HDM-4 Calibration
Dec 22, 2015
HDM-4 Calibration
2
• How well the available data represent the real conditions to HDM
• How well the model’s predictions fit the real behaviour and respond to prevailing conditions
Reliability of Results Depends On:
3
• Depends on Level of Calibration (controls bias)
• Depends on accuracy and reliability of input data (asset & fleet characteristics, conditions, usage)
• HDM-4 has proved suitable in a range of countries
• As with any model, need to carefully check output with good judgement
How Credible are HDM-4 Outputs?
4
3
• Input dataMust have a correct interpretation of the
input data requirementsHave a quality of input data appropriate for
the desired reliability of results
• CalibrationAdjust model parameters to enhance the
accuracy of its representation of local conditions
Approach to Calibration
5
• Need to appreciate importance of data over calibration
• If input data are wrong why worry about calibration?
D ata
C alib ra tion
'The D epth o f the S ea andthe H e ight o f the W aves'
Data & Calibration
6
• Road User EffectsPredict the correct magnitude of costs and
relativity of components - dataPredict sensitivity to changing conditions -
calibration
• Pavement Deterioration & Works EffectsReflect local pavement deterioration rates and
sensitivity to factorsRepresent maintenance effects
Calibration Focus
7
Un-calibrated Calibrated
Actualdeterioration
Model
We attempt tominimize the
"mistake"
Time
Ext
ent
of
Def
ect
(%)
Actual Progression
Pre
dic
ted
Pro
gre
ssio
n
Actual Progression
Pre
dic
ted
Pro
gre
ssio
n
Estimating Calibration Coefficients
8
General PlanningQuick Prioritisation
Preliminary Screening
Coarse Estimates
Field Surveys
ExperimentalSurveys and
Research
Desk Studies
ResourcesRequired
Time Required
Weeks
Months
Years
Limited Moderate Significant
Project AppraisalDetailed Feasibility
Reliable Estimates
Research andDevelopment
Hierarchy of Effort
9
• Level 1: Basic ApplicationAddresses most critical parameters ‘Desk Study’
• Level 2: CalibrationMeasures key parametersConducts limited field surveys
• Level 3: AdaptationMajor field surveys to requantify relationshipsLong-term monitoring
Calibration Levels
10
• Required for ALL HDM analyses
• Once-off ‘set-up’ investment for the model
• Mainly based on secondary sources
• Assumes most of HDM default values are appropriate
Level 1 - Basic Application
11
• Makes measurements to verify and adjust predictions to local conditions
• Requires moderate data collection and moderate precision
• Adjustments entered as input data, typically no software changes
Level 2 - Calibration
12
• ComprisesStructured research, medium termAdvanced data collection, long term
• Evaluates trends and interactions by observing performance over long time period
• May lead to alternative local relationships/models
Level 3 - Adaptation
13
• Calibrate over full range of values likely to be encountered
• Have sufficient data to detect the nature of bias and level of precision
• High correlation (r^2) does not always mean high accuracy: can still have significant bias
• Primary aim: minimize bias (mean observed values / mean predicted values)
Important Considerations
14
Observed
Pre
dict
ed
Data
Observed = Predicted
Low BiasLow Precision
Observed
Pre
dict
ed
Data
Observed = Predicted
Low BiasHigh Precision
Observed
Pre
dict
ed
Data
Observed = Predicted
High BiasHigh Precision
Observed
Pre
dict
ed
Data
Observed = Predicted
High BiasLow Precision
A B
C D
Bias and Precision
15
Observed
Pre
dic
ted
Data
Observed = Predicted
Rotation andTranslation
Translation
Rotation
Observed
Pre
dic
ted
Data
Observed = Predicted
Translation
Translation
Observed
Pre
dic
ted
Data
Observed = Predicted
Rotation
Rotation
BA
C
Calibration Adjustments
16
• Used to correct for bias
• Two types of factors
Rotation (CF = Observed/Predicted)
Translation (CF = Observed - Predicted)
• Rotation factors adjust the slope
• Translation factors shift the predictions vertically
Correction Factors
17
HDM-4 Road DeteriorationCalibration Factors
All relationships have a calibration factor - ‘K’ factor
Used to adjustpredicted to observed
18
ICA = Kcia{a0 exp[a1SNP + a2(YE4/SNP2)]}
CalibrationFactor
ModelCoefficients
Initiation of Cracking
Typical Relationship
19
CalibrationFactor
DeteriorationModel
Kddf Drainage FactorKcia All Structural Cracking - InitiationKciw Wide Structural Cracking - InitiationKcpa All Structural Cracking - ProgressionKcpw Wide Structural Cracking - ProgressionKcit Transverse Thermal Cracking - InitiationKcpt Transverse Thermal Cracking - ProgressionKrid Rutting - Initial DensificationKrst Rutting - Structural DeteriorationKrpd Rutting - Plastic DeformationKrsw Rutting - Surface WearKvi Ravelling - InitiationKvp Ravelling - ProgressionKpi Pothole - InitiationKpp Pothole - ProgressionKeb Edge BreakKgm Roughness - Environmental CoefficientKgp Roughness - ProgressionKtd Texture Depth - ProgressionKsfc Skid ResistanceKsfcs Skid Resistance - Speed Effects
Road Deterioration Calibration Factors
20
Crack Initiation
0
10
20
30
40
50
60
70
80
90
100
0 5 10 15 20
Years
Per
cen
t A
rea
of
Cra
ckin
g
Kci = 1.00 Kci = 1.80 Kci = 0.55
Cracking Initiation Calibration
21
Crack Progression
0102030405060708090
100
0 5 10 15 20
Years
Per
cent
Are
a of
Cra
ckin
g
Kcp = 1.0 Kcp = 2.0 Kcp = 0.4
Cracking Progression Calibration
22
• Simulation of Past Since Construction
take sample of roads with historical data (traffic, design, etc.)
simulate with HDM-4 the deterioration from construction time to current age
compare the simulated results with actual road condition at current age
deal with the uncertainty regarding the road conditon at construction time
Road Deterioration Calibration (1)
23
• Simulation from Two Points in Time
take sample of roads with road condition data available for two years (e.g. roughness measurements surveyed in two different years)
simulate with HDM-4 the deterioration from the first year to the second year
compare the simulated results with the actual road condition at the second year
Road Deterioration Calibration (2)
24
Kazakhstan Calibration Example
Roughness surveys three years apart
Without Calibration Scenario
0.00
2.00
4.00
6.00
8.00
10.00
12.00
14.00
0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00
Predicted Roughness Values (IRI, m/km)
Obs
erve
d R
ough
ness
Val
ues
(IR
I, m
/km
)
Bias = Mean Observed / Mean Predicted = 1.14
Roughness Environmental Factor = 1.0Cracking Initiation Factor = 1.0
With Calibration Scenario
0.00
2.00
4.00
6.00
8.00
10.00
12.00
14.00
0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00
Predicted Roughness Values (IRI, m/km)
Obs
erve
d R
ough
ness
Val
ues
(IR
I, m
/km
)
Bias = Mean Observed / Mean Predicted = 1.03
Roughness Environmental Factor = 1.5Cracking Initiation Factor = 0.6
25
• Controlled Studies
collect detailed data over time on traffic, roughness, deflections, condition, rut depths, etc.
sections must be continually monitored
long-term (5 year) commitment to quality data collection
Road Deterioration Calibration (3)
26
• HDM-III has about 80+ data items and model parameters; HDM-4 has more.
• Sensitivity of each item has been classified by sensitivity tests
• Simplify effort for less-sensitive items
What to Focus On?
27
Impact SensitivityClass
ImpactElasticity
High S-I >0.50
Medium S-II 0.20-0.50
Low S-III 0.05-0.20
Negligible S-IV <0.05
Sensitivity Classes
28
SensitivityClass2/
ImpactElasticity
Parameter Important forTotal VOC3/
Parameter Important forVOC Savings4/
S-I > 0.50 kp - parts model exponentNew Vehicle Price
kp - parts model exponentNew Vehicle PriceCSPQI - parts modelroughness termC0SP - parts model constantterm
S-II 0.20 - 0.50 RoughnessE0 - speed bias correctionAverage Service Life AverageAnnual UtilisationVehicle Weight
E0 - speed bias correctionARVMAX - max. rectifiedvelocityCLPC - labour model exponent
S-III 0.05 - 0.20 Aerodynamic Drag CoefficientBeta - speed exponentBW - speed width effectCalibrated Engine SpeedCLPC - labour model exponentC0SP - parts model constanttermCSPQI - parts modelroughness termCrew/Cargo/Passenger CostDesired SpeedDriving PowerEnergy Efficiency FactorsFuel CostHourly Utilisation RatioInterest RateProjected Frontal Area
Beta - speed exponentVehicle Age in kmC0LH - labour model constanttermLabour CostHourly Utilisation RatioBW - speed width effectsNumber of tires per VehicleNew tire CostLubricants CostCrew/Cargo/Passenger CostVehicle WeightNumber of Passengers
S-IV <0.05 All Other Variables All Other Variables
Sensitivity Classes
29
Sensitivity Impact Parameter Outcomes Most Impacted Class Elasticity Pavement
Performance Resurfacingand Surface
Distress
EconomicReturn on
Maintenance S-I > 0.50 Structural Number 2/
Modified Structural Number2/
Traffic Volume
Deflection3/
Roughness
S-II 0.20 - Annual Loading
0.50 Age
All cracking area
Wide cracking area
Roughness-environment factor
Cracking initiation factor
Cracking progression factor S-III 0.05 - Subgrade CBR (with SN)
0.20 Surface thickness (with SN)
Heavy axles volume
Potholing area Rut depth mean Rut depth standard deviation Rut depth progression factor Roughness general factor
S-IV < 0.05 Deflection (with SNC) Subgrade compaction
Rainfall (with Kge) Ravelling area Ravelling factor
Sensitivity Classes
30
IQL-1
P erfo rm ance
S truc tu re C ond ition
R ide D is tressFric tio
n
IQL-5
IQL-4
IQL-3
IQL-2
System Perform anceM onitoring
P lanning andPerform anceEvaluation
Program m eAnalysis orD etailedP lanning
Project Level orD etailedProgram m e
ProjectD etail orR esearch
HIGH LEVEL DATA
LOW LEVEL DATA
Information Quality Levels
31
• IQL-1: Fundamental Researchmany attributes measured/identified
• IQL-2: Project Leveldetail typical for design
• IQL-3: Programming Levelfew attributes, network level
• IQL-4: Planningkey management attributes
• IQL-5: Key Performance Indicators
Information Quality Levels
32
IQL-2 IQL-2B IQL-3 IQL-4Lane roughness (m/km IRI) Roughness (6 ranges) Ride quality (class)All Cracks Area (% area) Cracking (score, or
Universal CrackingIndex, UCI)
Wide Cracks Area (% area)Transverse thermal cracks
(no./km)Ravelled Area (% area)
Potholes Number(units/lane-km)
Disintegration (score) Surface Distress Index(SDI)
Pavement Condition(class)
Edge-break area (m2/km)Patched Area (% area)Rut Depth Mean (mm) Deformation (score)
Rut Depth Standard Dev.(mm)
Macro-texture depth (mm) Surface texture (class)Skid Resistance (SF50) Friction (class) Surface Friction (class)
Adapting Local DataRoad Condition
33
EstablishingReliable Input
Data40%
ModelCalibration
10%
Treatments,Triggers and
Resets20%
Running HDM-410%
Verification ofOutput20%
Time Spend on Different Phases of Analysis
34
23
• Yes, if sufficiently calibrated
• HDM-4 has proved suitable in a range of countries
• As with any model, need to carefully scrutinize output against judgement
• If unexpected predictions occur, check: Data usedCalibration extentCheck judgment of the expert
Can We Believe HDM-4 Output?
35
For Further Information
•A guide to calibration and adaptation
•Reports on various HDM calibrations from:www.lpcb.org