Field and laboratory stress-wave measurements of asphalt concrete

Construction and Building Materials 126 (2016) 508–516

Contents lists available at ScienceDirect

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

Field and laboratory stress-wave measurements of asphalt concrete Henrik Bjurström a,⇑, Anders Gudmarsson b, Nils Ryden a, Josefin Starkhammar c a

Department of Civil and Architectural Engineering, KTH Royal Institute of Technology, Brinellvägen 23, 114 28 Stockholm, Sweden Peab Asfalt AB, Drivhjulsvägen 11, 126 30 Hägersten, Sweden c Department of Biomedical Engineering, Lund University, Ole Römers väg 3, 221 00 Lund, Sweden b

h i g h l i g h t s
A 48 MEMS sensor array, providing clear surface wave data, is presented.
Good fit between laboratory and non-contact field measurement regarding stiffness.
Non-contact field measurements showing high repeatability.

a r t i c l e

i n f o

Article history: Received 30 March 2016 Received in revised form 13 July 2016 Accepted 18 September 2016

Keywords: Asphalt concrete Seismic testing Non-contact measurements MEMS receivers Surface waves Dynamic modulus Master curve

a b s t r a c t Non-contact surface wave measurements are performed on a new asphalt concrete (AC) pavement using 48 micro-electro-mechanical system (MEMS) sensors as receivers to estimate the real part of the dynamic moduli of the AC top layer. Laboratory measurements of core samples, extracted from the field measurement positions, are used to construct master curves for comparison with the field measurements. The real parts of the dynamic moduli from the two test methods are consistent at the field measurement temperatures, and the non-contact field measurements are highly repeatable. These results indicate a possible application for quality assurance of AC based on mechanical properties. Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction Material stiffness is one of the most important parameters in pavement design. The dynamic modulus is directly linked to structural capacity [1] and it strongly affects the pavement’s lifespan and deterioration. The modulus for asphalt concrete (AC) is highly dependent on both frequency and temperature; the stiffness of AC decreases with increasing temperatures and/or longer loading times (lower loading frequencies). It is therefore essential to characterize and account for the time and temperature dependency of asphalt in stiffness-based methods used for quality assurance/ quality control (QA/QC). The dynamic modulus in the field can be estimated using a falling weight deflectometer (FWD) that measures the deflection at various offsets from an impact. These measurements can be used to study the deflection from a complete (multi-layered) pavement

⇑ Corresponding author. E-mail address: [email protected] (H. Bjurström). http://dx.doi.org/10.1016/j.conbuildmat.2016.09.067 0950-0618/Ó 2016 Elsevier Ltd. All rights reserved.

structure. However, backcalculation of the mechanical properties from the deflection data is subject to uncertainties. Additionally, results have indicated that this approach is more sensitive to the moduli in the base and subgrade layers and less sensitive to that of the thinner AC layers [2]. It also can be difficult to link a representative frequency to this type of measurement. In a recent study, Varma and Kutay [3] evaluated a frequency- and temperaturedependent dynamic modulus from repeated FWD tests at different temperatures. They presented a dynamic modulus master curve that was backcalculated from the deflection time history. The linear viscoelastic behavior of the AC layer and the nonlinear elastic behavior of the unbound subgrade were introduced in the same model to more accurately characterize the master curve of the AC layer and the material parameters of the unbound underlying layer. In recent years, seismic field measurements on pavement were compared to data from seismic laboratory tests in order to conduct material characterization. Nazarian et al. [4] applied the spectral analysis of surface waves (SASW) method to collect in situ surface

H. Bjurström et al. / Construction and Building Materials 126 (2016) 508–516

wave data and construct a continuous dispersion curve. They successfully compared the in situ measurements to laboratory seismic test results on core samples and showed small differences between the two. Ryden and Park [5] used a method similar to SASW, called the multichannel analysis of surface waves (MASW) method [6], to construct a multichannel data record used to calculate multimodal dispersion curves and subsequently estimate the material properties and layer thickness. Recently, Lin et al. [7] determined the in situ moduli through surface wave measurements and compared them to laboratory moduli from indirect tension testing. However, all previous measurements on AC layers were performed using contact receivers (accelerometers), which present several difficulties. There must be sufficient coupling between the surface and the contact receiver, and this can be difficult to achieve on rough surfaces or for multiple measurements in various locations. Stationary measurements are also costly and labor intensive for large-scale scanning, since a new setup is needed for each individual measuring position. Thus, there is a need for a faster test method for measuring the dynamic field modulus that can cover larger areas. Non-contact sensors (microphones) hold great potential for faster data acquisition when performing large-scale testing, since it is not necessary to set up each individual measurement. Several papers over the past decades have described the use of noncontact receivers for data acquisition in seismic wave testing. Uses include material characterization [8–10], detection of bridge deck delamination [11], and determination of surface crack depths [12]. Bjurström et al. [13] presented measurements obtained using air-coupled microphones that were rolled over a concrete surface. This method provided reliable results for the Rayleigh wave velocity that were comparable to stationary accelerometer measurements. The dynamic modulus can be measured in the laboratory by applying cyclic loading to AC specimens. This conventional laboratory testing for the dynamic modulus is usually performed at 50 micro-strains over a narrow and limited number of loading frequencies, and it is repeated at several different temperatures [14] to capture the viscoelasticity of AC. However, these tests are time consuming and have high costs. Furthermore, there are no in situ test methods that link the results to these conventional laboratory measurements. New, rapid non-destructive testing (NDT) methods are needed that can link results from the field with those from the laboratory. These test methods need to be performed at well-defined temperatures and frequencies and could ideally lead to QA/QC based on the measured mechanical properties of the materials instead of using bulk density and/or void ratio data. Recently, a new and cost-efficient laboratory modal test method, based on the backcalculation of simple frequency response measurements, was developed by Gudmarsson et al. [15]. The assumption of a thermo-rheologically simple material and the time-temperature superposition principle allowed the complex modulus to be expressed over a wide range of frequencies and temperatures (i.e., a master curve) [15]. Then, laboratory measurements of the complex modulus performed by modal testing at the same strain levels (107) as the field measurements can enable a direct comparison between laboratory and field testing. This paper presents a study that evaluates the surface wave velocity of the top AC layer using acoustic non-contact field measurements. The dynamic modulus is then determined using the surface wave velocity through fundamentally correct elastic relationships. An air-coupled receiver array, consisting of 48 microelectro-mechanical system (MEMS) sensors, is constructed and employed for in situ data acquisition to enable fast and effective measurements. The tests are performed in five different sections of a newly built pavement, and each section is constructed using a unique set of layers and mixtures. Once the in situ tests are com-

509

pleted, core samples are extracted from the five test locations to perform modal testing on the same material volumes in a controlled laboratory environment. The laboratory measurements are performed over a range of temperatures and frequencies in order to characterize the master curves, allowing the dynamic moduli to be shifted and presented over a wider range of temperatures and frequencies. Using this approach, the dynamic modulus can be expressed at an arbitrary reference temperature and frequency. Comparisons with field measurements at the measurement temperatures show good agreement between the laboratory and field test results. These results indicate that the presented method can be used for asphalt concrete QA/QC based on the seismic dynamic modulus.

2. Methodology 2.1. Test site The field testing in this study was performed on a newly built highway (riksväg 40) close to the town of Ulricehamn, Sweden. A 2 km portion of highway was built for research purposes and was divided into five sections that were constructed using different designs (see Table 1). The five different test sections were examined during this investigation. The road was not opened to traffic during the in situ measurements. The width of the road is 10.75 m (two driving lanes in one direction) and the lengths of each section are 500 m for the reference section and 375 m for the four sections P1-P4. All tests were performed in the central part of the respective section in order to avoid errors due to the boundaries. The reference section was constructed with four layers of conventional hot mix asphalt (HMA) at a total thickness of 19 cm with unmodified bitumen (penetration grade of 70/100 in examined top layer), according to Swedish specifications of bituminous layers [16]. Section P1 was constructed with the same asphalt mixtures but with three layers of HMA, resulting in a total thickness of 14 cm. Sections P2, P3, and P4 were constructed using the same layer thicknesses as used in P1, but they all contain different bitumen in the asphalt mixtures. Section P2 has a stiffer unmodified bitumen (penetration grade of 50/70 in top layer), Section P3 has a 4% SBS-polymer modified bitumen (PMB), and Section P4 has a 7.5% SBS-polymer modified bitumen (penetration grades of 45/80 and 25/55 respectively). Although all sections were constructed using multiple AC layers, this study was limited to estimates of the dynamic moduli of the top layer (4 cm, highlighted in Table 1) for each respective test section using a wavelength filter (described in Section 2.3). Asphalt concrete is at small strains considered as a linear viscoelastic material. In this study, the field data are presented in terms of wave propagation velocity, thus linear elastic theory. In order to fully characterize the material, the attenuation of surface waves can be analyzed to determine the viscous properties of the AC mixture [17]. The collected data contain information about the viscous properties; however, it is in practice difficult to quantify from in situ measurements. The laboratory measurements also showed that the imaginary parts of the moduli are small compared to the real parts. The attenuation is therefore omitted from this study. Future studies can possibly include attenuation analyses to fully characterize the viscoelastic behavior. In the presented study, all AC mixtures are considered to be isotropic. It cannot be excluded that anisotropy could affect the results to some degree. Prior authors have presented laboratory test results showing anisotropy in AC specimen [18]. Both field (Rayleigh waves) and laboratory measurements used in this study are influenced by both vertical and horizontal stiffness, and can

510

H. Bjurström et al. / Construction and Building Materials 126 (2016) 508–516

Table 1 Overview of the test sections showing the hot mix asphalt (HMA) thickness reduction and the different types of HMA. This study is limited to estimating the dynamic modulus of the top AC layer.

thus be assumed to represent an average dynamic modulus in the case of an anisotropic material. 2.2. MEMS measuring system A new field measuring system was constructed for this study. The system contained 48 MEMS microphones (SPM0408LE5H, Knowles, Itasca, IL, USA) mounted on a straight array with a constant increment dx of 1.0 cm as measured from center-to-center of the acoustic ports. Six circuit boards were constructed, each carrying eight MEMS microphones. The SPM0408LE5H model is operational in the 100 Hz – 10 kHz frequency range, with a sensitivity of about 18 ± 3 dBV/Pa at a 94 dB sound pressure level at 1 kHz. Each circuit board had its power supplied by two 1.5 V AA batteries. The microphones were driven with unity gain (0 dB), resulting in a signal level that matched one of the optional dynamic ranges in the A/D converters (±1 V). The system employed 48 individual A/D converters in the form of six NI PXI-5105 digitizer boards (National Instruments, Austin, TX, USA), each with eight simultaneously sampled channels. The digitizer boards were mounted in a NI PXI-1042 rack and controlled by an embedded controller, NI PXI-8106. (Simultaneous sampling on all channels and boards was realized through the NI PXI-star trigger bus.) The acoustic signals from the 48 microphones were sampled at 300 kHz for a duration of 20 ms and stored to disc for further signal processing in Matlab (The Mathworks, Natick, MA, USA). 2.3. Field measurements The dynamic modulus E of the top AC layer relates to the shear wave velocity VS according to Eq. (1):

E ¼ 2qV 2S ð1 þ mÞ

ð1Þ

where q is the material density and m is the Poisson’s ratio. The shear wave velocity is slightly higher than the Rayleigh wave velocity VR, which in practice can be estimated by Eq. (2) [19].

VR ¼ VS

0:87 þ 1:12m 1þm

ð2Þ

Using Eqs. (1) and (2), the dynamic modulus can be obtained by determining the Rayleigh wave velocity once the density and Poisson’s ratio are known or can be estimated. All field data acquisition was performed using the air-coupled MEMS array described in Section 2.1. The circuit boards holding the MEMS units were mounted on a stiff rod that kept the MEMS sensors in a straight line. The receiver array was always placed parallel to the paving direction, and the MEMS units were located 35 mm above the pavement surface. The array was slightly rotated (see Fig. 1(b)) along its longitudinal axis in order to avoid multiple acoustic reflections in the air, between the pavement surface and the flat circuit boards. A vertical impact was applied in line with

the microphone array using a normal size (0.5 kg) hammer (see Fig. 1(a)) to generate high amplitude waves in the studied frequency range (0–30 kHz). Due to the roughness of the surface, the impact could not be applied at exactly the same distance from the receiver array in the different test sections. However, the offset variation was kept to a minimum and the impacts were applied between 3 cm and 6 cm from the microphone closest to the impact source. The data acquisition was performed using a LabVIEW (National Instruments, Austin, TX, USA) routine in which the microphone closest to the impact was set to trigger the acquisition. All 48 MEMS sensors acquired data simultaneously, creating a multichannel data record that was normalized for each signal individually. The data are plotted in gray in Fig. 2(a). In the data post-processing, a tapered cosine window was applied with a taper to constant value of 0.7, where 1.0 represents a Hann window and 0 is a rectangular window. The time window was defined in order to suppress the direct sound wave from the impact through the air and to accentuate the surface wave. The window was automatically placed relative to the maximum amplitude of the first signal in each multichannel record, and was set to a length of 0.35 ms at zero offset from the impact source. The start times for the window were then adjusted for each signal separately using a velocity of 4000 m/s. Similarly, the end times for the window were adjusted for each signal using a velocity of 400 m/s, just above the sound velocity through the air. This type of increasing window length was used to reduce the risk of introducing artifacts in the phase velocity-frequency domain. The windowed data are plotted in black on top of the raw data in Fig. 2(a). Again, each signal was normalized individually before being plotted so that the Rayleigh wave could be clearly seen. The data were then transformed into the frequency domain using the MASW approach. The frequency and phase velocity resolutions in this transform were set to 10 Hz and 1 m/s, respectively. A wavelength filter within the MASW routine set the amplitudes to zero if the wavelengths were shorter than twice the increment between two adjacent receivers (2 * 1.0 = 2.0 cm) (Nyquist-Shannon sampling theorem) or longer than twice the length of the receiver array (2 * 47 = 94 cm). The wavelength filter was applied to avoid spatial aliasing and a too low spectral resolution. An example of the data presented in the frequency domain is shown in Fig. 2(b). The amplitude was normalized with the maximum amplitude marked as black and the minimum amplitude marked as white. The amplitudes were also normalized for each frequency (vertical line in Fig. 2(b)) individually so that the maximum amplitude at each frequency was equally significant in further analyses. The data presented in Fig. 2 were obtained from a single impact from the reference section in the present study. The Rayleigh wave velocity was characterized by the nondispersive high-frequency asymptote of the fundamental Lamb modes (A0 and S0) [20]. The MEMS units in the receiver array measured the air pressure variation from the leaky energy over time, proportional to the out-of-plane displacement caused by the surface wave. Gibson and Popovics [21] described the out-of-plane

H. Bjurström et al. / Construction and Building Materials 126 (2016) 508–516

511

Fig. 1. (a) The field measurement equipment. (b) The MEMS array set up at the site.

Fig. 2. (a) Multichannel data record from the 48 MEMS units. The red lines represent the time window applied to suppress the direct sound wave through the air and to accentuate the surface wave. (b) MASW imaging of frequency domain data from a single impact together with fundamental dispersion curves. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

displacement as being dominated by the fundamental antisymmetric Lamb mode (A0) when a transient impact is applied vertically to the surface. Consistently, the exemplified MASW dispersion curve data in Fig. 2(b) clearly show a good fit to the theoretical A0 and S0 dispersion curves. These dispersion curves correspond to realistic input data for the shear wave velocity VS, plate thickness d, and Poisson’s ratio m of this particular AC (VS = 1750 m/s, d = 0.19 m, and m = 0.30). Multiple theoretical dispersion curves can be fitted to such MASW data field in order to determine the mechanical properties and thickness of plate structures [5]. Note that the Lamb modes are not important in this specific study since only the surface wave velocity is estimated. However, the A0 and S0 modes are plotted in Fig. 2(b) to demonstrate how they coincide with the Rayleigh wave velocity at higher frequencies. Once the settings for the calculations (described above) were chosen, the data from each measurement (impact) were evaluated objectively and automatically in order to estimate the Rayleigh wave velocity. Ten individual measurements were obtained at the same position in each of the five test sections. The estimated results from the individual measurements were then compared to verify good repeatability. The standard deviation for the evaluated Rayleigh wave velocities was calculated for the ten measurements in each test section, and the results are provided in Section 3 below. However, to obtain the most robust results over a wide frequency range, the data record in the frequency domain (Fig. 2(b)) from the ten individual measurements were normalized and

stacked according to Fig. 3. The results from these ten impacts are thus shown in a single figure as a single Rayleigh wave velocity. The dominating A0 mode is visualized through the maximum surface wave amplitude marked in green at each individual frequency using an example field data record. The Rayleigh wave has a limited penetration depth and the particle motion is confined close to the surface. Any significant particle motion is limited to one-half to one-third of the wavelength (k/2  k/3) [22]. Since the aim of this study was to estimate the stiffness of the top layer (4 cm, see Table 1), only a limited wavelength range was included in the estimation of the Rayleigh wave velocity. Wavelengths equal to twice the top layer thickness, 2 * 4 cm = 8 cm ± 20% (thus, 6.4 cm 6 k 6 9.6 cm), are highlighted within the red wedge shape in Fig. 3. The mean amplitude within this wavelength range was then calculated for each individual phase velocity image (horizontal line in Fig. 3). Finally, the phase velocity having the maximum amplitude within the highlighted wavelength range was taken as the Rayleigh wave velocity at each test section. Other possible methods to detect the Rayleigh wave velocity are available. A commonly used method for estimating the surface wave velocity is to study two or more signals in the time domain. This involves identifying the first high-amplitude peak, which characterizes the Rayleigh wave, in the beginning of the wave package for each signal, and dividing the known offset between the receivers by the time difference. However, summing the amplitudes after the MASW approach in the frequency domain as

512

H. Bjurström et al. / Construction and Building Materials 126 (2016) 508–516

Fig. 3. Frequency domain data from the individual impacts are normalized and stacked to create a compiled image containing 10 measurements. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

described above is a robust alternative method that is less sensitive when multiple modes are present. Due to the frequency dependency of AC, a representative frequency is needed to compare with the laboratory measurements. The representative frequency for the field measurements is then considered to be determined from the studied wavelength (8 cm) and the calculated Rayleigh wave velocity. It should be emphasized that the measured phase velocity is an average over this microphone array length (47 cm). To account for the temperature dependency when comparing the field measurement results to the laboratory results, the temperature was measured at each of the five test sections in conjunction with each field measurement. A small hole was drilled in each test section to enable temperature measurements at a depth of 3 cm in the asphalt. The holes were allowed to cool for more than 15 min before the measurement to allow any heat generated during drilling to dissipate. 2.4. Laboratory testing

Fig. 4. Laboratory measurement setup needed for modal testing [24].

Modal testing developed by Gudmarsson et al. [15,23] was used here to determine the master curves of the AC cores extracted from the different sections in the field. In contrast to conventional testing methods, modal testing is applicable to specimens with arbitrary geometries and dimensions, which simplifies the comparison between field and laboratory testing. The core samples in this study had a diameter of approximately 150 mm and a height of approximately 30 mm, and they were taken from the exact same line as the microphone array used in the field measurements at each test section. The principle of the modal test method is to measure the free vibration of a specimen that is excited by an impact. An accelerometer (PCB model 35B10) was used to measure the acceleration and an instrumented hammer (PCB model 086R80) measures the applied load. A computer, a signal conditioner (PCB model480B21), and a data acquisition device (NI USB-6251M Series), all needed to perform the modal testing, are shown in Fig. 4. The measured acceleration and load were used to determine the frequency response functions, H (f), through Eq. (3):

Hðf Þ ¼

n 1X Y k ðf Þ  X k ðf Þ n k¼1

!,

! n 1X  X k ðf Þ  X k ðf Þ ; n k¼1

ð3Þ

where Y (f) is the measured acceleration, X (f) is the measured applied force, and X⁄(f) is the complex conjugate of the applied force. The frequency response functions (FRFs) include information on both the elastic stiffness and the viscous damping. The complex

modulus and the complex Poisson’s ratio can therefore be characterized by matching the finite-element-computed FRFs to the measurements. Isotropic linear viscoelastic behavior was assumed for the AC, and the frequency- and temperature-dependent complex modulus and complex Poisson’s ratio were expressed by the Havriliak-Negami model (Eqs. (4) and (5), respectively) [25,26] and the Williams-Landel-Ferry (WLF) shift factor equation (Eq. (6)) [27]:

E ðx; TÞ ¼ E1 þ 

ðE0  E1 Þ

v  ðx; TÞ ¼ v 1 þ  log aT ðTÞ ¼

a b

1 þ ðixaT ðTÞsÞ ðv 0  v 1 Þ

1 þ ðixaT ðTÞsP Þ

c1 ðT  T ref Þ ; c2 þ T  T ref

;

a b

ð4Þ

;

ð5Þ

ð6Þ

where E0 and v0 are the low-frequency values of the complex modulus and complex Poisson’s ratio, respectively. E1 and v1 are the corresponding high-frequency values, x is the angular frequency, a governs the width of the loss factor peak, and b governs the asymmetry of the loss factor peak. s = 1/x0 is the relaxation time which describes the position of the loss factor peak along the frequency axis where x0 is the frequency at the loss factor peak, and s and

513

H. Bjurström et al. / Construction and Building Materials 126 (2016) 508–516

sp govern the complex modulus and the complex Poisson’s ratio, respectively. The master curve was determined by estimating unique values of the parameters in Eqs. (4)–(6) by fitting the computed FRFs to the measurements. More specifically, the computed FRFs were optimized to find values of each the eight parameters (E1, a, b, s, sp, v1, c1 and c2) that resulted in a good fit to all of the measured FRFs at different temperatures. An assumption was made that the same values of a, b, c1, and c2 were valid for both the complex modulus and the complex Poisson’s ratio. This simplification in determining the complex Poisson’s ratio has been shown to provide a good match between the computed and measured FRFs. The low-frequency parameters, which did not affect the FRFs for realistic values of AC, were assumed to be E0 = 100 MPa and v0 = 0.5 [23]. The use of shift factors in Eqs. (4) and (5) to determine computed FRFs (master curve parameters) that match all of the measured FRFs, is equivalent to determine master curves by shifting conventional measured data in a narrow frequency range to a continuous master curve [14]. A comparison of the modal test to conventional cyclic loading showed small differences in the dynamic modulus at higher temperatures and lower frequencies, which was not surprising due to the different strain levels applied in the two tests [28]. This difference decreased with decreasing temperatures and increasing frequencies, and despite the different strain levels, the two laboratory test methods resulted in a similar complex Poisson’s ratio.

3.3. Laboratory results Fig. 5 shows the complex modulus determined through modal testing of the core specimen from section P4 where the highly modified bitumen containing 7.5% SBS-polymers was used. The figure presents the modulus determined over a limited frequency range and at different temperatures, being shifted to a continuous master curve representing the modulus at a reference temperature of 15 °C. The Figs. 5(a)–(c) show respectively the dynamic modulus, the phase angle, and the complex modulus in the Cole-Cole space. Since the complex modulus plotted in the Cole-Cole space is independent of time and temperature shifting, Fig. 5(c) can serve as an indication of the accuracy of the estimated master curve. If the data from each temperature fit a unique curve, it is a good indication that the assumption of a thermo-rheologically simple material is correct. The good fit between the master curve and the moduli of the different temperatures in Fig. 5(c) (R2 = 0.995) shows that, even in the case of a highly modified asphalt mixture, the modal test method is able to characterize the complex master curve with high accuracy. This is probably due to the low strain levels and high loading frequencies applied in the test. Accurate estimations of the complex modulus master curves were also obtained for the other test sections. 3.4. Comparison of field and laboratory results The obtained field measurement results (Rayleigh wave velocities) were transformed into dynamic moduli using Eqs. (1) and (2). Laboratory observations of specimens provided the densities and Poisson’s ratios used in Eqs. (1) and (2) and are shown in Table 2. Since the core samples must be extracted from the pavement to perform the laboratory measurements, these two parameters are most accurately determined in a controlled environment. The densities for the different test sections were almost constant, and they only varied 1.3% between the maximum and the minimum. Table 3 shows that the Poisson’s ratios for all test sections were mostly similar to each other, except the Poisson’s ratio for P3 was a little lower. However, changing this Poisson’s ratio to the highest value found in all sections (0.340 for Section P1) only affected the dynamic modulus by 3.4%. Thus, it was concluded that the Poisson’s ratio has low impact on the stiffness. The laboratory master curves and field measurement results are presented at their respective field temperatures in Fig. 6. The exact results and the discrepancies between the field and laboratory measurements are also shown in Table 3. Since the measured Rayleigh wave velocity was evaluated as a real constant without an imaginary component (damping), the dynamic moduli from the field were compared with the real part of the dynamic moduli (storage moduli, E0 ) from the laboratory. The dynamic moduli measurements from the laboratory and field varied from 0.5% to 6.4% across the different test sections. These discrepancies were small compared to those in alternative tests. Varma and Kutay [3] compared the modulus obtained from the creep compliance curve to that obtained using a backcalculation of FWD deflection data and demonstrated higher uncertainties. The Swedish National Road and Transport Research Institute [29] also found large differences in conventional cyclic loading tests between AC core samples and

3. Results 3.1. General setup Data were collected from five different test sections as described in Section 2.1, and they were evaluated following the procedure in Section 2.2 in order to extract one surface wave velocity value for each section. These values were then transformed into dynamic moduli using Eqs. (1) and (2) to enable a comparison between the field and laboratory tests.

3.2. Field measurement results All collected field data were processed automatically and objectively (as described in Section 2.3) in order to estimate the Rayleigh wave velocity. Table 2 shows the obtained velocities and temperatures at which the field measurements were performed. The standard deviation is shown for the Rayleigh wave velocity of the individual impacts, and it indicates how repeatable the measurements are in terms of estimating the surface wave velocity. In this study, the coefficients of variation (i.e. standard deviation divided by the mean value) ranged from 0.3% to 1.2%, hence indicating high repeatability. Also, note that the representative frequency was slightly different in the five test sections depending on the evaluated Rayleigh wave velocity (according to Section 2.3).

Table 2 Evaluated Rayleigh wave velocity for each test section. The standard deviation indicates the repeatability of the individual impacts. Test section

Ref.

P1

P2

P3

P4

Temperature (°C) Rayleigh wave velocity, VR (m/s) Standard deviation (m/s) Representative frequency (kHz)

18.1 1745 9.4 21.8

26.2 1634 19.5 20.4

20.9 1717 6.5 21.5

24.0 1645 10.6 20.6

26.7 1392 4.4 17.4

514

H. Bjurström et al. / Construction and Building Materials 126 (2016) 508–516

Fig. 5. The complex modulus of the P4 core specimen with 7.5% SBS-polymers. The master curve is covering a wide frequency range by shifting measurements, performed at different temperatures over a narrow frequency range, to a chosen reference temperature of 15 °C. (a) Dynamic modulus, (b) phase angle, and (c) Cole-Cole diagram.

Table 3 Evaluated dynamic moduli from laboratory and field measurements. Poisson’s ratio and density are also determined in the laboratory.

Elab (MPa) Efield (MPa) Diff. Efield/Elab (%) Poisson’s ratio Density (kg/m3) * DT (°C) *

Ref.

P1

P2

P3

P4

20,708 21,603 4.3 0.315 2336.3 +2.1

18,375 19,263 4.8 0.340 2332.1 +1.9

22,554 21,113 6.4 0.310 2345.7 3.0

18,910 19,010 0.5 0.265 2359.5 +0.2

14,124 13,852 1.9 0.328 2362.0 0.6

Dynamic modulus ratio expressed as a temperature difference, as described in Section 4.

AC specimens constructed in a laboratory. There were large differences within the group of specimens as well as high variance in the results from the same specimen when it was tested in different laboratories. 4. Discussion The field measurements results correspond very well with the laboratory results, with differences between 0.5% and 6.4%. After the field measurements were performed, AC samples were cored from the same locations. Thus, the field and laboratory measurements should be directly comparable. However, there was a difference in scale in the material volumes, as the laboratory study examined an AC disc with 15 cm diameter, and the field result was based on the average surface wave velocity over the entire receiver array length, 47 cm. The comparison also relied on the assumption that there was no significant variation within the field tested volume. However, from Fig. 2(a) it appears that the Rayleigh wave velocity between the 48 equally spaced receivers was constant, which could indicate a homogenous body in terms of Rayleigh wave velocity. Despite the identical top layer thicknesses and mixtures in the reference section and section P1, it was concluded that their moduli were different. This discrepancy can most certainly, beside the difference in temperature, be assigned the different number of layers in the pavement. It should be noted that by using surface wave

testing, the underlying layers should not be affected nor influence the results. However, the degree of compaction is different depending on the underlying layers. These layers can thus be said to influence the top layer modulus indirectly, although this difference is assigned the paving process and not the test method. In order to consider the temperature dependency of AC in these measurements, the temperature was measured using a digital thermometer at a depth of 3 cm. It is difficult in general to determine the correct temperature within the AC layer since there may be a temperature gradient in the field. This temperature gradient could be included in future studies to attempt to improve the results. In the present study, there is a possible uncertainty as the temperature was measured at a single depth. To investigate how a different temperature would affect the results, the field temperature alone was changed to find that temperature at which the laboratory measurement and field measurement match exactly. The temperature of the specimen in the laboratory remained unchanged, so it was possible to show if any discrepancies between the laboratory and field results could be due to a wrongly measured temperature. Table 3 shows the temperature change corresponding to the discrepancy between the laboratory and field measurement results at the respective test sections. The differences in measured dynamic moduli corresponded to temperature changes between 0.2 °C and 3.0 °C. Due to the relatively high temperatures of the AC at the time of the measurements (see Table 2), it can be assumed that the temperature gradient was high, which

H. Bjurström et al. / Construction and Building Materials 126 (2016) 508–516

515

Fig. 6. Master curves constructed using a reference temperature of 15 °C and shifted to the field measurement temperature at each section. The obtained field data are plotted as squares for comparison.

made it more difficult to estimate the temperature correctly. No certain conclusions can be made regarding whether the differences in the results were caused by misinterpreted field temperatures; however, it is believed that any inaccuracies in the temperature would not be the sole cause of the differences. The presented study demonstrates a novel test method for top layer AC moduli, showing great potential for faster QA/QC in the future where non-contact field measurements can be utilized. However, the measurements presented are limited to five different constructions. Although they present small differences between field and laboratory, more measurements are required to fully validate the test method. Future testing should also include a wider temperature range than what was possible for this study. 5. Summary and conclusion Non-contact surface wave measurements obtained from a newly built highway were analyzed to estimate the dynamic moduli. The testing was performed using a microphone array constructed with 48 MEMS units. Core samples were extracted from the same measuring line as used in the field tests. Seismic laboratory measurements were then performed to characterize the top AC layer. Master curves were constructed over a wide frequency range at the field test temperatures. Comparisons between the field and laboratory test results showed small differences in the dynamic moduli. Furthermore, it was shown that the field measurements are repeatable in terms of the dynamic modulus. Ten individual measurements were evaluated for the Rayleigh wave velocity, and a maximum coefficient of variation of 1.2% indicated highly repeatable testing.

The minor differences observed between the two presented test methods indicated that seismic measurements may enable QA/QC of pavement AC layers based on mechanical properties such as the dynamic modulus.

Acknowledgments The Swedish transport administration (Trafikverket) and Swedish construction industry’s organization for research and development (SBUF) are gratefully acknowledged for their financial support. Valuable recommendations from Prof. John Popovics and Dr. Suyun Ham on the type of MEMS sensors used are greatly appreciated.

References [1] S.F. Said, Test and design parameters of elastic stiffness and permanent deformation, VTI särtryck, vol. 242, Swedish National Road and Transport Research Institute (VTI), Lecture at the Course on Bituminous Pavements: Materials, Design and Evaluation at Uleåborg, Finland, 25–28 April 1995. [2] M.F. Aouad, K.H. Stokoe II, J.M. Roesset, Evaluation of flexible pavements and subgrades using the spectral-analysis-of-surface-waves (SASW) method, Research Report 1175-7F, Research Project 2/3-18-88/1-1175 Development of dynamic analysis techniques for falling weight deflectometer data, Center for Transportation Research, The University of Texas at Austin, 1993. [3] S. Varma, M.E. Kutay, Backcalculation of viscoelastic and nonlinear flexible pavement layer properties from falling weight deflections, Int. J. Pavement Eng. 17 (5) (2016) 388–402, http://dx.doi.org/10.1080/ 10298436.2014.993196. [4] S. Nazarian, D. Yuan, V. Tandon, Structural field testing of flexible pavement layers with seismic methods for quality control, J. Transp. Res. Board 1654 (1999) 50–60, http://dx.doi.org/10.3141/1654-06.

516

H. Bjurström et al. / Construction and Building Materials 126 (2016) 508–516

[5] N. Ryden, C.B. Park, Fast simulated annealing inversion of surface waves on pavements using phase velocity spectra, Geophysics 71 (4) (2006) R49–R58, http://dx.doi.org/10.1190/1.2204964. [6] C.B. Park, R.D. Miller, J. Xia, Multichannel analysis of surface waves, Geophysics 64 (3) (1999) 800–808, http://dx.doi.org/10.1190/1.1444590. [7] S. Lin, J.C. Ashlock, R.C. Williams, Nondestructive quality assessment of asphalt pavements based on dynamic modulus, Constr. Build. Mater. 112 (2016) 836– 847. [8] M. Lukkala, P. Heikkila, J. Surakka, The wave resonance – a contactless test method, Ultrasonics 9 (4) (1971) 201–208, http://dx.doi.org/10.1016/0041624X(71)90387-8. [9] M. Castaings, P. Cawley, The generation, propagation, and detection of Lamb waves in plates using air-coupled ultrasonic transducers, J. Acoust. Soc. Am. 100 (5) (1996) 3070–3077, http://dx.doi.org/10.1121/1.417193. [10] J. Zhu, J.S. Popovics, Non-contact detection of surface waves in concrete using an air-coupled sensor, AIP Conf. Proc. 615 (2002) 1261–1268, http://dx.doi. org/10.1063/1.1472940. [11] S.-H. Kee, T. Oh, J.S. Popovics, R.W. Arndt, J. Zhu, Nondestructive bridge deck testing with air-coupled Impact-Echo and infrared thermography, J. Bridge Eng. 17 (2012) 928–939, http://dx.doi.org/10.1061/(ASCE)BE.19435592.0000350, 928-939. [12] S.-H. Kee, J. Zhu, Using air-coupled sensors to determine the depth of a surfacebreaking crack in concrete, J. Acoust. Soc. Am. 127 (3) (2010) 1279–1287, http://dx.doi.org/10.1121/1.3298431. [13] H. Bjurström, N. Ryden, B. Birgisson, Non-contact surface wave testing of pavements: comparing a rolling microphone array with accelerometer measurements, Smart Struct. Syst. 17 (1) (2016) 1–15, http://dx.doi.org/ 10.12989/sss.2016.17.1.001. [14] D. Perraton, H. Di Benedetto, C. Sauzéat, B. Hofko, A. Graziani, Q.T. Nguyen, S. Pouget, L.D. Poulikakos, N. Tapsoba, J. Grenfell, 3Dim experimental investigation of linear viscoelastic properties of bituminous mixtures, Mater. Struct. 49 (11) (2016) 4813–4829, http://dx.doi.org/10.1617/s11527-0160827-3. [15] A. Gudmarsson, N. Ryden, B. Birgisson, Characterizing the low strain complex modulus of asphalt concrete specimens through optimization of frequency response functions, J. Acoust. Soc. Am. 132 (4) (2012) 2304–2312, http://dx. doi.org/10.1121/1.4747016. [16] Trafikverket, Bitumenbundna lager, document ID: TDOK 2013:0529, Swedish Transport Administration, 2013.

[17] C.-W. In, J.-Y. Kim, K.E. Kurtis, L.J. Jacobs, Characterization of ultrasonic Rayleigh surface waves in asphaltic concrete, NDT&E Int. 42 (7) (2009) 610– 617, http://dx.doi.org/10.1016/j.ndteint.2009.04.007. [18] H. Di Benedetto, C. Sauzéat, P. Clec’h, Anisotropy of bituminous mixture in the linear viscoelastic domain, Mech. Time-Depend. Mat. 20 (3) (2016) 281–297, http://dx.doi.org/10.1007/s11043-016-9305-0. [19] L. Bergmann, Ultrasonics and their Scientific and Technical Applications, John Wiley Sons, New York, 1948. [20] H. Lamb, On waves in an elastic plates, Proc. R. Soc. Lond. Ser. A 93 (648) (1917) 114–128, http://dx.doi.org/10.1098/rspa.1917.0008. [21] A. Gibson, J.S. Popovics, Lamb wave basis for Impact-Echo method analysis, J. Eng. Mech. – ASCE 131 (4) (2005) 438–443, http://dx.doi.org/10.1061/(ASCE) 0733-9399(2005). [22] W. Lowrie, Fundamentals of Geophysics, Cambridge University Press, United Kingdom, 1997, ISBN 0-521-46164-2. [23] A. Gudmarsson, N. Ryden, H. Di Benedetto, C. Sauzéat, N. Tapsoba, B. Birgisson, Comparing linear viscoelastic properties of asphalt concrete measured by laboratory seismic and tension-compression tests, J. Nondestr. Eval. 33 (4) (2014) 571–582. [24] A. Gudmarsson, Resonance testing of asphalt concrete (PhD thesis), KTH Royal Institute of Technology, Stockholm, 2014. ISBN 978-91-87353-50-5. [25] S. Havriliak, S. Negami, A complex plane analysis of a-dispersions in some polymer systems, J. Polym. Sci. C 14 (1966) 99–117, http://dx.doi.org/10.1002/ polc.5070140111. [26] W.M. Madigosky, G.F. Lee, J.M. Niemiec, A method for modeling polymer viscoelastic data and the temperature shift function, J. Acoust. Soc. Am. 119 (2006) 3760–3765, http://dx.doi.org/10.1121/1.2195292. [27] M.L. Williams, R.F. Landel, J.D. Ferry, The temperature dependence of relaxation mechanisms in amorphous polymers and other glass-forming liquids, J. Am. Chem. Soc. 77 (1955) 3701, http://dx.doi.org/ 10.1021/ja01619a008. [28] A. Gudmarsson, N. Ryden, H. Di Benedetto, C. Sauzéat, Complex modulus and complex Poisson’s ratio from cyclic and dynamic modal testing of asphalt concrete, Constr. Build. Mater. 88 (2015) 20–31, http://dx.doi.org/10.1007/ s10921-014-0253-9. [29] L. Viman, H. Hakim, S. Said, Styvhetsmodul – jämförande provning, Swedish National Road and Transport Research Institute (VTI – Statens väg- och transportforskningsinstitut), Report Reference Number: 2004/0636-29, 2007.