Influences of intelligent compaction uniformity on pavement performances of hot mix asphalt

Construction and Building Materials 30 (2012) 746–752

Contents lists available at SciVerse ScienceDirect

Construction and Building Materials journal homepage: www.elsevier.com/locate/conbuildmat

Influences of intelligent compaction uniformity on pavement performances of hot mix asphalt Qinwu Xu a,⇑, George K. Chang a, Victor L. Gallivan b, Robert D. Horan c a

Pavement Research, The Transtec Group Inc., Austin, TX 78731, USA Federal Highway Administration, Indianapolis, IN 46204, USA c Asphalt Institute, Lexington, KY 40511, USA b

a r t i c l e

i n f o

Article history: Received 14 October 2011 Received in revised form 18 December 2011 Accepted 23 December 2011 Available online 16 January 2012 Keywords: Intelligent compaction Semivariogram M-E PDG model Uniformity Pavement performances

a b s t r a c t Conventional pavement analysis and design methods are based on the homogeneous or uniform material model such including the multi-layered analysis program and AASHTO design methods. With the Intelligent Compaction (IC) technology on hot mix asphalt (HMA) involved in recent years, the compaction uniformity of material property can be quantified. This paper intends to study the effects of compaction uniformity on pavement performances using the Bomag IC Evib – a measurement of elastic moduli with 100% coverage of the compaction area. The three dimensional (3-D) finite element (FE) model was built to simulate pavement responses with the heterogeneous HMA moduli derived from the field IC measurements. Then the Mechanistic-Empirical Pavement Design Guideline (M-E PDG) models were used to predict HMA performances of rutting and fatigue life. The geostatisical semivariogram model was studied to evaluate the uniformity of predicted performances. Different from conventional pavement analysis and designs, spatial-distributed heterogeneous moduli of the asphalt layer were considered in this work. Results show that spatial uniformity of material moduli affects pavement performances in terms of the distress severity and uniformity. Less uniform material moduli result in higher rutting depths and shorter fatigue lives. For the case study in this paper, the mean and peak values of fatigue lives for the heterogeneous model are 38.2% and 0.1% of those for the uniform model. A pavement section with overall lower material moduli does not necessarily correspond to inferior performances as the effects from uniformity of material property may dominate other factors. Therefore, it is recommended that the uniformity of pavement layer properties that emulate the typically more variable service condition be considered in future pavement designs and performance predictions. Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction Uniformity of field compaction has long been considered qualitatively as a factor on pavement performance. A non-uniform compaction results in non-uniform pavement material properties such as moduli. Field compaction uniformity has long been considered an important qualitative characteristic for desired pavement responses and long term performance. However, for the conventional pavement analysis and design method, homogeneous or uniform material properties (including the anisotropy property at different direction) are used for the same pavement layer through spatial distribution. These methods include the currently used multi-layered analysis program, the American Association of State Highway and Transportation Officials (AASHTO) pavement design methods, etc. For the conventional quality control (QC) and quality assurance (QA) method used for field construction, usually the random spot ⇑ Corresponding author. Tel.: +1 512 451 6233; fax: +1 512 451 6234. E-mail address: [email protected] (Q. Xu). 0950-0618/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.conbuildmat.2011.12.082

tests for material or structural properties are conducted. For example, the AASHTO’s method [1] measures the nuclear gauge (NG) and/or coring densities to check if it reaches the target with a certain percentage (e.g. 80% samples exceed the target of 92% Gmm – the theoretical maximum specific gravity). However, this QC/QA method based on very limited spot number of random tests is unable to evaluate the compaction uniformity. Uniformity measures how uniform or variable the factor is. Meanwhile, some weak spots or zones might be missed using the conventional QC/QA method. It is not until in the past decade that uniformity of field compaction can be quantified with Intelligent Compaction (IC) technologies that can measure levels of compaction for the entire compacted areas. IC technology was initially developed in Europe and Japan and used more than one decade, and introduced to the US in late 2000s [2,3]. Though successfully used for the earthwork and soil compaction, IC as an emerging technology is still immature and involving for the hot mixture asphalt (HMA) [2]. IC uses vibratory rollers equipped with accelerometer-based measurement system, temperature sensors, global position system

747

Q. Xu et al. / Construction and Building Materials 30 (2012) 746–752

HMA Overlay:h= 10.16 cm (4.0 inch), E is non-uniform, ν = 0.35 Exiting HMA: h=15.24 cm (6.0 inch), E = 489,115 kPa, ν = 0.35 Base layer: h= 17.28 cm (7.0 inch), E = 3,481,177 kPa, ν = 0.2 Soil: E=23,866 kPa, ν = 0.45 Fig. 1. Pavement structure and material properties (h: depth, E: elastic modulus, v: Poisson’s ratio).

2100 1950 1800 1650 1500 1400 1300 1200 1100 1000 900 800 700 600 500 400 300 200 75

(GPS), and real-time onboard display of measurements [2]. The IC measurement value (ICMV) is an index value relating to the pavement layer stiffness and moduli. From the Federal Highway Administration (FHWA)’s IC demonstration study [4–8], the ICMV exhibits fair linear correlation with layer moduli as back-calcalculated from the falling or light-weight deflectometer (FWD or LWD). Also, ICMV is usually non-uniform or heterogeneous in field compaction conditions [2]. Non-uniform stiffness in field conditions may result in premature distresses [2,4], or inferior long-term

pavement performances. In the FHWA’s generic IC specifications [9], IC is considered as a quality control (QC) tool. However, the uniformity is not considered in the QC yet in those specifications. Quantifying the effect of compaction uniformity on the pavement performance remains as a research gap. Alkasawneh et al. [10] studied the influence of vertical heterogeneous property of materials on pavement response. The horizontal heterogeneous property is more common due to non-uniform support and compaction conditions as identified by the IC technology. The objective of this research is to study the influences of compaction uniformity of hot mixture asphalt on pavement performances such as rutting and fatigue cracking, using the field measured ICMVs. This research may provide rationale and support for including uniformity requirements in future IC specifications. To achieve this goal, following research efforts were performed: (1) IC experiment on the HMA overlay; (2) three-dimensional (3-D) finite element (FE) modeling of pavement responses with the IC measured non-uniform HMA elastic moduli with 100% coverage on the compaction area; (3) mechanistic-empirical (M-E) pavement design guideline (M-E PDG) modeling of pavement performances (rutting and fatigue cracking); and (4) geostatistical study of the

0m

100m

50m Section I

Section II

150m Section III

MPa

200m Section IV

235m Section V

14.65 cm

800 Loading pressure (Kpa)

39.62 cm

Loading area

Fig. 2. Kriged HMA elastic moduli Evib for the 235 m compaction lane (total five sections).

700 600

3.1

g vin o M

din loa

m

g

500 400 300

HMA overlay

200 100

Existing HMA

0 0

0.01 0.02 0.03 0.04 0.05 0.06 Time (s)

235

Base

m 3.5 m

Loading pulse

Soil

Fig. 3. Finite element model as deformed under loading (max deflection: 9.68  104 m). Note: the grids of the HMA overlay are volume bodies rather than elements as to possess heterogeneous elastic moduli with ANSYS modeling techniques.

748

Q. Xu et al. / Construction and Building Materials 30 (2012) 746–752 3.0

Semivariogram

2.5 2.0 Sill

Experimental semivariogram

1.5

Exponential model Gaussian model

1.0 0.5 Range

Nugget

0.0 0

50

100

150

200

Lag distqance (m) Fig. 4. Experimental semivariogram and model fit.

uniformity of pavement performances as related to the non-uniform material elastic moduli of HMA layer. 2. Methodologies Traditional pavement performance prediction and design methods are normally based on homogeneous or uniform layer properties, including the multi-layer pavement analysis program, the AASHTO design method, etc. [11]. In order to capture the effects of non-uniform or heterogeneous material elastic moduli on pavement performances, this research uses the M-E PDG models for predicting pavement performances, with response inputs simulated from the 3-D FE modeling method. The geostatistical semivariogram model is implemented in the developed Software to evaluate the uniformity of compaction and predicted pavement performances. 2.1. IC experiment and ICMV The IC construction project was an HMA overlay on the existing asphalt pavement, located on US 52 between the junctions with US 231 and with Cumberland Ave, Lafayette, Indiana, USA. One experimental lane with a length of 235 m and

width of 3.1 m was utilized as the study objective. FWD test was performed to back-calculate the existing layer properties. Fig. 1 presents the pavement structure and back-calculated elastic moduli of existing layers. It shall be noted that the IC was not used for the construction of earthwork and base layer, and thus the nonuniformity of the unbound layers are not studied in this study. The Bomag ICMV of Evib (MN/m2 or MPa), is a measurement of the vibration (dynamic) modulus of elasticity for pavement materials [12]. Though Evib has an influence depth which may go deeper under HMA layer dependent on the specific pavement structure [1,2], it was close to the actual HMA moduli and then used as the elastic modulus inputs for the HMA overlay in the FE and M-E PDG models, and named as elastic modulus (moduli) for all discussions in the rest of this paper. Kriging technique was performed on the Evib measurements of the HMA overlay to ‘‘smoothen’’ the data and estimate missing data at some coordinates, with results shown in Fig. 2. Kriging uses generalized linear regression techniques to estimate the missing data by minimize the estimation variance defined from a prior model for a covariance [13]. The entire compaction length of 235 m was divided into 5 sections (see Fig. 2), with 50 m for each of the first four sections and 35 m for the last (fifth) section, according to their geospatial uniformity as identified by VedaÒ software as discussed later. 2.2. Finite element model Traditional pavement modeling and design methods are normally based on homogeneous or uniform layer properties, which may not accurately represent the real pavement conditions. As found by IC study, the pavement conditions (e.g. ICMV and density) are actually non-uniform or heterogeneous. The ANSYS 12.1Ò software [14] was used to perform the FE simulation accounting for the heterogeneous material properties of spatial distribution. The FE model simulates one experimental compaction lane (235 m long and 3.1 m wide), consisting of 105,798 elements and 117,110 nodes. The loading area has a fine element size of 14.65 cm by 13.21 cm. Two models were built for comparison: a uniform model with mean value of IC-measured HMA elastic moduli (704 MPa) of Evib, and a non-uniform model with IC-measured elastic moduli distribution assigned to each element of the HMA overlay. The existing HMA, cement (treated) concrete base, and soil have uniform material elastic moduli (see Fig. 1). The sinusoidal loading pulse [11] was applied at the center of the lane, with a cycling period of 0.6 s, a standard peak value of 9000 lb (40 KN) and pressure of 100 psi (689.475 kPa), and a rectangular loading area of 14.65 cm long by 39.62 cm wide as shown in Fig. 3. The moving loading is simulated on the experimental lane at a constant standard highway speed of 60 mph (26.8 m/s), with the step loading applied on element by element in ANSYS (see Fig. 3). The peak pavement responses of each loading cycle (deflection on the top and tensile strain at the bottom of

Fig. 5. A screenshot of Veda software for the semivariogram analysis.

749

Q. Xu et al. / Construction and Building Materials 30 (2012) 746–752 HMA overlay) were recorded for the use of pavement performance prediction. It shall be noted that the dynamic motion was not considered in this simulation. The highperformance parallel computing was used to accomplish the entire simulation by eleven steps, with four 3.5 GHz processors and 9 GB memory.

this example); C 1 is the  0:1039  hac þ 2:4868  hac  17:342; C 2 is the 0:0172 2 hac  1:7331  hac þ 27:428 hac is the thickness of HMA layer in inch (10.16 cm or 4 in. for this example). Consequently, the rutting depth of HMA can be calculated by multiplying the plastic strain with HAM layer thickness as follows:

2.3. M-E PDG performance prediction

RDhma ¼ ep  hac

Given the FE modeled pavement responses, rutting depth and fatigue cracking can then be predicted using the M-E PDG models developed by the National Cooperative Highway Research Program (NCHRP). The M-E PDG rutting model for the HAM layer is described as follows [15]. The plastic strain of HMA layer is calculated as follows:

ep ¼ er  k1  103:4488 T 1:5606 N0:479244

ð1Þ

where ep is the accumulated plastic strain at N repetitions of load; er is the resilient vertical strain of asphalt material; T is the reference temperature (77 °F in this example); N is the loading repetition (or Equivalent Sing Axles, ESALs); k1 is the (C1 + C2 ⁄ depth) ⁄ 0.328196depth; depth is the depth of pavement in inch (43.18 cm or 17 in. for

2

ð2Þ

The fatigue life is estimated as follows [15]:

Nf ¼ 0:00432C

 3:291  0:854 1 1 et E

where C is the 10M; M is the 4:84

ð3Þ 

Vb V a þV b

  0:69 ; E is the elastic modulus of HMA, in

unit of psi; et is the tensile strain at the bottom of HMA layer; Vb is the effective binder content (5% for this example); Va is the air void of HMA (4% for this example). It has to be noted that the M-E PDG models presented above [15] are based on the US units and the results are converted back to the SI units. An ESALs of 90,000,000 is assumed to compute the rutting depth of HMA overlay in Eq. (1). Only the rutting of HMA overlay (not including that of the base/subbase and soil) are

Table 1 HMA layer elastic moduli and performances. Model

Uniform model

Nonuniform model

(a) Elastic moduli and performances for entire compaction lane Variable Mean Moduli (MPa) 704 Rutting (mm) 2.16E03 Fatigue life 3.86E+08 Statistics

95% Upper limit 713 2.27E03 1.52E+08

95% Lower limit 695 2.24E03 1.42E+08

Section 1

Section 2

Section 3

Section 4

Section 5

(b) Elastic moduli and performances of non-uniform model by sections HMA moduli (MPa) Mean 727 Flitting depth (mm) Peak 2.18E03 Mean 2.13E03 Fatigue life Peak 1.31E+08 Mean 2.16E+08

740 2.73E03 2.16E03 1.38E+06 1.79E+08

600 2.60E03 2.26E03 3.26E+06 1.04E+08

701 3.99E03 2.32E03 3.31E+05 1.50E+08

754 3.50E03 2.47E03 8.73E+05 6.42E+07

Deflection on top of HMA (mm)

Variable

Mean 704 2.25E03 1.47E+08

-0.4 -0.5 -0.6 -0.7 Nonuniform model deflection (mm)

-0.8 Uniform model deflection

-0.9 -1 0

50

100

150

200

250

200

250

Distance (m)

(a) Deflection on the top of HMA layer Distance (m) 0

50

100

150

HMA layer rutting (mm)

2.00E-03

2.50E-03

3.00E-03

Nonuniform model rut

Uniform model rut

3.50E-03

4.00E-03

4.50E-03

(b) Rutting depth of HMA layer and deflection on top of HMA layer Fig. 6. Simulated pavement deflection and predicted HMA layer rutting depth.

750

Q. Xu et al. / Construction and Building Materials 30 (2012) 746–752

Tensile strain at HMA bottom

1.0E-03 9.0E-04 8.0E-04

Nonuniform model tensile strain

7.0E-04

Uniform model tensile strain

6.0E-04 5.0E-04 4.0E-04 3.0E-04 2.0E-04 1.0E-04 0.0E+00 0

100

50

150

200

250

Distance (m)

Fatigue life

(a) tensile strain at bottom of HMA layer

3.5E+08

Nonuniform model fatigue life

3.0E+08

Uniform model fatigue life

2.5E+08 2.0E+08 1.5E+08 1.0E+08 5.0E+07 3.3E+05 0

100

50

150

250

200

Distance (m)

(b) fatigue life of HMA Fig. 7. Simulated pavement deflection and predicted fatigue life of HMA. predicted, considering that only the IC-measured elastic moduli of HMA overlay are available, and this study only focuses on the HMA material. Meanwhile, it must be aware that it is a simplified method without considering the multiple environmental and axle loading conditions as compared to the M-E PDG software since it is unable to capture the heterogeneous material properties [15]. 2.4. Semivariogram model Semivariogram is a function describing the degree of spatial dependence of a random field or stochastic process. It is defined as the expected, squared increment of the values between two adjacent locations [16]. It is a common tool used in geostatistics to describe spatial variation or uniformity. For a given geo-space with a defined direction, the experimental variogram, r(h), for the separation or lag distance of h, is defined as the average squared difference of values Z(u) separated approximately by h for all possible locations of u [16]:

2cðhÞ ¼ Ef½ZðuÞ  Zðu þ hÞ2 g

The lag distance h, is the separation between geo-spaces with given direction, and the calculation of variogram is repeated for many different values of lag distances as long as the sample data will support. As result, with the increase of lag distance value, the semivariogram value usually increases until reaching a relatively stable value (see Fig. 4). One moves from one node (data point) to the next to determine the Z(u)  Z(u + h) value for the calculation of variogram. Eq. (4) can be re-expressed as follows for the sample data:

2cðhÞ ¼

cðhÞ ¼

where u is the location; h is the lag distance (it could be a constant for the same lag areas); Z(h) is the value such as ICMV in this paper.

Section 1 Section 2 Section 3 Section 4

⎡ ⎛ 3h ⎞ ⎤ r(h) = c ⎢1 − exp ⎜ − ⎟ ⎥ ⎝ a ⎠⎦ ⎣

Section 5

1 X ½ZðuÞ  Zðu þ hÞ2 2NðhÞ NðhÞ

ð6Þ

Semivariogram for Rut of HMA Layer 3.0E-07

Semi-variogram (mm2)

2

Semi-variogram (MP )

240000 220000 200000 180000 160000 140000 120000 100000 80000 60000 40000 20000 0

ð5Þ

where N(h) is the number of pairs for lag distance h of a specific lag area. The semivariogram is then determined as half of the variogram:

ð4Þ

Elastic modulus

1 X ½ZðuÞ  Zðu þ hÞ2 NðhÞ NðhÞ

Section 1

⎡ ⎛ 3h ⎞ ⎤ r(h) = c ⎢1− exp ⎜ − ⎟ ⎥ ⎝ a ⎠⎦ ⎣

2.5E-07

Section 2 Section 3 Section 4

2.0E-07

Section 5

1.5E-07 1.0E-07 5.0E-08 0.0E+00 0

0

5

10

15

20

25

5

10

15

20

25

Lag distance (m)

Lag distance (m) Fig. 8. Experimental semivariogram and exponential model fit for HMA moduli Evib.

Fig. 9. Experimental semivariogram and exponential model fit for HMA layer rutting.

751

Q. Xu et al. / Construction and Building Materials 30 (2012) 746–752

   3h rðhÞ ¼ c 1  exp  a

ð7Þ

Based on Eq. (7), the nugget value is zero for the exponential model (r(h) = 0 at h = 0). 2.5. Numerical implementation The research team has implemented the semivariogram models into the VedaÒ software for the analysis of IC measurements. Veda is a software used for geospatial data analysis and geared toward IC technology, as required by the FHWA’s generic IC specification, recently developed by the Transtec Group [18,19]. This approach was verified by comparing the results from the VedaÒ software and the commercial SurferÒ software [20]. Fig. 5 presents a screenshot of VedaÒ [18], showing an interface with a semivariogram sample, Google map, functions, etc. The univariate statistics includes the mean ( x), peak (maximum or minimum value), 95% confidence limits pffiffiffi for n samples ( x  1:96r= n), etc. are also analyzed and implemented

in the Veda software.

3. Results and analysis 3.1. Predicted pavement performances Table 1 summarizes the univariate statistical values of elastic moduli and predicted pavement performances. Fig. 6 presents the predicted pavement rutting depth together with the FE simulated deflection on the top of HMA layer along the highway distance. Fig. 7 presents the predicted HMA fatigue life together with the FE simulated tensile strain at the bottom of HMA layer along the highway distance. The main findings include the followings: (1) Obviously the non-uniform pavement model results in nonuniform pavement performances (see Figs. 6 and 7), e.g. many peak values appear along the highway distance. (2) The trends of the rutting of HMA overlay follow that of the deflections on the top of HMA (see Fig. 6), which could be explained by the linear function of the M-E PDG rutting model (see Eq. (1)). However, the trends of the fatigue life of HMA overlay does not well follow that of the tensile strain at the bottom of HMA (see Fig. 7), and its distribution is more scattered (less concentrated). This can be indicated by the nonlinear function of M-E PDG fatigue model since the elastic modulus itself also compensates portion of the fatigue life (see Eq. (3)). (3) Overall the non-uniform HMA model results in significantly lower fatigue life than the uniform model. For example, the mean value, the upper 95% confidence limit, and the peak value of the non-uniform model is only 38.2%, 39.5%, and 0.1%, respectively, of that of the uniform model (see Table 1a). (4) Rutting depth of the non-uniform model is higher than that of the uniform model. Overall for the non-uniform model, more portions of the rutting depths fall ahead (55.1%) of the uniform model than that fall behind (44.9%). The mean value, lower 95% confidence limit, and peak value of nonuniform model are 4.4%, 37.7%, and 84.7% higher than those of the uniform model, respectively (see Table 1a); (5) Peak values of performances appear at those sections with less uniform and weaker elastic moduli Evib distribution (see Figs. 6 and 7 in comparison with Fig. 2);

(6) A pavement section with lower mean value of layer moduli does not correspond to inferior performances since the effects from uniformity of material property may dominate other factors. For example, section 3 has the lowest mean elastic modulus and higher portion of lower moduli than some other sections (i.e. mean Evib is 14.5% and 20.4% lower than that of sections 4 and 5, respectively); however, its pavement performances are obviously better than that of sections 4 and 5 that have lower uniformity (see Fig. 6). For example, its peak value of rutting depth is 34.7% and 25.6% lower, and its fatigue life is 8.86 and 2.73 times higher than that of sections 4 and 5, respectively. This result has further validated that the uniformity or spatial distribution of material property plays a critical role. 3.2. Uniformities by semivariogram models In order to quantify the uniformity, the semivariogram is studied for the geospatial data of ICMVs. Since IC data is geospatial based with 100% of coverage for the entire compaction area in real time with millions of data points recorded, it is regarded that the geostatistics would be a better method than the univariate statistics such as coefficient of variation to evaluate the variability and uniformity [21]. Figs. 8–10 present the semivariogram for the elastic moduli Evib, rutting depth of HMA layer, and fatigue life of HMA layer, respectively. Table 2 summarizes the semivariogram model sill values, a geospatial ‘‘variances’’. The rutting and fatigue performance semivariograms for sections 4 and 5 shows more scattered than that of the HMA layer moduli, which would be due to the nonlinearity and the less uniform distribution of performances. It shows that the variation trends of performances are basically consistent, e.g. section 4/5 has the least uniformity or the highest sill value, then section 2/3, and section 1 has the highest uniformity (lowest sill value) due to its more uniform spatial distribution of HMA elastic moduli (see Table 2). Meanwhile, the performance uniformity trends of sections basically follow that of the HMA layer elastic moduli, e.g. section 4 and 5 have a higher level of sill values than other sections, then sections 2

Semivariogram for Fatigue Life of HMA Layer 2.5E+16

⎡ ⎛ 3h ⎞⎤ r(h) = c ⎢1 − exp ⎜ − ⎟⎥ ⎝ a ⎠⎦ ⎣

Section 1 Section 2

2.0E+16

Semi-variogram

Consequently, theoretical semivariogram models can be fitted to the experimental semivariogram with models such as: power model, exponential model, Gaussian model, etc. [17]. The fitted parameters from the theoretical semivariogram are indicators of uniformity: sill (C + C0), range (R), and nugget (C0) as shown in Fig. 4 as an example. Sill is defined as the plateau that the semivariogram reaches, which represent the value of variance by considering the geospatial dependency. A lower ‘‘sill’’ indicates better uniformity, while the opposite indicates a less uniform condition [16,17]. The exponential model fit was used in this study as it could better fit the experimental semivariogram for the ICMV data [2,8]:

Section 3 Section 4 Section 5

1.5E+16

1.0E+16

5.0E+15

0.0E+00 0

5

10

15

20

25

Lag distance (m) Fig. 10. Experimental semivariogram and exponential model fit for HMA layer fatigue life.

Table 2 Sill values of fitted semivariogram models. Variable

Section 1

Section 2

Section 3

Section 4

Section 5

Evib (MPa2) Rutting (mm2) Fatigue life

49,502 6.48E10 2.41E+15

77,180 1.48E08 3.00E+15

95,248 1.44E08 6.50E+15

11,791 2.03E07 1.48E+16

196,091 1.90E07 1.08E+16

752

Q. Xu et al. / Construction and Building Materials 30 (2012) 746–752

and 3, and section 1 has the highest uniformity. The lower uniformity corresponds to higher peak rutting depth and lower peak fatigue life (see Table 1b, Figs. 6 and 7). However, some discrepancies between the uniformities of moduli and performances are also observed, e.g. for the moduli Evib section 5 has the lowest uniformity, while for the performances section 4 has a little bit lower uniformity than section 5. This could be due to other influence factors including the nonlinearity, e.g. the fatigue performance model accounts for both responses and elastic moduli in a non-linear fashion. 4. Conclusion and recommendation In this paper, the influences of non-uniform or heterogeneous HMA elastic moduli, derived from field measurements with a Bomag IC roller, on pavement performances were evaluated. The M-E PDG performance models were used to predict the rut depths and fatigue lives of the HMA overlay, using the pavement response inputs simulated from a 3-D FE model. The geostatistical semivariogram model is implemented into the developed software to evaluate the uniformity of pavement performances as related to the heterogeneous elastic moduli distribution. The key element of this research is to compare the pavement performance prediction between the heterogeneous HMA layer moduli and uniform HMA layer modulus. Different from conventional pavement analysis and design method, spatial-distributed heterogeneous moduli of the asphalt layer were considered in this work. The pavement performance results indicate that uniformity of material property significantly affect the pavement performance values in terms of severity and uniformity. The spatially distributed non-uniform elastic moduli, Evib measured by Bomag, in the FE model have resulted in inferior performances compared to those from the uniform material model. The case study from the Indiana IC project has shown that the mean and peak fatigue lives for the heterogeneous model are 38.2% and 0.1% of those resulted from the uniform model. A pavement section with overall lower layer moduli does not correspond to inferior performances as the effects from uniformity of material property may dominate other factors. Currently, pavement performance prediction models and structural design procedures are generally based on the assumption of uniform material properties with pavement layers. From this study, it is evident that uniformity of material property should be considered to improve pavement performance prediction and to develop design procedures that reflect material properties in field conditions. Future research is recommended as follows: (1) evaluate pavement performances with non-uniform materials properties for both asphalt and underlying materials; (2) incorporate the geospatial uniformity of compaction into the IC specification for better quality control and assurance; and (3) account the geospatial uniformity of compaction moduli into the pavement performance prediction and design practices with the integration of construction and performance-based pavement design in the future.

Acknowledgments In this research authors would like to acknowledge the support from the FHWA and Transportation Pooled Funding (TPF) project, ‘‘Accelerated Implementation of Intelligent Compaction Technology for Embankment Subgrade Soils, Aggregate Base, and Asphalt Pavement Materials’’ for this study. The IC field demonstrations were made possible by the assistance from the TPF State departments of transportation, IC roller vendors, GPS vendors, and construction contractors. References [1] AASHTO. Hot-mix asphalt paving handbook. 2nd ed.; 2000. [2] Chang GK, Xu Q, Horan RD, Michael L, White D, Vennapusa P. Accelerated implementation of intelligent compaction technology for embankment subgrade soils, aggregate base, and asphalt pavement materials. Final Report, Federal Highway Administration /Transportation Pooled Fund (TPF) No. 954: Washington, DC; 2011. [3] Mooney MA, Rinehart RV, Facas RW, Musimbi OM, White DJ, Vennapusa P. Intelligent soil compaction systems. Final report, National Cooperative Highway Research Program (NCHRP) 676: Washington, DC; 2010. [4] Xu Q, Chang GK, Vennapusa P, White D, Horan RD, Michael L, Gallivan VL. Hot mix asphalt intelligent compaction – a case study. CD-ROM. In: Proceeding of the 89th annual transportation research board meeting, Washington, DC; 2010. [5] Gallivan VL, Chang GK, Horan RD. Practical implementation of intelligent compaction technology in hot mix asphalt pavements. J Assoc Asphalt Paving Technol 2011;80:1–32. [6] Xu Q, Chang KG, Gallivan VL, and Horan RD. Data analysis for hot mix asphalt intelligent compaction. CD-ROM. In: Proceeding of the 90th annual transportation research board meeting. Washington, DC; 2011. [7] Gallivan VL, Chang GK, Xu Q, and Horan RD. Validation of intelligent compaction measurement systems for practical implementation. CD-ROM. In: Proceeding of the 90th annual transportation research board meeting. Washington, DC; 2011. [8] Intelligent compaction website; 2011 . [9] FHWA. Generic IC specifications for HMA, intelligent compaction technology for HMA applications; 2011 . [10] Alkasawneh W, Pan E, Han F, Zhu R, Green R. Effective modulus variation and field distributions with depth- using a multilayered pavement program. Int J Pavement Res Technol 2011;1(2):41–7. [11] Huang YH. Pavement analysis and design. 1st ed. Prentice-Hall, Inc.; 1993. [12] DipI.-Ing, Adam D. Standardization, design, quality assurance and monitoring of earth works in road engineering in Austria (powerpoint presentation). Utépítés és geotechnika – szabályok és tapasztalatok, MAKADÁM-Klub, }ház u, vol. 15; 2007. p. 1–33. Budapest, Lövo [13] Isaaks EH, Srivastava RM. An introduction to applied geostatistics. Oxford Univ. Press: New York; 1989. [14] ANSYS Inc., ANSYISÒ software; 2011 . [15] Applied Research Associate, Inc. Guide for mechanistic-empirical design. NCHRP 1-137A Final Report; 2004 . [16] Wackernagel H. Multivariate geostatistics. 3rd ed. Springer; 2003. [17] Clark I, Harper W. Practical geostatistics. 3rd ed. Columbus, OH: Ecosse North America LLC; 2002. [18] The Transtec Group, Inc. VedaÒ software; 2011 . [19] Chang GK, Xu Q, Dick J, Rutledge J. Technical manual for Veda software. Final report, minessota department of transportation and federal highway administration; 2011. [20] Goden Software. SurferÒ 10; 2011 . [21] Chang GK, Xu Q, Rassmusen R, Merritt D, Michael L, et al. Intelligent compaction data management guide. Research report, federal highway administration; 2011.