Damage investigation of thin flexible pavements to Longer Heavier Vehicle loading through instrumented road sections and numerical calculations

Longer Heavier Vehicles provide an improvement in energy efficiency and environmental performance compared to traditional Heavy-Duty Vehicles. In Sweden, the maximum permissible vehicle gross weight has been increased from ∼64 to ∼74 tonnes without increasing the axle load limits. The consequence of this is investigated in this study. Response from two instrumented thin flexible pavements subjected to loading from three types of heavy vehicles (∼64, ∼68 and ∼74 tonnes) has been measured and the recordings were compared with numerical calculations based on 2D multilayer elastic calculations. Pavement damage contribution by the three vehicles was thereafter investigated. As long as the number of axles is increased to compensate for the increased vehicle loading and dual wheels are used, ∼74 tonnes vehicle are not more aggressive to the two thin pavement structures compared to the lighter vehicles with fewer axles but higher average axle loads and tyre pressure.


Introduction
The popularity of Longer Heavier Vehicles (LHVs) has increased worldwide in recent years. LHVs offer more efficient and environmentally friendly transportation of goods compared to conventional trucks (Christidis & Leduc, 2009;de Saxe et al., 2019). Most analyses on LHVs focus on the economic aspects of the transportation of goods and vehicle-related costs without considering the effects on the pavement structures (Christidis & Leduc, 2009;Ortega et al., 2014;Pålsson et al., 2017). A quantitative assessment of any pavement damage related to the increase in gross vehicle weight (GVW) magnitude could assist in predicting more accurately the pavement lifetime and the pavement-related economic costs. Road structures are usually designed for the heavy traffic that is expected in the forthcoming 20 years. Introducing new truck configurations that were not considered in the design procedure of the road structure requires an evaluation of the impact of the new load combination for the preservation of the road assets.
Empirical methods based on the long-term experience have been traditionally used for the design of flexible pavements. Newer, mechanistic-empirical design methods capable of predicting the behaviour of pavements more accurately are being developed and implemented worldwide (ARA Inc., 2004). Their development requires accurate initial assumptions, reliable data collection, method validation and calibration. The development of new pavement design methods is typically supported

The test road site
Two in-service thin flexible pavement structures located in the Piteå municipality in the northern part of Sweden along the roads Lv373 and Lv515 were instrumented in the autumn of 2017. Both road structures were old and have had some maintenance history. The site on Lv373 is located east of the Långträsk village at N65.33741, E20.4675, while the site on Lv515 is located south of the intersection between Lv373 and Lv515 at N65.31813, E20.29687. The test sites are located 10 km away from each other. Their geographical location is shown in Figure 1. The filled square shows the location of the BWIM station and the filled triangles show the locations of the two weather stations. A bridge weigh-inmotion (BWIM) station is located on the Lv373 in the vicinity of the test sites across Lake Storlångträsk. A meteorological station is located approximately 15 km north of the test sites and the Swedish transport administration operates a weather station at 8 km distance west from the Lv373 site. Both sites are located in a relatively flat marshland type terrain. The test site at Lv373 is 6.5 m wide and has an annual average daily traffic (AADT) of about 630 vehicles, while the test site at Lv515 is 6 m wide and has an AADT of around 100 vehicles.
The installation procedure in the existing roads was identical at both sites (Erlingsson & Carlsson, 2018). The asphalt concrete was removed along a 20 m long stretch along the lane. Thereafter two excavations 1.2 m wide were made at 4 m distance from each other down to 1.4 m depth. The excavated material was stored layer by layer. All existing layer thicknesses were determined during the excavation process. The sensor installation was performed from bottom to top, while the removed material was placed subsequently in their original position and compacted with the unbound base course. Finally, a new asphalt concrete (AC) layer was repaved. The reason for two trenches was to get two set-ups of the instrumentation for both sites.
The instrumentation for the mechanical response measurements included asphalt strain gauges (ASG) to obtain the transversal and longitudinal strain at the bottom of the asphalt layer, strain measuring unit coils ( MU) to measure the vertical strains in the structure, and soil pressure cells (SPC) to measure the vertical stresses in the granular layers. The MU coils were installed in one side of the vertical walls of the excavation trenches along a vertical line in the outer wheel path. The SPCs were installed on the other side of the vertical excavation walls, also in the outer wheel path. The ASGs were finally placed on top of the base course material and the asphalt layer was placed in position. The MU coils and the SPCs were therefore installed inside the excavation trench while the ASGs were installed on top of the unbound base course outside the excavated part. A hand vibrator compaction equipment was used in 10 cm thick layers when refilling the trenches. A normal road construction equipment was used to place the AC layer. The installation procedure of the instrumentation for the cross-section of road Lv515 is illustrated below in Figure 2. The same installation procedure was implemented in the case of road Lv373.
After the installation of the pavement mechanical sensors, the installation of the instrumentation for the monitoring of the climatic variables was carried out. It consisted of three temperature sensors installed in the asphalt layer, a frost rod (VTI Tjäl 2004) made of 41 temperature sensors placed on a 200 cm log rod with 5 cm interval spacing to measure the temperature distribution in a vertical cross-section, and five time-domain reflectometer (TDR) sensors installed on a 200 long cm PVC rod to measure the moisture content in the granular layers and the subgrade. Figure 3 shows the layer thicknesses and material types at both sites with a schematic view of the placement of the mechanical instrumentation.
As seen in Figure 3 the test site on road Lv515 is a very thin and simple structure with only a 4 cm AC layer on top of a 15 cm granular base resting on a sandy subgrade. The site on road Lv373 is in an old part of the road that has been overlaid two times. Originally this was a cut-back asphalt concrete road that has since been rebuilt by placing a surface treatment layer and unbound base course on top of it. Later a new 47 cm thick pavement structure was added, consisting of two AC layers and an unbound base course and subbase layer.

Falling weight deflectometer (FWD) backcalculations
Falling weight deflectometer (FWD) measurements were performed for both pavement structures to obtain an estimation of the mechanical properties of the layers by backcalculating the stiffness  modulus. The FWD measurements were performed on the surface on top of each sensor. Two separate deflection bowls were observed, one corresponding to the ASG sensors located outside of the excavation trenches and the other corresponding to the MUs and the SPCs located inside the excavation trenches. The reason for this is attributed to the disturbances caused in the granular materials and the upper part of the subgrade during the instrumentation process by the excavation and the differences in the degree of compaction between the disturbed and undisturbed materials.
The FWD measurements were performed at three different load levels, 30, 50, and 65kN. The values obtained from the measurements at 50 kN loading were averaged and compared to deflection bowls obtained from MLET backcalculations. An iterative error minimisation algorithm was implemented for the backcalculation procedure. Upper and lower boundaries for the iteration were set for the stiffness values of each layer based on catalogue values with a tolerance of ±15%. The root mean square error (RMSE) was calculated comparing the measurements against the calculated deflection values for the assumed stiffness values at each step of the iteration. Thereafter a new set of layer stiffness parameters within the boundaries was tested. The parameters giving the minimum RMSE were then reported and thereafter used for comparison of the calculated and observed deflection bowls from the 30 and 65 kN load levels to ensure a good fit for a broad range of load levels.
As the FWD measurements were made during two consecutive dates, slightly different temperatures of the AC layers were observed. The stiffness of the AC layer was therefore iterated in the ±15% range to the value obtained by the following formula where E T is the calculated stiffness of the AC at the target temperature T, T ref is the reference temperature (10°C), E T,ref is the asphalt concrete stiffness 6500 MPa at 10°C, and b is a regression constant of 0.065 (Erlingsson, 2012). The resulting stiffness values obtained from the backcalculation procedure are shown in Table 1 for the cross-section at road Lv515 and in Table 2 for the cross-section at road Lv373.
At the test site Lv373 the temperature during the FWD measurements varied between 13°C and 16°C during the testing. At Lv515 the temperature varied between 15°C and 18°C, therefore a slightly lower stiffness modulus was backcalculated for the AC layer at Lv515. The recorded volumetric moisture content values at the time of measurement for Lv515 were 9.12% at 36 cm depth, 7.59% at 76 cm depth, 9.86% at 116 cm depth, 11.94% at 156 cm depth, and 5.40% at 196 cm depth. For Lv313, the recorded volumetric moisture content was 10.93% at 36 cm, 13.71% at 76 cm depth, 18.31% at 116 cm depth, 47.64% at 156 cm depth, and 37.69% at 196 cm depth. No precipitation occurred during the measurement campaign.
The FWD measurements revealed that the observed deflection bowls were larger within the trenches than outside, results attributed to the differences in compaction. Thus in all analyses here, stiffness values within the trenches were used for the modelling of the MU coils and SPCs, whereas the values outside the trenches were used when the ASGs are modelled.  The observed deflections and the calculated deflection bowls using the ERAPave software (Erlingsson & Ahmed, 2013) and the stiffness values from Tables 1 and 2 are shown in Figure 4 for all three load levels (30, 50 and 65 kN).
The stiffnesses were validated by comparing the observed registrations of the induced vertical strain of the MU coils from the 50 kN FWD measurements with the MLET calculations. The results are shown in Figure 5. Multiple FWD measurements were carried out for all load levels at both locations on top of each sensor group. The figure shows the average value of the 50 kN measurement over the depth intervals and the error bars show one standard deviation.
As shown in Figure 5, the values of vertical strain measured by the MU coils were relatively similar to the MLET-based calculated values except lower registrations were observed in the subbase layer in the Lv373 test site over the depth range 25-39 cm. The reason for this is probably a malfunction of the sensor. A similar accuracy was obtained for the 30 and 65 kN loading. Since the degree of the fit was considered acceptable, a linear elastic material model was used for all layers of both structures in the response calculations for the heavy vehicle loadings.

The heavy vehicle testing
The responses of three different LHV with a gross weight of approximately 74 tonnes (LHV1), 64 tonnes (LHV2), and 68 tonnes (LHV3) were measured to assess the effect the vehicles imposed on the two structures as they drove over the test sections with a speed of 60 km/h. The vehicle drivers were instructed to drive in such a manner that the outer wheel would follow the wheel path. It is therefore assumed in the calculations that the vehicles passed directly over the top of the sensors which  were marked on the surface, although minor deviations in position due to lateral wander might have occurred.
Images of the vehicles directly before entering the test site Lv515 are shown in Figure 6. The yellow marks on the road surface in front of the vehicles were used to mark the position of the instrumentation in the structure which is required for accurate positioning of the loading placement. The schematics of the half-axial configurations of the three LHVs are shown in Figure 7 where distances between axles are given in centimetres. The axles of the vehicles have been divided into four groups and a number has been assigned within the groups, as a naming convention. The axle loading and tyre pressure are further given in Tables 3 and 4. The tyre pressure was measured by a pressure meter, and axle loads were obtained by static weighing of each axle of the vehicles. The pavement response measurements at the two sites were performed on two consecutive dates. The same vehicles were used both days but they were slightly differently loaded each day, therefore there is a difference in axle loads but not in tyre pressures.
LHV1 had 9 axles, total weight approximately 74 tonnes and a c/c distance between the centre of the first and last tyre was 20.50 m. LHV2 had 7 axles with a total weight of approximately 64 tonnes and a length of 19.74 m. Finally, LHV3 had 8 axles, total weight approximately 68 tonnes and a 20.80 m distance between the centre of the first and last tyre.
There were noticeable differences in tyre pressures, axial load magnitudes and configuration between the three vehicles. Overall, LHV1 had the lowest variation in axle loads and the lowest average axle load. It had further generally the lowest tyre pressures. LHV2 had the lowest number of axles and the highest average axle load but high tyre pressures in the trailer tyres. LHV3 had the largest single axle load but the average axle load was lower than LHV2. Furthermore, LHV3 had single tyre configurations on all multiple axles except for the front-rear driving axle of the tractor where dual tyre configuration was used.

Response modelling
Multilayer elastic theory-based software ERAPave (Erlingsson & Ahmed, 2013) was used for the response modelling. The stiffness values obtained from the FWD backcalculation process were used as inputs in the modelling process. Linear elasticity was assumed for all the layers on both road cross-sections. All axle contact areas were assumed to be circular and the tyre pressure was assumed to be constant at 850kPa, which is close to the average tyre pressure of all tyres. Typical comparisons between the measured signals and the MLET simulated pavement responses generated by ERAPave are shown in Figures 8-10 for LHV1, 2 and 3, respectively. The best fit between the measured and the calculated values was observed for the longitudinal ASG, and the SPC sensors, (a), (b) and (e), (f) in Figures 8-10. The largest deviations between the sensor observations and the calculations were associated with the transversal strain gauges (ASG) at the bottom of the AC layer. The transversal tensile strain measurements were highly sensitive to the positioning of the vehicles and slight lateral deviations as the tyres passed over the sensors led to inaccurately recorded values, typically providing lower registrations than the expected (calculated) values.  The recordings from the MU sensors were noisy and mainly provided the peak values after a noise filtering process; hence only measured peaks are shown along with the calculated curve for the vehicle response. It should be noted that recordings from two deepest MU sensors in the Lv373 structure were omitted as they gave unrealistic values. Figure 11 shows all calculated peak values against all measured peak values for both structures and all four sensor types. The total number of observed data points is given in the figure as well as the coefficient of determination R 2 which provides the degree of accuracy of the modelling approach.
In the case of longitudinal tensile strain the agreement between the calculated strain and the ASG recordings is quite acceptable with R 2 = 0.950. The points for the two test sites are clustered together since recordings were made by sensors located at the same depth for each structure. The strain values were higher at Lv515 as the AC layer is thinner (4 and 11 cm depth, respectively). The variations in the registered sensor values are due to different tyre pressures (700-920 kPa) and axle configurations (single vs. dual tyres) and to some extent differences in axle load level.
The degree of accuracy between recordings obtained from transversal ASG sensors and calculations was lower due to the difficulties in maintaining the correct transversal position of the vehicle. The measured strains were generally lower than the calculated strains, as shown in Figure 11(b). For the calculated value it is always assumed that the sensor is under the middle of the tyre for single tyre axles and between the two tyres for dual configurations and therefore providing the peak values. However, due to inaccurate positioning as the vehicles pass over the sensors, the measurements g frequently gave lower values. Calculations show that only a few centimetres of lateral movement had a great impact on the tensile strain observations and it is difficult even for experienced drivers to keep that accuracy at a speed of 60 km/h on a narrow road. The degree of prediction accuracy for the SPC sensors was relatively high (R 2 = 0.917), however, and the points were dispersed equally around the equality line on the chart. The SPC sensors were located at three depths providing stress levels over the entire range up to 120 kPa. The same was observed in the case of MU coils, though the accuracy here was slightly lower (R 2 = 0.882). The graph shows registrations from eight and seven sensors from structures Lv515 and Lv373, respectively.

Damaging effect of different vehicles
Two failure criteria were used in order to compare the damaging effect of the three different LHV's to the pavement structures. They correspond to the allowable number of load repetitions to prevent bottom-up fatigue cracking of the AC layer N f and limiting the accumulated permanent deformation in the subgrade N d , respectively expressed as (Chatti et al., 2009;Hajek & Agarwal, 1990) where t and v are the tensile strain at the bottom of the AC layer and the compressive strain on top of the subgrade respectively and f 1 , f 2 and n 1 , n 2 are material parameters.
By using Miner's rule a damage factor D f with respect to bottom-up fatigue life for each axle group for the three LHV's can be defined through the summation over all sub-axles in the axle group as where j stands for the number of sub-axles in the axle group and j end is the total number of sub-axles in the axle group. Further is l = 1, 2, 3, 4 the axle group number and k = 1, 2, 3 stands for the three LHV under consideration. A similar expression for a damage factor D d can be received for the subgrade deformation criteria by replacing the tensile strain t with the vertical compression strain v and f 1 and n 1 with f 2 and n 2, respectively . Now we can define in a similar manner a relative damage factor D rf for the axle group with respect to fatigue cracking using the four axle groups of LHV2 ( ∼ 64 t vehicle) as normalisation as Again, a similar expression D rd can be received for the subgrade deformation criteria. Calculated critical strain values using the values from Tables 3 and 4 for the axle loads and configuration along with the tyre pressures and distances between axles from Figure 7 are given in Figure 12. These values were further used to estimate the relative damage factor of the different axle groups of the three vehicles as shown in Tables 5 and 6. Here the exponent n = n 1 = n 2 = 4 was used for both criteria. The depth for the subgrade criteria was assumed as 118 cm for the Lv373 structure but a 60 cm depth was assumed for the Lv515 structure as the structure is unusually thin and a 19 cm depth was not considered representative for the Swedish conditions for a low volume road in a cold region (see Figure 3).
Obviously there is a large variation in the damaging effect of the different axle groups. It is further difficult to compare the different axle groups together as the combined effect of tyre pressures, subaxle loads and configurations have an impact on the distress contribution and therefore the damaging effect. By looking at the difference in the damaging effect of LHV1 compared to LHV2 there are two things that stand out. The effect on the steering axle of the LHV1 on the subgrade criteria is about 50% higher than for the LHV2. This is attributed to the higher axle loading of that axle. As axle group 2 of LHV1 is a triple axle, of which the last sub-axle has a single tyre configuration, it had a significant impact   on the LHV's1 2nd axle group on the fatigue cracking life of the structure compared to the comparable axle group of LHV2. The main contributor here was the third sub-axle of LHV1 as well as the lower tyre pressure of LHV2. This effect could be mitigated by replacing the single tyre of LHV1 with dual tyres and distributing the loading evenly over all sub-axles. Finally, the LHV3 had a much larger damaging impact on the two pavement structures than both LHV1 and LVH2 except for axle groups 3 and 4 for the subgrade criteria of Lv373. The main reason was that single tyre configuration was mainly used on that vehicle with relatively high tyre pressure. In a similar manner, a damage factor with respect to bottom-up fatigue life for each vehicle can be defined through the summation over all axle groups for the respective vehicle. In addition, normalizing as before with the LHV2 then revealed a relative damage factor for the entire vehicles as Again, a similar expression can be used for the subgrade criteria. Results where values from Tables 5  and 6 have been used are given in Table 7. The exponent n = n 1 = n 2 = 4 was used as before.
As shown in Table 7, the damaging effect of LHV1 was about 22% and 39% higher compared to the LHV2 with respect to bottom-up fatigue cracking for the two structures. The subgrade criteria were highly dependent of the depth where it was applied. At Lv515 the LHV1 courses caused about 8% less damage than the LHV2, whereas at Lv373 the LHV2 courses about 6% more damage. This is attributed to the fact that axle group 4 of LHV1 is a tridem axle group and heavier that the corresponding tandem axle group of LHV2. Further, the subgrade was reached at considerable depth (118 cm) at Lv373 but the effect of superposition of loading from closely spaced axles increased with depth. Therefore this is seen in structure Lv373 but not in Lv515 where the subgrade criteria were applied at a 60 cm depth. The LHV3 in all cases had the highest damage effect and for the bottom-up fatigue cracking the effect was quite detrimental. The main reason for this was that all sub-axles except one had a single tyre configuration with relatively high tyre pressure.

Conclusions
GVW up to 64 tonnes have been allowed on the road network in Sweden since 2015. In 2018 the allowed load limits were increased to 74 tonnes on part of the road network. In this paper, the response of two in-service instrumented thin flexible pavement structures subjected to loading by FWD and three types of LHV ( ∼ 74, ∼ 64 and ∼ 68 tonnes, respectively) were analysed in order to investigate the effect of LHV with different axle configurations, axle loads, and tyre pressures have on the damage accumulation. FWD measurements and 2D axisymmetric MLET-based backcalculation software were performed to obtain the stiffness of the pavement layers. The obtained stiffness values were used in the response modelling of the LHV loading. By comparison between the measured and the calculated response values, the modelling strategy was proven to be accurate. The damaging effect of the different vehicles was compared, first by axle groups and thereafter for the three vehicles. Two criteria were examined: the bottom-up criteria of the AC layer and the permanent deformation criteria of the subgrade.
The LHV1 ( ∼ 74 tonnes vehicle) resulted in being more aggressive to the two pavements than the previously allowed 64 tonne vehicles (LHV2). It was mainly the axle group 2 that caused a high contribution to the fatigue life reduction of the two structures. This could be mitigated by replacing the last sub-axle single tyre of LHV1 with dual tyres and distributing the loading evenly on all sub-axles in the axle group.
There was also a 50% higher impact of the steering axle of the LHV1 on the subgrade criteria compared to LHV2 due to the higher axle loading. The fourth axle group of LHV1 was a triple axle and also had a great impact on the subgrade criteria in the deeper layers as seen by their effect superimposed with depth at the test site on structure Lv373. This was seen as a 13.5% higher damage effect on the subgrade for LV373 than at LHV1. By applying the criteria at even greater depth this damaging difference between the two vehicles would have increased even further (not shown here).
The highest damaging factor values were calculated for LHV3 (68 tonne vehicle) due to the high number of axles with single tyres and relatively high tyre pressure, and hence the loading applied over a smaller contact area resulting in high impact on the pavement structures.
This was a limited study including only two pavement structures and three LHV. Other combinations of axle load configurations, load distribution and tyre pressures might reveal different results. Furthermore, only two criteria are looked at, fatigue cracking and rutting accumulation in the subgrade. Other distress mechanisms such as accumulation of plastic deformation in the AC layer or the granular materials were not part of this study.