AUTOMATED TRACKING OF STRUCTURAL STEEL MEMBERS AT THECONSTRUCTION SITE1

Karen M. Furlani and Lawrence E. Pfeffer

Construction Metrology and Automation GroupNational Institute of Standards and Technology

Gaithersburg, MD 20899, USAkfurlani@nist.gov pfeffer@nist.gov

Abstract: This paper presents further development and experimental testing ofComp-TRAK, a prototype system for identification and spatial tracking of structural steelcomponents on a live construction site. Comp-TRAK is a web-based system for rapidtracking of construction components with compact, field-rugged sensors and computers,interoperability protocols for data transmission, and a 3-D site visualizer that reflects thecurrent state of tracked components on the construction site. The result is a graphically-driven system that provides real-time identification, position, and orientation data aboutcomponents to users in the field and at remote locations.

Keywords: bar code, component tracking, construction, laser positioning system, metrology,real-time tracking, RFID, wireless data transfer.

1 Official constribution of the National Institute of Standards and Technology; not subject to copyright in theUnited States

1. INTRODUCTION

The Construction Metrology and AutomationGroup (CMAG) at the National Institute of Standardsand Technology (NIST) has on-going research inreal-time construction component tracking to addressthe problem of identifying and tracking discreteconstruction components and sub-assemblies on aconstruction site. Discrete components, in thecontext of the present discussion, are manufacturedprefabricated singular items, such as structural steel,which carry a means of identification, such as a barcode label.

Although many technologies exist to aid in theautomation of component tracking systems, their usesare limited by a lack of construction industrystandards supporting inter-operability betweenvarious hardware and software systems. Additionally,many products are not designed to function andsurvive in the hostile and dynamic environment of atypical construction site. To help the constructionindustry overcome these obstacles and achieve a fullyintegrated and automated tracking system, the CMAGis developing and testing Comp-TRAK, a prototypesystem for tracking manufactured components at thejobsite. The Comp-TRAK system involves the use ofRFID and bar code identification systems, 3D long-range (50 m radius) coordinate measurementtechnologies, local data processing, wirelesscommunication, networking, temporal project

databases, web-based data analysis, and 3D userinterfaces to provide real-time access to part statusupdates at the jobsite and to remote, off-site officeusers. With this approach, CMAG is focusing ondeveloping the technical infrastructure to supportreal-time tracking of sub-assemblies and componentsat the jobsite.

2. PROJECT SCOPE

In earlier work, we proposed an overall systemarchitecture [1] and developed a prototype systemthat experimentally demonstrated 3D spatial trackingof discrete components in real-time and undercontrolled conditions to viewers at a remote locationover 1 km away [2]. Most recently, the prototypesystem was field tested at the Building 205 EmissionControl System (ECS) addition - a US$6 millionproject on the NIST, Gaithersburg campus. A subsetof the ECS structural steel framework was tracked.This paper discusses integration and implementationissues encountered while developing and testing theComp-TRAK system for data collection on aconstruction site.

3. COMP-TRAK SYSTEM

3.1 System Overview

Structural steel components scheduled for arrivalon the construction site are tagged with bar codeand/or radio-frequency identification (RFID) tags atthe galvanizing plant prior to shipment. Via a web-based graphical user interface, the identificationinformation encoded on the tag is scanned directlyinto a ruggedized, wearable computer and used toquery a project database for additional informationrelating to the scanned item, including a 3D CADmodel of the part. This model, coupled with user-friendly web browsing software, guides field workersthrough the measurement of key fiducial points usinga long-range 3D coordinate measurement system toobtain x, y, z point data. We use the term fiducialpoint here to mean pre-specified points, typically atreadily identified features of interest such as corners.The fiducial points selected comprise a sufficient setof locations to establish the position and orientationof a component in 3-D space. The collected part IDand x,y,z fiducial point data is sent wirelessly to aremote server hosting a part locator service, whichdetermines the object’s position and orientation in thejob site’s global coordinate system (see figure 1.)The database then uses this data to track the currentstate of the object.

Figure 1: After scanning a part’s bar code or RFIDtag for the unique ID, the project database returns a3D model with visual queues for the location of keyfidicial points. Above, the field inspector hasmeasured coordinate data for Points A-D, which arethen wirelessly relayed to the part locator service.

3.2 Field Data Collector

The field data collector (see figure 2), initiallydeveloped and tested in a controlled laboratoryenvironment, was modified and upgraded for the fieldtesting at the Bldg. 205 ECS addition. Originallyhosted on an IBM laptop computer, the datacollection system was re-configured for use with a

VIA II belt-wearable computer [3]. A PCMCIARangeLAN2 Ethernet Adapter by Proxim and aQuatech DSP100 Dual Serial I/O Card providewireless communications and two additional RS-232D serial ports, respectively.

For part identification, a bar code laser scannerwith built-in keyboard wedge software integrates viathe USB port, and a RFID integrates via a RS-232Dserial port. To measure 3D coordinate data forfiducial point collection, a Vulcan 3D measurementsystem from ArcSecond, Inc. [4] integrates via aRS232 serial port using a C program written at NISTto interface the field computer with the Vulcansensing tool. The Vulcan system creates a 3-Dmeasuring environment within the intersecting area oftwo rotating fanned laser beam transmitters. Amobile “receiving” measurement tool collectscoordinate point data within the working area createdby the transmitters.

The entire web-based user interface runs throughan Internet Explorer 4.1 graphical user interfacewritten in Microsoft FrontPage98, driven byVisualBasic Scripts. Active Server Pages interactwith the project database, and a LiveView [5]Common Gateway Interfaces (CGI) script providesthe protocol for the transmission of both the part IDand its coordinates to the part locator service forposition and orientation determination.

Figure 2. Field operator collecting fiducial pointcoordinate data with the Vulcan receiving toolintegrated with the VIA belt-wearable computer andbar code scanner. A manlift was used to facilitatedata collection once the steel tower was erected andinstalled is its final location.

4. FIELD TEST

4.1 Part Identification

A subset of the ECS structural steel framework,Tower #2 (see figure 3), was tagged with bar codes atthe galvanizing plant. Each bar code tag has theunique part identifier encoded in Code 3-of-9symbology and also printed in human readable fontalong with some additional descriptive informationrelated to the tagged part.

During pre-construction meetings with the steelfabricator, it was verified that piece marks, typicallystamped on a steel member at the fabricators foridentification purposes, would be located on the farlside of a steel part as typically shown on thefabricator’s drawings. Since this provided an easilyfindable and repeatably consistent location on thesteel, the bar code tags were attached adjacent toeach steel part’s piece mark to aid yard workersattaching the bar codes at the fabricator as well asfield workers trying to scan the tags at theconstruction site

A member of the CMAG research team visited theyard crews at the galvanizing plant to explain thepurpose and desired placement of the bar code tags.After the first couple tags were attached, theremainder were left with crew leader to finish. Afield inspection of the tagged steel after delivery tothe jobsite yielded no instances of unmarked ormismarked steel. During a follow-up conversation,the crew leader remarked that the tagging process wasvery simple and took little time to accomplish, notingthat it only involved matching up the tags (whichwere organized in numerical order) with the piecemarks on the steel.

The bar code tags, printed on “Stick Tuff” byIntegrated Labeling Systems, Inc., survived the fulllength of the project; they stayed attached andreadable from the time the steel parts were tagged atthe galvanizers through final erection at the job site.As mentioned above, each steel part was tagged witha bar code tag located adjacent to the fabricator’sstamped piece mark. Although this simplifiedattaching and locating a tag, there were instanceswhen access to the one tag was difficult, if notimpossible, due to the stacking of steel in thelaydown area or the location of the tag once a partwas installed. In future applications, attaching morethan one tag on a steel part will provide a fieldoperator with additional options in case ofinaccessible or damaged tags.

Attempts to use RFID tags for part identificationwere not successful due to difficulties in permanentlyattaching the tags to the structural steel members.The tags used during the field test were encased in aplastic housing, attachable by either wire or epoxyglue. Due to the nature of most steel members, thereis no good way to attach RFID tags using wire.

Although bolt holes might work initially, they wouldbecome ineffective at the time of erection. Attachingthe tags with epoxy resulted in all the tags falling offor being easily knocked off within days of theirattachment.

Figure 3. Elevation view of Bldg. 205 ECS Tower#2. The columns, angles and beams were trackedduring the field test. The fabricator’s ID shownadjacent to each steel member served as the basis forthe unique part ID assigned to each part in the projectdatabase. The unique ID was encoded on the barcode and RFID tags for auto identification at the jobsite and used to query the project database to return aVRML model of the part to guide users throughfiducial point measurements.

4.2 Fiducial Point Assignment Convention

Two basic structural steel shapes were tracked inthe Tower #2 subset: W shapes (I-beams andcolumns) and angles (see figure 4.) A standard set ofdefault fiducial point locations were established foreach of the three basic geometric shapes. The upper,

left-hand corner above the piece mark was designatedfiducial point #1 on all the parts. Since the piecemark was always stamped on the steel in the samelocation, this provided a relatively simplemethodology for the field test without requiring that aseparate VRML model be created for each individualsteel part. In general, this methodology worked wellwith the sampling of tracked steel, since the onlygeometric difference between structural steel pieceswith identical shapes is length. However, difficultiesarose when steel parts arrived with pre-attachedplates or were erected into groups on the jobsite, andwhen access to at least three of the pre-definedfiducial points was not possible. Work in follow-onyears will seek to identify potential solutions to suchcomplex measurement issues as when certain partsare identical and interchangeable with others on aproject, when parts are symmetrical, and when partsare theoretically identical with others according tofabricators drawings, but have been shortened due toa field clearance (see Section 5.0.)

Figure 4. An enhanced VRML model of the structuralsteel angle (top) and I-Beam (bottom) with fiducialpoint markings and fabricator’s steel mark help guidefield workers through the acquisition of x, y, zcoordinate data for a minimum of three fiducialpoints for position and orientation determination of apart at the jobsite.

4.3 Coordinate Measurement System Registration

To obtain registered 3-D coordinate measurementdata using the Vulcan system, a known coordinate

frame must first be established. There are two basicoptions available, either:1) The field measurement sensor can be configured

to transform its measurements into a known, site-reference frame using the Vulcan re-sectioncalibration; or,

2) Data sufficient to uniquely determine the sensor’sworking coordinate frame can be recorded, so thatdata in the working frame can be transformedafter the field measurements are taken using theVulcan’s quick calibration.

The re-section calibration method requires thecollection of four points known in the desired frameand obtained within the area where fieldmeasurements will be taken. These points must bewithin range and line-of-sight of the transmitters andaccessible by the field inspector using the fieldmeasurement tool at all times during the project. Inaddition, the field measurement tool must be levelledwhen collecting these points. While this methodallows the transmitters to be set-up at unknownlocations on an as needed basis, it would be nearlyimpossible to guarantee continued access to a set ofknown survey points within the construction areawithout continually surveying new points asconstruction progresses through the lifecycle of theproject. In addition, the likelihood that equipmentand materials would block access and/or line-of-sightduring measurements is fairly high.

Due to these limitations, the quick calibrationmethod was used during the field test. This methodestablishes a frame based on the location andorientation of the transmitters. While this stillrequires a known set of locations for the transmitters,and requires that the transmitter serving as the originbe levelled, they can be located outside the actualwork area and remain undisturbed duringconstruction.

To obtain known points for the transmitterlocations, ten benchmarks were established toprovide adequate coverage for measurements inboth the laydown and construction areas. The newbenchmarks were surveyed with a total station usingpermanent USGS (United States Geological Survey)benchmarks located at the NIST site for reference.All but one location was easily accessible for set-upand removal on an as needed basis. A semi-permanent set-up with weatherproof housing wasinstalled for the one transmitter that was difficult toaccess [see figure 5.]

A total of four transmitters were available for useduring the field test. Although the receiving tool onlyuses two transmitters at a time during measurements,having two “extras” significantly reduced the numberof times transmitters needed to be moved. Prior to ameasurement session, the field operator notes whichbenchmarks the beacons are set-up on to provide thesite visualization system with the necessarycoordinate frame information.

Figure 5. A pulley system and a long power cordallows the field inspector at ground level,approximately 12 m below, to lift the weatherproofcover and attach a rechargeable battery to begin usingthe transmitter for measurements.

4.4 Position and Orientation Determination

The position and orientation of a part (referredto, collectively, as the part’s “pose”) is computedfrom the measured positions of 3 or more non-co-linear fiducial points and the corresponding, nominalpositions of those fiducial points in the part’scoordinate frame (according to a pre-defined,geometric model of the part.)

The part-locator acts as both an informationaggregator and a data-interpreter. It mergesinformation from two different sources (sensor datafrom the field data collector) and part geometry (thefiducial-point location, from the geometric model.)From those two sets of data, the locator determinesthe part’s position and orientation, which are derivedquantities (and thus an interpretation of the “raw”data.) The part locator acts as separate service thatcan run on a different computer, communicating overa local-area network.

Each time the part locator service is used, it doesthree things:1. Determines the part’s pose (in some site-

referenced frame) from locations (x, y, z) offiducial points measured in that frame andcorresponding data from the part’s geometricmodel.

2. Computes a metric for the uncertainty in thepose.

3. Transforms the pose into the project’s preferredcoordinate frame (and representation) for storagein the project database.

In the locator, the part’s pose is primarily representedas a pose-parameter vector of 6 real numbers, x,y,z(Cartesian translation) together with θx, θy, θz(rotations about site-fixed axes, with the rotationsapplied in θx, θy, θz, order.)

The use of a 3-term representation for orientationhas a potential problem common to 3-parameterrepresentations of orientation -- numerical problemsat/near the “gimbal lock” orientation (a singularitycaused when the rotations result in alignment of twocorresponding body-fixed axes, and the resultant lossof angular freedom about another axis.) Thisproblem has not arisen in our data, so far. Use of a 4-term representation of orientation (e.g. quaternions)would require use of constrained optimization, toenforce the constraint required to keep a 4-termrepresentation geometrically valid.)

The parameter search technique for the “best-fit”pose parameters is based on unconstrained numericaloptimization. In practice, the “best-fit” poseparameters do not represent a perfect fit of a posevector to the measured data, due to measurementerror, part variation (from the designed geometry),and finite-precision computation.

The cost function chosen for the minimization-based search is a sum of squared errors between themeasured and the corresponding (translated androtated) model-derived points, as denoted in Eq. (2):whereJ(X) is the cost function,

n is the number of measured fiducial points,T(X) is a homogeneous transform that combinestranslation and rotation

This cost is used for two reasons: First, a closed-form solution for the gradient of the cost function isknown; this can substantially speed up theoptimization-based search. Second, this cost leadseasily to a metric -- the root-mean-square (RMS)error (between measured and “best-fit” ofcorresponding points) that has a physically-intuitivemeaning. The RMS error indicates how far themeasured fiducial-point locations are (on average)from a perfect fit. As such, the RMS error is our

[ ] ( )1,,,,, TzyxzyxX θθθ=

( )21

2))()()(()( mod∑ ⋅−=

=n

iipXTipXJ meas

tmeasurementheis ),,()( zyxiip thmeas

modelthefromdatatheis ),,()(mod zyxiip th

current candidate for a metric of the uncertainty in thepart’s pose.

The best-fit pose parameter vector (X*) isdetermined, using unconstrained optimization tosolve the minimization problem:

(3)*

))(( XJMinXX

=

Eq. (3) is solved numerically to determine theoptimum X, denoted by X*. The cost at this best fit,J* = J(X*), is thus uniquely determined. This cost isthen converted from the sum-of-squares-error to RMS(by normalizing by the number of points and takingthe square root.) Finally, the pose is converted intothe coordinate frame and representation required bythe project’s database and is then sent to the database.

5.0 FUTURE WORK

Upcoming work will begin to address some of thetechnological limitations experienced during fieldtesting of the Comp-TRAK system on a liveconstruction site. Research will focus on threespecific areas:

1) Metrics for registration of multiple fanninglaser transmitters: Development of a standardmethod for automatically registering large-scalelaser-based coordinate measurement systems to theWGS-84 (World Geodetic System 1984) coordinatesystem. This work is essential to providing the basisfor uniform and accurate coordinate registration.

2) Automated methods for fiducial pointdetermination: Development of automated methodsfor extraction of easy-to-use fiducial points forstructural steel elements. The current method ofmanually identifying fidicial points is a timeconsuming process, and difficulties arise when pre-defined points are inaccessible at the jobsite.Automated methods are needed to mine thisinformation from existing data representations ofstructural steel parts.

3) Measures of performance for part locatoralgorithms: Data flow protocols being developed atNIST for transmission of sensor data from aconstruction site deliver a series of raw threedimensional coordinates for a structural componentthat has been placed on a construction site. Theremay be three (or more) 3-D points involved in anindividual identification yet, by themselves, they donot describe the state of the object – i.e. its positionand orientation (pose). Numerical algorithms havebeen developed at NIST to solve this problem usingnumerical optimization. Other approaches may bedeveloped by industry in time. However, there is nostandard method by which the accuracy of positionand orientation (pose) determination can be measuredor compared. Future work will focus on developing

standard measures of performance for posedetermination.

Acknowledgements and Notices

Special thanks to Jim Claytor, Chuck Helm, andMike Dalton of Green Contracting Company, andK.C. Patel and Carrie Kelly of the NIST PlantDivision.

Certain company products are mentioned in thetext in order to adequately specify the equipmentused. In no case does such an identification implyrecommendation or endorsement by the NationalInstitute of Standards and Technology, nor does itimply that the products are necessarily the bestavailable for the purpose.

REFERENCES

[1] Furlani, K.M, and Stone, W.C., “Architecture forDiscrete Construction ComponentTracking,” Proceedings, 16th IAARC/IFAC/IEEEInternational Symposium on Automation andRobotics in Construction, September 22-24, 1999,Madrid, Spain, pp 289-294.[2] Stone, W.C., Furlani, K., and Pfeffer, L.,“Automated Part Tracking on the Construction Site,”Proceedings, 4th International Conference onRobotics for Challenging Situations andEnvironments (Robotics 2000), Albuquerque, NewMexico, February 27-March2, 2000.[3] VIA, Inc., 12550 W. Frontage Rd. #201,Burnsville, MN, 55337,http://www.flexipc.com.[4] Arc-Second (1999), ArcSecond, Inc., 44880Falcon Pl. #100, Dulles, VA 20166,http://www.arcsecond.com.[5] Pfeffer, L.E., and Latimer, D.T., (1999),“Toward Open Network Data-Exchange Protocols forConstruction Metrology and Automation:LIVEVIEW,” Proceedings, 16th IAARC/IFAC/IEEEInternational Symposium on Automation andRobotics in Construction, September 22-24, 1999,Madrid, Spain, pp 117-122.[6] Furlani, K., Pfeffer, L., Latimer, D., and Stone,W.C., (1999), NIST Construction AutomationProgram Report No. 5: Real-Time Vehicle andDiscrete Component Tracking for Construction,National Institute of Standards and Technology,Gaithersburg, MD 20899, USA.