**4. Laboratory Tests**

The mechanical sensors come out with a predefined set of characteristics given by the manufacturers. However, it was considered important to verify the properties of the sensors, for typical pavement loading conditions, before installing them on an actual test site. In the application for pavement instrumentation, the sensitivity of the sensors is of utmost importance, due to the low displacement and acceleration levels. In addition, the signal to noise ratio of the sensors, with the actual acquisition system, and in real measurement conditions, needed to be determined. These tests were performed with different signal amplitudes and frequencies corresponding to typical heavy vehicle loading.

The tests were carried out at IFSTTAR using a vibrating table, with a hydraulic actuator, which applies the displacement to the horizontal plate with sensors installed on it. The responses of the different sensors were compared with the measurements of a reference Keyence laser displacement sensor. The following characteristics were determined:


The tests on the vibrating table were carried out using controlled displacement signals, reproducing the deflection created by a 5-axle semi-trailer truck. These deflection signals were defined using pavement calculations, performed with the Alize pavement design software for the pavement structure in Table 1. A signal of realistic shape was thus produced, and then applied at different amplitudes and frequencies, to simulate different vehicle loads and speeds.

Four sensors can be tested at the same time on the vibrating table, as the acquisition system has four channels. Out of four, one channel was reserved to the Keyence laser sensor, used as reference to measure the displacements of the vibrating table. The measurements of the geophones and accelerometers require some treatment in order to convert them into appropriate values of vertical displacements or deflections. The measurements are to be integrated from velocity and accelerations into vertical displacements (deflections). For analyses purpose, Figure 3 represents a raw geophone signal (measured displacement velocity), and the signal obtained after a simple integration. The same trend is observed for the values of acceleration after integration. It can be seen that after a simple integration process, the displacement signal obtained does not correspond exactly to the reference signal applied to the vibrating table (signal calculated with Alize). As can be seen in Figure 3, the amplitudes of the peaks differ from the reference signal, and the shape of the signal is also different, showing some positive (upward) displacements. These upward movements are not realistic as only a downward displacement is applied to the sensors. The differences between the reference signal and the sensor signal are due to two main reasons:


signals, which do not represent the actual vertical displacement of the pavement and needs to be filtered [14]. signals, which do not represent the actual vertical displacement of the pavement and needs to be filtered [14]. *Infrastructures* **2020**, *5*, 25 6 of 22

signals, which do not represent the actual vertical displacement of the pavement and needs to

*Infrastructures* **2020**, *5*, 25 6 of 22

**Figure 3.** Example of raw and integrated geophone GS11D signal, for a 5-axle truck loading. **Figure 3.** Example of raw and integrated geophone GS11D signal, for a 5-axle truck loading.

### **5. Signal Processing and Filtering of Sensor Responses 5. Signal Processing and Filtering of Sensor Responses Figure 3.** Example of raw and integrated geophone GS11D signal, for a 5-axle truck loading.

### *5.1. Scheme to Convert the Sensor Response into Vertical Displacement and Improving the Accuracy 5.1. Scheme to Convert the Sensor Response into Vertical Displacement and Improving the Accuracy* **5. Signal Processing and Filtering of Sensor Responses**

One of the objectives of this work is to find an appropriate method to convert the vertical velocities and accelerations into realistic values of vertical pavement deflections, which can then be used for back-calculation of the pavement layer moduli, as described in Figure 4. For that, a signalprocessing method has been tested and a methodology to correct the measurements has been developed. This procedure has been applied to the measurements of the two types of geophones used (Geospace GS11D and Ion LF-24) and two types of accelerometers (MEMSIC and CXL04GP1). In the following figures, explaining the treatment procedure, only the responses obtained with the geophone GS11D are shown for illustration. This methodology is described in the following steps. One of the objectives of this work is to find an appropriate method to convert the vertical velocities and accelerations into realistic values of vertical pavement deflections, which can then be used for back-calculation of the pavement layer moduli, as described in Figure 4. For that, a signal-processing method has been tested and a methodology to correct the measurements has been developed. This procedure has been applied to the measurements of the two types of geophones used (Geospace GS11D and Ion LF-24) and two types of accelerometers (MEMSIC and CXL04GP1). In the following figures, explaining the treatment procedure, only the responses obtained with the geophone GS11D are shown for illustration. This methodology is described in the following steps. *5.1. Scheme to Convert the Sensor Response into Vertical Displacement and Improving the Accuracy*  One of the objectives of this work is to find an appropriate method to convert the vertical velocities and accelerations into realistic values of vertical pavement deflections, which can then be used for back-calculation of the pavement layer moduli, as described in Figure 4. For that, a signalprocessing method has been tested and a methodology to correct the measurements has been developed. This procedure has been applied to the measurements of the two types of geophones used (Geospace GS11D and Ion LF-24) and two types of accelerometers (MEMSIC and CXL04GP1). In the following figures, explaining the treatment procedure, only the responses obtained with the geophone GS11D are shown for illustration. This methodology is described in the following steps.

**Figure 4.** Data conversion scheme. **Figure 4.** Data conversion scheme. **Figure 4.** Data conversion scheme.

### Step1: Filtering of the Signals Step1: Filtering of the Signals Step1: Filtering of the Signals

The first step consists in suppressing the lower frequency components of the signal using a recursive filter. The advantages of using Infinite Impulse Response (IIR) filter are that the noise suppression of the filtering is very low and the computation is faster [15]. One of the downsides of using the IIR filter could be having unequal phase delays at each frequency component. For this reason, a zero-phase shifting filter is used, which uses forward and reverse filtering processes to compensate for the delay. As the output is created with a significant delay, it is fed back to the filter, to reverse the data points in time and align the signal to its original phase. The first pass performs the The first step consists in suppressing the lower frequency components of the signal using a recursive filter. The advantages of using Infinite Impulse Response (IIR) filter are that the noise suppression of the filtering is very low and the computation is faster [15]. One of the downsides of using the IIR filter could be having unequal phase delays at each frequency component. For this reason, a zero-phase shifting filter is used, which uses forward and reverse filtering processes to compensate for the delay. As the output is created with a significant delay, it is fed back to the filter, to reverse the data points in time and align the signal to its original phase. The first pass performs the The first step consists in suppressing the lower frequency components of the signal using a recursive filter. The advantages of using Infinite Impulse Response (IIR) filter are that the noise suppression of the filtering is very low and the computation is faster [15]. One of the downsides of using the IIR filter could be having unequal phase delays at each frequency component. For this reason, a zero-phase shifting filter is used, which uses forward and reverse filtering processes to compensate for the delay. As the output is created with a significant delay, it is fed back to the filter, to reverse the data points in time and align the signal to its original phase. The first pass performs

the forward direction filtering using IIR filters and bypassing this signal through a zero-phase shift filter the backward filtering is achieved. In this work, an IIR highpass Chebyshev filter gave the best results, as it allows certain frequencies to pass and suppresses the lower frequencies containing noise and unwanted frequency content. It was observed that the cut-off frequency of the filter has a huge impact on the filtering process, and needs to be adapted to the vehicle speed. For low speeds, a cut-off frequency of 4.5 Hz is selected, which is increased as the speed is increased. For the maximum speed (20 m/s), a cut-off frequency of 20 Hz is used. The Figure 5 shows the signal after filtering. forward direction filtering using IIR filters and bypassing this signal through a zero-phase shift filter the backward filtering is achieved. In this work, an IIR highpass Chebyshev filter gave the best results, as it allows certain frequencies to pass and suppresses the lower frequencies containing noise and unwanted frequency content. It was observed that the cut-off frequency of the filter has a huge impact on the filtering process, and needs to be adapted to the vehicle speed. For low speeds, a cutoff frequency of 4.5 Hz is selected, which is increased as the speed is increased. For the maximum speed (20 m/s), a cut-off frequency of 20 Hz is used. The Figure 5 shows the signal after filtering. forward direction filtering using IIR filters and bypassing this signal through a zero-phase shift filter the backward filtering is achieved. In this work, an IIR highpass Chebyshev filter gave the best results, as it allows certain frequencies to pass and suppresses the lower frequencies containing noise and unwanted frequency content. It was observed that the cut-off frequency of the filter has a huge impact on the filtering process, and needs to be adapted to the vehicle speed. For low speeds, a cutoff frequency of 4.5 Hz is selected, which is increased as the speed is increased. For the maximum speed (20 m/s), a cut-off frequency of 20 Hz is used. The Figure 5 shows the signal after filtering.

*Infrastructures* **2020**, *5*, 25 7 of 22

*Infrastructures* **2020**, *5*, 25 7 of 22

**Figure 5.** Geophone GS11D signal after filtering. **Figure 5.** Geophone GS11D signal after filtering. **Figure 5.** Geophone GS11D signal after filtering.

### Step 2: Signal Amplification Step 2: Signal Amplification

It was observed that when the cut off frequencies are kept lower than the values indicated above, the noise was not sufficiently reduced. Hence, the shape of the signal was deteriorated. With high cut off frequencies, filtering improved the overall shape of the signal but reduced the amplitude of the signal. When the final filtered signal was compared to the reference signal, the amplitude was lower. For this reason, after filtering, the signal had to be amplified (by a linear constant amplification factor), to keep the initial amplitude as shown in Figure 6. Step 2: Signal AmplificationIt was observed that when the cut off frequencies are kept lower than the values indicated above, the noise was not sufficiently reduced. Hence, the shape of the signal was deteriorated. With high cut off frequencies, filtering improved the overall shape of the signal but reduced the amplitude of the signal. When the final filtered signal was compared to the reference signal, the amplitude was lower. For this reason, after filtering, the signal had to be amplified (by a linear constant amplification factor), to keep the initial amplitude as shown in Figure 6. It was observed that when the cut off frequencies are kept lower than the values indicated above, the noise was not sufficiently reduced. Hence, the shape of the signal was deteriorated. With high cut off frequencies, filtering improved the overall shape of the signal but reduced the amplitude of the signal. When the final filtered signal was compared to the reference signal, the amplitude was lower. For this reason, after filtering, the signal had to be amplified (by a linear constant amplification factor), to keep the initial amplitude as shown in Figure 6.

**Figure 6.** Signal amplification after filtering. **Figure 6.** Signal amplification after filtering. **Figure 6.** Signal amplification after filtering.

### Step 3: Signal Integration and Detrending Step 3: Signal Integration and Detrending

After filtering, the vertical velocity is integrated to be converted into vertical displacement and the vertical acceleration is double integrated in a similar manner, to get the displacement. Simply integrating and double integrating the velocity and acceleration would increase small frequencies in the signal as well. Hence pre-processing is necessary before the integration of the signals. After filtering, the vertical velocity is integrated to be converted into vertical displacement and the vertical acceleration is double integrated in a similar manner, to get the displacement. Simply integrating and double integrating the velocity and acceleration would increase small frequencies in the signal as well. Hence pre-processing is necessary before the integration of the signals. Step 3: Signal Integration and DetrendingAfter filtering, the vertical velocity is integrated to be converted into vertical displacement and the vertical acceleration is double integrated in a similar manner, to get the displacement. Simply integrating and double integrating the velocity and acceleration would increase small frequencies in the signal as well. Hence pre-processing is necessary before the integration of the signals.

As mentioned above, the integration step adds a continuous component to the signal as described by Equations (1) and (2). Hence, a detrending function is applied to suppress this continuous component. The signal after integration and detrending is shown in Figure 7. As mentioned above, the integration step adds a continuous component to the signal as described by Equations (1) and (2). Hence, a detrending function is applied to suppress this continuous component. The signal after integration and detrending is shown in Figure 7.

*Infrastructures* **2020**, *5*, 25 8 of 22

$$V(n) = \sum\_{n=1}^{1} (A(n-1) + A(n)) \* \Delta t. \tag{1}$$

$$\Delta t = \sum^{1} \dots \tag{n}$$

$$d(n) = \sum\_{n}^{1} (V(n-1) + V(n)) \star \Delta t \tag{2}$$

where V is the velocity, A is the acceleration, d is the displacement and ∆t is the difference in time. where V is the velocity, A is the acceleration, d is the displacement and ∆t is the difference in time.

**Figure 7.** Geophone GS11D signal after integration process. **Figure 7.** Geophone GS11D signal after integration process.

### Step 4: Application of the Hilbert Transform to the Signal Step 4: Application of the Hilbert Transform to the Signal

Due to the oscillations observed in the signals of the geophones and accelerometers, it seemed appropriate to use the Hilbert transform, to eliminate these oscillations from the sensor response. Time and frequency data analysis using the Hilbert transform specifically for structural health monitoring application has shown favorable results for interpreting structure response [16]. Other applications of the Hilbert transform for vibration analysis include machine diagnosis, signal decomposition and industrial applications like health monitoring of powering systems [17]. The common application of the Hilbert transform is the demodulation operation for extracting the envelope of the initial signal. As the oscillations detected in the signal from the sensors recall those of modulated signals, hence the Hilbert transform is used in our study. This function shifts the phase component of the signal by ± 90 degrees, and thus the Hilbert transform of a sine signal would be a cosine signal with the same magnitude. Hence, when the signal is combined with its Hilbert transform the periodical oscillations in the signal are removed and the upward lifts in the signal are eliminated. In the time domain, the signal is the convolution of the signal and the Hilbert transform of the signal as expressed in Equation (3), where X(t) is the Hilbert transform H of the signal x(t) Due to the oscillations observed in the signals of the geophones and accelerometers, it seemed appropriate to use the Hilbert transform, to eliminate these oscillations from the sensor response. Time and frequency data analysis using the Hilbert transform specifically for structural health monitoring application has shown favorable results for interpreting structure response [16]. Other applications of the Hilbert transform for vibration analysis include machine diagnosis, signal decomposition and industrial applications like health monitoring of powering systems [17]. The common application of the Hilbert transform is the demodulation operation for extracting the envelope of the initial signal. As the oscillations detected in the signal from the sensors recall those of modulated signals, hence the Hilbert transform is used in our study. This function shifts the phase component of the signal by ±90 degrees, and thus the Hilbert transform of a sine signal would be a cosine signal with the same magnitude. Hence, when the signal is combined with its Hilbert transform the periodical oscillations in the signal are removed and the upward lifts in the signal are eliminated. In the time domain, the signal is the convolution of the signal and the Hilbert transform of the signal as expressed in Equation (3), where X(t) is the Hilbert transform H of the signal x(t)

$$\mathbf{X}(\mathbf{t}) = \mathbf{H}[\mathbf{x}(\mathbf{t})] = \frac{1}{\pi} \times \frac{\mathbf{x}(\tau)}{(t-\tau)} d\tau = (\mathbf{h} \ast \mathbf{w})(\mathbf{t}) \tag{3}$$

where h(t) = <sup>ଵ</sup> where h(t) = 1 π*t* .

of the real signal.

గ௧ In the frequency domain, it is obtained with the Fourier transform and corresponds to the multiplication of the complex signal by –jsgn(w), where the signum function (sgn) is defined by Equation (4) [18]. This multiplication produces a phase shift of the signal by ±90° or <sup>±</sup> <sup>గ</sup> ଶ , to generate the Hilbert transform. Hence, in the frequency domain, the Hilbert transform is actually a phase shift In the frequency domain, it is obtained with the Fourier transform and corresponds to the multiplication of the complex signal by –jsgn(w), where the signum function (sgn) is defined by Equation (4) [18]. This multiplication produces a phase shift of the signal by ±90◦ or ± π 2 , to generate the Hilbert transform. Hence, in the frequency domain, the Hilbert transform is actually a phase shift of the real signal.

$$\text{FT}\{\mathbf{x}(\mathbf{t})\} = \mathbf{X}(\mathbf{w})[-\text{jsgn}\mathbf{w}]\text{Sgn}(\mathbf{w}) = \begin{cases} -1 \text{ for } w < 0 \\ 0 \text{ for } w = 0 \\ 1 \text{ for } w > 0 \end{cases} \tag{4}$$

*Infrastructures* **2020**, *5*, 25

The principle of the Hilbert transform is to combine a signal that is the real part (analytical signal) of the signal and an imaginary part of the signal used to extract the envelope of the signal. Hence, the corrected signal is given by: Sgn(w) = ቐ−1 ݂ݎ ݓ > 0 0 = ݓ ݎ݂ 0 0 < ݓ ݎ݂ 1 (4)

*Infrastructures* **2020**, *5*, 25 9 of 22

$$\mathbf{S}\_{\mathbf{a}}(t) = \mathbf{s}(t) + \mathbf{j}\hat{\mathbf{s}}(t) \tag{5}$$

The corrected signal at the end of the process S<sup>a</sup> (t) corresponds to the envelope of the signal s(t),. This analytical signal is determined by the amplitude (A) and phase Ø, defined by the signal and an imaginary part of the signal used to extract the envelope of the signal. Hence, the corrected signal is given by:

$$\mathbf{A(t)} = \sqrt{\left| \text{Real } \text{parf} \{ \mathbf{S(t)}^2 \} \right| - \left| \text{imaginary } \text{part} \left( \mathbf{S(t)} \right)^2 \right|} \tag{6}$$

$$=\sqrt{\mathcal{S}^2(t) + \hat{\mathcal{S}}^2(t)} = \left| \mathbb{S}\_a(t) \right| \tag{7}$$

And the phase is given as

And the phase is given as

$$\mathcal{Q} = \tan^{-1} \left[ \frac{i \,\text{maginary} \,\text{component}}{real \,\text{part}} \right] \tag{8}$$

(7) |(ݐ)ܵ| = (ݐ)ଶܵ +) ݐ)ଶܵට=

This function A also describes the dynamical behavior of amplitude modulation of the signal. This is why it is also known as the amplitude of the envelope. The function of phase is a measure of the signal at an instant in time and is therefore known as the instantaneous phase [19]. In this case, this treatment drastically improves the shape of the signal, and leads to a response close to the theoretical deflection under a five-axle vehicle, as evident in Figure 8. (8) ൨ ݐݎܽ ݈݁ܽݎ This function A also describes the dynamical behavior of amplitude modulation of the signal. This is why it is also known as the amplitude of the envelope. The function of phase is a measure of the signal at an instant in time and is therefore known as the instantaneous phase [19]. In this case, this treatment drastically improves the shape of the signal, and leads to a response close to the theoretical deflection under a five-axle vehicle, as evident in Figure 8.

**Figure 8.** Final signal of geophone GS11D after application of the Hilbert transform. **Figure 8.** Final signal of geophone GS11D after application of the Hilbert transform.

### *5.2. Comparison of the Reference Laser Sensor with the Improved Measurements 5.2. Comparison of the Reference Laser Sensor with the Improved Measurements*

Finally, the proposed correction procedure has been applied to the geophone and accelerometer signals obtained at different speeds and amplitudes, and the results have been compared with the reference deflection values obtained with the Keyence Laser sensor (Figure 9). Generally, the procedure improves the geophone and accelerometer measurements significantly and leads to realistic signal shapes and displacement amplitudes, as described in the section below. Finally, the proposed correction procedure has been applied to the geophone and accelerometer signals obtained at different speeds and amplitudes, and the results have been compared with the reference deflection values obtained with the Keyence Laser sensor (Figure 9). Generally, the procedure improves the geophone and accelerometer measurements significantly and leads to realistic signal shapes and displacement amplitudes, as described in the section below.

## *5.3. Results*

This section presents examples of laboratory results, namely displacement values obtained for a speed of 70 km/h (approx. 20 m/s), for the two geophones and two accelerometers (see Figure 10). These values are treated in the same manner as described in Section 5.1. The treated signals are compared with the response of the Keyence laser displacement sensor and it is important to note that the treatment procedure is the same for all the amplitudes and speeds. The parameter which is different is the cut-off frequency of the filter, which needs to be increased as the speed increases. With

0

**6. Accelerated Pavement Tests** 

*6.1. Experimental Setup* 

deflection response.

different cut off frequencies, the amplification factors are varied and the optimal shapes of the signals are obtained, which are very close to the reference signals. *Infrastructures* **2020**, *5*, 25 10 of 22

*Infrastructures* **2020**, *5*, 25 10 of 22

**Figure 9.** Example of comparison between the geophone signal obtained with and without the improved processing method. **Figure 9.** Example of comparison between the geophone signal obtained with and without the improved processing method. different cut off frequencies, the amplification factors are varied and the optimal shapes of the signals are obtained, which are very close to the reference signals.

(**a**) *Infrastructures* **2020**, *5*, 25 11 of 22

**Figure 10.** (**a**): Displacement measurements with geophones at a speed of 70 km/h and amplitude of 0.8 mm. (**b**): Displacement measurements with accelerometers at a speed of 70 km/h and amplitude of 0.8 mm. **Figure 10.** (**a**): Displacement measurements with geophones at a speed of 70 km/h and amplitude of 0.8 mm. (**b**): Displacement measurements with accelerometers at a speed of 70 km/h and amplitude of 0.8 mm.

The sensors described in Section 2 (two geophones and two accelerometers) were embedded in a pavement structure tested on the IFSTTAR accelerated pavement testing (APT) facility. A reference, anchored deflectometer was also installed, to serve as a reference for the measurement of the

The IFSTTAR APT, or fatigue carrousel, is a circular outdoor testing facility that consists of a circular test track of 40 m diameter and a central loading system (Figure 11). The carousel has four identical loading arms, each equipped with a wheel carriage which can carry loads up to 13 tons.

The wheel carriages can comprise different wheel arrangements. In this experiment, dual wheels were used. The wheel dimensions are shown in Figure 12, and the tire-pavement contact stress was equal to 0.6 MPa. The sensors were installed in one of the six test sections of the carousel, called Section S1 (Figure 13). This pavement section was 22 m long and consisted of an 11 cm thick asphalt concrete layer, and a granular base. The positions of the geophones, accelerometers and anchored deflectometer on the pavement section are shown in Figure 14. The geophones and accelerometers were placed 1 cm below the pavement surface, in the center of the wheel path. Each sensor placed in small boreholes drilled in the asphalt layer, and sealed with resin. A small trench was also cut in the pavement, up to the pavement edge, and the sensor cable was inserted in this trench and sealed with resin. The anchored deflectometer consists of an LVDT which is connected to a rod, anchored at a significant depth (here three meters), and which measures the total vertical displacement between

**Figure 11.** carrousel APT at IFSTTAR.

the pavement surface and the bottom of the rod, supposed fixed.

deflection response.

of 0.8 mm.

*6.1. Experimental Setup* 

**6. Accelerated Pavement Tests** 



Deflection mm

0

0.2

### **6. Accelerated Pavement Tests** were used. The wheel dimensions are shown in Figure 12, and the tire-pavement contact stress was equal to 0.6 MPa. The sensors were installed in one of the six test sections of the carousel, called

### *6.1. Experimental Setup* Section S1 (Figure 13). This pavement section was 22 m long and consisted of an 11 cm thick asphalt

The sensors described in Section 2 (two geophones and two accelerometers) were embedded in a pavement structure tested on the IFSTTAR accelerated pavement testing (APT) facility. A reference, anchored deflectometer was also installed, to serve as a reference for the measurement of the deflection response. concrete layer, and a granular base. The positions of the geophones, accelerometers and anchored deflectometer on the pavement section are shown in Figure 14. The geophones and accelerometers were placed 1 cm below the pavement surface, in the center of the wheel path. Each sensor placed in small boreholes drilled in the asphalt layer, and sealed with resin. A small trench was also cut in the pavement, up to the pavement edge, and the sensor cable was inserted in this trench and sealed with

*Infrastructures* **2020**, *5*, 25 11 of 22

**Displacement for accelerometers at speed 70 km/h and amplitude 0.8mm** 

(**b**) **Figure 10.** (**a**): Displacement measurements with geophones at a speed of 70 km/h and amplitude of 0.8 mm. (**b**): Displacement measurements with accelerometers at a speed of 70 km/h and amplitude

0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5

Time(s)

Reference (laser)

ACC\_SD ACC\_CX

The sensors described in Section 2 (two geophones and two accelerometers) were embedded in a pavement structure tested on the IFSTTAR accelerated pavement testing (APT) facility. A reference, anchored deflectometer was also installed, to serve as a reference for the measurement of the

The IFSTTAR APT, or fatigue carrousel, is a circular outdoor testing facility that consists of a circular test track of 40 m diameter and a central loading system (Figure 11). The carousel has four

The wheel carriages can comprise different wheel arrangements. In this experiment, dual wheels

The IFSTTAR APT, or fatigue carrousel, is a circular outdoor testing facility that consists of a circular test track of 40 m diameter and a central loading system (Figure 11). The carousel has four identical loading arms, each equipped with a wheel carriage which can carry loads up to 13 tons. resin. The anchored deflectometer consists of an LVDT which is connected to a rod, anchored at a significant depth (here three meters), and which measures the total vertical displacement between the pavement surface and the bottom of the rod, supposed fixed.

**Figure 11.** carrousel APT at IFSTTAR. **Figure 11.** Carrousel APT at IFSTTAR.

The wheel carriages can comprise different wheel arrangements. In this experiment, dual wheels were used. The wheel dimensions are shown in Figure 12, and the tire-pavement contact stress was equal to 0.6 MPa. The sensors were installed in one of the six test sections of the carousel, called Section S1 (Figure 13). This pavement section was 22 m long and consisted of an 11 cm thick asphalt concrete layer, and a granular base. The positions of the geophones, accelerometers and anchored deflectometer on the pavement section are shown in Figure 14. The geophones and accelerometers were placed 1 cm below the pavement surface, in the center of the wheel path. Each sensor placed in small boreholes drilled in the asphalt layer, and sealed with resin. A small trench was also cut in the pavement, up to the pavement edge, and the sensor cable was inserted in this trench and sealed with resin. The anchored deflectometer consists of an LVDT which is connected to a rod, anchored at a significant depth (here three meters), and which measures the total vertical displacement between the pavement surface and the bottom of the rod, supposed fixed. *Infrastructures* **2020**, *5*, 25 12 of 22

**Figure 12.** Dual wheel loading dimensions. **Figure 12.** Dual wheel loading dimensions.

S3 S4 The characteristics of the pavement structure where the sensors were installed are given in Table 2. The characteristics of the pavement layers were determined using the following tests:

• Thicknesses of the different layers were determined from the controls made during and after construction

S1

Instrumented area

S2

19m

S5

S6

**Figure 13.** APT section dimensions

Access

**Figure 14.** Instrumented pavement structure.

• Thicknesses of the different layers were determined from the controls made during and after

• Bituminous material complex moduli were determined from laboratory tests on the field

• The moduli of the granular layer were determined by back-calculation, from FWD

• The moduli of the subgrade were determined from Benkelman beam deflection measurements.

2. The characteristics of the pavement layers were determined using the following tests:

construction

produced mix

measurements

The characteristics of the pavement structure where the sensors were installed are given in Table

• Bituminous material complex moduli were determined from laboratory tests on the field produced mix

*Infrastructures* **2020**, *5*, 25 12 of 22

*Infrastructures* **2020**, *5*, 25 12 of 22


**Figure 13.** APT section dimensions **Figure 13.** APT section dimensions

**Figure 14.** Instrumented pavement structure. **Figure 14.** Instrumented pavement structure.


**Figure 14.** Instrumented pavement structure. The characteristics of the pavement structure where the sensors were installed are given in Table **Table 2.** Characteristics of the pavement structure.

produced mix • The moduli of the granular layer were determined by back-calculation, from FWD measurements • The moduli of the subgrade were determined from Benkelman beam deflection measurements. • The moduli of the granular layer were determined by back-calculation, from FWD measurements • The moduli of the subgrade were determined from Benkelman beam deflection measurements. The tests were conducted at different speeds and positions of the wheels with respect to the sensor positions, as shown in Figure 15. The wheels can move in the transverse direction, over 11 different positions, with a spacing of 10.5 cm. Position 6 corresponds to the central position, where the sensors are placed between the two wheels. For positions 4 and 8, the sensors are placed under the center of one wheel.

• Bituminous material complex moduli were determined from laboratory tests on the field

### *6.2. Tests Results*

produced mix

The tests were carried out at three load levels (45, 55, and 65 kN), at speeds varying from 6 m/s to 20 m/s and at surface temperatures of 19 ◦C to 20 ◦C. For each test condition, the deflections measured, after signal processing, with the two types of geophones and two types of accelerometers (defined in Figure 1) were compared with the anchored deflectometer measurements. In this section, in Figures 16–19 the following acronyms are used for the sensors: Geo\_ion for Geophone ion, Geo\_GS11D for Geophone

the center of one wheel.

GS11D, Acc\_SD for Accelerometer Silicon Design and ACC\_CX for accelerometer CXL04GP1. Figure 16 shows the results obtained for speeds of 8 m/s and 20 m/s and position 6 of the wheels. different positions, with a spacing of 10.5 cm. Position 6 corresponds to the central position, where the sensors are placed between the two wheels. For positions 4 and 8, the sensors are placed under the center of one wheel.

sensor positions, as shown in Figure 15. The wheels can move in the transverse direction, over 11

*Infrastructures* **2020**, *5*, 25 13 of 22

The tests were conducted at different speeds and positions of the wheels with respect to the sensor positions, as shown in Figure 15. The wheels can move in the transverse direction, over 11 different positions, with a spacing of 10.5 cm. Position 6 corresponds to the central position, where the sensors are placed between the two wheels. For positions 4 and 8, the sensors are placed under

*Infrastructures* **2020**, *5*, 25 13 of 22 **Table 2.** characteristics of the pavement structure. **Pavement Layer Thickness (cm) Modulus (MPa)**  Bituminous concrete 11 9441(15 °C and 10 Hz) Granular base 30 145 Subgrade 260 110

> **Table 2.** characteristics of the pavement structure. **Pavement Layer Thickness (cm) Modulus (MPa)**  Bituminous concrete 11 9441(15 °C and 10 Hz) Granular base 30 145 Subgrade 260 110

**Figure 15.** Different wheel positions with respect to the sensors. **Figure 15.** Different wheel positions with respect to the sensors. position 6 of the wheels.

**Figure 16.** (**a**): Deflection for 45 kN and speed of 8 m/s. (**b**): Deflection for 45 kN and speed of 20 m/s. (**c**): Deflection for 55 kN and speed of 8 m/s. (**d**): Deflection for 55 kN and speed of 20 m/s. (**e**): Deflection for 65 kN and speed of 8 m/s. (**f**): Deflection for 65 kN and speed of 20 m/s **Figure 16.** (**a**): Deflection for 45 kN and speed of 8 m/s. (**b**): Deflection for 45 kN and speed of 20 m/s. (**c**): Deflection for 55 kN and speed of 8 m/s. (**d**): Deflection for 55 kN and speed of 20 m/s. (**e**): Deflection for 65 kN and speed of 8 m/s. (**f**): Deflection for 65 kN and speed of 20 m/s

(**a**) (**b**)

16 m/s and the maximum deflection is obtained at positions 4 and 6.

Figure 17 shows the deflection responses corresponding to different lateral positions of the

(**c**) (**d**)

(**e**) (**f**) **Figure 16.** (**a**): Deflection for 45 kN and speed of 8 m/s. (**b**): Deflection for 45 kN and speed of 20 m/s. (**c**): Deflection for 55 kN and speed of 8 m/s. (**d**): Deflection for 55 kN and speed of 20 m/s. (**e**):

Figure 17 shows the deflection responses corresponding to different lateral positions of the wheels with respect to the sensors. In position 6, the sensors are placed between the wheels whereas,

Deflection for 65 kN and speed of 8 m/s. (**f**): Deflection for 65 kN and speed of 20 m/s

16 m/s and the maximum deflection is obtained at positions 4 and 6.

**Figure 17.** (**a**): Deflections at position 6 and 45 kN load. (**b**): Deflections at position 4 and 45 kN load. (**c**): Deflections at position 6 and 55 kN load. (**d**): Deflections at position 4 and 55 kN load. (**e**): Deflections at position 6 and 65 kN load. (**f**): Deflections at position 4 and 65 kN load. **Figure 17.** (**a**): Deflections at position 6 and 45 kN load. (**b**): Deflections at position 4 and 45 kN load. (**c**): Deflections at position 6 and 55 kN load. (**d**): Deflections at position 4 and 55 kN load. (**e**): Deflections at position 6 and 65 kN load. (**f**): Deflections at position 4 and 65 kN load.

The results of Figures 16 and 17 show that the deflections obtained with the two geophones and two accelerometers, using the proposed signal processing method, are very close to the reference deflections measured with the deflectometer. The four sensors also give very similar results. Figure 18 shows the variations of the maximum deflections with increasing speed, from 6 to 20 m/s at position 6. In general, a slight decrease of deflections is observed when the speed increases, Figure 17 shows the deflection responses corresponding to different lateral positions of the wheels with respect to the sensors. In position 6, the sensors are placed between the wheels whereas, in position 4, the sensors are under the center of one wheel, as shown in Figure 15. The speed is kept 16 m/s and the maximum deflection is obtained at positions 4 and 6.

except for the speed of 6 m/s, for which the results present more scatter. Figure 19 shows the variation of the maximum deflections with different wheel positions and a speed of 16 m/s. For all positions, the results obtained with all the sensors are very similar. The results of Figures 16 and 17 show that the deflections obtained with the two geophones and two accelerometers, using the proposed signal processing method, are very close to the reference deflections measured with the deflectometer. The four sensors also give very similar results.

**Figure 18.** (**a**): Variation of deflections with the loading speed (from 6 to 20 m/s) for load levels 45 kN. (**b**): Variation of deflections with the loading speed (from 6 to 20 m/s) for load levels 55 kN. (**c**): Variation of deflections with the loading speed (from 6 to 20 m/s) for load levels 65 kN.

Figure 18 shows the variations of the maximum deflections with increasing speed, from 6 to 20 m/s at position 6. In general, a slight decrease of deflections is observed when the speed increases, except for the speed of 6 m/s, for which the results present more scatter. Figure 19 shows the variation of the maximum deflections with different wheel positions and a speed of 16 m/s. For all positions, the results obtained with all the sensors are very similar. **Figure 18.** (**a**): Variation of deflections with the loading speed (from 6 to 20 m/s) for load levels 45 kN. (**b**): Variation of deflections with the loading speed (from 6 to 20 m/s) for load levels 55 kN. (**c**): Variation of deflections with the loading speed (from 6 to 20 m/s) for load levels 65 kN.

*Infrastructures* **2020**, *5*, 25 17 of 22

(**b**)

**Figure 19.** (**a**): Evolution of deflections with the position of the wheels, for 45 kN load and 16 m/s speed. (**b**): Evolution of deflections with the position of the wheels, for 55 kN load and 16 m/s speed. (**c**): Evolution of deflections with the position of the wheels, for 65 kN load and 16 m/s speed.

To evaluate more precisely the difference between the anchored deflectometer signal, and the signals of the evaluated sensors, an error indicator R was defined. This indicator is expressed by Equation (9). It is defined as the mean relative difference, in percentage, between the values measured by each evaluated sensor and the reference anchored deflectometer, divided by the total number of measurement points N of the signal.

$$R = \frac{1}{N} \sum \frac{References\ deflection - sensor\ deflection}{Max(References\ deflection)} \times 100\tag{9}$$

Table 3 summarizes the values of the percentages of difference (error indicator R) obtained between the signals of the geophones and accelerometers, and the signals of the reference anchored deflectometer, for the different test conditions. A mean value of error indicator R is also given for the measurements made for the same load level, and different speeds. This indicator gives an estimate of the accuracy of the measurements for this particular application. It is possible to conclude that:



**Table 3.** relative differences in percentage between the reference deflections (anchored displacement sensors) and those measured by the geophones and accelerometers, for 45, 55, and 65 kN loads.

### **7. Data Analysis**

### *7.1. Back Calculation Methodology*

The procedure used for the back-calculation of pavement layer moduli is based on the linear elastic pavement design software ALIZE [12]. This software is used to calculate the response of the experimental pavement (Table 2) under the APT dual wheel loading, and the corresponding deflection basin. This theoretical deflection basin is then compared with the measured basin, and an iterative method is used to adjust the layer moduli, until the best match between the calculated and measured response is obtained. The methodology used for the optimization of the layer moduli is the following:


**Table 4.** Limit values of layer moduli defined for the back calculation.

