Journal of Transportation Engineering, Part B: Pavements

Abstract

Pavement maintenance causes an instant change in the pavement’s material or structural properties and affects the subsequent development of pavement distresses and performance. The occurrence of maintenance action significantly limits the applicability of predictive models that rely heavily on the continuity of time and pavement exposure conditions. Using rutting of asphalt pavement as an example, this study treated the partitioned rutting development as short-term time series. The proposed average cosine similarity of pavement rutting development effectively integrated the data collected within a characteristic length in the longitudinal direction of the pavement. Integration of the raw data effectively mitigated the data inconsistency caused by measuring errors and simplified the model construction. We employed convolutional neural network (CNN) and long short-term memory (LSTM) as two typical deep-learning (DL) models to capture the characteristics of rutting development from limited data and make corresponding predictions. The effects of model hyperparameters and input type on the model performance (e.g., accuracy and stability) were investigated to identify the optimal setting for various modeling data. The comparisons with two statistical models—exponential smoothing (ES) model and autoregressive integrated moving average (ARIMA) model—indicated the potential of applied DL models in accurately predicting rutting development in field pavement. Finally, three strategies of improving model performance were explored and discussed for future applications, i.e., increasing the input length, input dimensionality, and model complexity.

Practical Applications

This study stands in the position of transportation engineers to solve a practical problem with computer-aided tools, which is the prediction of asphalt pavement performance with deep-learning (DL) models. The data utilized in this study were collected from the field, resulting in inherently challenging characteristics such as being short-term and noisy. This study investigated corresponding strategies to allow DL models to provide accurate and stable predictions. Besides, this study makes DL models more accessible to practicing engineers. A significant amount of model information obtained in this study can be extended to various types of performance indices. Finally, DL models constructed in this study were compared with classic and currently applied models using the same data. The results reveal that DL models can serve as reliable and promising tools for predicting asphalt pavement performance in the natural deterioration stage of the pavement.

This investigation examines the influence of braking on the reliability of bituminous pavements. For this purpose, different loading conditions of a two-axle truck are simulated using vehicle dynamics software. Individual response surface methodology models to predict critical strain values are developed for various thickness and modulus values and loading combinations. Monte Carlo simulations are then used to generate random variabilities of input parameters, and the reliability is estimated accordingly. Three different cases are simulated. In the first case, it is assumed that all the vehicles impact the pavement by braking, and the reduced reliability is found to vary between 99.8% and 1.2% for the different loading and braking conditions. In the second case, the allowable load repetitions for fatigue damage are computed for a given reliability of 90% and are found to vary between 24 and 1 million standard axle (MSA). In the third case, different combinations of braking conditions are investigated, and the reliability is estimated.

Two-dimensional digital image correlation (2D-DIC) is a widely used optical technique to measure displacement and strain during asphalt concrete (AC) testing. An accurate 2D-DIC measurement can only be achieved when the camera’s principal axis is perpendicular to the planar specimen surface. However, this requirement may not be met during testing due to device constraints. This paper proposes a simple and reliable method to correct errors induced by nonperpendicularity. The method is based on image feature matching with scale-invariant feature transform (SIFT) and rectification. No additional equipment is needed. A theoretical error analysis was conducted to quantify the effect of a nonperpendicular camera alignment on measurement accuracy. The proposed method was validated numerically using synthetic images and experimentally in an AC fracture test. It achieved relatively high accuracy, even under considerable camera rotation angles and large deformation. As a preprocessing technique, the proposed method showed promising performance in assisting the recently developed CrackPropNet for automated crack propagation measurement under a nonperpendicular camera alignment.

Rigid airport pavements are a critical part of the Australian airport pavement inventory. The thickness design of rigid airport pavements in Australia was previously based on Australian developed charts and a conservative characteristic concrete strength, but in recent years this has changed to reliance on the software FAARFIELD provided by the US. Australian designers have used FAARFIELD in combination with the historical and conservative definition of concrete strength. That conflicts with the intended use of a less conservative Australian concrete strength value for inputting into FAARFIELD. The unintended consequence of this change in Australian practice is an additional 44 to 74 mm of concrete slab thickness, depending on the aircraft type, subgrade support, and concrete strength used. This additional thickness costs an additional AUD 185,000 to AUD 429,000 and 154,000 to $\mathrm{333,000}\mathrm{kg.}{\mathrm{eCO}}_2$ for a modest $100\times 100\text{-}\mathrm{m}$ rigid airport pavement development. This additional financial and environmental burden is not justified, and Australian airport pavement designers should adjust their thickness design calculations to avoid this unintended overdesign in the future. This example also serves as a lesson regarding the potential consequence of mixing elements from different systems for pavement thickness design.

Asphalt mixture rutting may involve vertical compaction, volumetric dilation, and shear-induced lateral movement, depending on many factors. Both in the field and in the laboratory [e.g., Hamburg wheel tracker (HWT) and asphalt pavement analyzer (APA) tests], lateral movement of a mixture creates a heave region and a vertical permanent deformation along the wheel path. Typical laboratory wheel tracking tests such as HWT and APA measure the depth of pure vertical permanent deformation; field measurements measure rut depth and heave height. Currently, HWT and APA test devices are not equipped to measure heave height. The objective of this study was to develop an image-based heave height measurement methodology for HWT-tested asphalt specimens. The method can easily be extended to other wheel tracking tests, such as the APA. The study also included the generation of a three-dimensional (3D) model of HWT-tested asphalt specimens, the extraction of 3D coordinates, and the calculation of specimen heave height and rut depth. The proposed methodology was verified by comparing the calculated rut depths with laboratory-measured rut depths. The proposed methodology measured the heave height of the HWT-tested specimens with reasonable accuracy. Furthermore, a linear relationship was observed between the rut depth and heave height of the tested specimens. This research will help researchers evaluate asphalt mixture heave resistance and will extend state-of-the-art research in the domain.

Abstract

The addition of fibers to asphalt mixtures has the potential to improve the resistance of asphalt mixtures to cracking and permanent deformation (rutting). However, the state (distribution) of fibers in the asphalt mix is critical parameter in determining performance enhancement. This study evaluated the impact of different laboratory mixing methods (mainly Hobart and bucket) on fiber distribution and selected a method comparable to plant-produced fiber-reinforced mixes. The laboratory performance of plant-mixed lab-compacted (PMLC) and lab-mixed lab-compacted (LMLC) fiber-reinforced mixtures was compared. Four types of laboratory and plant mixtures [unreinforced, polyolefin and aramid (PFA) fibers at 0.05% dosage, and Sasobit-coated aramid (SCA) fibers at 0.01% and 0.02% dosages] were produced to compare the laboratory performance of asphalt mixtures. The cracking resistance [evaluated using indirect tensile strength (ITS) and semicircular bend (SCB) tests], rutting susceptibility [evaluated using flow number (FN) and the Hamburg wheel tracking test (HWTT)], durability (evaluated using Cantabro loss), and fatigue (evaluated using a uniaxial fatigue test) performance of control and fiber-reinforced asphalt mixtures were evaluated. Among laboratory mixing methods, a bucket mixer was selected to produce LMLC samples because it produced the maximum fiber distribution (87% of fibers in the individual state). Laboratory performance testing results showed that PFA-reinforced mixtures enhanced the rutting performance regardless of mixing method. The mixtures with PFA 0.05% and SCA 0.01% fibers were highly durable and had better dynamic modulus ($|E^{*}|$) at low frequency and high temperature. Cracking performance was improved with the addition of SCA fibers into the mix; however, fiber-reinforced mixes (PFA 0.05% and SCA 0.01%) had higher fatigue damage tolerance, except SCA 0.02% reinforced mix. The laboratory bucket mixing method is representative of plant-produced fiber-reinforced mixtures, and laboratory performance results were consistent for laboratory bucket and plant produced mixtures.

Practical Applications

This paper showcases laboratory mixing method equivalent to fiber distribution that produce fiber-reinforced asphalt mixtures comparable to those produced at an asphalt plant. Different laboratory mixing equipment, including Hobart and bucket mixers, was used to produce fiber-reinforced asphalt mixtures in the laboratory. Aramid fibers were added into these mixers after separation by hand. In addition, the same fiber-reinforced mixtures were produced at a batch plant for comparison purpose. Fibers and binder were extracted from all mixtures and individual fibers (indicating better fiber distribution) were quantified. According to the results, the use of a bucket mixer led to better fiber distribution within the produced asphalt mixtures than did a Hobart mixer. Laboratory performance comparison results showed that the mixes produced using the bucket mixer had identical results to plant-produced mixtures. Hence, using a bucket mixer and introducing aramid fibers after hand separation is an effective way to produce plant-representative fiber-reinforced asphalt mixtures. This is a major step toward designing and producing fiber-reinforced asphalt mixtures and enhancing the cracking performance of asphalt mixtures.

An excessive amount of differential settlement was observed during the construction of taxiways at Changi Airport in Singapore. The pavement exhibited severe settlement (up to 200 mm). The first cracks appeared in April 2018, 11 months after the pavement construction was completed. The cracks stayed in the asphalt concrete layer, but 15 months after the emergence of the first cracks, they extended downward and propagated beyond a subgrade layer. From the analyses performed on the pavement structure using the finite element method, it was found that tensile strain occurred at the surface of the pavement. The amount of tensile strain was higher at the joint between the main pavement and the shoulder pavement than in any other location on the pavement. Fracture analyses were conducted and indicated that cracks would propagate from the top of the pavement due to the tensile strain and move to its bottom. To identify an indication of settlement, heavy weight deflect (HWD) tests were conducted over three pavements experiencing three levels of settlement—severe, moderate, and no settlement. Backcalculation analysis was done for the three pavements using backcalculation programs BAKFAA and BAKSIMPLEX, and the closed-form solution. It was observed that the backcalcuated asphalt modulus was highly affected by root mean square (RMS) %. It is hence less reliable when the asphalt concrete modulus was determined at a higher RMS %, whereas the correlation became stronger when it was determined at an RMS % lower than four. Meanwhile, subgrade modulus backcalculated even at higher RMS % values was less affected by the error because most errors occurred and accumulated near the loading area of HWD. From the statistical analysis (t-test) performed on subgrade moduli backcalculated, it was concluded that the means of subgrade moduli obtained from the three different pavements differed from the hypothetical means of interest at a 95% level of confidence. This indicates that if a pavement undergoes settlement, the indication of settlement could be captured from HWD tests and the backcalculation of subgrade moduli.

Pavement markings are essential traffic control devices that enhance safety for motorists during nighttime. Numerous statistical learning models have been developed in prior studies to predict the retroreflectivity of the markings, but the applicability of these models is questionable in terms of accuracy. The key objective of this study was to develop a machine learning–based framework that can be used by US transportation agencies to reliably predict the retroreflectivity of their pavement markings over a period of 3 years utilizing the initially measured retroreflectivity and other key project conditions. The random forest (RF) algorithm was used in this study to develop the proposed framework considering seven types of marking materials in three different US climate zones. A total of 49,632 transverse skip retroreflectivity measurements were retrieved from the National Transportation Product Evaluation Program (NTPEP) and 11 RF models were developed to sequentially predict retroreflectivity at different prediction horizons. The models were trained with randomly selected 80% of the total data points, and the remaining 20% data points were utilized for testing the predictive performance of the developed models. The RF models predicted the retroreflectivity with a superior level of accuracy ($R^2$ ranging between 0.88 and 0.99) than the models proposed in prior studies. These models are expected to aid transportation agencies in reliably determining the effective service lives of their marking products.

Resilient modulus ($M_R$) serves as a fundamental material property utilized for characterizing unbound pavement materials in pavement design. While the repeated load triaxial test (RLT) is the standard method to determine $M_R$, it can be impractical due to the required test facilities and expertise. As a result, various constitutive models have been proposed to predict $M_R$ of soil based on the stress state and soil properties. This study evaluates the performance of 10 such constitutive models based on their accuracy in representing the relationship between $M_R$ and stress conditions for RLT test data taken from the long-term pavement performance (LTPP) database. The 10 constitutive models consisted of models recommended by various researchers and institutions, including the one proposed by the National Cooperative Highway Research Program (NCHRP) and adopted by the Mechanistic-Empirical Pavement Design Guide (MEPDG). A multivariable nonlinear curve fitting technique was used to fit the data to the models and obtain model coefficients. The performance of these models was then measured by using the coefficient of determination (${\mathrm{R}}^2$), the root mean square error (RMSE) and the mean absolute error (MAE) values as metrics. The results show that the models proposed by NCHRP, Puppala as well as Ni performed best, with Puppala’s model showing slight superiority. On the other hand, the models proposed by Rahim and George, Kim, and Moossazadeh and Witczak produced poor results. The results also indicate that models having the general form of NCHRP and expressed in terms of confining pressure and deviator stress perform better. The findings of this study will provide insights into the reliability of existing models and considerations that should be made for developing better alternative models in the future.

Overloaded traffic is one of the major effects that cause more early damage to asphalt pavement than regular traffic. Moving overloaded traffic not only needs to consider dynamic coefficient and axle load coefficient when carrying out pavement mechanics calculations, but also accelerates pavement failure and boosts the cost of construction and maintenance. Although asphalt pavement of a stabilized base with inorganic binder has a strong bearing capacity, the shortcoming of reflective cracking also seriously restricts its performance. This paper describes the mechanical response and crack propagation of asphalt pavement of stabilized base with inorganic binder which has a crack, under the coupling effect of moving overloaded traffic and temperature. Extended finite-element method (XFEM) was conducted to simulate reflective cracks propagating freely from a stabilized base with inorganic binder to asphalt overlay. Dynamic analysis was also carried out by varying the axle weight and tire pressure, which could help highway builders understand the stress distribution of pavement under design traffic and the crack propagation law of pavement structures. Results indicated that the mechanical response of asphalt pavement was comprehensively related to tire pressure, axial weight, and its active position and that overloaded traffic accelerates crack propagation more than regular traffic. This research achievement contributes to providing roadway engineers with a more profound perception to potentially advance pavement design practitioners for specific traffic characteristics.

The objective of this paper is to assess the impact of mineral grain size and grain size distribution of aggregates on the deterioration of pavement surface microtexture. For the study, aggregates were collected from nine different sources to cover a wide variation in mineralogical type. The grain size parameters were obtained by detecting mineral grain boundaries using scanning electron microscope electron back scattered diffraction (SEM EBSD) analysis on thin sections prepared from the aggregates. Scanning electron microscope (SEM) analysis was also carried out on thin sections of all aggregate types to determine their mineralogical composition. Microtexture deterioration was measured by collecting the British pendulum number (BPN) of the aggregate after every hour of polishing for six hours using an accelerated polishing machine. From the mineral composition analysis, it was observed that the aggregates in the study belonged to three rock types: granite, amphibolite, and basalt (three sources of each type). The BPN values of the polished stones indicated that aggregates with finer grains offer more resistance to polishing. Statistical analysis of the test results also suggests that a wide variation in grain sizes in aggregates positively contributes to resistance against polishing. The study confirmed that, for a given rock type, coarse aggregates with finer grains and greater variation in grain size can better withstand microtexture deterioration caused by mechanical polishing from vehicular loads when used in the surface course of pavements.

The integrity of road markings has a significant impact on driving safety, especially in the emerging automated driving scenario. Road markings are susceptible to wearing damage; it is therefore important for road agencies to examine and maintain markings periodically. This study presents a novel method for detection of road marking defect via an unmanned aerial vehicle (UAV)–laser radar (LiDAR) platform. The key idea is to evaluate the damage rate of road markings by point reflectivity. The method consists of three major steps: point cloud registration, marking segmentation, and defective marking detection. A high-intensity prioritized raster method is proposed to extract whole marking regions within defect area, and the damage rate of markings is defined to evaluate damage severity. Field test results show that the precision and recall of marking extraction are over 90%, and the precision of marking defect detection is over 93%. The failed detections were attributed to subjectivity of ground truth and neglect of small defects.

Existing approaches to infrastructure management neglect the impact of infrastructure condition on safety performance, even though poorly maintained roadways can contribute to increased crash frequencies. This is perhaps because very few studies have quantified the impact of infrastructure condition on safety performance. Hence, this work develops safety performance function (SPF) models that relate pavement condition—specifically, roughness, as measured using the international roughness index (IRI)—and other roadway factors with observed crash frequencies. These SPFs are then combined with pavement roughness deterioration models to better understand how maintenance decisions can be used to improve roadway safety performance. Data from two-lane rural roadways in Pennsylvania are used for this effort. The SPF estimation results reveal that accounting for IRI in SPF development significantly improves model fit, and that IRI has a significant and positive impact on crash outcomes. The results further suggest that the IRI has a different impact on total crash frequency as compared to fatal and injury crash frequency, and rear-end crashes. This is likely due to how roughness of the roadway can impact travel speeds. Finally, the results also show that maintenance activities that reduce IRI can have significant safety benefits. This suggests that pavement management decisions should also consider the anticipated safety benefits, in addition to other factors.

Pavement management systems (PMS) are considered to be crucial tools for agencies in managing their pavement networks and using accessible budgets in an appropriate technique. In this paper, a newly developed condition rating index [modified pavement condition rating (MPCR)] was used to assess the current and future pavement performance. The pavement future performance in terms of MPCR was predicted using two models: (1) a stochastic Markov chain model, and (2) an artificial neural network (ANN) model. In Markov chain modeling, the Long-Term Pavement Performance (LTPP) database was used to develop the transition probability matrix (TPM). An ANN model for future pavement performance prediction was developed based on the same LTPP data. For both models, 245 data points from the LTPP data were used for deterioration modeling. Both models were verified using LTPP data and data obtained from the General Authority of Roads, Bridges and Land Transport (GARBLT), Egypt. The Markov chain and ANN results were compared. The Markov chain model performed better than the ANN model, with a coefficient of determination, $R^2$, of 0.9 and RMS Error (RMSE) of 0.1268. the ANN model yielded an $R^2$ of only 0.53 and a RMSE of 0.2205. Increasing the number data points did not lead to a significant improvement in the ANN model accuracy. A multiobjective particle swarm optimization (PSO) model is used for fund allocation to maximize the average pavement condition and minimize the maintenance cost. The developed optimization model was tested on two benchmark optimization problems to ensure its validity for real-life applications. The developed optimization model was used on a network of roads and was found to be capable of generating optimal or near-optimal solutions for the maintenance decisions to keep the pavement in good workable condition. As such, the current study presents a comprehensive development tackling different modules of a pavement management system.

Pavement roughness has long been linked to both vehicle fuel efficiency and pavement structural degradation. Perceptions of pavement roughness, however, may vary among users of different sociodemographic features and therefore potentially impact the allocation of resources for highway maintenance and rehabilitation when equity is considered. It is necessary to investigate the major factors that influence the user perception of pavement roughness. Considering the complexity in structure and relationship of field data, this study applied three state-of-the-art machine learning and statistical methods, including a classification tree, a random-parameter ordered-probit model with a random effect (Model 1), and a correlated random-parameter ordered-probit model (Model 2), to the analysis of user perceived roughness. Data were collected from a prior study that conducted in-vehicle tests involving individual user, pavement, and vehicle. The analysis identified more key factors influencing roughness ranking than previous research and accounted for the heterogeneity in individuals and interactions among random parameters. The results indicate that, whereas physical measurements of pavement roughness (e.g., International Roughness Index), visible distresses such as patches and faulting, and joints provided a strong indication of user roughness ranking, other factors (i.e., particular regularly used route, participants’ age, income, and gender, number of household infants, and interior vehicle noise levels) were also statistically significant. Two-way group random effects were statistically significant in the data, which should be accounted for in future studies. Results from this study fill an important gap between making accurate prediction and uncovering underlying causality in research of physical infrastructure measurement with user perceptions of infrastructure conditions.

Fracture energy from semicircular bending tests is commonly used to measure the fracture resistance of asphalt mixtures. A study was conducted to develop master curves for the macrocracking of asphalt mixtures. The study examined four testing temperatures, five loading rates, two binder types, and two nominal maximum aggregate sizes (NMASs). Fracture energy was captured at different loading rates and temperatures, and a preliminary master curve was built using the dynamic modulus master curve approach. However, the initial plot of fracture energy versus reduced loading rate did not show a consistent trend. Instead, a parabolic shape was found in the local fracture energy when increasing the loading rate, and peak values were plotted for each temperature to produce the master curve for the macrocracking of asphalt mixtures. The maximum fracture energy was unique for each combination of NMAS and asphalt binder type and occurred at a specific temperature, indicating a new methodology for classifying the cracking resistance of asphalt mixtures: macrocracking master curves. Maximum values fell between typical low-temperature and intermediate-temperature fracture testing, highlighting a gap in understanding regarding macrocracking behavior.

As a step toward revolutionizing and optimizing freight transport, autonomous truck platooning technology (i.e., a set of connected heavy-duty trucks traveling closely at specific headway intervals) has been the center of research over last few years. Connected autonomous vehicles (CAV), in the form of truck platooning, can be implemented as a novel contribution to sustainable freight transport with the potential of offering the following synergistic benefits, among others: road safety improvements, enhanced economic prospects, and environmental preservation. In this study, the impacts of fixed-path truck platooning on pavement performance were modeled and numerically quantified using elastic and dynamic-viscoelastic finite-element (FE) methods. Factors including pavement layers’ geometrical and mechanical characteristics, traffic characteristics and counts, axle configuration, and seasonal temperature are analyzed to address the platooning effect in this study. Yet, other possible wandering patterns between the fixed-path and normally distributed wandering patterns, as well as the effects of speed variations, driving behaviors, and tire pressure (i.e., variation in dynamic loading), can be further studied. The mechanical responses (namely displacements, strains, and stresses) obtained from the FE modeling were used to predict the effects of truck platooning on the pavement performance due to limited wandering (lateral movement of truck tires). The FE modeling results indicated that the channeling effects (i.e., limited tire wandering) of truck platooning have negative effects on the pavement performance with respect to decay in fatigue life and increase in permanent deformation. With fixed-path platooning, the fatigue life is reduced in the range of 13.9% to 34.5% in terms of the number of load cycles over the 20-year design life. The permanent deformation (rutting) value over a fixed period is increased by a factor of 1.2 to 2.9.