Abstract
The reasons for the disparity between predicted and measured ground movements are discussed. The factors which affect the ratio of in situ effective stress, K[o], are reviewed together with the methods for evaluating K[o]. Similarly, the factors which affect the stress-strain properties of a soil and the methods of obtaining the stress-strain relationship of a soil are reviewed. On the basis of evidence presented a case is made for achieving similarity between the laboratory test model and the field prototype. The extents to which this similarity is both possible and necessary is explored in a series of tests on normally consolidated coarse sand. The effects of undrained total stress relief sampling disturbance is examined together with the effects of isotropic and anisotropic reconsolidation, and the subsequent application of the conventional triaxial test stress path or of an idealised construction stress path measured at the site of an embankment dam. The development and operation of an automatic programmable triaxial test system which was used to perform the tests is described. In the system a desk-top computer is used to acquire and process the test data and to control the test to allow conventional triaxial tests,and the simultaneous application of varying axial and radial stresses to simulate field conditions. The in situ soil is modelled by carrying out K[o] consolidation in the conventional triaxial cell. By comparison with the modelled in situ soil, it is confirmed that the undrained stress-strain relationships are made significantly more realistic by anisotropic reconsolidation to the in situ stresses, and it is further shown that the drained stress-strain relationships are made significantly more realistic by applying the field construction stress path. In addition it is shown that stress relief sampling disturbance does affect the deformation moduli determined from both undrained and drained triaxial test data. A speculative method is proposed whereby the in situ state of effective stress may be estimated from undisturbed triaxial test specimens.