT OF STATEWIDE AXLE LOAD SPECTRA DATA USING CLUSTER ANALYSIS Fig. 10. Verification of SH 114 Test Section: (a) Comparison of VCD, (b) Comparison of Axle Load Spectra highway test section along US 59 (Atlanta District) was employed since most of input data for MEPDG execution were available including material, climate, traffic, and construction, with the exception of axle load spectra data. Results indicated that the predictions based on the default axle load spectra data tend to overestimate performance when compared to the case of using the cluster-analysis established axle load spectra as provided from this study. Fig. 13 shows the comparison of class 9 tandem axle load spectrum obtained from this study and MEPDG default value. The estimated axle load spectra tend to give higher axle load spectra 133 and 142 kn, but generally the MEPDG default value yields higher values. It should be also noted that the portion of high axle weight greater than 151 kn, Fig. 12. Application of US 59 Test Section: (a) Comparison of Rutting, (b) Comparison of Fatigue Cracking which is a legal limit of tandem axle, seems to be greater in the MEPDG default value. The researchers are of the opinion that further verifications and refinement is needed along with field performance data validation. 8. Conclusions The truck traffic characteristics relevant to the TxM-E flexible pavement design program were analyzed based on a large WIM database. Cluster analysis was employed to group WIM sites in terms of the VCD and Class 9 tandem axle load spectrum, and to consequently propose a guideline/procedure to generate axle Fig. 11. Verification of IH 35 Test Section: (a) Comparison of VCD, (b) Comparison of Axle Load Spectra Vol. 19, No. 7 / November

8 Jeongho Oh, Lubinda F. Walubita, and Joe Leidy load spectra data for use in the TxM-E flexible pavement design program. Some of the main findings of this study can be summarized as follows: 1. Truck Classes 5 and 9 represent the predominant truck traffic types. The change of truck traffic composition was found to be insignificant with seasonal variation. 2. The increase of Class 9 trucks and AADTT is generally followed by the transition of functional class of roadway from FM to IH. 3. The distribution of the steering and single axle load spectrum is somewhat different, and it is expected to have an impact on the pavement evaluation/performance, an aspect that is not considered in the MEPDG. The TxM-E flexible pavement design program is being developed to take into account this aspect. 4. The number of axles per truck estimated from the WIM data generally shows good agreement with MEPDG default values. 5. K-Means cluster analysis was found to be an appropriate methodological approach to group WIM sites with respect to vehicle class distribution and Class 9 tandem axle load spectrum. 6. This study proposes a guideline and procedures to establish an axle load spectra of a given roadway for locations where VCD data is available. 7. The guideline is tentatively verified on two highway test sections (SH 114 and IH 35) where WIM data are available. The measurement of VCD using a traffic tube system employed in this study seems to be promising as a supplement to WIM data from limited locations. 8. Flexible performance predictions on one highway test section (US 59) using the MEPDG program with default axle load spectra versus the estimated values based on the recommended methodology exhibited a different trend in the development of rutting and fatigue cracking during service lives. Inevitably, this warrants the future need to comparatively verify these performance predictions with actual field performance data. Acknowledgements The authors greatly acknowledge the cooperation of Texas DOT s Transportation Planning and Programming Division (TPP) for providing valuable traffic data used in this study. Fig. 13. Application of US 59 Test Section: (a) Comparison of Rutting, (b) Comparison of Fatigue Cracking References American Association of State Highway and Transportation Officials (AASHTO) (1993). Guide for design of pavement structures, Washington, D.C. Applied Research Associates (ARA) (2004). Guide for mechanisticempirical design of new and rehabilitated pavement structures, NCHRP Report I-37A, TRB, National Research Council, Washington, D.C. Hajek, J. J., Selezneva, O. I., Mladenovic, G., and Jiang, Y. J. (2005). Estimating cumulative traffic loads, Volume II: Traffic data assessment and axle load projection for the sites with acceptable axle weight data, Final Report, FHWA-RD , Federal Highway Administration, McLean, VA. Hardle, W. and Simar, L. (2003). Applied multivariate statistical analysis, Springer-Verlag, New York. Li, S., Nantung, T., and Jiang, Y. (2005). Assessing issues, technologies, and data needs to meet traffic input requirements by mechanisticempirical pavement design guide. Transportation Research Record: Journal of the Transportation Research Board, No. 1917, Transportation Research Board of the National Academies, Washington, D.C., pp , DOI: / Lu, Q. and Harvey, J. T. (2006). Characterization of truck traffic in california for mechanistic-empirical design. Transportation Research Record: Journal of the Transportation Research Board, No. 1945, Transportation Research Board of the National Academies, Washington, D.C., pp , DOI: / Lu, Q. and Harvey, J. T. (2009). Estimation of truck traffic inputs for mechanistic-empirical pavement design in California. Transportation Research Record: Journal of the Transportation Research Board, No. 2095, Transportation Research Board of the National Academies, Washington, D.C., pp , DOI: / Texas Department of Transportation (2003). A strategic plan for weighin-motion compliance, Austin, TX. Tran, N. H. and Hall, K. D. (2007). Development and significance of statewide volume adjustment factors in mechanistic-empirical pavement design guide. Transportation Research Record: Journal of the Transportation Research Board, No. 2037, Transportation Research Board of the National Academies, Washington, D.C., pp , DOI: / Wang, K. C. P., Li, Q., Hall, K. D., Nguyen, V., and Xiao, D. X. (2011). Development of truck loading groups for the mechanistic-empirical pavement design guide. Journal of Transportation Engineering, Vol. 137, No. 12, pp , DOI: /(ASCE) TE KSCE Journal of Civil Engineering