| dc.description.abstract | The geologic structures such as faults, joints, bedding planes, inter-bedding, and foliation, are very common in Taiwan. The rock formation associated with these features, of course, yields a mechanical behavior totally different from that of a homogeneous, intact mass. The main objective of this thesis is to numerically study the bearing mechanism of a rigid strip foundation of width (B) sitting on a mass of two layered formations, and to establish a predictive model of bearing capacity for such a case, which has never arrived at a unified form in the literature.
The bearing behavior of such a mass was simulated by the FLAC code, in which finite-difference scheme is employed for both the spatial and time domains. The foundation loading mode used is strain-controlled, and the ultimate bearing capacity (qu) is determined from the loading versus settlement curve and the disturbed zone beneath the foundation is localized according to the displacement vector plots. For understanding the distinct bearing behavior of a two-layered formation system, two cases were selected: Case A with a hard sandstone layer (HS) of thickness (H1) overlying a soft shale (SS), and Case B with SS overlying HS, each material assigned with typical mechanical properties. Besides, the neural networks analysis with backward propagation algorithm (NNAB) was adopted to develop a general bearing capacity formulas for a two-layered formation system, and about nine hundreds of cases were run by FLAC with formation properties (for common soils): the friction angle (f) varying from 0 to 30° and cohesion (c) from 0 to 1MPa. Three situations were considered: both layers cohesionless, one cohesionless and another cohesive soils, and both cohesive.
The simulation results of HS-SS system show that qu increases with H until H approaches 4B for Case A, qu decreases with H until H approaches 2.5B for Case B, and both (f,c) values of two formations affect qu to some certain extent (but not merely qu of each formation). The NNAB established a fair predictive model of qu for a two-layered soil system, with a prediction error less than 10%. In its linear model analysis, the most influential factor for each situation was also identified, and for instance, such a factor is the friction angle of the top layer (f1) for a two-layered cohesionless system with a weaker top layer. | en_US |