Integrate the FEM and ANN into the Dynamic Model of Rigid Pavement Backcalculation
Date Issued
2007
Date
2007
Author(s)
Gao, Cyuan-hong
DOI
zh-TW
Abstract
The use of falling weight deflectometer (FWD) in pavement strength inspection is getting matured and popular. Also, different kinds of software have been developed in order to backcalculate more precisely the pavement stiffness. However, the setting of parameters, such as initial stiffness, layer thicknesses, and depth of stiff layer, would influence the accuracy of backcaluation results. Thus, the finite element software, ABAQUS, was used in this research for simulating pavement responses to dynamic loadings and for evaluating the relationships between different parameters under various types of pavement compositions.
Deflection bowls and rebounding time were analyzed to observe the relations with the depth and stiffness of stiff and surface layers as well as the stiffness of subgrade. It was found that rebounding time had close relationships with stiff layer depth and subgrade stiffness, and they were used to develop the equation for the calculation of stiff layer depth. In addition, subgrade stiffness could be obtained by the equation composed of BDI and F2. By comparing the backcalculated subgrade stiffness obtained with three different methods, i.e. by the iteration of two equations mentioned above, by setting different stiff layer depths in static backcalculation software, MODCOMP, and by using different layer stiffness in ABAQUS, it was found that the results of the first and the last methods were closest. In those of the second method, the result without considering stiff layer thickness was quite different. Therefore, it was shown that the depth of stiff layer could not be neglected while using softwares to backcalculate. Besides, the simulation results of the finite element software were also used as the training data of artificial neural network (ANN) to build dynamic backcalculation models. The backcalculation results of the dynamic ANN backcalculation models were compared with those of other common backcalculation softwares, and it was found that the former ones were more stable and closer to the laboratory values, especially for those of subgrade stiffness. Therefore, the dynamic ANN backcalculation model built in this research can backcalculate pavement stiffness more precisely and faster.
Deflection bowls and rebounding time were analyzed to observe the relations with the depth and stiffness of stiff and surface layers as well as the stiffness of subgrade. It was found that rebounding time had close relationships with stiff layer depth and subgrade stiffness, and they were used to develop the equation for the calculation of stiff layer depth. In addition, subgrade stiffness could be obtained by the equation composed of BDI and F2. By comparing the backcalculated subgrade stiffness obtained with three different methods, i.e. by the iteration of two equations mentioned above, by setting different stiff layer depths in static backcalculation software, MODCOMP, and by using different layer stiffness in ABAQUS, it was found that the results of the first and the last methods were closest. In those of the second method, the result without considering stiff layer thickness was quite different. Therefore, it was shown that the depth of stiff layer could not be neglected while using softwares to backcalculate. Besides, the simulation results of the finite element software were also used as the training data of artificial neural network (ANN) to build dynamic backcalculation models. The backcalculation results of the dynamic ANN backcalculation models were compared with those of other common backcalculation softwares, and it was found that the former ones were more stable and closer to the laboratory values, especially for those of subgrade stiffness. Therefore, the dynamic ANN backcalculation model built in this research can backcalculate pavement stiffness more precisely and faster.
Subjects
落重撓度儀(FWD)
鋪面反算
堅硬層
有限元素法
類神經網路
falling weight deflectometer (FWD)
artificial neural network (ANN)
stiff layer
finite element method
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-96-R94521513-1.pdf
Size
23.31 KB
Format
Adobe PDF
Checksum
(MD5):bda0731914a8c223b7307b99afc05aea