The prediction and back analysis of excavation behaviour in Oxford Clay.

The work described in this thesis involved the prediction and back-analysis of ground
movements and pore water pressures around a temporary excavation, lOm deep, 105m long and
35m wide (50m long and lOm wide at its base) in the heavily overconsolidated Oxford Clay.
An experimental programme was carried out which was designed to investigate the nonlinearity and the anisotropy in the soil's response. A series of computer controlled triaxial stress
path tests were carried out on specimens of Oxford Clay trimmed from block samples. The
initial anisotropic stress state resulted in a highly stress path dependent stiffness, and the stressstrain behaviour was closely linked to the time at which the specimen had been held at constant
stress prior to a change in loading.
In the triaxial apparatus, shear wave propagation tests using square wave input functions into
bender element apparatus were significantly affected by near-field effects and by additional
wave components. Tests carried out using sine wave inputs provided a far more consistent
output allowing correlation analyses and easier visual identification of the travelling shear
wave. The shear modulus of the soil at very small strains could then be determined.
A stress path dependent, non-linear, cross-anisotropic elastic model was developed and
implemented into the finite element program CRISP. A genuine prediction of the Elstow
excavation was carried out while the instrumentation data from the site investigation were
temporarily withheld. It was found that the horizontal displacements were modelled
satisfactorily but that the vertical displacements were in error by as much as 2 or 3 times. This
discrepancy was attributed to volumetric changes suggested by the instrumentation data. A
parametric analysis was carried out in which the effects of the initial stress state, the degree of
anisotropy, and the degree of non-linearity were investigated. This showed that although it was
possible to improve the accuracy of the prediction locally, it was not possible to improve on the
overall pattern of behaviour predicted by the first non-linear cross-anisotropic analysis.