You are on page 1of 20

4th International Conference on Managing Pavements (1998)

THE CALIBRATION AND USE OF HDM-IV PERFORMANCE MODELS IN A PAVEMENT MANAGEMENT SYSTEM

Gustav T Rohde*, Fritz Jooste**, Elzbieta Sadzik*** and Theunis Henning***

* ** ***

Africon, P O Box 905, Pretoria, South Africa. Transportek, CSIR, P O Box 395, Pretoria, South Africa. Gauteng Provincial Government, Department of Transport & Public Works, Private Bag X3, Lynn East, Pretoria, South Africa.

Abstract Performance modelling has become a key element of a Pavement Management System (PMS). The Gauteng Province of South Africa has established 36 Longterm Pavement Performance (LTPP) sections which are monitored annually to improve and calibrate PMS performance models. This paper describes how the LTPP sections were set out, monitored, and how the collected data influenced the pavement performance models captured in the PMS. Initially the HDM-III pavement performance models were calibrated and captured in the PMS. Subsequently improved models have been proposed for inclusion in HDM-IV. These proposed models are evaluated on the LTPP data. It is also shown how the inclusion of the calibrated performance models has influenced PMS decisions. 1. INTRODUCTION

In southern Africa, highway agencies are undergoing a fundamental change in activities. The emphasis has changed from cost-effective construction of new pavements to maintenance of the existing road network. Effective decisions regarding the timing and type of maintenance activities require predictive models to forecast future road conditions and the consequences of deferred maintenance. The need for predictive models becomes even greater in the current scenario where available funds are inadequate and where road maintenance budgets must be justified. The need for the inclusion of performance prediction models into pavement management systems (PMS) became apparent in the early nineties. At that time (and even today) no paved road performance prediction models were available for direct application in southern Africa. Based on initial investigations by Kemp (1989), Kannemeyer and Visser (1993) the World Banks Highway Design and Maintenance Model, HDM-III (Watanatada et al, 1987) appeared best suited for local use. The HDM-III models were derived from a broad empirical base, are versatile, and can be adapted to local conditions with relatively little effort. The first country in southern Africa to actively pursue the inclusion of the HDM-III pavement performance models into their road management system was Botswana (Pinard, 1992; VWL-TTCS, 1994). Subsequently, locally calibrated HDM-III
TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

performance models have been included and used in the pavement management systems of several countries and provinces in southern Africa. This paper describes how the models were calibrated and included in the PMS of the Gauteng Province of South Africa. In 1993 the province decided to verify, and if required, calibrate the HDM-III pavement performance models for use in the provincial PMS. For this purpose a factorial of Long-term Pavement Performance monitoring (LTPP) sections were selected and monitored annually. The calibration was done by comparing predicted with actual pavement performance on the 36 LTPP sections. It was decided to calibrate and implement the models as soon as possible and to improve the models further as deficiencies in the models became evident while in use. A procedure was also implemented to monitor the sections annually in order to improve the PMS predictions. The annually collected data has recently been used to evaluate some of the performance models proposed for HDM-IV. Initial findings are reported in this paper. 2. ESTABLISHMENT OF LTPP SECTIONS FOR PMS MODELLING

After selecting the HDM-III models for use in the provincial PMS it was decided that the accuracy and suitability of the models should be established. Accordingly a factorial of LTPP monitoring sections was established. 2.1 Selection of calibration sections Pavement sections were selected to represent pavement conditions found in the province. The selection criteria consisted of: 3 2 2 3 Traffic Levels (Low, Moderate, High) Base Types (Granular, Stabilised) Environments (Dry, Moderate) Conditions (Poor, Moderate, Good)

36 Total Sections (3 x 2 x 2 x 3) For each cell in the matrix a 500 metre long pavement section was selected. This section was divided into 20 subsections (10 in each direction) on which detailed distress data was collected. The selected sections covered a wide range of conditions as evident from Table 1 below: Table 1: Range of the LTPP section characteristics Parameter
Number of Sections Number of Subsections Length of Section Pavement Age (years) Current Surface Age (years) Annual Rainfall (m/year) Cumulative Traffic Load (10 ESALs)*
6

Range
36 720 10 x 50m x 2 directions 6 2 0.48 0. 05 40 19 0.81 3.43

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

Traffic Loading Rate (106 ESALs/lane/year) Traffic Volume (veh/day) Total Surface Thickness (mm) Total Pavement Thickness (mm) Maximum FWD Deflection (mm) Benkelman Beam Deflection (mm) Modified Structural Number (inch) Insitu Subgrade CBR (%)

0.003 772 10 280 0.15 0.173 2.3 3

0.22 9894 100 850 0.8 1.180 6.5 80

ESALs - Equivalent Single Axle Load

2..2 Selection of representative test positions Pavement materials and their performance vary considerably even throughout a "uniform" section. Since it was not feasible to conduct many Dynamic Cone Penetrometer tests (DCPs) and testpits per uniform section, measured surface deflections were used to find a representative test position per calibration section. The pavement deflections collected at 50 m intervals were analysed to determine the average maximum deflection. A single test position where the measured peak deflection is close to the calculated average peak deflection was then selected as representative (See Figure 1).

700
MAXIMUM DEFLECTION (um)

600
AVERAGE DEFLECTION

500 400 300 200 100 0 0 0.1 0.2 0.3 0.4 0.5 0.4 0.3 0.2 0.1
STATION (km) REPRESENTATIVE TEST POSITION

Figure 1: Selection of representative test position from measured deflections 2.3 Definition of pavement condition The present pavement condition for each calibration section was monitored in detail. The distresses, in HDM-III and TMH9 (TMH9, 1992) terms, were recorded in 50 m subsections. For the visual survey a detailed evaluation procedure was developed (Van Zyl 1994). Data was collected systematically in as much detail as possible. The visual evaluation form used is shown in Figure 2. Road roughness was measured and rutting was recorded using both a 2 m and 1.2 m straight edge.
TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

LTPP VISUAL EVALUATION


ROAD NR:__________ SEGMENT: KM:from _______ DISTRICT: CLIMATE: to ___________ DIRECTION: +/ASSESSOR:________

DATE:____________

ROAD WIDTH:__________

___________________________________________________________ Very Wet Wet Moderate Dry TERRAIN: Mount Rolling Flat

SURFACING DEGREE 1 SURFACE DISINTEGRATION Ravelling : Ravelling : Aggr Disint Mech Damage Shoving BINDER CONDITION BLEEDING/FLUSHING Active Loss Stable 2 3 4 5 1 EXTENT 2 3 4 5 % AREA AVE DIA AVE DEPTH NR

STRUCTURALL REFLECTIVE CRACKING Block/Stab Cracking < 0.5 0.5-2.5 > 2.5 STRUCTURAL CRACKS Longitudinal Cracks Edge Cracking Wheel Track (road) Irregular (s/grade) Crocodile Cracks General Wheel Track THERMAL CRACKS TRANSVERSE CRACKS OTHER CRACKS Parabolic (shift of seal) Map (irregulat block) Length(m) 1 DEGREE 2 3 4 5 1 EXTENT 2 3 4 5 NARROW % <3mm WIDE % >3mm (spalling) SEALED (%)

PUMPING % AREA PATCHING Skin Patch Structural POTHOLING Potholes Edge Breaks Delamination AVE DIA AVE DEPTH NR

FUNCTIONAL DEGREE 1 RIDING QUALITY SKID RESISTANCE SURFACE DRAINAGE UNPAVED SHOULDERS COMMENTS REASON 2 3 4 5 1 EXTENT 2 3 4 5

RUTTING 1.2m STRAIGHT EDGE 2.0m STRAIGHT EDGE

10

15

20

25

30

35

40

45

50

Figure 2: Evaluation form used on TLPP sections

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

2.4 Definition of pavement strength The pavement strength in the HDM models is represented by the modified structural number (SNC). On the monitoring sections the modified structural number was determined for each of the 720 subsections. This was done using a procedure to determine SNC from FWD deflections documented elsewhere (Rohde, 1994). These were also compared to DCP results and traditional methods, utilising layer thicknesses and material types. 2.5 Definition of historic traffic The traffic since construction on each of the calibration sections was carefully determined. The available traffic data (manual and visual) and information from Weigh in Motion (WIM) studies were combined to calculate the past traffic loads. Data from the permanent counting stations of the province was also used. 2.6 Definition of maintenance history and environmental conditions The construction and maintenance history for each calibration section was determined from departmental records. Routine maintenance records were not available but generally all potholes are sealed as soon as they developed. The rainfall statistics recorded at nearby rain stations were also analysed to characterise the environment. Since the start of the monitoring all maintenance actions on the sections have been carefully recorded. 3. USE OF LTPP DATA TO IMPROVE PMS MODELS

The LTPP sections established by the province have been monitored annually since 1993. This growing database of performance, traffic and maintenance is becoming a valuable tool for the selection, calibration and improvement of the PMS models. 3.1 Calibration of the HDM-III performance models For the initial calibration of the HDM-III predictive models no time series data was available and it was decided to use condition data from the survey and conduct a subsequent back-analysis of the HDM-III calibration coefficients. This implied that the fundamental form of the individual predictive relationships was accepted, and that scaling rather than alterations in the models was foreseen. The calibration coefficients were obtained by determining the scaling factors that minimise the difference between the model predictions and actual monitored performance.
CRACKING RUTTING

RAVELLING

POTHOLES

ROUGHNESS

Figure 3: Interaction of the distress types as modelled in HDM-III


TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

The HDM pavement performance models which were calibrated are described in detail elsewhere (Watanatada et al 1987, Paterson 1987). HDM-III performance models exist for six distress modes as defined in Table 2. The complex interaction of these variables, which is one of the reasons why the models were originally selected, is graphically shown in Figure 3. Table 2: Distresses modelled by HDM-III Distress All Cracking Description The area of narrow and wide cracking (greater than 1 mm in width) as a percentage of the total carriageway area. The area of wide cracking (greater than 3 mm in width) as a percentage of the total carriageway area. The total area ravelled as a percentage of the total carriageway. The total area of open potholes (minimum depth of 25 mm, diameter of 150 mm) as a percentage of the total carriageway. The mean and standard deviation of rut depth as measured in mm under a 1.2m straight edge. Roughness in IRI as defined by Sayers et al (1986)

Wide Cracking Ravelling

Potholing

Rutting

Roughness

On each of the 36 calibration sections the performance since first construction was predicted using the HDM-III performance models. The prediction was compared to the actual observed performance and the HDM-III calibration coefficients were adjusted to minimise the difference between the predicted and measured distress. This process was described and documented by Rohde (1994). The coefficients resulting from the first calibration are listed in Table 3. Table 3: Calibration coefficients resulting from the first HDM calibration Calibration Order 1 2 3 4 5 6 7 Model Calibration Coefficient 0.8 0.2 1.2 1.0 0.5 0.3 1.0

Crack Initiation kci Crack Progression kcp Ravelling kri Pothole Progression kpp Rut Depth Progression krp Roughness Progression (Environment) kge Roughness Progression (Traffic) kgp

The calibrated HDM-III pavement performance was subsequently captured in the PMS and used very effectively. The LTPP sections were annually monitored to ultimately provide time-series data to improve the PMS models.
TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

3.2 Initial investigations into the use of the HDM-IV models In 1993 an international study was launched to improve HDM-III. A component of this study consisted of the review, and where required, revision and improvement of the HDM pavement performance models (NDLI, 1995). The proposed changes were subsequently reviewed by an international working party for inclusion into HDM-IV by the University of Birmingham. At the time this paper was written the models finally selected for HDM-IV had not yet been published. However, the proposed models were evaluated and calibrated on the LTPP data (Africon & Jooste, 1997). This calibration was restricted to thin surface seals, since the data for performance of asphalt surfacings were limited. 3.2.1 Cracking NDLI (1995) proposed three changes to the cracking models in HDM. These include: Modelling of 3 crack mechanisms - structural, reflective and thermal cracking. A single crack index rather than all and wide cracking as in HDM-III. Using traffic based crack progression models rather than time dependent models. 3.2.1.1 Calibration of structural crack initiation with the LTPP data The equation proposed for crack initiation in HDM-IV is: YE 4 ICX = a o .exp a1 . (1) SNC 2 .............................................................. where: ICX a0, a1 YE4 SNC = = = = time to structural crack initiation, in years; calibration parameters; traffic loading rate, in millions of ESAs per year; modified structural number.

For the calibration, structural cracking was defined as the sum of all crocodile and longitudinal wheel path cracks. For each 500 m section, the percentage structural cracking was determined by combining the average percentage cracked area recorded on the 20 subsections. Calibration was performed by minimisation of the following error function: Err = wi .(ICX - SAGE )2 ..................................................................... (2) where: Err = SAGE = wi = Error function, summarised over all 49 observations; Seal age at time of observation weighting factors: 0.0 if ICX > SAGE and pavement is uncracked; 1.0 if ICX < SAGE and pavement is uncracked; 1.5 if ICX > SAGE and pavement is cracked, and 1.0 if ICX > SAGE and pavement is cracked.

According to this method, no error is registered if a section is uncracked and the predicted crack initiation time is greater than the age of the seal. For all other
TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

conditions where ICX is not equal to SAGE, a normal squared error is registered. However, observations where the predicted crack initiation period is longer than the observed crack initiation time are regarded as more critical (as reflected by the higher weighting factor). The minimisation scheme resulted in the following values for a0 and a1: a0 = 12.2 a1 = -90.9 Figure 4 shows the observed versus predicted crack initiation times. It should be noted that the unfilled characters in Figure 4 denotes sections where the cracked situation (ie cracked or uncracked) was correctly predicted. The calibration factor a0 agrees well with the suggested default parameter. The factor a1, however, is significantly larger than the default value, indicating an increased sensitivity to traffic and structural numbers. This is most likely because the range of traffic loading rates (YE4) in the data set is relatively small, with a maximum YE4 value of 0.22. Figure 5 gives an indication of the shape of the predicted structural crack initiation time for different SNC values. The actual SNC values of the observed data ranged from 3.1 to 6.7. It can thus be seen from Figure 5 that, even though there is considerable scatter in the data, the calibrated HDM 4 equation predicts the time to crack initiation time reasonably well. It can also be seen from Figure 5 that, with the calibration factors derived here, a very short time to crack initiation is predicted for seals with annual traffic in excess of about 0.4 million ESAs, even for strong pavement structures. This is considered to be consistent with seal design practice in southern Africa.
15
SYMBOL OBSERVED SITUATION UNCRACKED PREDICTED SITUATION UNCRACKED CRACKED CRACKED UNCRACKED

12
CRACKED

OBSERVED SEAL AGE (yrs)

UNCRACKED

CRACKED

Figure 1.

Figure 4: Observed vs predicted crack initiation


TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

16
24 = OBSERVED AND CRACKED 17 SNC FOR SECTION 31 = 3.21 SNC FOR SECTION 24 = 6.06

14 OBSERVED SEAL AGE (yrs) 12 10 8 6 4 2 0 0


4 22

SNC FOR SECTION 23 = 3.49 SNC FOR SECTION 22 = 4.95

23

SNC FOR SECTION 20 = 4.77 SNC FOR SECTION 17 = 3.48 31 SNC FOR SECTION 5 = 5.09 SNC FOR SECTION 4 = 4.12 5 20

PREDICTED FOR SNC=6 PREDICTED FOR SNC=3

0.2

0.4 0.6 0.8 TRAFFIC LOADING (millions/lane/year)

1.2

Figure 5:

Observed crack initiation relative to the HDM-IV strength and traffic sensitive crack initiation models

3.2.1.2 Calibration of structural crack progression on LTPP data Calibration of structural crack progression was based on the behaviour observed on selected pavement sections where a clear crack progression pattern was recorded over the monitoring period. It was found that different seal types exhibit significantly different crack progression patterns and therefore conventional seals (i.e. those using unmodified bitumen as binder), bitumen rubber seals and slurry seals were separately calibrated. The crack progression model proposed for HDM-IV is (NDLI, 1995): CRX = 50. (1 - z) + z[z. a 0 . a 1 . NEci + z. o. 5 a 1 + (1 - z). 50 a 1 ] 1 / a 1 ........................ (3) where: CRX z a0 and a1 TCI NEci t50 = = = = = = the cracking area at time t; 1 for NEci < t50, otherwise z = -1; calibration parameters; time since cracking initiation; cumulative ESAs since cracking initiation; and a1 1 (50 - 0.5a )/(a0.a1).

The factor a0 is a function of the structural number, for which default values are as follows (NDLI, 1995): -4.25 Asphalt Surfacings : 3330.SNC -2.51 Surface Treatments: 1530. SNC Because of the relatively complex form of equation 3, and because of the high variability in observed data, a rigorous calibration using mathematical techniques
TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

was not considered feasible. Instead, the calibration was performed visually by means of graphs, until a satisfactory agreement between observed and predicted trends was observed. The results are listed in Table 4. Table 4. Calibration factors for structural crack progression Coefficient HDM* a0 Conventional Bitumen Rubber Slurry * 1530 SNC 1530 SNC 1350 SNC
-2.51 -2.51 -2.51

Proposed a1 0.25 0.25 0.25

Seal Type

in Coefficients Calibrated on LTPP a0 1530 SNC 330 SNC 1350 SNC


-2.06 -4.24 -2.51

a1 0.30 0.25 0.01

Applicable to granular and cemented base types

A comparison of measured and predicted crack areas, as well as a comparison of the calibrated HDM4 equations with the observed data is shown in Figure 6 . As shown in Figure 6, the calibration factors were chosen to simulate the following observed trends: Conventional Seals (Graphs 6A and 6B): The observed data indicate a smaller dependence of crack progression on structural number (as indicated by the decrease in a0) than is suggested by the default value. The factor a1 is similar to the default value but slightly smaller. Slurry Seals (Graphs 6C and 6D): The observed crack progression data suggests that crack progression on slurry seals takes place at a relatively high rate, even when the pavement structure is strong. This is shown in Figure 6D, which indicates that relatively low volumes of cumulative traffic can result in a large percentage cracked area. Bitumen Rubber Seals (Graphs 6E and 6F): There is a definite tendency for bitumen rubber seals to inhibit crack progression. The data suggest that bitumen rubber seals can accommodate relatively high traffic volumes without the structural crack area increasing very much. In some cases a definite decrease in cracked area was even recorded and verified. The chosen calibration factors for bitumen rubber are therefore identical to the default values suggested for asphalt surfacings.

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

A
PRED/OBSERVED CRACK AREA (%)
15 70 60 50

B
CONVENTIONAL SEALS
OBSERVED SNC for Section 22 = 4.95 PREDICTED 40 30 20
22

OBSERVED CRACK AREA (%)

CONVENTIONAL SEALS
12 9 6 3 0 0 3 6 9 PREDICTED CRACK AREA (%) 12 15

FOR SNC=4

10
22 22

22

PREDICTED FOR SNC=5 0.06 0.08 0.1

0 0.02

0.04

TRAFFIC SINCE CRACK INIT. (mill. ESA)

C
SLURRY SEALS
OBSERVED SEALAGE (YRS) PRED/OBSERVED CRACK AREA (%)

D
35 30 25 20 15 10 5
7 6 6 6 18 7 18 7

25 20 15 10 5 0 0 5 10 15 20 PREDICTED CRACK INITIATION TIME (yrs) 25

SLURRY SEALS
OBSERVED SNC for Section 6 = 4.11 SNC for Section 7 = 5.16 SNC for Section 18 = 4.46

PREDICTED FOR SNC=4

18

18 6

PREDICTED FOR SNC=5


7

0 0

0.02

0.04

0.06

0.08

0.1

TRAFFIC SINCE CRACK INIT. (mill. ESA)

E
15
OBSERVED CRACK AREA (%)

F
40
PRED/OBSERVED CRACK AREA (%)

BITUMEN RUBBER SEALS


12 9 6 3 0

BITUMEN RUBBER SEALS


30 OBSERVED SNC for Section 9 = 5.26 SNC for Section 16 = 5.51 SNC for Section 16 = 4.77 20 PREDICTED FOR SNC=4

10
9 16 9 9 16 20 16 20 16 20 20

PREDICTED FOR SNC=5

0 0 0.2 0.4 0.6 0.8 TRAFFIC SINCE CRACK INIT. (mill. ESA)

3 6 9 PREDICTED CRACK AREA (%)

12

15

Figure 6:

Observed and predicted crack progression on the LTPP sections

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

3.2.3 Calibration of reflection cracking on LTPP data The HDM4 equation for the modelling of reflection cracking is (ND LEA Int., 1995): ICR = a0 + a1.HNEW................................................................................. (4) where: ICR HNEW a 0 , a1 = = = The reflection crack initiation period, in years Thickness of new surfacing, in mm (<200 mm) Calibration parameters

The default calibration parameters for HDM4 are 0.5 and 0.02, for a0 and a1, respectively. This means that for an effective seal thickness of 15 mm, the predicted reflection crack initiation period is approximately 9 months. Very little data was available for the calibration of reflection cracking on surface seals. Three instances were observed where a bitumen rubber seal was placed over a cracked slurry surfacing (percentage cracked area varied from 0.01 to 13 per cent). In two of the three cases the cracks had reflected through the new bitumen rubber seal within two years. In the third case the cracks had reflected through in less than one year. In another instance a sand seal was placed over a cracked bitumen rubber seal with approximately 1 per cent surface cracks. The crack reflection time in this case was between 2 and 3 years. This longer crack initiation period may be partly attributed to the crack retardation properties of the bitumen rubber seals noted in the previous section. Based on the few available observations it can be concluded that crack reflection time for surface seals is likely to be below two years. It is therefore recommended that default calibration parameters for reflection crack initiation and reflection crack progression as suggested be retained. 3.2.1.3 Calibration of thermal cracking The thermal cracking models developed for deterioration of high stability asphaltic concrete surfacings in regions with harsh winters were not calibrated because the province does not experience serious freezing conditions. 3.2.2 Rutting The form of the rutting performance models used in HDM III and HDM IV are the same. The basic forms of the rutting equations are as follows. To predict the rut depth at the end of the first year, the following equation is used (Watanatada et al, 1987, NDLI, 1995): ERM a 0 [YE4x 10 6 ] RDM = K rp ........................................................ (5) SNC a 4 .COMP a 5 With: ERM = a 1 + a 2 DEF + a 3 MMP. ACX ........................................... (6)

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

Where:

Krp RH DEF COMP ai MMP ACX

= = = = = = =

Calibration factor for rut progression, Rehabilitation indicator with value of 1.0 for asphalt or open graded cold mix overlays, and 0.0 for all others, The Mean Benkelman Beam Deflection in mm, The relative compaction of the pavement (%), Constants defined in Table 5, Mean Monthly Precipitation, in m, The percentage cracking at the start of the analysis year.

After the first year, rutting is calculated by adding a calculated rut increment to the rutting at the start of the analysis year. For HDM IV, the incremental rut depth in the analysis period is calculated by (NDLI, 1995): RDM = K rs [ ( a 0 + a 1 DEF + a 2 MMP. ACX) + AGE3 a 3 MMP. ACX. ln [ max(1; AGE3.YE4)] _ RDMa

.................. (7)

Table 5. Coefficients used by HDM 4 in Equations 5, 6 and 7 Coefficient a0 a1 a2 a3 a4 a5 HDM IV (Eq 5, 6) 51740 0.09 0.0384 0.0016 0.502 2.300 HDM IV (Eq 7) 0.333 0.0494 0.0021 0.0285

Calibration was performed using only those sections where no rehabilitation was performed during the analysis period. Since the data set consisted mostly of old pavements for which the rut depths at the start of the analysis period were known, it was decided to focus on the rut depth increment relation (Equation 7). This was done by comparing the predicted rut depth increment over the three year analysis period with the observed average rut depth increment . The calibration was performed using known input values (i.e. actual values of ACX, dACX etc.) instead of projected input values. In this manner focus was placed on the rut increment prediction and not on the accuracy of projected input values. During the analysis period, it was found that a large number of subsections had experienced significant increases in rut depth. Although some of these rut depth increments are unexpectedly high, they were verified and can therefore be seen as a true reflection of the road deterioration due to rutting. There are two possible reasons for the unexpectedly high increases in rut depth: Firstly, a definite increase in wheel load as well as tyre pressure. Secondly, the
TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

unusually high rainfall experienced during 1996, which was the last year of the analysis period. Where possible, this higher rainfall was taken into account during the calibration by using the average rainfall recorded during the analysis period. The calibration showed that the HDM IV rutting equation has a definite tendency to under-predict the rut increment. This was also shown in an earlier study by Jooste (1997). Because of this, and also because of the high observed rut increments, a relatively large calibration factor of 30.0 had to be used in order to produce reasonable correlations between observed and predicted rut depth increments. Thus the value of Krs in equation 3 was set to 30.0. Figure 7 shows the predicted rut depth increments after calibration plotted versus the observed rut depth increments. It is clear from Figure 7 that the average observed rut depth increment is significantly higher than that predicted by HDM IV. A minimum rut depth increment of 1 mm per year seems appropriate for the observed data set. Figure 8 shows a comparison of the predicted and observed rut depths for all years after the start of the analysis period. For this comparison, the rut depths at the start of the analysis period were assumed to be known. That is, the predicted rut depths were projected from the known 1993 rut depth values using the predicted rut depth increment. Although the statistical analysis suggested a calibration factor of 30.0 over the short analysis period of 4 years, this factor will not lead to realistic predictions over longer prediction periods. For use in the provincial PMS it was subsequently decided to limit the incremental rutting on cracked sections to a maximum of 1 mm per year. On un-cracked sections it was decided to use a coefficient of 1.0. The calibration and model will be re-evaluated when the sections has been monitored over a longer period.
OBSERVED RUT RATE (mm/year) BEFORE CALIBRATION OBSERVED RUT RATE (mm/year) 6 5 4 3 2 1 0 0 1 2 3 4 5 PREDICTED RUT RATE (mm/year) 6 6 AFTER CALIBRATION 5 4 3 2 1 0 0 1 2 3 4 5 PREDICTED RUT RATE (mm/year) 6

Figure 7:

Observed versus predicted rut increments, before and after calibration (using a calibration factor Krs = 30.0)

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

35 OBSERVED RUT DEPTH (mm) OBSERVED RUT DEPTH (mm) BEFORE CALIBRATION 30 25 20 15 10 5 0 0 5 10 15 20 25 PREDICTED RUT DEPTH (mm) 30 35

35 30 25 20 15 10 5 0 0 5 10 15 20 25 PREDICTED RUT DEPTH (mm) 30 35 AFTER CALIBRATION

Figure 8:

Observed versus predicted rut depth (mm), before and after calibration

3.2.2.2 Calibration of the rut depth standard deviation model on LTPP data The rut depth standard deviation prediction model suggested for HDM4 is of the following form (NDLI, 1995): RDS = a0.RDM............................................................................... (8) Where: RDS = a0 = RDM = predicted rut depth standard deviation at the start of the analysis period; calibration coefficient , (Table 7) mean rut depth at the start of the analysis period

Table 7. Proposed Default Coefficients for Determining RDS from RDM


Range of rut Depth (mm) a0 (Proposed for HDM-IV) a0 (As calibrated on LTPP)

0-5 5-15 >15

0.8 0.5 0.3

0.42 0.34 0.24

Since the mean rut depth and standard deviation of each subsection for each analysis year was available, calibration of the coefficient a0 was done simply by minimising the squared error between observed and predicted RDS values. This minimisation process led to the coefficients shown in Table 7. Figure 9 shows a comparison between the observed and calculated RDS values after calibration. It can be seen from Figure 9 that there is a reasonable correlation between the observed and predicted RDS values, even for this very simple prediction model. Figure 10 shows the calibrated model plotted together with the observed RDS values.

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

12 BEFORE CALIBRATION 10 OBSERVED RDS 8 6 4 2 0 0 2 4 6 8 PREDICTED RDS 10 12

12 AFTER CALIBRATION 10 OBSERVED RDS 8 6 4 2 0 0 2 4 6 8 PREDICTED RDS 10 12

Figure 9:

Observed and predicted rut depth standard deviation (RDS) values, before and after calibration
12 OBSERVED DATA 10 UNCALIBRATED HDM4 PREDICTIONS CALIBRATED HDM4 PREDICTIONS OBSERVED RDS 8

0 0 5 10 15 OBSERVED RUT DEPTH (mm) 20 25

Figure 10:

Predicted and observed rut depth standard deviation (RDS) as a function of rut depth

4.

USE OF THE CALIBRATED MODELS IN THE PMS

The calibrated pavement performance models have been included in the PMS since 1994. For every link in the network the performance is predicted over the analysis for all feasible maintenance strategies. Figure 11 shows a typical prediction on a road segment.

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

100

CRACKING
75 Rutting (mm) % Cracking Do Nothing

14 12 10 8 6 4 2 0 16

RUTTING

50

Do Nothing

25 Treatment 0 1 6 11 Year

Treatment

11 Year

16

14 12 Roughness (IRI) 10 8 6 4 2 0 1 6

ROUGHNESS
Condition Index

Do Nothing

Treatment 11 Year 16

100 90 80 70 60 50 40 30 20 10 0 1

OVERALL CONDITION
Treatment

Do Nothing

11 Year

16

Figure 11: Typical road segment predictions in the PMS The inclusion of the calibrated performance models has improved the decision process significantly. Since 1994 the PMS decision process has changed from selecting maintenance and rehabilitation actions based on a ranking process, to a decision process based on optimisation. Long-term predictions are done on network condition, asset value and influence on road users. Typical output is shown in Figure 12. (Extracted from Rohde et al 1996.)

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

GAUTRANS
LONG-TERM CONSEQUENCES OF VARIOUS MAINTENANCE POLICIES FOR A 45 MILLION ANNUAL BUDGET
FIG 4a: AVERAGE VEHICLE OPERATING COST OVER 10 YEARS 1.5
1.48 1.46 1.44 1.42 1.4 1.38 WORST F IRST PRIOR INDEX AREA UNDER CURVE TTC

FIG 4b: SAVINGS IN VEHICLE OPERATING COST OVER 10 YEARS

RAND PER TRAVELLED KM

RAND PER TRAVELLED KM

0.17 0.15 0.13 0.11 0.09 0.07 WORST F IRST PRIOR INDEX AREA UNDER CURVE TTC

FIG 4c: EXCESS USER COST


3500 3000 2500 F IX W O R S T F I R S T USE PRIORITY INDEX AREA UNDER CONDITION CURVE T O T A L T R A N S P O R T A T IO N C O S T

FIG 4d: DECREASE IN ASSET VALUE: 1995 TO 2005


700 600 500 R'000 000 400 300 200

R'000 000

2000 1500 1000 500 0 1995

100 0 WORST F IR S T
1997 1999 2001 2003 2005

PRIOR INDEX

AREA UNDER CURVE

TTC

CLASS A,B

CLASS C

CLASS D,E

Figure 12: Typical PMS output made possible through the inclusion of performance models

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

5.

CONCLUSIONS

Since 1993 the Gauteng province of South Africa has monitored a factorial of pavement sections to select, improve and calibrate PMS performance prediction models. Data collected from the performance sections indicate that local calibration of models is essential. The models currently used are calibrated versions of the HDM-III pavement performance models. The following summarises provincial experience with the models: Time to crack initiation is predicted with reasonable accuracy. The crack progression model of HDM-IV can be calibrated to effectively differentiate between the performance of different seal types. The rutting models of both HDM-III and HDM-IV provide poor predictions for rut development observed in Gauteng. The riding quality prediction model which influences the calculation of road user effects, still requires local calibration.

The calibration and improvement of PMS performance models will continue for several years. Since first introduction in 1994, the prediction capabilities of the PMS have significantly changed and improved pavement management activities and strategic level decisions in the province. REFERENCES AFRICON and Jooste. F J, 1997. Initial investigations into the use of HDM-IV pavement performance Models in the Gautrans PMS, Working Document, PMS-Consortium, Pretoria, South Africa. GEIPOT, 1982. Research on the Interrelationship Between Costs of Highway Construction, Maintenance, and Utilization, Final Report, 12 volumes, Ministry of Transport, Brazil. Jooste, F J, 1997. Evaluation of the HDM Performance Prediction Models for use in the Gauteng Province. Gautrans, Pretoria, South Africa. (Contract Report CR-97/022, Prepared by Transportek, CSIR) Kannemeyer, L. and Visser, A T., 1993. The Evaluation of Overseas Pavement Deterioration Models for Application on South African National Roads, Annual Transportation Convention, Pretoria, South Africa. Kemp, M J. Curtayne, P C and York-Hart, M A.., 1989. A study of the Applicability of the HDM-III for Use in Assessing Maintenance Alternatives on South African Rural Roads, Annual Transportation Convention, Pretoria, South Africa. NDLI, 1995. Modelling Road Deterioration and Maintenance Effects in HDMFinal Report Asian Development Bank Project RETA 5549. ND Lea 4. International, Vancouver.
TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

4th International Conference on Managing Pavements (1998)

Paterson, W D O., 1987. The Highway Design and Maintenance Standards Model (HDM-III), Volume III, Road Deterioration and Maintenance Effects : Models for Planning and Management. Washington, D.C.: Transportation Department, World Bank. Pinard, M I., 1992. Potential of HDM-III for Adaption to southern African Conditions, Proceedings of HDM-III Review Meeting, Roads Department, Gaborone, Botswana. Rohde, G T., 1994. Calibration of the HDM-III Pavement Performance Models in three countries in southern Africa, International Workshop on HDM-IV, Institute Kerja Raya, Malaysia. Rohde, G T., 1994. Determining a Pavement's Structural Number from FWD Testing, Paper # 940351 presented at the 1994 Annual TRB meeting, Washington, DC. Rohde, G T., Pinard, M I and Sadzik, E., 1997. Long-term Network Performance - a Function of PMS Maintenance Selection Policy. Paper presented at the 1997 annual TRB meeting, Washington DC, USA. Sayers, M W., Gillespie, D and Paterson, W D O., 1986. Guidelines for Conducting and Calibrating Road Roughness Measurements, World Bank Technical Paper No 46, The World Bank, Washington D.C., USA. TMH9., 1992. Pavement Management Systems: Standard Visual Assessment Manual for Flexible Pavement. Department of Transport Van Zyl, G D., 1994. Visual Assessments for the calibration of HDM-III, Report prepared on behalf of VWL for the Transvaal Roads Branch, Report DPVT C/283, CSIR, South Africa. Watanatada, T and Harral, C G., et al. 1987. The Highway Design and Maintenance Standards Model, Vol 1, Description of the HDM-III model. Baltimore, MD; Johan Hopkins University Press. VWL-TTCS 1994. Customization of the dTIMS software as PMS for Botswana, Report IV, PRN Study, Gaborone, Botswana.

TRB Committee AFD10 on Pavement Management Systems is providing the information contained herein for use by individual practitioners in state and local transportation agencies, researchers in academic institutions, and other members of the transportation research community. The information in this paper was taken directly from the submission of the author(s).

You might also like