building a wim data archive for improved modeling, design, and rating
DESCRIPTION
Building a WIM Data Archive for Improved Modeling, Design, and Rating. Christopher Monsere Assistant Professor, Portland State University Andrew Nichols Assistant Professor, Marshall University. NATMEC 2008, Washington, D.C. August 6, 2008. Outline. Data Almanac PORTAL - PowerPoint PPT PresentationTRANSCRIPT
Building a WIM Data Archive for Improved Modeling, Design, and Rating
Christopher Monsere Assistant Professor, Portland State University
Andrew Nichols Assistant Professor, Marshall University
NATMEC 2008, Washington, D.C. August 6, 2008
Outline Data Almanac PORTAL WIM Archive Quality Control Sample Uses of the WIM Archive
Sensor Health and CalibrationUses for LFR and MPDEGSystem Performance and InformationPlanning Data
2
DATA ALMANACDescribe the data source
3
Data Almanac 22 reporting WIM sites
All upstream of weigh stationsAll are CVISN sitesApril 2005 - March 2008
2007 - 12,054,552 trucks
Intermittent data outages and problemsData quality and accuracy?
4
Data Almanac These WIM sites provide
Axle weightsGross vehicle weightSpacingVehicle classUnique transponder numbers
5
7
Axle Weight Sensors Single load cells Sensors weigh vehicles
traveling at normal highway speeds
Weight measurement affected by many factors Site characteristics Environmental factors Truck dynamics
High SpeedWIM Sorting Site
(with Overhead AVI)
NotificationStation(in-cab)
TrackingSystem
RampSorterSystem
StaticWeighing
1 mileR.F. Antenna
Trucks BypassWIM Layout
9
Sorting Lanes WIM installed on
entrance ramp to determine if vehicle is close to legal weight
Legal trucks directed to exit facility without being weighed statically
Exit LaneWeigh Lane
10
Indiana WIM Site Layout
InductiveLoop
InductiveLoop
PiezoSensor
Load CellSensor
Loops – vehicle presence & speed measurement Load Cell – weight & speed Piezo – speed & axle spacing
11
WIM Classification Algorithm Portion related to 5-axle
vehicles shown Works like a sieve Min/Max thresholds for
# of axles axle spacing axle weight gvw
Currently only use axle spacing
Vehicle Type 19 20 21 22 23 Vehicle Class 7 9 9 11 9 # of Axles 5 5 5 5 5 Min GVW 0 0 0 0 0 Max GVW 221 221 221 221 221 1 Min Weight 3 3 3 4 3 1 Max Weight 50 50 50 50 50 1 Axle Marking s s s x x 1-2 Min Spacing 0 0 0 0 0 1-2 Max Spacing 40 40 40 14.2 40 2 Min Weight 0 0 0 4 0 2 Max Weight 50 50 50 50 50 2 Axle Marking x d d x x 2-3 Min Spacing 0 0 0 0 0 2-3 Max Spacing 5.8 5.8 5.8 40 40 3 Min Weight 0 0 0 4 0 3 Max Weight 50 50 50 50 50 3 Axle Marking x d d x x 3-4 Min Spacing 0 0 0 0 0 3-4 Max Spacing 5.8 40 40 40 40 4 Min Weight 0 0 0 4 0 4 Max Weight 50 50 50 50 50 4 Axle Marking x d x x x 4-5 Min Spacing 0 0 0 0 0 4-5 Max Spacing 5.8 5.8 11.7 40 40 5 Min Weight 0 0 0 4 0 5 Max Weight 50 50 50 50 50 5 Axle Marking x d x x x
Electronic screening
12
Three types of tagsHeavy Vehicle
Electronic License Plate (HELP)’s PrePass program
North American Pre-clearance and Safety System (NORPASS)
Oregon Green Light Program
All RF tags
State operated/developed; compatible with NORPASS
PrePassNORPASS
J. Lane, Briefing to American Association of State Highway and Transportation Officials (AASHTO), 22 February 2008
freight.transportation.org/doc/hwy/dc08/scoht_cvisn.ppt
PORTAL 101
What is PORTAL PORTAL - Postgresql database
13
QUALITY CONTROLDescribe Quality Control Metrics
14
15
Outline Summarize Existing Quality Control Metrics
Speed Accuracy Weight Accuracy
System-Wide Monitoring vs. Site-based Monitoring SQL ServerExtract Pivot Tables (.CUB)
Move Beyond Heuristic (sometimes visual) Rules to a Statistically Solid Decision Making Criteria Learn from Manufacturing World of Statistical Process
Control
16
Data Quality Assessment Obtain data from WIM sites (nightly or weekly) Upload individual records to SQL database Macroscopic view to identify potential problems Microscopic view to investigate causes
Differentiate between problems & random noise using statistical process control (individual lanes)
Schedule maintenance/calibration accordingly
17
Past Quality Control Hurdles Emerging WIM quality control metrics Enormous amounts of data Many vendors’ standard reports do not
provide level of detail necessary for effective quality control, particularly comparing multiple sites
Most DOTs do not have time or resources to perform QA/QC
19
Accuracy Metrics Illustrated
Vehicle Classification Axle Spacing
Speed
Axle WeightWheel weight
Accuracy Metric
Accuracy Metric
20
Speed Accuracy Metric Class 9 drive tandem axle spacing Currently used by many agencies Target values based on manufacturer
sales data Individual trucks
4.25-4.58 feetPopulation average
4.30-4.36 feet
21
2002 Sales Data Data obtained from 3 of top 6 truck
manufacturers Verified at Indiana WIM sites with accurate
speed calibration
Weighted Average = 4.33 feet
Modal Value (0.01)’ would be even better than average
22
WIM Speed Control Chart Class 9 Drive Tandem Axle Spacing Limits based on sales data and data from
sites with good calibration Limits same for all sites
Upper Control Limit (UCL) = 4.36 feetCenterline (CL) = 4.33 feetLower Control Limit (LCL) = 4.30 feet
23
Speed & Axle Spacing WIM calculates speed by measuring
vehicle time between two sensors
WIM speed used to calculate axle spacing based on time between consecutive axles
sensor
sensorwim time
distancespeed
wimaxleaxle speed * time distance
Calibration factor
24
Speed Accuracy Importance Used to calculate axle spacing
Vehicle classificationBridge formula compliance for enforcement
Used to apply weight calibration factors in some WIM systems
25
Why Unclassified? All axle spacings in this lane were
overestimated by ~15% The axle spacing 3-4 is exceeding the
defined thresholds in the vehicle classification file (40’ max)
Axle 3-4 Spacing30-35 feet typical
26
Weight Accuracy Metric Class 9 front axle weight
Individual trucks 8,000-12,000 lbs Population average 9,000-11,000 lbs
Currently used by many agencies This range is too big for detecting subtle drifts
Class 9 WIM Accuracy Metrics Steer Axle Weight
8.5 kips (GVW < 32 kips)9.3 kips (GVW 32-70 kips)10.4 kips (GVW > 70 kips)
Steer Axle Weight and Axle 1-2 SpacingLogarithmic relationshipSteer Axle Weight/Axle 1-2 Spacing
Steer Axle Wheel Weight Gross Vehicle Weight Distribution
28-32 kips unloaded70-80 kips loaded
Dahlin’s GVW Criteria
0%
2%
4%
6%
8%
10%
20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80 84 88
GVW Bin (kips)
Fre
quen
cy
Site 1 Site 2
Unloaded Peak 28-32 kips
Loaded Peak 70-80 kips
3rd Peak?
Class 9 GVW Distribution Modal distributions Look for peak mode shifts Trend identification difficult Very difficult to automate (QC)
Step function to find max GVW binsBin width affects precision
Mixture Models Fitting a continuous function to GVW
distribution – combination of normal distributions
Use algorithm to fit “mixture” of normal distributionsObtain mean and covariance for each
component
)(...)()()()( 22111
yfyfyfyfyf gg
g
iii
GVW Fit – 1 Component
0%
1%
2%
3%
4%
5%
20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
GVW Bin (kips)
Fre
qu
ency
0%
1%
2%
3%
4%
5%
20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
GVW Bin (kips)
Fre
qu
ency
1-Component
Mean 1 = 56 kipsCovariance 1 = 310 kipsProportion 1 = 100%
GVW Fit – 2 Components
0%
1%
2%
3%
4%
5%
20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
GVW Bin (kips)
Fre
qu
ency
0%
1%
2%
3%
4%
5%
20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
GVW Bin (kips)
Fre
qu
ency
2-Component
Mean 1 = 74 kipsCovariance 1 = 11 kipsMean 2 = 48 kipsCovariance 2 = 222 kipsProportion 1 = 33%Proportion 2 = 67%
GVW Fit – 3 Components
0%
1%
2%
3%
4%
5%
20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
GVW Bin (kips)
Fre
qu
ency
0%
1%
2%
3%
4%
5%
20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
GVW Bin (kips)
Fre
qu
ency
3-Component
Mean 1 = 75 kipsCovariance 1 = 9 kipsMean 2 = 52 kipsCovariance 2 = 193 kipsMean 3 = 30 kipsCovariance 3 = 11 kipsProportion 1 = 31%Proportion 2 = 57%Proportion 3 = 12%
34
WIM Speed/Axle Spacing Accuracy Common speed accuracy metric is
Class 9 drive tandem axle spacing No consensus on proper values Verified with manufacturer sales numbers
Individual trucks 4.25-4.58 feet
Population average 4.30-4.36 feet
SENSOR HEALTHDescribe from Quality Control Metrics About Sensor Health
35
36
37
38
LFR AND MPDEGDescribe Higgins Work
39
SYSTEM PERFORMANCEDescribe Sample PMs
40
Freight performance metricsAverage travel times on key corridorsTon-miles on each corridor by various
temporal considerationsOverweight vehicles on corridors by temporal
variation (measuring change)Seasonal variability in loading, routes, and
volumesPercent trucks with tags on each corridorOrigin-destination matrix
41
Freight performance metrics
42
Freight performance metrics
43
Using Federal Highway Administration (FHWA) / American Transportation Research Institute (ATRI) proprietary truck satellite data.
Farewell Bend
Emigrant HillWyeth
Next Steps
49
Acknowledgements Dave McKane and Dave Fifer, ODOT Kristin Tufte and Heba Alawakiel, PSU
50