STRESS WAVE PROBLEMS IN PILING ENGINEERING

Abstract

This thesis addresses two aspects of stress wave problems in piling engineering, namely (a) to investigate the feasibility of using artificial neural networks for the prediction of the performance of pile foundations from dynamic measurements, and (b) to formulate an improved rational wave equation model for the driving analysis of open-ended pipe piles. Two applications of artificial neural networks for the prediction of performance of pile foundations using dynamic methods are presented. The first application uses a neural network approach as an alternative to pile driving formulas which are widely used by geotechnical engineers. The main advantage of the neural network approach over pile driving formulas is that it need not make any a priori assumptions, either in terms of physical laws or regression models. It is found that the overall output results of the neural network approach are better than those obtained by a pile driving formula using the same input parameters. The second application of neural networks is to assess the performance of pile foundations by the stress wave matching technique. The neural network approach can avoid the time-consuming process of iterative adjustment employed by the conventional manual or automated matching approaches, making it feasible to determine the pile capacity in real time in the field. The abilities of the proposed neural network approach for the prediction of the performance of pile foundations are illustrated in analysing seventy dynamically tested concrete bored piles and seventy- two steel H piles. A new one-dimensional wave equation model is proposed for driving analysis of open-ended pipe piles. In this model, the soil plug is modelled based on a theoretical analysis of the dynamic reaction of the soil plug inside the pile. It is shown that the effect of the soil plug during driving should theoretically be represented only by frequency-dependent springs unlike that of the external pile-soil reaction which can be closely approximated as a series of frequency-independent springs and dashpots. As shown in the case studies, an average frequency-independent stiffness determined based on the spectral distribution of the driving energy can be used to replace the frequency-dependent spring stiffness representing the soil plug effect for pile driving analysis in the time domain. The shortcoming of existing soil plug models which fail to represent the physical behaviour of the soil plug column inside the pile during driving rationally has been surmounted in the proposed wave equation model. The capability of the new model in the analysis of open-ended pipe piles subjected to impact driving is illustrated by two case studies.