Click for next page ( 58


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 57
57 compute vehicle classification. While these outputs are the key 5.8.5 A Final Word on Calibration calibration measure, problems with the estimation of these and Equipment Installation values are often a function of poor measurement of the sen- sor layout. Adjustment of these parameters may be needed to Calibration is a key part of the initial equipment set up. make the classifier correctly report vehicle speed and conse- However, calibration alone will not compensate for a poorly quently axle spacing or vehicle length. (Note that, depending installed piece of equipment. Poorly installed sensors often on the classifier technology used, other device parameters produce inconsistent signal outputs, making calibration either may also require adjusting to produce correct device output.) impossible or unstable over time, as sensor performance The accuracy of vehicle speeds should be determined by declines over time. Poor installation also leads to early sys- comparing device output against independent measures of tem failure and significant increases in both sensor downtime vehicle speed collected using a calibrated radar gun or sim- and maintenance costs. A key part of installation and cali- ilar device. Vehicle length and axle spacing computations bration efforts is to ensure that the sensors that have been should be compared by comparing these outputs against inde- installed are producing consistent signals. pendently collected axle spacing and vehicle length data. Correctly classifying a vehicle requires more than an accu- 5.9 CONDUCT PREVENTIVE rate measurement of vehicle speed and the distance between AND CORRECTIVE MAINTENANCE axles (or overall vehicle length). The conversion of these vehi- cle characteristics into an estimate of what vehicle type is Preventive maintenance keeps equipment operating. Per- represented by that set of attributes is the function of the clas- haps more importantly, preventive maintenance helps keep sification algorithm used by the equipment. While many clas- data quality problems to a minimum by reducing the number sification errors are caused by poor sensor input, many errors of strange axle detections that occur during early phases of are simply the result of a classification algorithm that incor- sensor and pavement failure. rectly associates a given set of vehicle characteristics with the Preventive maintenance includes tasks such as cleaning wrong class of vehicle. With axle classification, this occurs in electronics cabinets, replacing parts that are showing wear part because some vehicles from different classes have iden- but that have not yet failed, and even doing minor pavement tical axle spacings. For example, FHWA Class 2 (cars) and repairs that are designed to improve pavement smoothness FHWA Class 3 (light-duty pickups) have considerable axle- and life such as crack sealing on approaches to sensors. spacing overlap. Many small pickups have shorter axle spac- Corrective maintenance is the process of bringing a sensor ings than larger cars. Thus, a considerable number of errors that is not performing properly back into correct operation. occur when classifiers try to differentiate between these two Corrective maintenance can include physical changes to sen- classes of vehicles. Recreational vehicles (especially those sors or pavement (e.g., sealing cracks in the pavement or pulling trailers or other vehicles) are another class of vehi- repairing the bond between sensors and pavement), replace- cles that have axle spacings that frequently cause them to be ment of failing or failed electrical components, or simply misclassified, even when the classifier is working. adjustments to sensor calibrations used for computing speed, Errors associated with poor classification must therefore weights, or overall vehicle length. be separated into those due to poor sensor output and those Proper, timely maintenance increases sensor life, improves related to the combination of a poor classification algorithm data quality, and decreases overall system life-cycle cost. and/or vehicle characteristics that prevent a given set of sen- States that have snow and ice conditions need to consider the sor outputs (axle number and spacing or vehicle length) from added maintenance needed for traffic monitoring systems correctly classifying a vehicle. Poor sensor output can be dealt that may be affected by sand, snowplows, and corrosive anti- with at the time of equipment installation, set up, and calibra- icing materials. Other specific types of environmental condi- tion. The other two problems must be dealt with through the tions, such as dust storms and lightning, are also renowned rigorous design and testing of the classification algorithm for frequently causing equipment problems that need to be used by the agency and through an extensive equipment addressed through timely preventive and corrective mainte- acceptance-testing program that should be performed when nance activities. Understanding when these types of envi- a given brand or model of equipment is first selected for use. ronmental conditions are occurring and causing equipment (That is, the agency needs to make sure the equipment will or site damage, and timing site reviews to coincide with these classify correctly if installed properly before purchasing conditions, will decrease the amount of data lost to these con- large quantities of a given device. Only once the classifier has ditions and increase the quality of the data available from been shown to work as desired should the device be pur- data collection devices. chased in quantity. If, during the acceptance testing, a device Maintenance activity needs to be tied to data quality con- appears to be working correctly but cannot classify specific trol systems. Data being collected often provide early warn- types of vehicles, the agency should carefully examine the ing signs that minor corrective maintenance is needed at classification algorithm being used to determine if the algo- a site. Quick intervention when data quality first becomes rithm itself needs to be fixed.) questionable both increases the amount of good data that are

OCR for page 57
58 collected and decreases the staff time spent examining ques- sor failures. This information can be used both in the budget tionable data. planning process and as part of sensor deployment planning If a site visit indicates that minor corrective action is efforts. For example, if it is known from maintenance activ- unlikely to resolve problems, the data collection site can be ities that pavement at a given site has degraded, data collec- shut down until more significant repairs can be performed. tion staff can plan to move electronics to a new location, This dramatically reduces the effort wasted in retrieving and where pavement conditions are conducive to accurate data reviewing invalid data. Ongoing preventive maintenance collection, until pavement maintenance activities upgrade also provides excellent input into the performance of differ- conditions at the original site. This ensures the most produc- ent vendors' equipment and early warning of impending sen- tive use possible from the available data collection resources.