U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590

Skip to content
Facebook iconYouTube iconTwitter iconFlickr iconLinkedInInstagram

Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations

This report is an archived publication and may contain dated technical, contact, and link information
Back to Publication List        
Publication Number:  FHWA-HRT-07-040    Date:  October 2011
Publication Number: FHWA-HRT-07-040
Date: October 2011


Falling Weight Deflectometer Calibration Center and Operational Improvements: Redevelopment of The Calibration Protocol and Equipment



Deflection data from falling weight deflectometers (FWDs) are used for structural evaluation of pavements and can be used to calculate remaining pavement life, load-supporting capacity, and the required thickness of structural overlays. The results of these calculations can be used to manage pavement systems at the project or network level.

Various mechanistic models of pavements are used in analysis including the elastic layer theory, the finite element theory, the discrete element theory, etc. These analytical methods utilize the moduli of elasticity of the pavement layers as an input. Pavement deflections, combined with backcalculation methods, can provide the layer moduli. However, the accuracy of the deflections has a major influence on the accuracy of the backcalculated moduli and subsequent analyses.(2)

FWD data have the following sources of error:

Repeated deflection measurements plot as a bell-shaped frequency distribution as shown in figure 1.

Figure 1. Illustration. Random error and systematic (bias) errors. A bell-shaped curve is superimposed over a vertical line representing measured deflection. Another vertical line on the left side of the figure represents true deflection. Offset between the two lines and the curve illustrate the relationship between the systematic error, or bias error, which is the difference between the true deflection and measure deflection (less than 2 percent) and the random error (plus or minus 0.08 mil (plus or minus 2  m)).

Figure 1. Illustration. Random and systematic (bias) errors.

Seating Errors

When an FWD first occupies a new test point, seating errors arise from the mechanical placement of the sensors (i.e., geophones or seismometers) on the pavement. Because of roughness and loose debris on the road surface, the sensors are not always seated firmly. Deflection sensors can be seated fairly easily by dropping the FWD mass two times without recording data. It is seldom necessary to perform more than two drops to achieve good seating. However, extensively cracked pavements, such as alligator cracking, may pose a seating problem regardless of the number of seating drops.

In addition to the deflection sensors, the FWD load plate must also be seated squarely on the pavement. For most types of FWDs, all that is needed is to keep the load plate swivel well lubricated.

Random Errors

Random errors are mainly attributable to the digitization of the analog (voltage) signal from the deflection sensors. Analog-to-digital devices convert the voltage so it can be read and processed by a computer. Using an operating system (also known as the field program) provided by the FWD manufacturer, the digitized signal is converted to displacement in engineering units (mils, microns, and other units). Digital conversions include a small amount of error, typically one or two least significant bits. Equipment manufacturers state that the random error is on the order of 0.04–0.08 mil (1–2 μm) per reading. The magnitude of the error varies from one drop to the next independent of the peak amplitude.

Thus, each deflection reading includes a small component of random error. Because they are random, some of the errors are positive and some are negative. For any drop, it is impossible to know the magnitude of the random error or whether it is positive or negative. However, by averaging the results of several deflection peaks taken at essentially the same load level on the same test point, the random error of the average is reduced (but not totally eliminated).

To avoid changing the properties of the pavement by performing too many drops on a single point, typically no more than three to five drops per load level are taken and averaged. If the random error in one measurement is X, the random error of the average of n drops is X/n 0.5. Specifically, the random error of the average is equal to the random error of one observation divided by the square root of the number of observations. For example, by performing and averaging four drops, the random error can be reduced by half (since √4=2).

The random error exhibited by a set of deflection sensors is quantified by the calibration procedure outlined in this report.

Bias Errors

Bias errors (also known as systematic errors) arise from the need to calibrate the sensors and are directly proportional to readings. For example, if the deflection doubles, the bias error in the reading will be approximately twice as large in magnitude. Bias errors can occur for any number of reasons; however, regardless of the reason, FWD manufacturers generally state that they will comprise no more than ±2 percent of the total reading. Therefore, bias errors could be zero, slightly positive, or slightly negative. Unlike random errors, they will be constant (as a percentage of the reading) for a given sensor from one drop to the next.

The main reason for performing FWD calibration is to reduce the bias error of each load and deflection sensor using the procedure outlined in this report. The procedure uses a reference sensor (i.e., a reference load cell for load calibration and a reference accelerometer for deflection calibration) that is independently calibrated to a high degree of accuracy and reliability. The peak result from the FWD is compared with the peak recorded by the reference sensor. The ratio (reference device divided by the FWD reading) becomes the calibration factor (also referred to as the gain factor).

Deflections must be fairly large to be able to detect bias error. In figure 1, if the measured deflection is close to the true deflection, the random error will mask the bias error. For example, if measured deflection is 2 mil (50 μm) and the bias error is 2 percent, the bias error is only 0.04 mil (1 μm). With a random error of 0.08 mil (2 μm), the bias error is so small that it is nearly impossible to detect using statistical methods.

Similarly, there are limits to how much bias error can be reduced, even with a large deflection. In the calibration procedure, there is typically an average deflection of about 20 mil (500 μm). When the bias error is reduced to 0.3 percent, the difference between the measured deflection and the true deflection is only about 0.06 mil (1.5 μm). Again, in this range, bias error is difficult to detect reliably, so 0.3 percent is effectively the practical lower limit that FWD calibration can achieve.

Significance of Errors

In 1988, Irwin et al. reported on the effect of random deflection errors on backcalculated layer moduli.(2) They found that random error, without any additional bias error, could cause the backcalculated modulus of the surface layer to be off by a factor of two or more. The effect on base and subgrade layer moduli was less dramatic.

When a deflection is greater than 4 mil (100 μm), a 2 percent bias error in an uncalibrated FWD will be larger than the ±0.08-mil (±2-μm) random error. Thus, the combined effect of random and bias errors could be substantially more detrimental to the backcalculated moduli than random error alone.


In 1988, the dramatic effect of FWD errors on backcalculated moduli led the Strategic Highway Research Program (SHRP) to begin developing an FWD calibration procedure with the goal of reducing the bias error to as close to 0.3 percent as possible. This goal was achieved by the SHRP/LTPP FWD Calibration Protocol.(3) Reducing the bias error to 0.3 percent ensures that a 0.08-mil (2- μm) random error is larger than the bias error for deflections up to 26 mil (667 μm). Deflections measured on most good-quality roads are seldom that large, especially for sensors at or beyond 12 inches (300 mm) from the load plate. For a calibrated FWD, random error, which is not preventable, controls the accuracy of backcalculated layer moduli in most circumstances.

Four regional FWD calibration centers were established by SHRP in 1992 and were operated by the State transportation departments in Pennsylvania (PennDOT), Texas (TxDOT), Minnesota (Mn/DOT), and Nevada. The western center in Nevada was later moved to Colorado. Since their inception, the demand for FWD calibration services has risen steadily.(4) In August 2001, the American Association of State Highway and Transportation Officials (AASHTO) Subcommittee on Materials adopted a resolution supporting the continued operation of the calibration centers.(5)

Because of changes in technology, the original calibration hardware and software became obsolete. There was a particular need to update the FWDREFCL software.(6) It was originally written as a disk operating system (DOS) program and utilized a Metrabyte DAS-16G data acquisition board (DAQ). DOS has been replaced by Microsoft Windows®, and the industry standard architecture (ISA) bus used by the Metrabyte board has been phased out by the computer industry.

Orr and Wallace reported that the 1992 SHRP calibration procedure had several shortcomings.(4) Among them were the following:

The procedure was not field portable; hence, the FWDs had to go to a calibration center, and the need for out-of-State travel posed a problem for some agencies.


Pooled Fund Study Tasks

In 2002, the Federal Highway Administration (FHWA) initiated a pooled fund study, TPF-5(039), with the financial support of 17 State transportation departments (see the acknowledgements section) to overcome the problems noted above. Overall, the primary goal of the study was to modernize and streamline the calibration procedure compared to the earlier SHRP approach without reducing the accuracy and precision of the results.

To meet this goal, the various FWD manufacturers and FWD calibration centers were contacted for suggestions and ideas. Based on their suggestions as well as those from annual meetings of the FWD User Group, a series of tasks was developed. The tasks accomplished during the pooled fund study were as follows:


The four original SHRP calibration centers have been outfitted with new equipment, and their personnel have been trained and certified. In addition, two new calibration centers have been established, operated by the Montana Department of Transportation (MDT) and the California Department of Transportation (Caltrans). The calibration center operated by the Indiana Department of Transportation (INDOT) has also been updated, but it is not available to outside agencies.

While they are not part of the pooled fund study project, calibration centers have also been set up at other manufacturing facilities: Foundation Mechanics Inc. (JILS) in El Segundo, CA; Dynatest® in Starke, FL; and Carl Bro in Kolding, Denmark. Additionally, the calibration center operated by Main Roads Western Australia in Perth, Australia, has been updated, and new calibration centers have been established at the Australian Road Research Board in Melbourne and Brisbane, Australia, and at Fugro PMS in Hamilton, New Zealand.

The new equipment and procedure are fully portable. Onsite calibrations conducted by certified technicians have been performed in Hawaii and Chile, and nine different FWDs have been calibrated at remote locations in Australia. Four agencies (Dynatest®, Foundation Mechanics Inc. (JILS), Carl Bro, and Fugro PMS) are currently offering onsite calibrations, and others have expressed interest in doing so.

A spinoff product of the project arose from the need to electronically transfer calibration data from the FWD computer to the calibration computer. Each different type of FWD has its own native file format, and the formats are not interchangeable. This incompatibility led to the development of PDDXconvert, which is a software program that converts each native FWD data file to the pavement deflection data exchange (PDDX) standard (see appendix C).(7,8) It is used integrally with WinFWDCal to seamlessly transfer FWD data to the calibration computer. PDDXconvert can also be used in a standalone mode in support of other software, such as those used for backcalculation. For example, the MODTAG backcalculation program uses PDDXconvert to process all types of FWD data to determine pavement layer moduli.

Quality Assurance Procedures

Quality assurance (QA) procedures have been developed to annually recertify the calibration center operators. Procedures and checklists have been applied successfully and updated several times during four annual cycles of QA visits. Currently, all FWD calibration centers are operated by certified technicians.

At the conclusion of the pooled fund project in October 2010, the AASHTO Materials Reference Laboratory (AMRL) in Gaithersburg, MD, took over the responsibility of certifying calibration technicians. In preparation of this responsibility, the AASHTO personnel worked and trained with Cornell University staff for more than 1 year.

Adoption of the Calibration Procedures by AASHTO

The SHRP FWD calibration protocol was originally adopted by AASHTO as AASHTO R32-03.(9) The AASHTO R32-03 procedure has been revised to reflect the changes brought about by the pooled fund study, and in January 2009, the AASHTO Subcommittee on Materials approved and adopted the revised procedure. The revised procedure was published as AASHTO R32-09.(1)

There have been additional revisions and improvements in the calibration procedure since 2009. As of October 2010, a revised AASHTO R32-09 procedure was under review by AASHTO, and it is expected to be published in 2011.(1)

The AASHTO R33-03 for annual calibration of the reference load cell has also been updated.(10) It was submitted for review and approval by AASHTO in May 2010. As of October 2010, it was under review by AASHTO, and it is expected to be published in 2011.(10)

Revision of the PDDX Standard

In April 1998, AASHTO adopted a standard for PDDX.(7) The initial intent of the standard was to facilitate electronic input of FWD data to the DARWin pavement design program. At the 2001 meeting of the FWD User Group in Gulfport, MS, all four FWD manufacturers confirmed they would provide an output file in their field programs that complied with the PDDX standard.

As noted previously, the pooled fund study used the PDDX standard output as the means for electronic exchange of FWD calibration data. However, several problems became evident. First, the PDDX file input requirements for DARWin were not fully compliant with the 1998 AASHTO PDDX standard. In addition, the AASHTO PDDX standard did not provide a placeholder for all of the FWD identification data that are needed for FWD calibration.

To overcome these problems and continue with the project, the PDDXconvert computer program was created.(8) In addition, a revised version of the PDDX standard was created to ensure compatibility of the PDDX files and satisfy the original intent of the AASHTO PDDX standard (see appendix C).

Development of an Informational Video

An 11-min video was produced to familiarize FWD owners with the new calibration procedure and encourage State transportation departments to utilize the calibration facilities. The video is available at the Long-Term Pavement Performance (LTPP) Web site, and CD-ROM copies can also be obtained from FHWA.

Long-Term Calibration Center Support Stewardship

Responsibility for the long-term support for the calibration centers is being transferred to AMRL. From March 2009 to June 2010, training was provided to two AMRL technicians and their supervisor. The training placed emphasis on the QA visit and operator certification processes (see appendix D). Each of the nine calibration centers was jointly visited one or more times by both AMRL and Cornell personnel. AMRL was slated to take over the QA visit responsibility starting in late 2010. Cornell expects to continue to support the centers’ needs for technical assistance, reference load cell calibrations, and software upgrades for at least 1–2 years.