High-Speed Weigh-in-Motion System Calibration Practices (2008)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

7A thorough review of the literature was undertaken. It was conducted utilizing the Transportation Research Informa- tion Services (TRIS), Research in Progress, and International Transport Research Documentation online databases. In addi- tion, a multitude of papers from past proceedings of the North American Travel Monitoring Exhibition and Conference and the International Conference on WIM were reviewed. Finally, information on current practices was obtained from existing department of transportation (DOT) documentation and by telephone interviews of state DOT staff. The review, which focused on WIM calibration methods and practices, as well as research advancements that will allow these methods and practices to evolve, covers the following: • Current standards related to WIM system calibration, • Historic WIM calibration practices in the United States, • Current WIM calibration practices in the United States, • WIM-related research in the United States, • European WIM calibration practices, • European WIM-related research and, • Discussion. CURRENT STANDARDS RELATED TO WEIGH-IN-MOTION SYSTEM CALIBRATION Weigh-in-Motion Calibration Standard (ASTM E1318-02) The ASTM E1318-02 standard (6) describes test methods for evaluating and calibrating WIM systems using test vehicles of known static weights and dimensions. WIM system eval- uation encompasses on-site activities for ascertaining com- pliance of WIM system measurements to error tolerances. Calibration involves determining “factors that will be sub- sequently applied within WIM system calculations to corre- late the observed vehicle speed and tire-force signals with the corresponding tire-load and axle-spacing values for the static vehicle.” Both evaluation and calibration require two test trucks of known static weights and dimensions making multi- ple runs over the WIM system sensors at prescribed speeds in each lane. This standard allows the user to modify WIM system performance requirements through the equipment procurement process. The following four generic types of WIM systems are dis- tinguished in the ASTM E1318-02 standard on the basis of operational and performance requirements: • Type I has the ability to collect each axle’s left and right load data at vehicle speeds ranging from 16 to 130 km/h (10 to 80 mph). • Type II has the ability to collect individual axle load data at vehicle speeds ranging from 24 to 130 km/h (15 to 80 mph). • Type III has a load enforcement screening or sorting func- tion and operates at vehicle speeds from 16 to 130 km/h (10 to 80 mph). They are installed on the approaches to truck inspection stations, either on freeway lanes or ramps, to identify trucks that are likely to be over the legal load limits and need to be weighed statically. It is noted that before the E1318-00 version (i.e., E1318-94) the standard described the speed range of Type II systems from 24 to 80 km/h (15 to 50 mph), thus limiting their use to in-station ramp sorting. • Type IV, which is intended for load enforcement at vehicle speeds up to 16 km/h (10 mph) is not yet used in the United States. WIM accuracy is evaluated using a minimum of two test trucks, one each of FHWA Classes 5 and 9, performing several runs over the system at each of three vehicle speeds, (i.e., min- imum and maximum operating speeds at a site and an inter- mediate speed). These test vehicles “shall have a suspension type (leaf spring, air, other) that is deemed by the user to be representative of most vehicles of their type operating at the site.” The static axle loads of all these vehicles are established through static weighing using National Institute of Standards and Technology certified static scales (31). Axle spacings of the test trucks are to be measured at a resolution of 0.03 m (0.1 ft). Static weights should be measured a minimum of three times. Limits are set for the range in replicate axle weight mea- surements (e.g., static tandem-axle weight measurements must be within ±3% from the mean). The percent error in indi- vidual measurements, e, is defined with reference to the static measurements using the following equation: where WIM and static are the measurements obtained with the WIM system and the static scale, respectively. Calibra- tion consists of adjusting the WIM output to achieve a zero mean for the errors. The standard does not specify the actual measurement element(s) to be used for this computation. WIM accuracy is defined in terms of the probability that individual e = −WIM static static 100 1( ) CHAPTER TWO LITERATURE REVIEW

axle load measurement errors are within prescribed limits, as shown in Table 1. Each WIM type is to meet the specified load tolerances, provided that the pavement at the WIM site satisfies certain smoothness requirements. The latter estab- lishes the essential obligation of the customer in supplying a site that will allow the manufacturer/vendor to install a sys- tem that can meet the prescribed tolerances. Smoothness is specified for a length of 60 m (200 ft) upstream from the WIM sensors and a length of 30 m (100 ft) downstream of them. For a new installation, smoothness is measured using a 6.1-m (20-ft) long straightedge and a 0.15 m (0.5 ft) diameter, 3-mm (0.001-ft) thick circular plate. The pavement passes the smoothness requirement (i.e., meets the on-site acceptance requirements) if the plate does not fit under the straightedge, when positioned along the pavement between the edges of the lane, as described in Table 2. Before calibration, the loca- tion and magnitude of pavement surface deviations from the smoothness requirement should be documented. After ini- tial calibration, “alternative means of measuring the surface smoothness of the paved roadway . . . may be used to avoid closing the traffic lane. Data from suitable inertial profiling instruments analyzed by means of computer simulation of the 20-ft (6-m) straightedge and circular plate is suggested.” 8 The standard calls for calibration immediately following an initial WIM system installation and subsequent routine recalibrations at least annually. In addition, recalibration is recommended when • A system is reinstalled; • There are significant changes in system components, including software and their settings; • There is significant change in site conditions; or • There are significant changes in traffic data patterns. It is noted that this standard does not prescribe pavement stiffness or pavement material type requirements, although it states that “[e]xperience has indicated that a portland cement concrete (also called rigid) pavement structure generally retains its surface smoothness over a longer period of time than a bituminous (also called flexible) pavement structure under heavy traffic at a WIM site.” A WIM installation is deemed acceptable if it yields measurement errors within the prescribed tolerances for the particular WIM type. However, “the user can require quality of performance only in proportion to the quality of the site conditions provided.” In addition, all specified data- collection and data-processing features of a system should be demonstrated to function properly before it is accepted. A detailed outline of the main features of the ASTM E1318-02 standard, as applicable primarily to Type I and II WIM sys- tems, is given in Appendix A. When testing a new type of WIM system, its capabilities are ascertained through type approval testing. This requires testing in addition to that described previously, involving at least 51 traffic stream vehicles of known static weight and Tolerance for 95% Probability of Conformity Element Type I Type II Type III Type IV Wheel Load +25% —* +20% 2300 kg** 5,000 lb +100 kg +300 lb Axle Load +20% +30% +15% 5400 kg 12,000 lb +200 kg +500 lb Axle-Group Load +15% +20% +10% 11300 kg 25,000 lb +500 kg +1,200 lb GVW +10% +15% +6% 27200 kg 60,000 lb +1100 kg +2,500 lb Vehicle Speed +2 km/h Axle Spacing +0.15 m *Type II systems do not weigh individual wheels. **Lighter masses and associated loads are of no interest in enforcement. TABLE 1 WIM SYSTEM ACCURACY TOLERANCES PER ASTM E1318-02 STANDARD (6) Lane Edge Longitudinal Distance from Center of Sensors, m (ft) Right 6, 9, 13, 18, 23, 28, 33, 38, 43, 48, 53, 58, 62 (20, 30, 44, 60, 76, 92, 108, 124, 140, 156, 172, 188, 204) Left 6, 11, 16, 21, 26, 30, 35, 40, 45, 50, 55, 60, 65 (20, 36, 52, 68, 84, 100, 116, 132, 148, 164, 180, 196, 212) TABLE 2 PRESCRIBING STRAIGHTEDGE POSITIONING IN DEFINING WIM SITE SMOOTHNESS (6)

9dimensions (i.e., this number is the result of selecting a given number of vehicles from each vehicle class). For these tests, pavement smoothness needs to be exceptional. This allows establishing the capabilities of a system in terms of the toler- ance requirements listed in Table 1. Smoothness of Pavements at Approaches to Weigh-In-Motion Scales (AASHTO MP 14-05) AASHTO, in consultation with LTPP, currently is consid- ering adopting a provisional standard known as MP 14-05 for quantifying the pavement smoothness requirements at a WIM site (8). This standard is based on simulating WIM measurements as the dynamic axle load estimates of a fleet of 3S2 trucks obtained from a two-degree of freedom (bounce and pitch) vehicle simulation model over a number of repre- sentative pavement profile sections (9,29). A variety of pave- ment smoothness indices were considered and evaluated on the basis of their correlation with the 95th percentile tandem axle WIM error computed according to Eq. 1. Two pavement smoothness indices were identified, referred to as the short- range index (SRI) and the long-range index (LRI). They are computed on two segments of the pavement profile, one from −2.8 m to +0.5 m (−9.16 to 1.64 ft) and the other from −25.8 m to +3.2 m (−84.6 to 10.49 ft), where the minus/plus signs sig- nify locations upstream and downstream from the middle of the WIM sensor(s). Butterworth filters are applied to the pave- ment profile in these two segments to eliminate wavelengths outside the range of 1.6 m/cycle to 16.5 m/cycle (5.25 to 54.13 ft/cycle) and 1.1 m/cycle to 11.4 m/cycle (3.6 to 37.4 ft/cycle), respectively. The two resulting filtered profiles are summarized in terms of their average rectified (AR) veloc- ity (m/km) using the following: where Fi is the elevation at profile location i after Butterworth filtering, Wi is a weighing function (i.e., selected as equal to 1.0 for all locations), and N1 and N2 are the profile location limits identifying the profile range selected. The Butterworth filtering and the AR computations are performed in the dis- tance domain using a state-transition algorithm. AR N N FWi i i N N = − + = ∑1 1 22 1 1 2 ( ) In addition, a Peak SRI was defined to account for the potential localized roughness created by the installation of the WIM sensors themselves. The rationale is that this local- ized roughness needs to be considered, although it does not affect the overall SRI for the site. The Peak SRI is defined as the maximum SRI value for a distance ranging from −2.45 m to 1.5 m (−8 to 4.92 ft). These algorithms were implemented into a nonproprietary software package available though the LTPP product deliv- ery team. Thresholds for these three indices were established through a parametric study using the error tolerances for WIM Types I and II (Table 1) as a guideline. The results are given in Table 3 for Type I and II WIM systems. Sites with pave- ment roughness below the lower threshold are “very likely to produce an acceptable level of weighing error,” whereas sites with pavement roughness above the upper threshold are “very likely to produce unacceptable levels of weighing error.” It is recommended that a WIM site should be located on a pave- ment for which roughness is measured below the lower thresh- olds given in Table 3 over a length of at least 30 m (98.4 ft) upstream of the WIM sensors. European Weigh-In-Motion Specification Europeans developed WIM specifications as a result of a multi-year study that culminated in the European COST 323 report published in 2002 (3). A summary of the standard is given in Appendix I of the COST 323 report (7). It distin- guishes six main WIM Classes, designated as A, B+, B, etc., based of the width of the confidence interval of error toler- ated for each of the elements being measured, as shown in Table 4. Note that the designations A, B+, B, etc., are typi- cally followed by the confidence interval width for the GVW measurements in parentheses [e.g., A(5), B+(7), and so on]. Additional classes, for example, E, can be distinguished by extrapolating the requirements for the main classes. The vehi- cle speed requirements in Table 4 are not mandatory, although it is recognized that speed is used as an input to the load com- putation algorithms for some sensor types. The confidence level for the specified intervals depends on the type of reference loads and the methodology used for Lower Threshold m/km (in./mi) Upper Threshold m/km (in./mi) LRI 0.5 (31.68) 2.1 (133.06) SRI 0.5 (31.68) 2.1 (133.06) Type I WIM Peak short range 0.75 (47.52) 2.9 (183.744) LRI 0.9 (57.02) 3.8 (240.77) SRI 1.25 (79.2) 5.7 (361.2) Type II WIM Peak short range 1.6 (101.38) 6.6 (418.18) TABLE 3 ROUGHNESS INDEX THRESHOLDS FOR WIM SITES (8)

WIM system evaluation. Using trucks of known static weight allows defining four “repeatability levels” as follows: • r1: One test truck passing several times over the WIM system at the same speed and carrying the same weight. • r2: One test tuck passing several times over the WIM system at several speeds and loaded in several differ- ent ways. • R1: A small sample of test trucks, ranging from 2 to 10, making several passes over the WIM system at several speeds and loaded in different ways. • R2: A large sample (i.e., 10 to 200) of traffic stream trucks of known weight passing over the WIM system. Furthermore, three domains of “environmental repro- ducibility” are distinguished, depending on the time period over which the test truck WIM system evaluations take place. The rationale is that the properties of the pavement layers, the smoothness, and the sensor constants may exhibit temporal variations. These three domains are as follows: I: The test period is limited to a couple of hours within a day or spread over a couple of consecutive days. II: The test period extends at least over a full week or several days spread over a month. III: The test period extends over at least a year with actual test days distributed throughout this period. Confidence levels are defined as a function of repeatabil- ity level, environmental reproducibility, and sample size, as shown in Table 5 for environmental reproducibility levels I, II, and III. As an example, to classify a WIM system as Type B+, the accuracy requirement in weighing individual axles should be within ±11% of their corresponding static values. If 20 axles were used for this evaluation that took place during a single day (i.e., environmental reproducibility of I) and involved a single truck making repetitive runs at a given speed (i.e., repeatability level r1), the probability of conformity should be 97.2% per Table 5; that is, only 2.8% of the errors can be outside the ±11% range. 10 The WIM site conditions necessary for achieving these tol- erances are prescribed in terms of rutting, deflection (absolute and differential between left and right wheel paths), and smoothness [International Roughness Index (IRI) and Ana- lyzer of Longitudinal Profile], a pavement smoothness scale not used in the United States]. Limits established in each of these conditions define three WIM site classes (excellent, good, and acceptable), as shown in Table 6. It should be noted that the deflection limits apply to asphalt concrete- surfaced pavements only. Table 7 shows the relationship between site condition and its sufficiency in meeting the accuracy requirements (e.g., a site classified as Type II is insufficient to produce Class A and B+ tolerances). In addition to test trucks, the COST 323 standard provides for the use of alternative reference loads for evaluating and validating and calibrating WIM systems. These include the following: • Stationary loads being applied to the sensors. This approach may be feasible for some WIM sites, but clearly does not take into account the pavement roughness- induced load excitation at a particular WIM site. • Impact loads, such as those that can be applied with a falling weight deflectometer (FWD). As for stationary loads, applying such loads for evaluation or calibration may not be compatible with the data acquisition systems of some WIM systems. • Test trucks instrumented to measure dynamic axle loads. This approach allows computing WIM errors with respect to the dynamic axle loads applied to the WIM sensor, as measured from the instrumentation on-board an instru- mented vehicle. Instrumentation may consist of specially designed hub-mounted transducers or axles strain gauged in bending or shear. A number of research studies have made use of this technology for estimating “true” WIM accuracy (10,11). The COST 323 standard recommends a number of alterna- tive WIM calibration methods, depending on the repeatability Accuracy Classes and Tolerances; Confidence Interval Width δ (%) Element A(5) B+(7) B(10) C(15) D+(20) D(25) E GVW 5 7 10 15 20 25 >25 Axle Group 7 10 13 18 23 28 >28 Single Axle 8 11 15 20 25 30 >30 Single Axle within a Group 10 14 20 25 30 35 >35 Speed 2 3 4 6 8 10 >10 Axle Spacing 2 3 4 6 8 10 >10 Vehicle Volume 1 1 1 3 4 5 >5 TABLE 4 WIM SYSTEM ACCURACY TOLERANCES AS PER COST 323 STANDARD (7)

WIM Site Classes I Excellent II Good III Acceptable Rut Depth mm (in.) using a 3 m (9.84 ft) straightedge 4 (0.157) 7 (0.256) 10 (0.394) Semi-rigid pavement Mean Max Diff. L-R 15 (5.9) ±3 (1.18) 20 (7.87) ±5 (1.97) 30 (11.8) ±10 (3.94) All-bitumen pavement Mean Max Diff. L-R 20 (7.87) ±4 (1.57) 35 (13.78) ±8 (3.15) 50 (19.68) ±12 (4.72) Deflection 10-2 mm (10-3 in.) under quasi-static load of 13 t (28.6 kips) Flexible pavement Mean Max Diff. L-R 30 (11.8) ±7 (2.75) 50 (19.68) ±10 (3.94) 75 (29.52) ±15 (5.91) Semi-rigid pavement Mean Max Diff. L-R 10 (3.93) ±2 (0.78) 15 (5.91) ±4 (1.57) 20 (7.87) ±7 (2.75) All-bitumen pavement Mean Max Diff. L-R 15 (5.91) ±3 (1.18) 25 (9.84) ±6 (2.36) 35 (13.78) ±9 (3.54) Deflection 10-2 mm (10-3 in.) under dynamic load of 5 t (11.02 kips) Flexible pavement Mean Max Diff. L-R 20 (7.87) ±5 (1.97) 35 (13.78) ±7 (2.75) 55 (21.65) ±10 (3.94) IRI (m/km) (in/mi) 0–1.3 (0–82.4) 1.3–2.6 (82.4–164.7) 2.6–4.0 (164.7–253.4) Smoothness APL (SW, MW, LW) 9–10 7–8 5–6 TABLE 6 WIM SITE CONDITION SPECIFICATIONS AS PER COST 323 STANDARD (7) 11 Vehicle Sample Size Test Conditions 10 20 30 60 120 Environmental Reproducibility I r1 95 97.2 97.9 98.4 98.7 99.2 r2 90 94.1 95.3 96.4 97.1 98.2 R1 85 90.8 92.5 94.2 95.2 97.0 R2 80 87.4 89.6 91.8 93.1 95.4 Environmental Reproducibility II r1 93.3 96.2 97 97.8 98.2 98.9 r2 87.5 92.5 93.9 95.3 96.1 97.5 R1 81.9 88.7 90.7 92.7 93.9 96.0 R2 76.6 84.9 87.4 90.0 91.5 94.3 Environmental Reproducibility III r1 91.4 95.0 96.0 97.0 97.6 98.5 r2 84.7 90.7 92.4 94.1 95.1 96.8 R1 78.6 86.4 88.7 91.1 92.5 95.0 R2 73.0 82.3 85.1 88.1 89.8 93.1 Source: Reference 7. ∞ TABLE 5 CONFIDENCE LEVELS (%) FOR WIM EVALUATION

levels identified earlier and the intended application of the traffic load data being collected. These are as follows: 1. Setting the mean error of the GVW of all the test vehi- cles passing over the sensors to zero (i.e., each pass is considered a sample). This method is recommended for r1 level calibrations. 2. Setting the ratio of the sum of WIM GVW measure- ments divided by the sum of the static GVW of all the test vehicles passing over the sensors to one (i.e., each pass is considered a sample). This method is recom- mended only for estimating aggregate traffic loading data (e.g., for economic impact analyses). 3. Finding the slope of the WIM GVW measurements versus the static GVW measurements by performing zero-intercept regression analysis (i.e., each pass is considered an observation). This method minimizes the sum of squares of the errors. It is recommended for r2 and R2 level calibrations. 4. Using the method described in 1 or 2 above, but devel- oping calibration factors by truck class. This method is deemed applicable to R1 and R2 level calibrations. Traffic Monitoring Guide Appendix 5-A of the Traffic Monitoring Guide (TMG) (12) provides qualitative guidelines enhancing the procedures described by the ASTM E1318-02 WIM calibration stan- dard. These include the following: • Use of more than two test vehicles for the on-site evaluation, • Testing of WIM performance at various speeds, and • Testing performance under different environmental con- ditions (i.e., usually different temperatures). These enhancements will improve WIM calibration depend- ing on the following: • The specific scale technology being used. • The types of environmental conditions present at the WIM site. 12 • The type and structural capacity of the pavement, where the WIM sensors are installed. It is argued that using more than a few test trucks for calibration limits the bias created from the use of a partic- ular truck type. A common method for evaluating the extent of bias after calibration is to examine traffic stream summary weight data output by the scale and compare it with the known weights of trucks commonly found on the road. The TMG also advocates WIM system calibration moni- toring using traffic stream data QC principles. These consist of monitoring several traffic stream properties over time, including the following: • Front axle weights (FAWs) of five-axle tractor semi- trailer trucks; • GVW distribution of five-axle tractor semi-trailer trucks, following the technique developed by Dahlin (13); • Spacing of tandem axles on five-axle tractor semi-trailer trucks; and • Traffic volumes and distribution by classification. 1998 Long Term Pavement Performance Protocol The LTPP protocol (14) describes a method for calibrating WIM systems located in the vicinity of truck inspection sta- tions equipped with static weight scales. It is based on direct comparisons of WIM measurements to the static axle load of a large number of traffic stream trucks (i.e., at least 150) of known static weight. Where this is not possible, the LTPP protocol provides for WIM system calibration using test trucks. The recommended methods involve a number of refinements to the ASTM E1318-02 standard (6) as follows: • One of the two test trucks must be a five-axle semi- trailer (i.e., 3S2) and, preferably, be equipped with an air suspension. • The other test truck must have a different configuration or at least a different suspension type. • Three- or four-axle single-unit trucks should not be used. • The test trucks must have tires with conventional high- way tread patterns. • The 3S2 vehicle must have a GVW between 320 kN and 356 kN (72,000 lb and 80,000 lb). • A minimum of 40 passes must be made (20 for each vehicle) at highway speeds. This protocol also suggests monitoring WIM data cali- bration using the GVW distribution patterns of traffic stream vehicles. This consists of periodic comparisons of the fre- quency distribution of the GVW of 3S2 trucks to the frequency distribution established at a WIM site following test truck calibration. Significant differences between these frequency distribution patterns indicate a calibration drift. Figure 1 shows an example of such a drift indicated by the shifting to the right of the GVW peak weights of both empty and loaded trucks. – = insufficient; + = sufficient; (+) = more than sufficient. WIM Class A B+ B C D+ D I Excellent + + + (+) (+) (+) II Good + + – – (+) (+) III Acceptable + – – – + + TABLE 7 WIM SITE CLASSIFICATION AND CORRESPONDING WIM CLASSES (7)

13 16 14 12 10 8 6 4 2 0 18 26 34 42 50 58 Gross Vehicle Weight (Kips) 66 74 82 90 98 Expected Curve 1 Calibration Drift 1 Calibration drift using GVW for five-axle tractor semi-trailer trucks. Pe rc en ta ge o f T ru ck s FIGURE 1 Example of GVW frequency distribution changing patterns (14). FIGURE 2 Example of GVW WIM measurement errors versus test speed. 15.0 Out of Bounds Truck #1 Site #046 – Galt Tue Feb 15, 1994 *** TRUCK, CLASS 9, LANE 1 *** (#FILES = N/A) %ERROR VS. SPEED: GROSS WEIGHT 12.0 9.0 6.0 3.0 0.0 –3.0 % E rro r –6.0 –9.0 –12.0 –15.0 45 50 Speed, MPH : WSCALRPTS.DLL v1.3.1.0 55 60 65 HISTORIC WEIGH-IN-MOTION CALIBRATION PRACTICES IN THE UNITED STATES Information for this section is derived from the States’ Suc- cessful Practices Weigh-in-Motion Handbook (15) and reflects mid-1990s practices. California Department of Transportation Practice The California Department of Transportation (Caltrans) uses a variation of the ASTM E1318-02 standard for calibrating its WIM systems. The major departure from this standard is performing test truck passes at speed increments of 8 km/h (5 mph), plotting the resulting error in GVW measurements versus vehicle speed, and determining calibration factors for up to three distinct user defined vehicle speed “points.” An example of such a plot is Figure 2, which shows a test truck’s GVW WIM errors before calibration. Three speed points are selected, such as speeds of 72 km/h (45 mph), 88 km/h (55 mph), and 104 km/h (65 mph). Two straight lines are subjectively fitted, one providing best fit of the WIM errors between 72 km/h (45 mph) and 88 km/h (55 mph) and the other providing best fit of the WIM errors between 88 km/h (55 mph) and 104 km/h (65 mph).

The slope of these lines provides calibration factors for these two speed ranges. For the Figure 2 example, and given typical selected calibration factor speed points of 72 km/h (45 mph), 88 km/h (55 mph), and 104 km/h (65 mph), the resulting adjustment factors are as follows: • Increase the calibration by 9% for the 72 km/h (45 mph) speed, • No calibration adjustment for the 88 km/h (55 mph) speed, and • Decrease the calibration factor by 6% for the 104 km/h (65 mph) speed. For intermediate speeds in each speed range, WIM mea- surements are obtained through linear interpolation. After mak- ing these calibration factor adjustments, another set of test truck runs is performed to verify that the adjustments have resulted in GVW WIM errors closely concentrated around the 0% error axis. Caltrans field experience has shown that this approach results in lower traffic stream vehicle GVW WIM errors com- pared with a single calibration factor for all vehicle speeds. Caltrans use a traffic data QC approach to monitor WIM system performance over time. The data QC approach involves four parts as follow: 1. Knowledge of site characteristics, 2. “Real time” review, 3. First level data review and, 4. Second level data review. The first three steps ensure that a system is operating properly. The final step, second level data review, determines if cali- bration factor adjustments are needed and/or identifies equip- ment idiosyncrasies or subtle malfunctions not disclosed in the first level data review. Known operating characteristics of three traffic stream vehicle types can be used; namely, three- axle tractors with two-axle semi-trailers (i.e., 3S2s), three-axle single-unit trucks with two-axle full trailers, and two-axle trac- tors with one-axle semi-trailers and two-axle full trailers. This second level data review consists of the following two steps: 1. Evaluation of GVW data to obtain • Lane by lane comparisons and • Load frequency distribution patterns focusing on the location of the two peaks corresponding to loaded and empty trucks. 2. Evaluation of • Individual left and right weigh-pad output for both consistency and relative magnitude, • Effects of speed on the WIM weight output, and • Axle spacing and vehicle length accuracies. Certain key statistical elements from each review are entered onto a log sheet that is used for the following: • Documenting the effect of calibration factor changes on the WIM data for weight, axle spacings, and vehicle length; 14 • Establishing weight trends to determine seasonal varia- tions and truck operating characteristic variations; and • Determining whether or not calibration “drifts” over time. This second level data review was initially performed using an off-the-shelf database program. Subsequently, a program was written in Basic to perform these functions. Finally, a much more comprehensive software program called the CTWIM Suite was developed in-house (16). It can be downloaded from www.dot.ca.gov/hq/traffops/trucks/ datawim/install.htm. It accepts as input WIM data in ASCII, formatted according to Caltrans specifications. Although CTWIM was designed for WIM systems with dual weigh- pads (e.g., bending plates), it can be used for QC purposes with data from other types of WIM systems. Minnesota Department of Transportation Practice The Minnesota DOT used a test truck for initial WIM system calibration and autocalibration to maintain their calibration over time. The latter was used primarily for its bending plate systems, as described in States’ Successful Practices Weigh- in-Motion Handbook (15). Initial calibration is carried out using a Class 9 test truck. Once concluded successfully, traf- fic stream data on Class 9 vehicles is monitored over a week- long period to verify that the GVW peaks occur at the fol- lowing reasonable locations: • Loaded peak between 329 and 347 kN (74 and 78 kips) and • Unloaded peak between 125 and 133 kN (28 and 30 kips). If these data are deemed satisfactory, they form the basis for the autocalibration, which is based on the FAWs of five-axle semi-trailer trucks (i.e., Class 9). Site-specific reference values for FAW weights are established for three distinct GVW groups of Class 9 vehicles. An example of such values is given in Table 8. The autocalibration routine determines the differences between the average FAW WIM load for each GVW class and calculates adjustment factors. Weights are assigned to these adjustment factors as a function of the vehicle sample size used for computing them. These range from 0.2 for a sin- gle sample to 0.95 for 100 or more samples. An example of this weighing procedure is shown in Table 9 (i.e., the adjust- ment factor of +4.23% was computed by multiplying 4.7%, the deviation from reference, by 0.90, which is the weight factor that corresponds to a sample size of 59). CURRENT WEIGH-IN-MOTION CALIBRATION PRACTICES IN THE UNITED STATES The main source of information for this section is recent national and state documentation of WIM calibration proce- dures, as well as personal telephone interviews of a number of state DOTs that have lengthy experience in managing

15 GVW kN (kips) <142 (32) 142–311 (32–70) >311 kN (70) Percent Deviation Allowed from Reference Weight Minimum Number of Monitoring Hours Minimum Sample Size of Class 9 Vehicles Reference FAW kN (kips) 37.8 (8.5) 41.4 (9.3) 46.3 (10.4) 3.5% 48 250 TABLE 8 EXAMPLE OF AUTOCALIBRATION REFERENCE VALUES FROM MINNESOTA DOT (15) GVW kN (kips) Sample Size Deviation from Reference Adjustment Factor Weight Adjustment Factor Calibration Adjustment <142 (32) 59 +4.7% 0.90 +4.23% 0.958 142–311 (32–70) 112 +4.3% 0.95 +4.09% 0.959 >311 kN (70) 79 +4.8% 0.90 +4.32% 0.957 Average Adjustment Factors — — — 0.958 Source: Reference 16. TABLE 9 EXAMPLE OF MINNESOTA DOT’S WIM AUTOCALIBRATION CALCULATIONS (16) WIM systems. Its focus is on procedures that significantly differ from those of the ASTM E1318-02 standard. Long Term Pavement Performance Specific Pavement Study Traffic Data Collection Pool Fund Study The objective of this pool fund study is to determine the acceptability of the data being generated by existing WIM systems, to accelerate the installation or replacement of new systems, and to ensure the quality of the traffic data being produced for the LTPP experiments known as Specific Pave- ment Study (SPS) 1, 2, 5, 6, and 8 (17). The study is being carried out in two phases by two different contractors. The Phase I contractor is responsible for the evaluation of any existing or new systems installed by a state using its own procurement procedures, as well as newly installed systems under the pooled fund study. The Phase II contractor is respon- sible for the procurement, installation, and maintenance of the new pooled fund study systems. It is anticipated that ultimately this study will result in 9 existing and new state-installed WIM systems and 18 new Phase II contractor-installed systems. The ASTM E1318-02 standard’s functional performance requirements for Type I systems is being used for the purposes of this study. Certain other provisions of ASTM E1318-02 specifications are also used, enhanced by the provisions of the LTPP protocol as described earlier under the 1998 Long Term Pavement Performance Protocol (14). Additional refine- ments to these procedures were made by the TRB expert task group on LTPP traffic data collection and analysis. These refinements are as follows: • A minimum of two test trucks are used, typically one 3S2 with a GVW between 338 and 356 kN (76,000 and 80,000 lb) and another 3S2 with a GVW between 267 and 285 kN (60,000 and 64,000 lb). The heaviest of the two trucks must have air suspensions for both tractor and trailer tandems. • The ASTM E1318-2 tolerance requirements must be met not only by the entire dataset, but also by subsets of this dataset grouped by speed and ambient temperature. • Static weighing of the test trucks is conducted before and after each set of test truck runs, and the axle weights and the GVWs are averaged. Evaluation takes place in two stages. In the first stage, each test truck performs a minimum of 20 runs over the WIM system. If a change in calibration constants is deemed neces- sary, the Phase I contractor in conjunction with either the state agency or the Phase II contractor, determines the cali- bration factor adjustments. Regardless of whether calibration factor adjustments were made, a second stage of testing is undertaken involving another 20 runs of each of the two test trucks to verify the calibration of the WIM system. For WIM systems installed by the Phase II contractor, this con- tractor is responsible for the initial evaluation/calibration. If the performance standards are met, the system is deemed acceptable and it is turned over to the Phase I contractor for independent verification. If performance requirements are not met, the Phase II contractor is given the opportunity to make adjustments to the calibration factors and repeat the entire test truck evaluation process once again. If this new set of runs is unsuccessful, the Phase II contractor must make any sensor, hardware, or software replacement deemed necessary

to correct the problem before performing a new performance evaluation. Once a WIM system is installed through this process and is operational, its performance is monitored by the Phase II con- tractor using a two-level review process. The first level review consists of weekly reviews of the daily data files to check a system for proper operation and to determine the validity of the data. The second level review consists of a monthly data analysis of a sample (typically, a minimum of seven consecu- tive days) of 3S2 trucks in the traffic stream to monitor each system’s maintenance of calibration and to identify any subtle operational problems. Several of the 3S2 truck properties are tracked and monitored, as shown in Table 10. In addition to the 3S2 properties noted previously, the fol- lowing properties are also included in the checks: • GVW frequency distributions regardless of speed and • GVW distributions by speed. These properties are tracked and monitored by means of graphs and tabular distributions. The noted properties for each month’s check are compared with reference properties established from traffic stream data collected over a period of one week shortly after a success- ful evaluation/calibration of a system using test trucks as described earlier. For the monthly sampling to be used for 16 calibration monitoring, it is important that the sampling consist of only data truly representative of the site’s typical truck oper- ating characteristics. Data from periods of time for which atyp- ical conditions existed (road construction, holidays, extreme weather, etc.) should not be used. It is also noted that the cali- bration monitoring procedure described earlier is not an LTPP requirement. It is a procedure developed by the Phase II con- tractor to ensure the data quality required by the contract. NCHRP Report 509 NCHRP Report 509: Equipment for Collecting Traffic Load Data (18) documents equipment and procedures for collecting traffic load data. It describes the on-site calibration approach using two test trucks standardized by ASTM E1318-02. Once calibration is verified, it is recommended that the traffic pat- terns established are used for monitoring WIM system cali- bration over time. This WIM data QC may involve the fol- lowing data elements: • Distribution of loaded and unloaded GVW peaks for the FHWA Class 9 trucks (or, potentially, other vehicle classes); • Consistency of mean FAW for loaded Class 9 trucks; • Consistency of percentage of weekday Class 9 trucks; • Changes in the percentage of unclassified vehicles; • Increases in equipment’s counting errors; • Consistency in load-relative magnitudes between right and left wheel path weighing sensors; AX1 RT, AX1 LT = Right wheel path and left wheel path loads of Axle 1 (1,000 lb or kips), resp. AXLE 1 = Axle load (1,000 lb or kips). AX SP = Axle spacing (ft). OAL = Over axle load limit (percent). UNEQ DET = System error messages indicating right and left weigh sensors did not detect same number of axles. TABLE 10 EXAMPLE OF MONITORING TRAFFIC STREAM PROPERTIES OF A WIM SYSTEM (17)

17 • Consistency of tractor drive tandem axle spacings; • Total number of vehicles within expected load ranges; • Changes in time-of-day traffic patterns; • Changes in hourly data volumes; and • WIM system error diagnostic messages. It is suggested that this WIM data QC process is essential and should not be replaced by autocalibration algorithms. The reason is that calibration drifting detected by data QC may be indicative of problems, such as pavement deterioration or sensor degradation, which cannot be rectified by simple calibration factor adjustments. Long Term Pavement Performance Weigh-In-Motion Data Quality Control Software This software was developed by LTPP (19). It implements a variety of traffic data QC tests, some of which relate to WIM data quality. This is conducted at two levels. The first level involves the raw Card-4 and Card-7 data obtained from the state WIM systems (i.e., they store individual vehicle clas- sification and weight data, respectively). These QC tests involve comparisons of daily summaries of the following: • Card-4 versus Card-7 data, • Traffic count data for selected state vehicle classes, and • GVW distribution data for selected state vehicle classes. At the conclusion of this QC level, feedback is obtained from state DOTs to decide whether particular discrepan- cies in traffic patters can be justified by known changes in local traffic. Data that have passed the first QC review level are converted into the FHWA vehicle classification scheme (i.e., 13 classes) and entered into the second level of QC pro- cessing. The data elements analyzed at this QC level depend on the type of experiment that generated them (e.g., GPS; SPS 1, 2, 5, and 6; and SPS-pooled fund study). In general, it includes analysis of data summaries over a user-selected time interval that could be monthly or annually and cover multiple year periods. The data elements analyzed graphi- cally include the following: • Vehicle counts by vehicle class, • Axle load distribution, • GVW distribution, • ESAL/vehicle, and • Error statistics. Clearly, these tests need to be performed by experienced users who must decide on the accuracy of the data on the basis of the software output graphs. Data that have successfully passed this QC process are labeled as “level E” and uploaded onto the Information Management System database. Florida Department of Transportation Practice The Florida DOT uses test trucks for WIM calibration. Before calibrating existing WIM systems, diagnostic checks are run to ensure that all the components for each WIM lane are oper- ating properly. At least one Class 9 test truck with air tractor and trailer suspensions is used. Test truck runs are conducted at preselected speeds at which calibration factors are to be established. WIM errors versus speed are plotted to establish calibration factors for each of these preselected speeds. In addition, the calibration of speed and axle spacing measure- ments is adjusted, if necessary. During this process, commu- nication with the test truck driver(s) is maintained by means of CB radio or cellular phone. Following this step, the test truck(s) are run over the WIM system a minimum of two times at each of the calibration speeds as well as once at 8 km/h (5 mph) increments between the high and the low preselected calibration speeds. To isolate the effects of speed and temperature during the calibration session, test runs are performed in a sequence of low to high speed, which is repeated in the same order. The WIM GVW percent error is plotted versus vehicle speed. This graph is analyzed to determine what adjustment needs to be made in the calibration factor for each preselected speed point to best reduce the overall GVW error to zero. If necessary, additional test truck runs are made to make final calibration factor adjustments. During initial calibration and verification, communications software is used to chronolog- ically record the initial calibration factors, the WIM data ele- ments for each test truck during each set of runs, any changes made to the calibration factors between test truck run ses- sions, the final calibration factors, etc. The calibration of new WIM systems is slightly different than the one described previously. The main differences are as follow: • No preliminary runs are made to adjust calibration factors. • Test truck(s) runs a minimum of three times at each calibration factor speed point and three times at each 8 km/h (5 mph) increment between the high and low pre-selected speeds. • The low to high speed incremental run sequence is per- formed three times. If, for any reason, it is necessary to make adjustments to the calibration factors for either speed/axle spacing or weight, then the accuracy testing truck run procedure must be con- ducted again. Not only must the data item accuracies meet the functional performance requirements of ASTM E1318-02, but the percent GVW error plots must be evenly distrib- uted around the zero error axis of the GVW versus vehicle speed plot. Following determination that the WIM system meets accuracy requirements, an analysis of the GVW ver- sus speed graph is made and, if deemed appropriate, the calibration factors are again adjusted. This procedure is per- formed for each individual lane instrumented with WIM sensors.

California Department of Transportation Practice Personal interviews with Caltrans personnel revealed that the methodology currently used for WIM calibration is essentially the same as the methodology followed by the Florida DOT. Texas Department of Transportation Practice The documentation submitted by the Texas DOT for calibrat- ing WIM systems using test trucks was very comprehensive. The basic steps involved are as follows: • A log file is initially created by the WIM system com- munication software to record all calibration events in chronological order. • A number of test truck runs is made to first calibrate for speed/axle spacing measurements before commencing runs for weight calibration. • The initial weight calibration factors are recorded for each speed. • A minimum of three test truck passes is made each at 80, 96, and 112 km/h (50, 60, and 70 mph). Where nec- essary, the test runs at 80 and 96 km/h (50 and 60 mph) may be combined, and the runs at 112 km/h (70 mph) may be conducted at the speed limit. • Calibration factors are computed for these distinct speeds using plots of GVW errors versus test speed. • All test data are documented and a copy is left inside the WIM system cabinet. • Finally, the log file is closed. Indiana Department of Transportation Practice The Indiana DOT uses a detailed WIM calibration procedure using test trucks. The main steps include the following: • First, it is verified that the WIM system components are working correctly. • The test truck drivers are briefed on the details of the test procedure to be used, including the sequence of speeds and lanes to be used as well as the signaling or cell phone/radio communication methods to be used. • Initial test runs are intended to verify/adjust the speed/ axle spacing calibration. • Subsequently, the overall test truck wheelbase is verified/ adjusted. • Then, the WIM system weight calibration by speed range is performed as follows: – Initial calibration factors are set using a single pass of a test truck; – The test truck WIM data are inserted into a calibration spreadsheet; – If the GVW WIM measurements are accurate, but the FAW WIM measurements are not, the Dynamic Compensation Factor of the WIM system is adjusted (the Dynamic Compensation Factor is a front-axle- specific calibration factor adjustment); 18 – The calibration process is concluded when both the GVW and the FAW WIM measurements collected for 10 consecutive test truck passes are within pre- scribed tolerances; and – Finally, the calibration factors are recorded and stored in a database. Montana Department of Transportation Practice The current Montana DOT WIM calibration practice applies to systems used for main-line weight enforcement screening for PrePass™ systems. WIM calibration is performed either quarterly to adjust calibration drift or as needed on the basis of the following weekly data: • PrePass™ system rates at which vehicles are being stopped for static weighing; • WIM data for performing QC (i.e., based on Class 9 vehicle GVW trends, changes in the number of unclas- sified vehicles, etc.); and • PrePass™ static axle weight measurements. For the latter, PrePass™ static axle load data are obtained for 25 Class 9 vehicles of each of the WIM lanes being mon- itored. These trucks exclude those that carry shifting cargo (e.g., liquid tankers and livestock haulers). WIM data are col- lected for the same trucks, which are identified manually on the WIM system printout to allow one-to-one comparison of static and WIM measurements. These reports are sent to the main office for analysis. To evaluate data and determine cal- ibration factor adjustments, the following procedure is used for each WIM lane: • The WIM data and the static weight data for each truck sampled are entered into a spreadsheet that calculates WIM error statistics. The tolerances are as follows: – Average GVW errors must be within 2% and – GVW error standard deviations must be lower than 6%. If the WIM GVW errors are within tolerances, there is no need for calibration. Otherwise, the following steps are performed: • A calibration factor correction is computed. • The new calibration factor is averaged with the old cal- ibration factor and the average is used to replace the old calibration factor by means of remote access (the reason for averaging is to effect small, rather than large, changes to system calibration). • The new system calibration is verified using the same procedure and another sample of 15 Class 9 traffic stream vehicles. Finally, if the WIM GVW error standard deviation is sig- nificantly higher than the specified tolerance, a visual inspec- tion of the site is done to ascertain any roadway or sensor prob- lems and, if deemed necessary, have these problems repaired.

19 New Jersey Department of Transportation Practice The New Jersey DOT uses a single 3S2 test truck equipped with air suspension tandems for WIM system calibration. The test truck is driven on each WIM lane until a minimum of five passes of consistent single-axle and GVW WIM mea- surements are obtained. The average of the measurements of these five test runs is used as the basis of calibration. The cal- ibration is validated by a minimum of another five consec- utive passes of a test truck and the average axle and GVW WIM measurements are computed. The verification is sat- isfactory if the WIM average values are within 10% of the sta- tic weights for axles and within 5% of the GVW static weight. If these tolerances are not met, the process is repeated. If two iterations of this process prove unsuccessful, corrective action is taken (i.e., repair or replacement of pavement and/or sensors) before repeating the calibration procedure. Utah Department of Transportation Practice Utah has 15 permanent WIM sites, which consist of 9 piezo- electric systems (traffic data collection) and 6 load cell systems [Port of Entry (POE) applications]. POE sites utilize load cell systems, and the WIM vendor performs the calibrations bi- annually. The procedure used is to compare the static weights from the POE’s static scale for particular trucks with their cor- responding WIM measurements. The traffic data collection sites use piezoelectric systems provided by two different vendors and calibrated by the Utah DOT. Currently, one vendor’s systems use autocalibration, but the other vendor’s systems do not (although the auto- calibration feature is available). As such, two different cali- bration procedures are used for calibrating these systems. Both procedures use prescribed Class 9 steer axle weight targets. The major difference is that for one vendor’s systems the calibration is automatically performed unattended by the systems, whereas for the other vendor’s systems, a biweekly manual procedure of processing downloaded data, determin- ing necessary factor adjustments based on steer axle weight criteria, and making the factor adjustments is used. It was recommended that the procedures for systems uti- lizing autocalibration be continued. For the systems currently not using the available autocalibration feature, it was recom- mended that repairs to temperature sensors and any other needed modifications be made to the systems such that their autocalibration features may be utilized. WEIGH-IN-MOTION-RELATED RESEARCH IN THE UNITED STATES NCHRP Project 3-39(02) On-Site Evaluation and Calibration of Weigh-in-Motion Sys- tems, the final report for NCHRP Project 3-39(02) (4), deals with the on-site evaluation and calibration of WIM systems. It examines the feasibility of two methods, one using a combi- nation of test trucks and vehicle simulation models, and the other using traffic stream vehicles of known static weight equipped with automatic vehicle identification (AVI) systems. The first method utilized a modified version of the vehicle simulation model VESYM (20), referred to as VESYMF, to estimate the dynamic axle loads exerted by test trucks at three WIM sites, each equipped with pressure cells, bending plates, and traditional piezoelectric sensors. At each site, the pavement roughness profile was measured using an inertial profilometer (30). A field experiment was conducted at each of these sites involving three test vehicles, namely a two- axle single-unit truck, a three-axle single-unit truck, and a 3S2 truck. Each truck performed 10 replicate runs at each of four speeds (i.e., 50, 70, 90, and 110 km/h [31, 43, 56, and 68 mph]). The WIM measurements were plotted for each axle as a function of speed, as shown in Figure 3. For a given speed, it was observed that individual axles produce very precise WIM measurements as a result of their repeatable dynamics (i.e., average coefficients of variation of 3.8%, 5.7%, and 3.8%, respectively). Efforts to predict the magni- tude of the dynamic load of individual axles over the WIM sensors using the vehicle simulation model were unsuccess- ful. Instead, the vehicle simulations were used for computing the distribution of axle load dynamics at the WIM site, given 3-axle truck; axle 2, static load=78.3 kN 0 20 40 60 80 100 0 20 40 60 80 100 120 SPEED km/h W IM L O A D k N FIGURE 3 Example of pressure cell WIM measurements of an axle versus speed (4).

the pavement profile. This approach provides a rigorous con- nection between pavement smoothness and WIM errors. This project also developed an automated method for com- paring WIM with static axle loads of traffic stream vehicles equipped with AVI. This method is based on static load data obtained for particular vehicles at truck inspection stations, which are subsequently identified by the AVI as they travel over WIM systems. The method was field-tested using the fixed AVI facilities of the Heavy Vehicle Electronic License Plate program on the I-5 corridor in Washington State, Oregon, and California. Software was developed automating the WIM calibration calculations. This method was shown to be feasible in calibrating WIM systems, where fixed AVI facilities are available. It was also shown to be feasible using transportable AVI equipment developed for this purpose. This equipment was field-tested at two WIM sites on I-94 in Minnesota. Multi-Sensor Weigh-In-Motion and Artificial Neural Networks A number of studies have used multi-sensor WIM (MS-WIM) configurations and focused on the use of statistics for improv- ing the resulting static load estimates (21). Recent work offers an alternative based on Artificial Neural Networks (ANN) (22). This approach involves measuring the WIM site pavement profile, using dynamic vehicle simulations (20) to model dynamic axle load and “train” the ANN algorithm to yield static axle loads. This approach was shown to significantly improve WIM accuracy, especially where pavement rough- ness was high. For example, WIM systems classified (see Table 4 for COST 323 WIM class designations) as B(10) by simply averaging the WIM reading of each of the multiple sensors, were improved to B+(7) and in some cases to A(5) by applying the ANN algorithm on the output of the multiple piezoelectric sensors. Indiana/Purdue Study on Weigh-In-Motion Data Quality Control Indiana DOT and FHWA cosponsored a study to explore innovative WIM QC methods (10). The motivation of this study, in addition to maintaining long-term data quality, was to more effectively manage WIM systems across the network. Formal QC techniques were followed, including: • Define, • Measure, • Analyze, • Improve, and • Control and Statistical Process Control. The following procedure describes how these techniques were applied in conducting WIM data QC. Two WIM system metrics are targeted, namely the axle spacing in drive tandem axles of Class 9 trucks and the dif- 20 ference between the left-hand-side and right-hand-side wheel loads of the steering axles of Class 9 trucks (referred to as “left-right residual”). Clearly, the latter is applicable only to Type I WIM systems. The statistical properties of the first metric were estab- lished from manufacturer data of Class 9 trucks in the United States. These data suggest four common axle tandem spac- ings, namely 130, 132, 137, and 149 cm (4.25, 4.33, 4.50, and 4.58 ft). These spacings are the result of standard drive axle suspension designs; they have been fairly constant in the past and are anticipated to remain relatively unchanged in the future. It was concluded that 99% of the population of axle spacings falls in the interval between 130 and 140 cm (4.25 and 4.58 ft) and have a weighted average of 132 cm (4.33 ft). As a result, the accuracy of WIM systems in mea- suring vehicle speed can be tracked by the spacing of the tandem axle of Class 9 truck tractors. The statistical properties of the second metric were estab- lished by extensive study of the configuration, mechanical arrangement, and dynamics of Class 9 truck tractors, as well as the result of the pavement cross-slope on the distribution of wheel loads between the left- and right-hand side. It was concluded that the latter differ less than 1% and hence can be used as a QC metric for WIM systems that separately mea- sure left and right wheel loads (i.e., Type I systems). The report presents the methodology used for down- loading WIM data and analyzing them according to the QC methodology described previously. The data are downloaded into a relational database, such as Microsoft SQL. Cube files were created using the online analytical processing tools from the Microsoft SQL analysis server. These files can be viewed as pivot tables and pivot charts by means of a Microsoft Excel connection to the analysis server database and can also be created offline for distribution to users who do not have online access to the database. These analysis cube files allow simple summary reporting while affording detailed data min- ing capabilities. The aforementioned analysis approach can be used to gen- erate accuracy graphs for spacing/speed and left/right weight of Class 9 trucks. These data are used to identify WIM sys- tems that are grossly off calibration and allow adjusting their calibration temporarily, until a proper calibration involving test trucks can be undertaken. The latter is warranted if the adjusted system generates stable data, which can be tracked through statistical process control charts (i.e., graphs indicat- ing the allowable confidence intervals for the properties being measured). Another innovation described in this study is the use of WIM systems as “virtual weigh stations,” which allow enforce- ment officers to view real-time truck records by means of a wireless link. Vehicles to be stopped for static weighing can be flagged automatically or this decision can be made by the

21 enforcement officers. Analysis determined that such virtual weigh stations are approximately 55 times more effective than static weigh stations alone in capturing overweight vehicles. Utah Department of Transportation/Brigham Young University Study on Weigh-In-Motion Calibration The Utah DOT sponsored a study on WIM calibration (23) intended to improve Utah’s commercial motor vehicle data collection program. To accomplish these objectives several tasks were undertaken, including the following: • Performance of a literature review to establish the state of the practice for commercial motor vehicle monitoring, • Collection of WIM data for the state of Utah, • Analysis of the WIM data collected, • Development of a calibration methodology for use in the state, and • Suggestions and conclusions based on this research. The literature review focused on: • WIM history, • Basic WIM concepts, • WIM technologies, • Weight data collection standards and their calibration, • Quality assurance methods, • TMG weight data collection, and • Discussion of the new AASHTO Pavement Design Guide. The literature review conducted for this study was extensive and each of the above listed topics is discussed in some detail. The report describes Utah’s 15 permanent WIM sites, which consist of 9 piezoelectric systems (traffic data collection) and 6 load cell systems (POE applications), and discusses the methods used to perform analyses of the data collected from these sites. Each site’s median and spread of the data, as well as the extent and nature of any departure from symmetry, were determined. The daily average steering axle weight for Class 9 vehicles was graphed for each site. Based on known static weight data, the average steering axle weight is expected to be approximately 49 kN (11,000 lb) with an acceptable site variance of ±20%. The drive tandem axle spacings for Class 9 vehicles were also found to be fairly constant, expected to average approximately 132 cm (4.33 ft), which provides a quality assurance measure for checking a WIM system’s speed and axle spacing outputs. The acceptable ranges for drive tandem spacing are ±15 cm (0.5 ft). The percent of overweight trucks (Class 5 and above) for each site was found to be another viable data analysis metric. A survey was conducted of 10 states with regard to system calibration procedures and methodologies. The report discusses the findings for each of the 10 states. The calibration and validation practices currently used in Utah were also reviewed. Recommended procedures were developed to bring Utah DOT practices in line with current WIM standards and practices of other states. Currently, there is no validation procedure for the piezo systems. It was recommended that a verification process be implemented and that it include runs with a single test truck (predominate class at site, typically Class 9) to verify the sys- tems’ autocalibrations on, as a minimum, an annual basis. The recommended verification procedure included the following steps to be taken before making test truck runs: • Evaluate the physical characteristics of the site, includ- ing pavement conditions 84 m (275 ft) before and 9 m (30 ft) after the WIM sensors. • Check the WIM system’s components for proper opera- tion and check for proper software settings (weigh sensor and inductive loop thresholds, etc.). • Use a radar gun to check speed calibration. The test truck procedure, as well as the verification of con- formance to performance requirements, is in general confor- mance to the ASTM E1318-02 standard for Type II systems, although there is some modification in the number of test truck runs and speeds. Quality assurance methods are recommended to assess the data quality and, in turn, WIM system performance. Included in such assessment is the creation of the following graphs: • Vehicle class histogram, • Daily average Class 9 steering axle weight, • Daily average Class 9 drive tandem axle spacing, and • Class 9 GVW histogram. The graphs are prepared on a quarterly basis. EUROPEAN WEIGH-IN-MOTION CALIBRATION PRACTICES The main source of information in this section is a recently completed report by the Commercial Motor Vehicle Size and Weight Enforcement scanning tour (24). It had two WIM related objectives, namely: • Emerging WIM technologies capable of accuracies that will allow their direct use in enforcement applications and • Novel uses or applications of WIM data to support pave- ment design, bridge/structural design, traffic engineer- ing, transportation planning efforts, or ongoing perfor- mance monitoring and evaluation of vehicle size and weight enforcement programs. The Commercial Motor Vehicle Size and Weight Enforce- ment scanning tour included six European countries (Belgium,

France, Germany, the Netherlands, Slovenia, and Switzerland). It produced a critical review of contemporary European prac- tices of WIM system utilization in enforcing commercial motor vehicle size and weight regulations. The majority of these systems use piezoelectric sensors. The basic WIM calibration standard used by these countries was developed by the COST 323 study (7). The actual implementation of this standard, how- ever, varies among countries. The highlights of the findings of this scanning tour are as follows: • France operates in excess of 170 WIM stations and relies largely on autocalibration techniques to keep them calibrated. • France and the Netherlands are developing MS-WIM systems capable of detecting axle weights at highway speeds with an accuracy rate that is sufficient for direct load enforcement. They anticipate that, with the appro- priate administrative changes, this goal will be reached within the next 10 to 15 years. • France has experimented with fiber-optic WIM sen- sors and has made progress in isolating the sensitivity of their output to steel in reinforced portland concrete pavements. Detailed documentation on the theoretical development and field performance of these sensors is provided in a Weigh-in-Motion of Axles and Vehicles for Europe (WAVE) report (25). • The Netherlands has developed an instrumented vehicle that allows WIM calibration using a reference dynamic rather than static axle loads. • Slovenia’s WIM calibration is conducted by either test trucks or traffic stream vehicles of known axle weights following COST 323 specifications. The pertinent docu- mentation and WIM data are available online (see http:// www.siwim.com/). • Switzerland reports WIM system maintenance prac- tices that involve annual inspections of sensors for wear and actual calibration according to the COST 323 stan- dard. Approximately 40 to 50 vehicles in excess of 3.5 metric tons are diverted from the local traffic stream to be statically weighed. They estimate that this requires approximately 15 staff for one to two days per year (e.g., approximately 120 to 240 person-hours). A number of innovative enforcement screening methods are used involving digital photographic records of vehicles. These are as follows: • WIM measurements and photographic records of vehi- cles that are likely overloaded are transmitted to mobile static weighing crews downstream (in Switzerland this is called WIM/VID). • Records of WIM data and photographic records are ana- lyzed to identify companies that habitually violate load limits. These companies are, in turn, contacted through the mail and given the opportunity to conform in the future. 22 EUROPEAN WEIGH-IN-MOTION-RELATED RESEARCH Weigh-in-Motion for Axles and Vehicles for Europe (WAVE) has been one of the most significant European WIM-related research studies (5,11,21,26). The WAVE project had the fol- lowing major objectives: • Develop innovative methods for WIM system evaluation/ calibration (11), • Develop innovative WIM systems with multiple sensors (i.e., MS-WIM) (21), and • Improve WIM data QC procedures (26). The major innovation of the WAVE project in WIM sys- tem calibration was an instrumented three-axle single-unit truck developed by the Technical Research Center of Finland (VTT). Its instrumentation consisted of strain gauges and accelerometers installed on the vehicle axles to detect bend- ing and vertical acceleration, respectively. The signals gen- erated were combined into measurements of dynamic axle loads on board the vehicle. This type of instrumentation was used in the mid-1980s by the National Research Council of Canada (NRCC) on a five-axle semi-trailer truck (32). The VTT vehicle was used for testing WIM accuracy with respect to dynamic, rather than static, axle loads. This approach was successfully tested by Papagiannakis et al. using the NRCC instrumented truck (27). The main challenge with this approach was identifying the segment of the dynamic load measurements that corresponded to the location where an instrumented axle is over the WIM sensor. The importance of synchronizing test truck dynamic load measurements and WIM measurements is illustrated in Figure 4. This figure shows the size of a bending plate sensor; that is, about 0.3 m (1 ft) wide, superimposed on the dynamic load measure- ments from three passes of the drive axle of the VTT vehicle at 80 km/h (50 mph). This figure also demonstrates that dynamic axle loads under the same speed exhibit repeatability in space, which explains the resulting lower variation in errors for a given speed and axle. The study also explores the idea of developing individ- ual calibration factors by axle type and vehicle configuration. This instrumented vehicle was also driven over an array of 16 piezoelectric WIM sensors arranged in various spacings to examine whether these multiple sensors could improve static load predictions. The development of MS-WIM (21) involved numerical simulations of the dynamic behavior of an idealized five- axle semi-trailer truck and of alternative arrangements of multiple sensor WIM installations. The simulation model for the five-axle semi-trailer truck was a theoretical two- dimensional model (i.e., pitch and bounce) that accepted the pavement profile as input. Similar models can be found in the U.S. literature (20). The dynamic axle load predictions

23 were simplified through several sinusoidal functions to describe the dominant dynamic load frequencies. This infor- mation was, in turn, used for estimating multi-sensor number and spacing to minimize the error between WIM and static axle loads. These results were finally field-validated using test trucks on several multi-sensor WIM installations. The WAVE project also dealt with WIM data QC (26). The main objectives of this part of the study were to standardize WIM data QC and storage (i.e., the EU-WIM database), as well as to develop software that allows extraction of data by a variety of traffic data users. It is comprised of two parts, a “low level” dealing with the functionality of the sensors and electronics themselves and a “high level” dealing with the quality of data being output and their statistics. The low- level data QC consists of evaluating the following: • Scale factor; that is, the maximum registered axle load (i.e., unusual or repeating values indicate that the sensor is malfunctioning or broken), • Total traffic volume, • Heavy vehicle traffic volume, and • Status error messages, if any. The high-level data QC is based on analysis of WIM data and their statistics using criteria such as the FAW and the GVW of certain truck classes. A computer program called MAR-TINE is used for these calculations. DISCUSSION The review of the national and international literature on WIM system calibration standards, practice, and related research reflects the significant effort expended in maintaining WIM traffic data quality. Conventionally, WIM errors are computed with reference to static load. As a result, errors are largely dependent on pavement roughness and the dynamic proper- ties of the vehicles used for evaluating them (vehicle class, suspension type, and speed). Another issue is that the stiff- ness of the pavement supporting the WIM sensors influences their signal output. This is more pronounced in asphalt con- cretes that exhibit temperature-dependent stiffness. This prob- lem is further compounded by the temperature sensitivity of some Type II WIM sensors. Error tolerances are set in the form of either • Fixed confidence intervals, given a confidence level and a prerequisite pavement roughness (empirically estab- lished), as in the ASTM E1318-02 standard, or • Variable confidence intervals, depending on the desired confidence level, available pavement roughness (IRI) and sample size, as in the COST 323 standard. In this light, the current U.S. WIM calibration standard, ASTM E1318-02, has the following limitations: • It does not consider pavement roughness as it relates to dynamic axle loads and, hence, the magnitude of the resulting WIM errors. The proposed AASHTO pro- visional standard MP 14-05 addresses this limitation through estimates of the dynamic response of a simu- lated fleet of trucks. This approach is an improvement over the COST 323 approach that sets empirical pave- ment roughness limits in terms of IRI, which is based on the relative displacement of the axle with respect to the frame of a passenger car. • It does not consider pavement stiffness in supporting WIM sensors, which, as described earlier, is being con- sidered by the COST 323 WIM standard. • It does not allow for a variable confidence level in estab- lishing error tolerances (i.e., confidence intervals). The approach taken by the COST 323 standard is more gen- eral, in as much as it provides for variable error tolerances as a function of the desired confidence level. • It assumes that errors are normally distributed in estab- lishing error tolerances. A Student-t distribution would be more appropriate for WIM errors, given the small 13000 12000 11000 10000 9000 8000 80 km/h 7 8 9 10 11 12 13 14 15 16 17 DISTANCE (m) 18 19 20 21 22 23 24 25 26 27 A XL E LO AD FIGURE 4 WIM evaluation using the VTT instrumented vehicle (28).

number of samples typically available during calibration. This is the approach used by the COST 323 standard, whereby error tolerance is defined as a function of sam- ple size. • Finally, it does not include WIM data QC methods. The latter was not one of the objectives of the ASTM E1318-02 standard, but it may need to be standardized. In addition to the main limitations described earlier, a number of enhancements may need to be considered in updat- 24 ing the national WIM calibration standard. These include data analysis by speed to quantify WIM system precision under replicate axle passes and to establish speed-specific calibration factors. Finally, a number of technological inno- vations may need to be considered for the future enhance- ment of WIM calibration, such as the use of instrumented test trucks and the use of traffic stream AVI-equipped trucks. Additional discussion on future research directions, which is based on the results of the survey questionnaire, is included in the following chapter.