Results of Oversampling Roughness Data on the Kansas Highway System

1
Results of Over-sampling Roughness Data on the
Kansas Highway System

• Road Profilers Users Group
• (RPUG)
• October 23, 2001

2
Over-sampled Data

• What is over-sampling?
• Multiple samples from same highway segment.
• Blind, statistically significant portion.
• Why is it useful?
• Estimates system error / repeatability
• Answers the question The IRI is 96 in/mile, with
  a 95 confidence interval of X.

3
KDOTs Over Sampling System

• Initiated in the 1997 Road Condition Survey
• Technicians were told to leave data collection
  system on at all times, even when backtracking.
• Resulted in
• No substantial time penalty in collecting data
• gt10 for all years since 1997

4
Kansas Over-sampled IRI Data 2001
5
Show years and number of miles of redundant data.
6
Distribution of 2001 IRI Data
7
Not Calibration Data!!!

• Calibration tests equipment - NOT SYSTEM
  VARIABILITY
• Controlled test on same portion of roadway
• Correlated results (not blind)
• For KDOT, calibration is performed weekly
• Typical years Calibration Results
• lt 3 For all weeks

8
Normalizing the Data

• All data is scaled to a mean of 100 IRI
• Example
• Sampled Data
• D1 40 in/mile D2 60 in/mile Average(D)
  50 in/mile
• Normalized Data
• D1 80 in/mile D2 120
  in/mile Average(D) 100 in/mile

9
Estimating Standard Deviation

• Estimating s for a small sample n, spop
  saverage/d2
• Combing estimates of s
• Estimating the error in s

10
Redundant Data Scatter Plot - 1997
11
Summary Statistics for All Years
12
Normal Distribution
68
95
s
s
2s
2s
Mean
13
Summary Statistics
14
Mean IRI Quartile Analysis
15
Analysis by Vehicle
16
Analysis by Pavement Type
17
Analysis by Direction
18
Analysis by Distress State
19
Sources of Variation Checked

• Sources
• Equipment
• Direction
• Pavement Type
• Other types of distress
• Level of roughness
• Year

• Effect or Correlation
• Some, depends on year
• Some, depends on year
• Some, 2000 2001
• Some, depends on year
• Scales with mean IRI
• Decreasing over 5 years

20
Error/Variability Sources or Why so much error?

• Equipment
• Variation in the roadway / wheelpath
• Lateral variation such as rutting and cracking
• Methods of calculation
• ! Consistent within the equipment
• Roadway changes
• Curling / dedris / etc.
• Other ?????

21
Inferences for one sample of IRI
68
95
72
88
64
96
80
IRI in inches per mile
22
Questions for one IRI Sample

• Is 9-14 standard deviation ..
• sufficient for network management, allocation of
  funds.
• Most likely YES
• sufficient to support project portfolio
  selection?
• ????????
• sufficient for new construction acceptance?
• Probably NOT

23
Conclusions

• IRI is functional, but not all that precise
• Variation in a single sample is 11
• Systematic over-sampling is only method for
  determining system variability
• Is it time to break from the old methods
• Spectrum analysis
• Robust error checking
• Full road scanning