Asphalt overlays are frequently employed to enhance the longevity and performance of existing pavements. However, the development of increased pavement cracking over time can result in water infiltration, which undermines the bond between the overlay and the underlying layers, ultimately compromising the integrity of the base and subgrade.
Numerous studies have utilized various methods, including nonlinear regression, artificial neural networks
(ANN), and genetic programming, to predict the occurrence of transverse cracks. While ANN is often characterized as a black box, nonlinear regression necessitates a predetermined formula structure. To formulate a model without a clearly established formula, some researchers have turned to genetic operation trees (GOT). This approach allows for the generation of model structures while iteratively identifying the formula that best aligns with the experimental data.
A study conducted by Machine Hsie et al., from the Department of Civil Engineering at National Chung Hsing University in Taiwan explores the prediction of transverse cracks in asphalt pavement using the Levenberg-Marquardt Genetic Operation Tree (LMGOT). This research draws on a comprehensive dataset derived from a 15-year study conducted by the Texas Department of Transportation (TXDOT) regarding pavement cracking. The framework of the Genetic Algorithm (GA) is inherently aligned with discrete optimization problems, while the Levenberg–Marquardt method (LM) is well-suited for addressing continuous optimization challenges. Consequently, this study effectively integrates both methodologies to derive an optimal formula that accurately fits the data on pavement crack lengths.
The study utilizes a dataset collected and prepared by the Strategic Highway Research Program (SHRP) in 1987 to assess long-term pavement performance through the Long-Term Pavement Performance (LTPP) program. This program conducts two main types of experiments: the General Pavement Study (GPS), which comprises nine categories aimed at creating a national pavement performance database, and the Specific Pavement Study (SPS), also consisting of nine categories, focusing on specially constructed or rehabilitated pavement sections.
The current study examines data from SPS-5, titled “Rehabilitation of AC Pavements.” The data used in the analysis is collected from overlay pavement experiments conducted by the Texas Department of Transportation (TxDOT) under SPS-5, specifically A502-509. The dataset consists of 88 samples, which were divided into 70% for training and 30% for testing to evaluate the generalization of the model. The parameters included in the dataset are: mill (surface preparation), material, and thickness.
The study evaluates the effect of milling on the performance of pavement overlays. Out of eight sections
analyzed, four underwent a milling process, which involved replacing the milled thickness with the same material designated for the overlay, a procedure commonly referred to as a mill-and-fill operation. It is important to note that the design thickness for the milled sections did not account for the material that was replaced, whereas the non-milled sections had the overlay material applied directly onto the existing pavement. The thickness of the overlay varied between 2.2 to 7.1 inches. This study utilized a maximum Reclaimed Asphalt Pavement (RAP) content of 35% in the asphalt concrete (AC) for some sections, while the remaining four sections, which did not contain RAP, were composed of “virgin” material.
The hyperparameter settings employed by LMGOT include a genetic algorithm population size of N = 100, a crossover rate of 0.9, a mutation rate of 0.001, and an evolution generation count of 1000. This study specifically addresses transverse cracking, recognized as the most common form of distress in pavements. The occurrence of transverse cracking is influenced by four key factors: duration of service, method of surface preparation (mill/no mill), type of material utilized (virgin or reclaimed asphalt), and the thickness of the overlay. Designers evaluate the allowable length of transverse cracks as a critical metric for assessing pavement service life.
The proposed model demonstrates that a 2-inch (50.8 mm) milling treatment has a minimal impact on mitigating transverse cracking in pavement. Research indicates that cracking can occur across various layers, including the asphalt layer, base, sub-base, and sub-grade. In the conducted experiment, it was observed that the ML = 1 (mill) surface treatment did not completely eliminate underlying cracks, which allowed for the propagation of transverse cracking into the asphalt concrete (AC) overlay.experiment, transverse cracking to propagate
Furthermore, the study examined the effects of using reclaimed asphalt concrete (RAP) in comparison to virgin asphalt concrete (AC) within a 2-inch thick overlay. After an 8-year service life, the crack length
in the overlay containing virgin AC was recorded at 44.4 meters, while the overlay with RAP exhibited a significantly longer crack length of 82.7 meters—approximately twice as extensive. This discrepancy is attributed to the higher viscosity of the aged asphalt binder found in RAP, which contributes to its reduced resistance to cracking when compared to virgin AC.
It is important to note that the length of cracking in pavement exhibits an inverse relationship with overlay thickness. To mitigate the occurrence of significant cracks after 8 years, the required overlay thickness would need to be 2.9 inches for virgin asphalt concrete (AC) and 7.8 inches for reclaimed AC. The findings illustrate the relationships between crack length, time, thickness, and reclaimed asphalt pavement (RAP), indicating that virgin AC (RAP = 0) demonstrates superior resistance to cracking compared to reclaimed AC (RAP = 1).
The LMGOT can generate a self-organized formula structure and optimize coefficients without relying on preconceived formula frameworks. This characteristic is particularly advantageous when the formula structure of certain material behaviors is unknown. Given that many material behaviors lack established formula structures, the LMGOT may present a more effective alternative to traditional non-linear regression methods in modeling these behaviors. Consequently, the LMGOT can assist researchers in developing a concise and physically meaningful formula.
Reference
Hsie, M., Ho, Y. F., Lin, C. T., & Yeh, I. C. (2012). Modeling asphalt pavement overlay transverse cracks using the genetic operation tree and Levenberg–Marquardt Method. Expert Systems with Applications, 39(5), 4874-4881.