Using NDT Data to Assess the Effect of Pavement Thickness Variability on Ride Quality

: Pavement condition largely determines its long-term behavior and is of paramount importance for rehabilitation and maintenance management. The use of non-destructive testing (NDT) systems to assess pavement condition has gained much popularity. Often, well-known NDT systems are combined to take full advantage of the capabilities of each system. Combining independent NDT systems to optimize the assessment process is a scientiﬁc challenge. With this in mind, the purpose of this paper is to investigate the extent to which data from two independent NDT systems can be combined: pavement thickness obtained with ground penetrating radar (GPR) and roughness data obtained with a road surface proﬁler (RSP). In particular, the objective of this study is to determine whether the expected variations in asphalt layer thickness, due to the construction process and the different pavement cross sections along the same road/highway road, may have an impact on pavement roughness as expressed in International Roughness Index (IRI) values. GPR and roughness data are collected, processed, and analyzed. The analysis results show that thickness variations are reﬂected in pavement roughness. The greater the variation in asphalt layer thickness, the greater the IRI values. Furthermore, it is argued that the GPR capabilities can be used for an initial assessment of the expected pavement quality.


Introduction
The deterioration rate of asphalt pavements depends on several factors, including not only material properties and pavement design, but also external factors such as the environment and traffic loading. Pavements are designed to withstand the effects of environmental and traffic stresses, and their long-term behavior determines the pavement's performance. Evaluating pavement condition through monitoring provides valuable information for predicting pavement behavior and thus performance.
Non-destructive testing (NDT) of in-service pavements has become the preferred and most accurate method for evaluating pavement performance. It has the advantage over conventional methods (e.g., core drilling) of not affecting pavement structure and being less time consuming. In addition, most NDT systems allow continuous measurements at steady or high travel speeds.
NDT systems used for pavement evaluation and monitoring include ground penetrating radar (GPR), profilometers, deflectometers, and skid resistance measurement systems. GPR is mainly used for crack detection, determination of pavement layer thickness, and detection of voids and moisture damage [1][2][3]. Deflectometers record the response of the pavement when subjected to an impulse load. Vertical deflections are measured with sensors placed at a specific distance from the center of load and can be used to assess the structural condition of pavements through appropriate processing and analysis. Profilometers are used to determine the roughness, rutting, and texture of pavements.
with foamed asphalt (FDR-FA). The combination of GPR and FWD proved to be extremely useful for locating and identifying asphalt delamination and moisture-prone areas [28].
The literature review revealed that the most common practice is to combine GPR with other NDT systems such as FWD to assess the structural condition of the pavement or damage within the pavement structure. However, the main function of pavements is to provide a smooth surface, i.e., a surface with low roughness. Roughness is directly related to ride quality and comfort, but also largely determines pavement life and maintenance costs [29]. Therefore, roughness is one of the different types of deterioration that measure the functional condition of pavements. Roughness can be affected by several parameters, such as road category (urban vs. rural), base type, HMA mix classification, construction and materials quality, age/traffic, and environmental parameters over time. Rural pavements have lower IRI values compared to urban, which can be associated with the fact that rural pavement construction generally has fewer interruptions, resulting in a smoother surface [30]. The HMA mix seems to have an effect on roughness especially in the case of SMA (stone mastic asphalt). SMA mixes are produced with a higher content of coarse aggregate than dense graded mixes, leading to a rough surface [31]. Open graded base courses are associated with higher level of roughness resulting from the absence of fine aggregates that poses a negative effect on workability and compaction [30]. According to [32], the traffic level (accumulated traffic with the passage of time) affects the roughness but does not affect the deterioration rate for a pavement constructed on a good quality subgrade. It was also demonstrated that the roughness deterioration rate under low temperature condition is higher than under medium temperature condition.
For the evaluation of pavement roughness, the International Roughness Index (IRI) is considered a reliable index [33]. The IRI is a measure for roughness based on the response of a generic motor vehicle to roughness of the road surface. Essentially, it describes the mounting and dropping of a lateral profile of the total investigated length. Therefore, it is determined based on the geometric characteristics of the road surface. Various measurement systems are available for the recording and measurement of IRI. According to [34], they are categorized into four levels depending on their accuracy. The most accurate are those that record the pavement surface profile and convert it to an IRI value. Systems that fall into this category are laser profilers, dipstick, walking profiler, profilometers, and inertial profilers.
The recording and evaluation of the IRI index has two objectives: acceptance of a new pavement as part of quality control and monitoring of the functional condition of the pavement during its service life in order to plan necessary maintenance measures.
However, as mentioned earlier, pavement thickness plays an important role in pavement quality and performance. Given the importance of evaluating pavement roughness on the one hand and the dependence of pavement performance on pavement thickness on the other, the question arises whether profilometer and GPR data can be combined to provide more comprehensive information on pavement quality. So far, not much attention has been paid to this research area, and there are only a few studies that address the relationship between IRI and pavement thickness. According to the analysis results of [30], the initial roughness of asphalt pavements is highly dependent on asphalt layer thickness. An increase in asphalt layer thickness leads to a decrease in IRI. This is consistent with another study [32], which investigated the factors affecting pavement roughness. Pavement thickness was calculated as the sum of the equivalent asphalt layer thickness for each layer of the pavement structure. Various conditions such as precipitation, temperature, traffic, and subgrade were considered. In almost all cases, the increase in pavement thickness had a noticeable effect on pavement roughness, in the sense that the thicker the pavement, the lower the IRI value. In a recent study, the factors contributing to pavement performance were investigated using different approaches to analyze Long-Term Pavement Performance Program (LTPP) data [35]. One of the main findings was that asphalt layer thickness is a statistically significant parameter affecting pavement roughness. In particular, the mean IRI value decreases with increasing asphalt layer thickness. A recent study [36] investigated the combined use of FWD, GPR, and profilometers to evaluate pavement roughness. GPR thickness data were used along with deflection data from FWD as input for estimating the structural condition of the pavement. The analysis of the integration of GPR, FWD, and profilometer data clearly showed that there is a relationship between the condition of the subgrade and the pavement quality.
All the afore-mentioned studies recognize that pavement has the most noticeable effect on roughness.
With this in mind, this study addresses the integration of GPR thickness data with roughness data to better assess pavement quality and performance. Along a highway section, the thickness of the asphalt layers varies due to the different pavement cross sections resulting from the pavement design. However, variations also occur within the same cross section mainly due to the construction process and poor-quality control procedures. The asphalt mixture construction depends on a variety of factors, including environmental conditions, characteristics of the mixture, and operating systems parameters (i.e., type of compaction system). In other words, the thickness of asphalt layers in paved condition can be expected to vary around the target thickness (design thickness) of a road section. Similar research has shown that there is a large variation in thickness between the designed thickness and the actual paved thickness [37,38]. These variations follow a normal distribution [39][40][41]. On the other hand, thickness variation along a road section appears due to alternative pavement cross sections. These variations are random.
Based on the above, this study aims to extend the results of previous studies by focusing on the effects of thickness variation (instead of absolute thickness) on pavement ride quality, expressed in IRI values. Thickness variations resulting either from expected variations between the design and paved condition or from the transition from one cross section to another along the same highway road were considered. Investigating the effects of thickness variation on pavement performance (in terms of ride quality) enables the assessment of performance models and equations.

Test Sections
To achieve the objective of the present study, a road test was conducted. Specifically, NDT measurements were made along two highway flexible pavement sections (A and B). The total length was 60 km. Both cross sections consist of the asphalt layers, the base, the subbase, and the subgrade. Asphalt layers consist of the wearing course and the asphalt base layer. The wearing course is a semi open graded mixture with 80-100 asphalt penetration modified with SBS 4% and limestone and steel slag aggregate. The asphalt base layer mixture is dense graded and was produced using 50-70 grade asphalt penetration and limestone aggregates. The base and subbase courses are dense graded and consist of unbound aggregates. The subgrade support is considered good (CBR > 20%). Each highway section was further divided in subsections with the same thickness design. The asphalt layers design thicknesses are given in Table 1. The climatic conditions prevailing in regions of the two highways do not show significant differences (Mediterranean climate). Highway sections A and B are part of a periodical monitoring system and as such a data base of the pavements characteristics is available and is updated with a new set of collected data on an annual basis. Roughness measurements are realized once in a year during the same period of the year. Therefore, in order to eliminate the effect of traffic in the analysis, the roughness data were retrieved from the database so that traffic (accumulated annual average daily traffic) was on approximately the same level for all sections.
Given that the parameters of materials, the traffic, and the climatic conditions are more or less common in all sections, the analysis will include only the effect of asphalt mix thickness variation in roughness index IRI.
For the measurements, a GPR system was used to determine the asphalt layer thickness and an inertial profiler called Road Surface Profilometer-RSP was used to determine the IRI index.

GPR Survey
In order to record the pavement stratigraphy and determine the layer thicknesses, an air-coupled GPR system was used with an antenna of 1000 MHz frequency Type 4108 Horn [42]. The fundamental operating principle of the system is based on the electromagnetic theory, since the signals that are emitted are electromagnetic waves that can be mathematically described by the equations of electromagnetic theory (Maxwell). The GPR system is positioned close to the under-investigation surface, so that the waves are directed to it. When they encounter material with different electromagnetic properties (in this case different pavement layers), on the one hand part of them is reflected back to the receiver and on the other hand the propagation velocity changes. Principally, the GPR system measures the two-way travel time of emitted and reflected EM wave. Based on this time and the properties of the materials, the thicknesses of the individual layers of the pavement are calculated. Figure 1 presents an example of the amplitudes and time intervals observed in case of a flexible pavement. The key component with a direct effect on the wave velocity is the dielectric constant of the materials. Pavement layers constitute of different materials and as such each layer has a different dielectric constant. The relationship between dielectric constant and velocity is inversely proportional. Thus, the higher the dielectric constant, the lower the electromagnetic wave velocity. Dielectric constant of the various pavement layers can be calculated considering the peak amplitudes (A 0 , A 1 , . . . , Ai) of the interfaces and the amplitude of the incident GPR signal [43]. For the determination of the incident GPR signal, a perfect electromagnetic reflector (copper plate) is used ( Figure 2).  The basic apparatus of a GPR system is the following: • The antenna that varies according to the survey needs. For the present investigation, an air coupled antenna of 1 GHz central frequency was used. • A special software for the recording and storage of the collected data [44]. • A special software for the processing and analysis of the recorder data [45].

RSP Survey
RSP is an inertial profiler that uses a combination of laser and accelerometer to measure the longitudinal elevation profile of the road for high degree accuracy at highway speeds [46]. Its basic unit is a rigid beam of 3.2 m length, where seven lasers and two accelerometers are embedded. The beam is adapted to the front of a properly configured vehicle ( Figure 3). RSP records the following parameters every 25 mm during the vehicle movement: • The vertical displacement of the beam from the road surface.

•
The vertical acceleration of the beam.

•
The time and distance that the two aforementioned parameters are recorded. The basic apparatus of a GPR system is the following: • The antenna that varies according to the survey needs. For the present investigation, an air coupled antenna of 1 GHz central frequency was used. • A special software for the recording and storage of the collected data [44]. • A special software for the processing and analysis of the recorder data [45].

RSP Survey
RSP is an inertial profiler that uses a combination of laser and accelerometer to measure the longitudinal elevation profile of the road for high degree accuracy at highway speeds [46]. Its basic unit is a rigid beam of 3.2 m length, where seven lasers and two accelerometers are embedded. The beam is adapted to the front of a properly configured vehicle (Figure 3). RSP records the following parameters every 25 mm during the vehicle movement:

•
The vertical displacement of the beam from the road surface.

•
The vertical acceleration of the beam.

•
The time and distance that the two aforementioned parameters are recorded.  The profile is then analyzed to calculate IRI values for road surface. It is calculated using a quarter-car vehicle math model, whose response is accumulated to yield a roughness index with units of slope (m/km). The quarter-car model is a theoretical system that consists of the vehicle wheel, the mass of the vehicle that corresponds to a single wheel, of the suspensiondamping system, and of the mass of the axle-wheel-tire system (Figure 4). The profile is then analyzed to calculate IRI values for road surface. It is calculated using a quarter-car vehicle math model, whose response is accumulated to yield a roughness index with units of slope (m/km). The quarter-car model is a theoretical system that consists of the vehicle wheel, the mass of the vehicle that corresponds to a single wheel, of the suspension-damping system, and of the mass of the axle-wheel-tire system (Figure 4). Specifically, the quarter-car model used for the calculation of IRI index consist of the following: • The sprung mass-ms (kg), which is the mass of the part of the vehicle that burdens the suspension and includes a percent of the weight of the suspension.

•
The unsprung mass-mu (kg), which is the mass corresponding to the weight that does not burden the suspension system but is supported by the wheel or the tire and follows its displacements.

•
The suspension spring with coefficient ks (kN/m).
For the calculation of the IRI index, the values of the aforementioned parameters need to be determined. The set of values as described in [47] that are used for the calculation of the IRI, through appropriate algorithm, constitute the "golden car". In other words, it is the quarter car with specific values of the parameters. The principal idea for the IRI calculation is to model the movement of the quarter car on the pavement surface and the calculation of the total displacement of the suspension. IRI values equal to zero correspond to a totally smooth pavement, while IRI values around ten (10 m/km) indicate a practically impassable road surface [48].
Profile elevations were measured in both wheel paths simultaneously. For the analysis, the average value of the IRI in the wheel paths was considered. The data collection field program [49] gives step-by-step guide for the calibration of lasers, accelerometers, and distance measuring instrument (DMI), making calibration simple and easy to perform.

Data Processing
There are various studies dealing with the methodology followed for the analysis of the GPR data [21]. The determination of the thickness through the time interval between the reflected pulses from different layers and the dielectric properties of the layer was Specifically, the quarter-car model used for the calculation of IRI index consist of the following:

•
The sprung mass-m s (kg), which is the mass of the part of the vehicle that burdens the suspension and includes a percent of the weight of the suspension.

•
The unsprung mass-m u (kg), which is the mass corresponding to the weight that does not burden the suspension system but is supported by the wheel or the tire and follows its displacements.

•
The suspension spring with coefficient k s (kN/m).
For the calculation of the IRI index, the values of the aforementioned parameters need to be determined. The set of values as described in [47] that are used for the calculation of the IRI, through appropriate algorithm, constitute the "golden car". In other words, it is the quarter car with specific values of the parameters. The principal idea for the IRI calculation is to model the movement of the quarter car on the pavement surface and the calculation of the total displacement of the suspension. IRI values equal to zero correspond to a totally smooth pavement, while IRI values around ten (10 m/km) indicate a practically impassable road surface [48].
Profile elevations were measured in both wheel paths simultaneously. For the analysis, the average value of the IRI in the wheel paths was considered. The data collection field program [49] gives step-by-step guide for the calibration of lasers, accelerometers, and distance measuring instrument (DMI), making calibration simple and easy to perform.

Data Processing
There are various studies dealing with the methodology followed for the analysis of the GPR data [21]. The determination of the thickness through the time interval between the reflected pulses from different layers and the dielectric properties of the layer was found to have an average error of about 5% when the core thickness was paralleled the corresponding data recorded by the GPR system [50]. The use of multi-offset array test (MOA) procedure resulted in low accuracy of the estimated thicknesses [51]. In the same study, the reflection amplitude coefficient methodology (RAC) procedure produced errors ranging from 5 to 9% with a tendency to increase for larger depths [51]. In another study [52], the common mid-point approach (CMP) was investigated for its accuracy in estimating the pavement thickness. For this purpose, two sets of antennas were used (air-coupled). Furthermore, the results of the CMP approach were compared to the ones of the surface reflection method (SRM). It was concluded that the CMP approach leads to more accurate results for asphalt layers of about 6 cm, since the average error was about 6% compared to 18% for the SRM. For asphalt layer thicknesses up to 13 cm, errors from both approached reached 7%. An extended CMP method was developed by using two air-coupled GPR systems [53]. The prediction error of the asphalt thickness ranged from 2 to 23%. Higher error was observed in the case of asphalt layer thickness up to 6.3 cm, while for larger thicknesses, the error decreased. Other procedures studied involve the use of a multi-offset array [54], image processing algorithms [53], and truncation algorithm [55].
In light of the above, it is evident that a lot of approaches have been developed for the analysis of GPR data in an effort to minimize the errors in the estimation of pavement thicknesses. In most of the cases, the minimum error achieved was around 5%. In this study, the thickness determination was achieved by combining the time interval and the dielectric constant calibrated with core data, as presented in [50], due to the simplicity of the method and the low average error.
GPR data processing and analysis was performed using appropriate software [56]. The GPR analysis consists of the following steps: preprocessing, processing, and interpretation.
In the preprocessing phase, GPR data are edited in such a way that the raw data itself will not be changed. In this phase, the GPR editing features normally used are file reversal, cutting and combining, scaling, and linking data to GPS coordinates or to different road address data bases [56]. During preprocessing, it must be ensured that GPR signal polarity and color scales follow certain rules. In a measured pulse from a metal plate in a single scan mode, the maximum peak should always point to the right, and it should be referred to as the positive reflection, and in a line scan format using a grey scale, it should be white. The two smaller peaks around the positive peaks are called negative peaks, and in a line scan format using a grey scale, they should be black [57].
In the processing phase, signals from raw pavement data, end reflection, and metal reflection are elaborated [57]. The end reflection is a pulse, which is removed from the actual road measurement data, so one could check that the gain and the data format correspond with those of the road measurement. It is only necessary to do this if the direct wave has clutter below the pavement surface level. The template pulse is needed to improve the vertical resolution in the upper part of the GPR data (pavement bottom), as well as for calculating the dielectric value of the pavement.
During the processing phase, data filtering must be also performed. Different filter settings can be tested in order to see if they improve the data quality. After the metal plate calibration and template subtraction, other filters can be used. A special "background removal" filter has also been proven to improve the data quality. In Figure 5, raw data as well as data after processing are presented.
improve the vertical resolution in the upper part of the GPR data (pavement bottom), as well as for calculating the dielectric value of the pavement.
During the processing phase, data filtering must be also performed. Different filter settings can be tested in order to see if they improve the data quality. After the metal plate calibration and template subtraction, other filters can be used. A special "background removal" filter has also been proven to improve the data quality. In Figure 5, raw data as well as data after processing are presented. GPR data analysis in the case of pavements is a complex procedure, due to the fact that it consists of different layers with non-homogenous materials. For the determination of their dielectric constant, the use of relevant bibliography is not recommended. In order to accomplish the maximum possible accuracy in estimating the thicknesses, limited core sampling was performed in all subsections. This way, it was possible to calculate the parameters that are needed for converting time to thickness. GPR data analysis in the case of pavements is a complex procedure, due to the fact that it consists of different layers with non-homogenous materials. For the determination of their dielectric constant, the use of relevant bibliography is not recommended. In order to accomplish the maximum possible accuracy in estimating the thicknesses, limited core sampling was performed in all subsections. This way, it was possible to calculate the parameters that are needed for converting time to thickness.
The thickness of a pavement layer using GPR can be determined by recording the time difference between the reflected pulses from different layers and the dielectric properties of the surveyed layer. The thickness of the nth layer could be computed according to the following equation [43]: where d n is the thickness of the nth layer, t n is the electromagnetic (EM) two-way travel time through the nth layer as shown in Figure 1, c is the speed of light free space (≈3 × 10 8 m/s), and ε r,n is the dielectric constant of the nth layer. In order for the layer thickness to be estimated based on the above-mentioned equation, the knowledge of the dielectric properties of layer materials is necessary. Table 2 presents the estimated through the GPR measurements asphalt layer average thickness per subsection and the standard deviation. Thickness data were exported at an interval of 10 m.
The average as built thicknesses are close to the design thicknesses in each subsection and the standard deviation is low for most of the cases. RSP measurements were continuous, and IRI data were exported at 10 m intervals. The data were recorded continuously and were exported at 10 m intervals. For the analysis, the AC thickness difference in segments of 100 m was calculated (i.e., ∆h AC = h AC at position 100 m-h AC at position 200 m). The IRI average value (IRI av ) was calculated for the same 100 m segments. Figure 6 presents an overall of the data used for the analysis. X axis is the number of the 100 m segments (600 segments in a total length of 60 km). Differences in h AC within the segments vary from 0 to 9 cm, while IRI av values range between 0.16 and 2.65. It is noted that in most cases, differences in h AC greater than 6 cm are due to the transaction from one subsection to another along the same highway section (i.e., the transaction from A1 to A2). The average as built thicknesses are close to the design thicknesses in each subsection and the standard deviation is low for most of the cases.

Results
RSP measurements were continuous, and IRI data were exported at 10 m intervals. The data were recorded continuously and were exported at 10 m intervals. For the analysis, the AC thickness difference in segments of 100 m was calculated (i.e., ΔhAC = hAC at position 100 m-hAC at position 200 m). The IRI average value (IRIav) was calculated for the same 100 m segments. Figure 6 presents an overall of the data used for the analysis. X axis is the number of the 100 m segments (600 segments in a total length of 60 km). Differences in hAC within the segments vary from 0 to 9 cm, while IRIav values range between 0.16 and 2.65. It is noted that in most cases, differences in hAC greater than 6 cm are due to the transaction from one subsection to another along the same highway section (i.e., the transaction from A1 to A2).  From a first glance, it is not evident whether the IRI values can be correlated with the thickness difference of the asphalt layers. However, a trend is scarcely noticeable in which IRI av values seem to be higher at higher ∆h AC values.

Results
Further analysis was conducted in which the ∆h AC values were grouped in three categories: 2 < ∆h AC ≤ 5 cm 3.
∆h AC > 5 cm The IRI av values that fall in each one of these ∆h AC categories are presented in Figure 7 in the form of boxplots. From a first glance, it is not evident whether the IRI values can be correlated with the thickness difference of the asphalt layers. However, a trend is scarcely noticeable in which IRIav values seem to be higher at higher ΔhAC values.
Further analysis was conducted in which the ΔhAC values were grouped in three categories: The IRIav values that fall in each one of these ΔhAC categories are presented in Figure  7 in the form of boxplots. The focal observation from Figure 7 is that there is the tendency for higher IRIav values with the increase in ΔhAC values. Table 3 contains the descriptive statistics of the IRIav values for the three ΔhAC categories. The high value of the coefficient of variance (>0.10) indicates that there is a large variation in IRIav values. This is expected since IRI values grouped in these categories do not represent the smoothness of a single pavement section, but rather IRI data from all sections and subsections. Thus, the riding quality cannot be represented by the mean value. On these grounds, it was investigated whether a distribution can be applied in order to The focal observation from Figure 7 is that there is the tendency for higher IRI av values with the increase in ∆h AC values. Table 3 contains the descriptive statistics of the IRI av values for the three ∆h AC categories. The high value of the coefficient of variance (>0.10) indicates that there is a large variation in IRI av values. This is expected since IRI values grouped in these categories do not represent the smoothness of a single pavement section, but rather IRI data from all sections and subsections. Thus, the riding quality cannot be represented by the mean value. On these grounds, it was investigated whether a distribution can be applied in order to describe the data. The generalized extreme value (Gen. Extreme) distribution was examined. This distribution is a model that combines the Gumbel, Frechet, and Weibull distributions [58]. The parameters of the density probability function are z = (x − µ)/σ, σ and k, where µ is the location parameter, σ is the scale parameter (σ > 0), and k is the shape parameter. Depending on the k value, we can have each of the following distributions: Figure 8 describes the distribution of the IRI av values. The parameters of the generalized extreme value distribution are presented in Table 2. For the goodness of fit, the Kolmogorov-Smirnov test was applied, and the results (test statistic along with significance level) are shown in Table 4. The location parameter µ is considered representative of the samples since it is the value with the maximum probability of occurrence.  Based on parameter µ value, the tendency for higher IRI av values with the increase in ∆h AC values becomes more evident and justified. To thoroughly elaborate the analysis results, the µ values are used to demonstrate the roughness level as well as the percentage difference of IRI values with asphalt layers thickness variation (Figure 9). Based on parameter μ value, the tendency for higher IRIav values with the increase in ΔhAC values becomes more evident and justified. To thoroughly elaborate the analysis results, the μ values are used to demonstrate the roughness level as well as the percentage difference of IRI values with asphalt layers thickness variation (Figure 9).  Figure 9 indicates that higher AC thickness variations (>5 cm) result in a greater percentage increase of IRI. It should be noted, however, that the cases in which the thickness varies by more than 5 cm correspond to the change in the pavement cross section along the same highway.

Discussion
The evaluation of the structural and functional condition of pavements requires the use of different NDT systems that operate according to different principles. Each of these systems provides specific data on its own, which limits their capabilities. The combination of NDT systems and the integrated analysis of their data can overcome these limitations and provide a more comprehensive pavement assessment.
In this study, the possible relationship between asphalt layer thickness variation and pavement roughness was investigated. GPR and profilometer measurements were performed in sections of two different highways. The analytical results presented in this paper show that the pavement performance in terms of pavement roughness is related to the variation of asphalt layer thickness. In particular, this study proves that it is useful to investigate the effects of asphalt layer thickness on roughness in addition to other studies, since the IRI values seem to be affected by the variation of asphalt layer thickness. The more the layer thickness varies, the lower the ride quality of the pavement surface.
Specifically, the variation in asphalt layer thickness was divided into three categories to investigate whether the effects on IRI remained constant. The IRI values in each of the three categories can be described by the generalized extreme value (Gen. Extreme) distribution. Based on the characteristic value, it can be concluded that asphalt layer thickness variations greater than 5 cm result in a larger percentage increase in IRI. In other  Figure 9 indicates that higher AC thickness variations (>5 cm) result in a greater percentage increase of IRI. It should be noted, however, that the cases in which the thickness varies by more than 5 cm correspond to the change in the pavement cross section along the same highway.

Discussion
The evaluation of the structural and functional condition of pavements requires the use of different NDT systems that operate according to different principles. Each of these systems provides specific data on its own, which limits their capabilities. The combination of NDT systems and the integrated analysis of their data can overcome these limitations and provide a more comprehensive pavement assessment.
In this study, the possible relationship between asphalt layer thickness variation and pavement roughness was investigated. GPR and profilometer measurements were performed in sections of two different highways. The analytical results presented in this paper show that the pavement performance in terms of pavement roughness is related to the variation of asphalt layer thickness. In particular, this study proves that it is useful to investigate the effects of asphalt layer thickness on roughness in addition to other studies, since the IRI values seem to be affected by the variation of asphalt layer thickness. The more the layer thickness varies, the lower the ride quality of the pavement surface.
Specifically, the variation in asphalt layer thickness was divided into three categories to investigate whether the effects on IRI remained constant. The IRI values in each of the three categories can be described by the generalized extreme value (Gen. Extreme) distribution. Based on the characteristic value, it can be concluded that asphalt layer thickness variations greater than 5 cm result in a larger percentage increase in IRI. In other words, it is proved that pavement sections with a high degree of asphalt layer thickness inhomogeneity tend to have poorer ride quality.
Consistent with the above, it is argued that GPR capabilities can be extended beyond pavement thickness determination and structural evaluation to an initial assessment of expected pavement ride quality (when surface elevation is considered).

Conclusions
In this research study, an approach was developed in which in situ collected pavement data (pavement profile and stratigraphy) were analyzed together to demonstrate a favorable quality control framework for identifying areas that may exhibit more rapid deterioration in ride quality during pavement operation. Variation in asphalt layer thickness was shown to affect pavement roughness. It was shown that, for a given pavement cross section, the layer thicknesses accumulate around the design value during the construction phase, especially during paving and compaction of the asphalt mix. In this study, it was demonstrated that these variations are reflected in the roughness of the pavement. Moreover, it is proved that the transition from a pavement cross section to another along the same road/highway section affects ride quality due to the thickness variations. In general, it can be stated that the more the thickness of the asphalt layer varies, the higher the IRI values.
Because roughness is an extremely important indicator of pavement performance that directly affects road user safety, areas with elevated IRI values should be monitored as part of a pavement monitoring system to determine rehabilitation actions in a timely manner. In this sense, the relationship between IRI and thickness change can be considered a useful tool for planning maintenance systems.
Overall, the approach developed is considered a robust method that optimizes the required testing. Further investigation is needed, and more pavement sections should be considered to support the numerical results. In addition, it should be investigated whether the variation of the base/subbase course thickness has an influence on the pavement roughness. The results could be further validated with a more systematic approach that considers a broader range of roads in terms of base type, asphalt mix classification, construction and material quality, age/traffic, and environmental parameters. In addition, various approaches to GPR data processing could be applied, including methods to handle in situ noise and mechanical vibration of the GPR antennas during high-speed measurements.
Author Contributions: Conceptualization C.P., K.G. and A.L.; methodology, C.P.; analysis, K.G.; writing-original draft preparation, K.G.; writing-review and editing, C.P. and A.L. All authors have read and agreed to the published version of the manuscript.