|FHWA > Engineering > Pavements > WIM Data Analysts Manual > Section 4. Extensive Data Analyses Utilizing Individual Vehicle Records|
WIM Data Analysts Manual
Section 4. Extensive Data Analyses Utilizing Individual Vehicle Records
The purpose of performing extensive analyses of the individual vehicle records is to attempt to isolate and identify system component problems not identifiable by routine real time reviews, data QC, or real time checks of a system's parameters and settings. Such analyses are typically necessary when system component problems are intermittent and/or subtle in nature.
The method for performing these analyses is to import the individual vehicle records into a spreadsheet or database program and perform search, filter, sort, or other procedures as well as to have the program generate tables and graphs necessary to find any pattern which might isolate intermittent system component malfunction. It may also well be that analyses of questionable data indicates such data is probably attributable to conditions other than component malfunction. The ability to import the individual vehicle records is, of course, somewhat dependent upon the data file format of the records. Appendix C provides guidance on data import using Excel and ASCII data files formatted in accordance with LTPP's model specifications.
Figure 51 displays a snapshot of a portion of a simple spreadsheet that was created by importing data from an ASCII text file as per the procedures contained in Appendix C. This spreadsheet includes only the data elements from the individual vehicle record included in the ASCII text file, although as displayed in Figure 51, it has been filtered for records of vehicles that could not be classified by the system using its classification algorithm. This system assigns a "Class 15" to its unclassified vehicles. It is often beneficial to delineate the Axle Spacing columns as has been done in this example. By performing sorting and filtering schemes on unclassified vehicles, the analyst may be able to detect flaws in the classification scheme or possibly an error in how the classification scheme was entered into the system's classification algorithm. Also, analysis of the records may indicate that the vehicles were not properly processed by the system (as would appear to be the case in this example).
It is noted that almost all of the records displayed in the Figure 51 sampling have a "VIOL" flag "21" which is this system's code for an "Unequal Axle Detection" (the right and left weigh sensors did not detect the same number of wheel hits for the vehicle). It is noted that when this system experiences a vehicle for which a wheel hit is detected on one side of an axle but not the other, the system "invents" the missed wheel's weight by copying the detected wheel's weight to the opposite wheel. In that this may or may not provide a reasonably valid estimate of the axle's static weight, the vehicle is flagged with the warning. Most of the vehicles flagged with the "Unequal Detection" flag would appear to have legitimate wheel and axle weights for the first four axles followed by weights and/or axle spacings that appear to be erroneous. This suggests a problem with a weigh sensor or its signal processing.
In looking at some of the records in Figure 51 from a classification only standpoint:
Therefore, an analysis of the unclassified vehicles in this sampling indicates the problem is in a weigh sensor, not the classification scheme or algorithm. This should prompt the analyst to perform simple diagnostics on the sensors, and if necessary, call for engineering or technical support to perform sensor signal analyses. The extent that this can be performed from the office would be subject to the features of the system.
This system's coding of system errors and warnings by type is of great benefit to the analyst. However, a detailed analysis of the records displayed in Figure 51 would still indicate a sensor problem even in the absence of the warning flag.
Figure 52 displays a plot of a day's Class 9 steer axle right and left wheel weight averages by hour of day. There is an obvious problem in that the system is generating right and left weights being very different, which is attributable to either improper calibration factors or one of the two sensors generating erroneous weight outputs. This situation will be discussed in detail in SECTION 5. The purpose of this analysis is to confirm that the right weigh sensor is outputting a consistent weight whereas the left weigh sensor is not. Between the hours of 6 AM and 6 PM the left sensor's weight outputs drop dramatically, suggesting that the sensor is at times generating only partial weights. This problem, in itself, is not related to calibration factor values. Although the Class 9 volume is at its highest during this same time frame, it is doubtful that the drop in the sensor's steer axle weights is due to the sensors having a "recovery time" problem, given that steer axle hits are relatively well separated from other axle hits. The problem would also not be caused by a large number of the Class 9s riding the shoulder stripe, which would only affect the right sensor's weights. Consideration might be given to the daytime and nighttime difference in temperature or moisture. Regardless, this plot points out a sensor problem and the need for further investigation of the sensor's operation. This type of check should be performed for additional days to see if the pattern is consistent.
Figure 53 displays a snapshot from a portion of the spreadsheet used to generate the Figure 52 graph with the "AX1LT" field sorted for ascending values. Many of the "AX1LT" weights are significantly less than the "AX1RT" weights even though the following tandem's "AX2LT" and "AX3LT" weights are reasonably close to their right wheel counterparts. The tandem's left wheel weights are generally somewhat lower than the right wheel weights, but based upon the consistency this is probably due for the most part to improper calibration factors.
Figure 54 displays a similar screen shot of the spreadsheet but with the "AX1RT" field sorted for ascending values. As is obviously apparent, the right weigh sensor is not displaying the partial weight problem noted for the left sensor.
It would be a virtual impossibility for a Class 9's left steer axle wheel to only partially hit the left sensor when the trailing tandem's left wheels fully hit the same sensor, as displayed in Figure 55.
The contractor performing data QC and system monitoring for the LTPP Specific Pavement Study (SPS) Traffic Pooled Fund study developed several procedures, utilizing Excel, for analyzing performance of the individual weigh sensors. The spreadsheet is set up to import individual vehicle records from data files that contain specified vehicle types included in Class 9, as well as the vehicles in Classes 10 through 13. This import process is described in Appendix C. For the data files used, columns A through AN of the spreadsheet are populated with the data elements for each imported vehicle record. Columns A through AM include the data elements required by the LTPP Model Specifications and column AN provides for a "Vendor Specific Optional Field". Although various analyses can be performed with this data utilizing only filter and sort procedures, several additional features were added to the spreadsheet's template for automating certain analyses, including the following.
Figure 56 displays the spreadsheet's calculated fields. An "X" flag is displayed in Column AO, labeled "IMBALANCE", for any record for which the vehicle meets the criteria for "Invalid Measurement" in accordance with LTPP's model specifications. The two conditions for this calculated field are:
There may be different interpretations as to how "...a difference of 40 percent or more" is calculated, but the intent is that the recorded weight of the lighter of an axle's right and left wheels must be at least 60 percent the weight of the heavier wheel for the axle weight to be deemed a valid measurement.
The actual wheel weight data in each record are used to determine if the "Invalid Measurement" criteria are met, regardless of whether or not a flag was assigned by the system. Cell AQ1 provides for user input of the percent difference value to utilize for condition (1) and Cell AS1 provides for user input of the wheel weight threshold to utilize for condition (2) of the specification. The analyst can experiment and play "what if" with these two values to determine what works best for each site. The default values may work well for sites which have a high percentage of trucks with loaded trailers, but may flag far too many vehicles at sites that have a high percentage of trucks with empty or very light trailers. For the spreadsheet displayed in Figure 55 the user has changed the condition (2) default value from 2.0 to 3.0. It must be remembered that the intent of the Invalid Measurement flag is to identify vehicles appearing to have one or more wheels that did not fully hit the appropriate weigh sensor. As was discussed following Figure 42, an empty trailer's right and left wheel weight difference is insignificant from both axle weight and gross weight perspectives.
For a site that has a large number of trucks with empty trailers, if vehicles flagged by a system as having potentially erroneous weights due to the right versus left axle weight imbalance are automatically discarded from weight reporting by the analyst, then many legitimately weighed vehicles may be discarded. This might drastically skew data utilized for both weight violation and loading analyses purposes. It is strongly recommended that for a system that has features allowing the analyst to program parameters for assignment of right versus left imbalance flags that the system be programmed not to flag vehicles that are obviously empty and have minor right versus left imbalances in terms of weights, not just percentage, for the trailer axles. Simply increasing the minimum wheel weight threshold from 2.0 kip to 3.0 kips might significantly decrease the percentage of vehicles flagged as meeting the criteria for Invalid Measurement.
Columns AT through AX in Figure 56 display the ratio of the right versus left wheel weight for each of axles 1 through 5 for each record. This provides the analyst with a "quick look" at each axle's ratio for the vehicles with flags in the "IMBALANCE" field. If all of the axles have a significant imbalance on the same side it is a good indication that the vehicle may not have been tracking well within the lane lines. If the AX1 imbalance is significant whereas the other four axles look normal, this is typically an indication that a sensor has not reported an accurate weight. If one of a tractor's tandem axles displays a significant imbalance whereas the other does not, this is also an indication that a sensor has not reported an accurate weight.
Figure 57 displays a screen shot of the spreadsheet's table that lists the counts and percentages of the sample's Class 9 vehicles flagged as having Invalid Measurement weights by GVW distribution. As used for this analysis, these flagged vehicles meet the criteria "Invalid Measurement" discussed following Figure 56. This table makes it evident that for this site Invalid Measurement weights are exhibited by lighter vehicles much more than by the heavier vehicles. The lighter the trailer, the more subject it is to effects of bouncing and crosswinds. If the heavier vehicles start to exhibit an increase in the percentage of Invalid Measurement weights, it might well be an indication that one of the sensors is starting to malfunction.
For any data analyst desiring to create spreadsheets with the enhanced analyses features, as displayed in Figures 56, 57, and 58, Excel ASCII Import workbooks and documentation are provided online at www.QualityWIM.com. It is noted that this spreadsheet also includes the additional calibration monitoring analyses features that will be discussed in SECTION 5, for Figure 68through Figure 72.
Figure 58 displays a snapshot of the spreadsheet's table showing summaries of counts and percentages utilizing the individual Axle 1 through 5 imbalance ratios from the Figure 56 spreadsheet (Columns AT through AX). The purpose of this table is not to identify the extent of the vehicles meeting the criteria for Invalid Measurement but to identify any pattern that might suggest one of the following:
The analyst can enter the desired percentage value to ascertain the threshold for determining what right versus left weight difference constitutes an axle imbalance. The percentage of imbalances for Axle 1 should always be quite low (unless the trucks at a particular site do travel with their right wheels on the shoulder stripe). On the other hand, at a site with a high percentage of empty trucks the imbalance percentage for Axles 4 and 5 might be quite high (as in the example, using a 25 percent threshold).As the analyst becomes familiar with the different axle weight imbalance patterns, he or she will be able to identify the more subtle sensor problems even when a significant percentage of vehicles are not flagged as meeting criteria for Invalid Measurement.
Is it really necessary to go through the effort of performing such extensive analyses on the right versus left weigh sensor outputs? For some sites or system types perhaps not, but in the absence of these analyses a weigh sensor may be intermittently reporting weights that are inaccurate but too subtle to be noticed by means of less extensive data QC procedures. These types of analyses can also be quite useful in determining whether sensor outputs which appear to be inaccurate estimates of static weights are caused by actual sensor malfunction or by conditions related to truck operating characteristics or crosswinds. Once a spreadsheet or database program has been set up to automatically produce the types of information shown in these examples, it takes very little effort to make quick checks to ensure a sensor's output is not changing. Additional procedures for monitoring individual sensor outputs will be addressed in SECTION 5.
It is noted that for most WIM sites the Class 9 is the predominant truck class and that the steer axle wheel weights are much less affected by a Class 9's loading than the wheel weights of its other axles. As such, many of the extensive data analyses utilizing individual truck records focus on the Class 9 vehicles (particularly the 3S2). In addition, the Class 9 steer axle weights are also a focus of analyses regarding individual sensor weight outputs.
There are many extensive analysis procedures that an analyst can perform, either on routine or ad hoc bases, other than those used in these few examples. Note that such analyses are time consuming and require knowledge of each site's traffic and data characteristics, knowledge of a spreadsheet or database program, and even some imagination. However, these analyses can be extremely beneficial in identifying, isolating, quantifying, and diagnosing a system's data problems.