Image-Based Geotechnical Constitutive Model Parameter Calibration

Abstract

In geotechnics, physical tests are often used to calibrate material parameters that are required to simulate soil behaviour through appropriately simplified mathematical models, commonly known as constitutive models. Conventional "element" tests (e.g. triaxial, simple shear) typically assume a homogeneous deformation response in a sample subject to externally applied stresses. While necessary due to current measurement technique limitations, there is evidence that this assumption is somewhat at odds with reality beyond small strains. Often, when conventional tests are ill-suited for calibrating a particular material parameter, multiple model-specific tests are proposed which increase the "cost" of the implementation of such constitutive models. However, advances in imaging techniques could supplement the calibration process in a quantitative manner by providing full-field measurement data, possibly allowing for a single test to calibrate and compare multiple complex constitutive models. One approach is to use a work-based objective function in which the difference between the external work (work transferred to the sample by boundary tractions and stresses) and internal work (work dissipated by the sample in strain) is minimised through the optimisation of the targeted parameters for calibration. Thus far, this technique has only been trialled numerically or experimentally with simplistic constitutive models. In this thesis, image-based geotechnical constitutive model parameter calibration has been experimentally trialled for realistic soil behaviour. To target strain-softening behaviour, episodic cyclic T-bar penetration testing was selected as a suitable boundary value problem as it provided large cyclic deformations in geometrically imposed plane-strain conditions. To compute the internal work from an experiment image series, a PIV/DIC analysis package was developed which attempted to maximise measurement accuracy and provided the necessary infrastructure following material point method (MPM) and strain path method (SPM) techniques. Once validated and evaluated for efficacy using artificially generated images, the PIV/DIC analysis package was trialled for calibration using the real-world test images. It was found that PIV/DIC measurement errors were prohibitive to successful calibration without some intervention. Heterogeneous deformation features, such as strain localisations typical of soil failure, were under-fitted by local PIV/DIC measurements at realistic experimental resolutions. Similarly, element-based strain interpolation was ill-suited for the deformations present. Significant erroneous volumetric strains were generated which required removal for the evaluation of internal work. On this basis, it was possible to calibrate the targeted strain-softening parameters. This suggested that the calibration approach could be implemented if the limitations of current PIV/DIC analysis methodologies can be addressed, whilst in the meantime the geotechnical experimentalist should be wary of expecting more than qualitative results from image-based measurements.