Cracking Models for Use in Pavement Maintenance Management
Adelino Ferreira1, Rui Micaelo2, and Ricardo Souza1
1 Department of Civil Engineering, University of Coimbra, Portugal {adelino,ricardosouza}@dec.uc.pt 2 Department of Civil Engineering, Universidade Nova de Lisboa, Portugal [email protected]
Abstract. With the recent approval of the Portuguese Law No. 110/2009 of 18 May, within the scope of road concession contracts, the concessionaires need to submit to the Portuguese Road Infrastructures Institute a Quality Control Plan (QCP) and a Maintenance and Operation Manual (MOM). These documents require the revision of current Pavement Management Systems to consider pavement performance prediction models for each pavement state parameter so that it permits time definition of maintenance and rehabilitation (M&R) interventions for the fulfilment of the values defined in the QCP in each year of the concession period. The QCP presents the admissible values for each pavement state parameter (cracking, rutting, roughness, etc.) that a concessionaire of highways needs to verify. Nevertheless, a concessionaire, beyond the annual pavement inspections to demonstrate the fulfilment of the QCP, wants to predict the proper time to apply M&R preventive interventions at a minimum cost for the complete concession period. This paper describes the state-of-the-art in terms of cracking models. The selected models evaluate the cracking area evolution for a set of representative Portuguese pavements structures and traffic conditions. The Indian and HDM-4 deterioration models were considered to be the most promising to implement in a new Portuguese Maintenance Optimisation System, i.e. to provide a good solution to the pavement maintenance management problem involving not only periodic maintenance but also routine maintenance (crack sealing, rut levelling, patching, etc.).
1 Introduction
A Pavement Management System (PMS) can be defined as a set of tools which helps a road network administration to optimize maintenance and rehabilitation (M&R) actions for keeping the pavements in good service condition. One of the modules of a PMS is the Pavement Performance Model (PPM), which is a mathematical representation that can be used to predict the future state of pavements, based on current state, deterioration factors and effects resulting from M&R actions [1]. Currently the PMS of Estradas de Portugal S.A., the Portuguese
A. Scarpas et al. (Eds.), 7th RILEM International Conference on Cracking in Pavements, pp. 429–439. © RILEM 2012 430 A. Ferreira, R. Micaelo, and R. Souza
Road Administration, uses for PPM the AASHTO pavement performance model that computes a global pavement condition index, the present serviceability index (PSI), based on several factors like the traffic, the material properties and the drainage and environmental conditions [2]. The extent and severity of distresses at the time that the pavement condition index reaches the warning level restricts the implementation of more cost-effective techniques [3]. In 2007 and 2009 the Portuguese Government published legislation [4, 5] established EP – Estradas de Portugal, S.A., as the global road network concessionaire and the basis of the concession contract. Within this contract it was established that concessionaires have to submit to the Portuguese Road Infrastructures Institute (InIR), the supervisor institution, on a regular basis, a Quality Control Plan (QCP) and a Maintenance and operation Manual (MOM). The QCP defines the limits of pavement condition parameters (rutting, cracking, roughness, etc.) than can be found at any time of the concession period. Therefore, these two documents, specially the first one, require knowing the pavement condition ahead, not only the general pavement service condition, but quantifying the extent and the magnitude of pavement distresses and the actions to be implemented in every situation. When a concessionaire does not fulfil the QCP, InIR can apply a contractual infraction, in which the global sum varies, according to its gravity, between €5,000 and €100,000, or daily values that can vary between €500 and €5,000 [4]. A concessionaire, beyond the annual pavement inspections to demonstrate to the InIR and the concessor (the Portuguese State or represented by the EP - Estradas de Portugal, S.A.) the fulfilment of the QCP, wants to predict the year when their pavements do not fulfil the admissible values for some state parameter. A concessionaire knowing this information can apply M&R preventive interventions at a minimum cost in order to effectively fulfil the QCP in all the remaining years of the concession period. This paper describes the state-of-the-art in terms of cracking models, and evaluates the performance of the selected models considering the cracking area evolution for a set of representative Portuguese pavements structures and traffic conditions.
2 Cracking Prediction Models
In this study, the methodology used was to analyse several pavement cracking deterioration models available in the literature with the objective of its integration in the Portuguese Pavement Management Systems. The models selected are the following ones: the Brazilian model; the PAVENET-R model; the HDM-4 model; the Ker Lee Wu (KLW) model; the Indian model; and the Austroads model.
2.1 Brazilian Model (1994)
This model was developed by Visser, Queiroz and Caroca [6] based on a long- term pavement monitoring program carried out in Brazil between 1975 and 1985. The road sections were unbound granular base flexible pavements in areas with tropical to subtropical climate with an average annual precipitation between 1200 Cracking Models for Use in Pavement Maintenance Management 431 and 1700 mm/year. Cracked area over time is predicted with Eqn. (1), developed with multiple regression and probabilistic time failure analysis, which depends on traffic volume, the pavement bearing capacity (load deflection) and age. For existing pavements with asphalt surfacing the model only applies to cracking progression prediction, while for asphalt overlays and slurry seals the model comprises two-phases, initiation time and progression prediction. Two-phase models give an extra opportunity to calibrate the model for cracking prediction over time. This model considers the Benkelman beam to measure pavement deflections, but this equipment is not used in Europe any more. However, some regressions relating the modified structural number with the Benkelman beam maximum deflection have been proposed by several authors such as Eqn. (2) by Paterson [7, 8]. The modified structural number is the evolution of the AASHTO structural number by considering the subgrade contribution. = ( × −2 )× ()(× + × )− − Ct B 10 log N80ct 0.0456 0.00501 Yt 18.53 C0 (1)
= × −0.63 SNC 3.2 B (2)
Where Ct is the pavement cracked area (class 2 cracking or worse, i.e. crack width 2 2 larger than 1 mm) in year t (m /100m ); Yt the age of pavement since original construction or since subsequent AC overlay (years); N80ct is the cumulative 80 kN equivalent single axle load (ESAL) at age t (ESAL/lane); B is the Benkelman beam maximum deflection for the existing pavement (mm); C0 is the cracking offset term calculated to ensure that predicted cracking conforms with the initial value at the start of analysis; SNC is the modified structural number.
2.2 PAVENET-R Model (1996)
The cracking prediction model defined by Eqn. (3) is used in the computer model PAVENET-R [9] aiming at the optimization of the maintenance-rehabilitation problem at the network level. The cracked area over time is predicted based on traffic and the pavement AASHTO structural number calculated using Eqn. (4). As for the previous model it is an only one phase model (just dealing with progression) and it does not include a variable that accounts for the existing cracked area at the beginning of the analysis. = × × −SN Ct 617.14 N80ct SN (3)
N = × e × d SN Hn Cn Cn (4) n=1
2 2 Where Ct is the total cracked area in year t (m /100m ); N80ct is the cumulative equivalent standard axle load (ESAL) at age t (million ESAL/lane); SN is a e d structural number; Cn is the structural coefficient of layer n; Cn is the drainage coefficient of layer n; and Hn is the thickness of layer n (mm). 432 A. Ferreira, R. Micaelo, and R. Souza
2.3 INDIAN Model (1994)
The Indian model was derived from a pavement performance study carried out during the 90’s with extensive monitorisation of pavement sections (145) along national and state highways in four Indian states [10]. The cracking prediction model is a two-phase model, considering the time to cracking initiation calculated using Eqn. (5), and the cracking progression calculated using Eqn. (6). As for the previous models, just two variables were included (traffic and the pavement structural number). The model is applicable to pavements with asphalt surfacing (excluding surface dressing and slurry seal). The climate where the pavement data was gathered varies from arid to humid subtropical, being far from the Portuguese Mediterranean climate.
n×N 80c −1.09× 1 = × SNC 2 (5) Tci 4.00 e
× ()− 0.65 = + × n N80ct N80cti × 0.32 × ()− Ct C i 4.26 SCi t ti (6) SNC
Where Tci is the time to structural cracking initiation (years) - Ct = 2.0%; Ct is the 2 2 total cracked pavement area in year t (m /100m ); N80ct is the cumulative 80 kN equivalent single axle load (ESAL) at age t (million ESAL/lane); N80cTci is the cumulative 80 kN equivalent single axle load (ESAL) at age of cracking initiation (million ESAL/lane); SNC is the modified structural number for the pavement; n is the number of lanes in the road section; Ci is the total cracked pavement area at the 2 2 beginning of the analysis period (m /100m ); ti is the time at the beginning of the 2 2 analysis period (years); SCi is the minimum of {Ci; 100 - Ci} (m /100m ).
2.4 HDM-4 Model (2000)
The Highway Development and Management (HDM-4) system uses a cracking pavement performance model applied in two phases [6, 10-14]: the time to structural crack initiation and the structural crack progression. The HDM-4 is the successor of the World Bank Highway Design and Maintenance Standards Model HDM-III, which has been used by various road agencies all over the world for the last 20 years. Eqn. (7) is used to calculate the time to structural cracking initiation (years). Eqn. (8), (9), (10) and (11) are used to calculate the percentage of cracking over time (class 2 or worse), designed as the “all cracking” model. This formulation is valid for flexible pavements with asphalt or surface treatment as surface course and granular or asphalt base course.
× − ⋅ ()× 2 = 2 × ⋅ 0.14 SNC 17.1 N80c8 / 8 SNC + Tci CDS 4.21 e CRT (7)
= + Ct Ci dC (8) Cracking Models for Use in Pavement Maintenance Management 433
= × × − + []{}− 0.45 Y 0.828 z (t t i ) min max(C i ;0.5); 100 max (C i ;0.5) (9)
CRP dC = × z × ()Y 0.45 − min {}max(C ;0.5); 100 − max (C ;0.5) if Y ≥ 0 (10) CDS i i
CRP dC = × ()100 − C if Y < 0 (11) CDS t
Where Tci is the time to structural cracking initiation (years) - Ct = 0.5%; Ct is the 2 2 total cracked pavement area in year t (m /100m ); Ci is the total cracked pavement 2 2 area at the beginning of the analysis period (m /100m ); N80ct is the cumulative 80 kN equivalent single axle load (ESAL) at age t (million ESAL/lane); SNC is the modified structural number for the pavement; Y is an auxiliary variable; CDS is the construction defects indicator in asphalt layers (0.5 to 1.5 according to real binder content – dry, normal and rich); CRT is the crack retardation time due to maintenance (years) (default value 0); CRP is the crack propagation indicator (1 - 0.12×CRT); z is an auxiliary variable (+1 if Ci ≤ 50 and -1 otherwise). The AASHTO structural number (SN), calculated using Eqn. (4), does not include the subgrade contribution as it is considered in the pavement design procedure through the resilient modulus. In opposition, HDM models (version III and 4) consider a different version of the structural number, the modified structural number (SNC), calculated using Eqn. (12), which takes into account the subgrade strength which is calculated using Eqn. (13) [7].
N = ()× e × d + SNC 0.0396 H n / 25.4 Cn Cn SNSG (12) n=1
= × ()− × [] ()2 − if CBR ≥ 3 SNSG 3.51 log CBR 0.85 log CBR 1.43 (13) e Where SNC is the modified structural number; Cn is the structural coefficient of d layer n; Cn is the drainage coefficient of layer n; and Hn is the thickness of layer n (mm). Thermal cracking is not considered to be an important source of cracking in Portuguese road pavements due to small number of days with subfreezing temperatures (Mediterranean climate). In a different position, reflexion cracking, which is the progression of cracks upwards to the surface from previously cracked asphalt pavements or cement stabilized materials, is important in every cracking deterioration model. It allows prediction of this pavement distress evolution after M&R actions have been taken, considering the common large concession periods. The HDM-4 model for reflexion cracking prediction, Eqn. (14) and (15), is based on Gulden and Malaysian studies, considering a two-phase model as for the other cracking models. {}−2 = 685 × −0.5 × − min H OV ;199 Tri B 1 (14) ADH 200 434 A. Ferreira, R. Micaelo, and R. Souza
2 0.5 H OV Cr = minCr + 0.0182 × ADH × B × max0; 1 − × ()t − tr ; C (15) t i 200 i OV
Where Tri is the time to reflexion cracking initiation (years); HOV is the thickness of the new asphalt concrete layer (mm) (< 200 mm); Crt is the total cracked pavement area in year t, limited to the amount of cracked area before overlay COV 2 2 (m /100m ); ADH is the average daily heavy traffic; tri is the time at the beginning of the analysis period; B is the average Benkelman beam deflection on both wheel-paths (mm).
2.5 Ker Lee Wu (KLW) Model (2008)
Ker, Lee and Wu [15] developed a fatigue cracking prediction model (Eqn. (16)) using LTPP data, for the pavements with asphalt surface course on granular or bound base. The model determines cracking area over time based on traffic, pavement age, climate (precipitation, air temperature and freeze-thaw cycles) and load pavement response (tensile strain). The model was developed to improve pavement design but it can be applied as a PPM.
− + × + × ()× 18.08 0.943 Yt 0.832 log 1000 N80 t = + × + × Ct exp 0.121 precip 0.869 temp (16) + 31.489 × ()ε ×1000 2 + 3.242 × log ()ft t 2 2 Where Ct is the total cracked pavement area in year t (m /100m ); N80t is the number of 80 kN equivalent single axle load (ESAL) applications in year t (million ESAL/lane); precip is the average annual precipitation (mm); temp is the mean annual temperature (°C); εt is the tensile strain at the bottom of the AC layer; ft is the yearly freeze-thaw cycles; Yt is the time since the pavement’s construction or its last rehabilitation (years).
2.6 Austroads Model (2010)
Austroads has recently developed road deterioration models for the roughness, rutting and cracking prediction of sealed granular pavements, which represents 85% of sealed pavements in Australia [16]. Cracking deterioration models were determined based on data collected with the RTA/NSW Road Crack equipment on arterial roads in South Australia between 1999 and 2004. As most of collected cracking data (1384 of 1675 samples) was on asphalt pavements, a cracking model was determined for this pavement type (Eqn. (17) and (18)). The two-phase model (cracking initiation and progression) was initially developed for sealed granular pavements and then adapted to asphalt pavements. Therefore, initiation of cracking is estimated using the same Eqn. to sealed granular pavements (seal life) that depends on climate (air temperatures), bitumen (ARBB test result) and Cracking Models for Use in Pavement Maintenance Management 435 maximum aggregate dimension. Cracked area over time is predicted based on pavement’s age, time since cracking initiation and climate (with Thornthwaite Moisture Index). It is referred that traffic and pavement bearing capacity were not considered statistically significant for the models because these variables could not be reliable assessed in the data set.
× − × + 2 = 0.158 TMIN 0.107 R 0.84 Tci (17) 0.0498×T − 0.0216× D − 0.000381× S 2
−1 0.682×()t−T ci − 3.5 200 TIi C = K × 100 − 200× 1+ e 25 (18) t
Where Tci is the seal life (years); TIi is the Thornthwaite Moisture Index for climate pavement conditions at year t; TMIN is the yearly average of the daily minimum air temperature (°C); TMAX is the yearly average of the daily maximum air temperature (°C); T is the average of TMAX and TMIN values; D is the ARRB Durability Test result; S is the nominal size of seal (nominal stone size, mm); R is the risk factor with a scale from 1 (very low risk) to 10 (very high risk); Ct is the total cracked pavement area in year t (m2/100m2); K is the calibration factor. This cracking prediction model has several drawbacks for the implementation in the Portuguese PMS, namely for having been developed from sealed granular pavements, which is not a common pavement type (at least on main roads), for including a variable from a lab test not used in Europe and for not including a traffic related variable that it is considered to induce most of pavements cracking (fatigue).
3 Cracking Prediction Models Testing
The selected cracking models were tested by comparing the evolution of cracked area over time (design period, taken usually as 20 years) for different pavement structures. The set of representative pavement structures were selected based on the structures proposed by the Portuguese Pavements Design Manual [17] as function of traffic level and foundation capacity. Table 1 presents the levels of the selected daily traffic (T1, T3 and T5) and the corresponding proposed pavement structures for a subgrade F3 (CBR ratio of 20%). In the analysis it was additionally considered a subgrade F2 (CBR ratio of 10%), which requires extra 40 mm of asphalt thickness in each case. The pavements were considered to be situated in central area of Portugal (Coimbra district). The temperature and precipitation values were determined with the weather data collected over the period 1971-2000 by the Portuguese Meteorology Institute [18]. 436 A. Ferreira, R. Micaelo, and R. Souza
Table 1. Pavement data
Pavements Parameter P4 P9 P14 300 800 2000 AADT (per way and lane) h (T5) (T3) (T1) Traffic Traffic growth rate (%) 3 4 5 Heavy vehicles damage factor 3 4.5 5.5 H (mm) 40 50 60 Asphalt surface layer n E (MPa) 4000 4000 4000 H (mm) 140 190 220 Asphalt base layer n Structure E (MPa) 4000 4000 4000 H (mm) 200 200 200 Granular sub-base n E (MPa) 200 200 200 Foundation CBR (%) 20 20 20 Average daily temperature (ºC) 15.5 15.5 15.5 Climate Average yearly precipitation (mm) 905.1 905.1 905.1
4 Results
Figure 1 shows pavement cracked area evolution predicted by all PPM for a pavement P9 (traffic level T3) and two subgrade levels (F2 and F3) during 20 years that is usually considered for the pavement design. It is considered that no M&R actions are implemented during the 20 years. HDM-4 predicts considerably larger values of cracked area during the analysis. At the end of the analysis period HDM-4 predicted cracked area is around the double of the second largest value and cracking is spread all over the road pavement (100%). In a lowest to largest predicted cracked area in year 20, the PPM sequence is the following: KLW; Austroads; Brazilian; Indian; PAVENET-R and HDM-4. HDM-4 predicts cracking initiation around years 5 to 6 and after that cracking spreads faster than predicted by any other method. In opposition, cracked area predicted by KLW and Austroads PPM is very low, less than 15% in year 20. When a lower bearing capacity foundation is considered (F2 subgrade instead of F3), with extra 40 mm of asphalt, cracked area evolution difference is almost imperceptible with the exception of using the PAVENET-R PPM. This PPM uses the AASHTO structural number, which does not consider the subgrade effect. As the pavement thickness is increased to compensate the subgrade strength, the SN value increases and predicted cracking is lower at any time. Most PPM results show that the extra 40 mm of asphalt thickness indicated in the design manual is adequate for this foundation bearing capacity variation. Figure 2 shows the cracked area evolution predicted by all models for two pavement structures (P4 and P14) with different traffic levels (T1 and T5) and F3 subgrade. Cracking Models for Use in Pavement Maintenance Management 437
100 PAVENET-R F2 F3
) BRAZILIAN F2 F3 2 80 INDIAN F2 F3 HDM-4 F2 F3
/100 m KLW F2 F3 2 AUSTROADS F2 F3 m 60
40
20 Cracking area ( Cracking
0 0 2 4 6 8 101214161820 Years Fig. 1. Cracking prediction for pavement P9 and subgrades F2 and F3
100 PAVENET-R P4 P14 BRAZILIAN P4 P14 )
2 INDIAN P4 P14 80 HDM-4 P4 P14 KLW P4 P14 /100 m 2 AUSTROADS P4 P14 m 60
40
20 Cracking area (
0 2 4 6 8 101214161820 Years Fig. 2. Cracking prediction for traffic levels T5 and T1 and pavements P4 and P14
The KLW, Austroads and Brazilian models predict similar values of cracked area during the analysis period. The Austroads method predicts exactly the same values as it does not consider traffic and subgrade in the model. The HDM-4 and Indian models predict larger cracked area values for pavement P14 (highest traffic level) and PAVENET-R predicts less cracked area. The largest difference between predictions is obtained with the Indian model. The cracking growing rate is very similar for both situations in the HDM-4 prediction and the extent of cracking 438 A. Ferreira, R. Micaelo, and R. Souza area at any moment is solely dependent on time for cracking initiation. The
PAVENET-R model predicts a 50% reduction of cracked area in year 20 as the pavement changes from P4 to P14, which allows concluding that the model is much more dependent on pavement bearing capacity than on traffic.
5 Conclusions
The Austroads model is not adequate to include into Portuguese PMS since it does not distinguish different pavement structures used in Portugal as well as traffic levels. The number of years to intervention is considered too optimistic. The KLW model is not satisfactory since the predicted cracked areas are extremely low during the design period, i.e. not in agreement to current Portuguese roads condition. The PAVENET-R model is considered not to be adequate because it amplifies too much the pavement bearing capacity (with exclusion of the foundation strength) and lessens too much the traffic influence. In the Brazilian model, the values obtained are logical though pavements sections used for the model development have different conditions (climate, soils) than can be found in Portugal. The Indian and HDM-4 models were considered to produce acceptable results and therefore it is recommended that a full verification and validation of both models should be conducted using Portuguese pavement condition time series data.
References
[1] Ferreira, A., Picado-Santos, L., Antunes, A.: Pavement performance modelling: state of the art. In: Proceedings of Seventh International Conference on Civil and Structural Engineering Computing, pp. 157–264. Civil-Comp Press, Oxford (1999) [2] Ferreira, A., et al.: Selection of pavement performance models for use in the Portuguese PMS. International Journal of Pavement Engineering 12(1), 87–97 (2011) [3] Lou, Z., Lu, J., Gunaratne, M.: Road surface crack condition forecast using neural network models, p. 93. University of South Florida, Florida (2003) [4] MOPTC, Portuguese Law No. 110/2009 of May 18. Ministry of Public Works, Transports and Communications, Daily of the Republic, 1a Série - No. pp. 3061-3099 (May 18, 2009) (in Portuguese) [5] MOPTC, Portuguese Law No. 380/2007 of November 13. Ministry of Public Works, Transports and Communications, Daily of the Republic 1st Series - No. pp. 8403-8437 (2007) (in Portuguese) [6] Visser, A., Queiroz, C., Caroca, A.: Total cost rehabilitation method for use in pavement management. In: Proceedings of 3rd International Conference on Managing Pavements, Texas, USA, pp. 37–44 (1994) [7] NDLI: Modelling road deterioration and maintenance effects in HDM-4, pp.351, International Study of Highway Development and Management Tools, ND Lea International Ltd., Vancouver, British Columbia, Canada (1995) [8] Bennett, C.R.: Comparison of loadman and Benkelman beam deflection Measurements, p. 38. Four States Pavement Management Project, India (1994) Cracking Models for Use in Pavement Maintenance Management 439
[9] Fwa, T.F., Chan, W.T., Tan, C.Y.: Genetic-Algorithm Programming of Road Maintenance and Rehabilitation. Journal of Transportation Engineering 122(3), 246–253 (1996) [10] Jain, S.S., Aggarwal, S., Parida, M.: HDM-4 Pavement Deterioration Models for Indian National Highway Network. Journal of Transportation Engineering 131(8), 623–631 (2005) [11] Mrawira, D., et al.: Sensitivity Analysis of Computer Models: World Bank HDM-III Model. Journal of Transportation Engineering 125(5), 421–428 (1999) [12] PIARC: Highway development and management, volume one – Overview of HDM- 4, pp.43, World Road Association, Paris, France (2000) [13] Odoki, J., Akena, R.: Energy balance framework for appraising road projects. Proceedings of the Institution of Civil Engineers - Transport 161(1), 23–35 (2008) [14] Ihs, A., Sjögren, L.: An overview of HDM-4 and the Swedish pavement management system. In: VTI - Infrastructure Maintenance, p. 31. Linköping, Sweden (2003) [15] Ker, H.-W., Lee, Y.-H., Wu, P.-H.: Development of Fatigue Cracking Prediction Models Using Long-Term Pavement Performance Database. Journal of Transportation Engineering 134(11), 477–482 (2008) [16] Austroads: Interim network level functional road deterioration models, p. 45. Austroads Ltd., Australia (2010) [17] JAE, Manual of pavement structures for the Portuguese road network, pp. 1–54. Junta Autónoma de Estradas, Portugal (1995) (in Portuguese) [18] IM: Climate Normal 71-00. Institute of Meteorology IP Portugal (2011), http://www.meteo.pt/en/oclima/clima.normais/006/ (cited September1, 2011]