Please use this identifier to cite or link to this item:
Type: Thesis
Title: Prediction of the Effectiveness of Rolling Dynamic Compaction Using Artificial Intelligence Techniques and In Situ Soil Test Data
Author: Ranasinghe, R. A. Tharanga Madhushani
Issue Date: 2017
School/Discipline: School of Civil, Environmental and Mining Engineering
Abstract: The research presented in this thesis focuses on developing predictive tools to forecast the effectiveness of rolling dynamic compaction (RDC) in different ground conditions. Among many other soil compaction methods, RDC is a widespread technique, which involves impacting the ground with a heavy (6–12 tonnes) non-circular (3-, 4- and 5- sided) module. It provides the construction industry with an improved ground compaction capability, especially with respect to a greater influence depth and a higher speed of compaction, resulting in increased productivity when compared with traditional compaction equipment. However, to date, no rational means are available for obtaining a priori estimation of the degree of densification or the extent of the influence depth by RDC in different ground conditions. In addressing this knowledge gap, the research presented in this thesis develops robust predictive models to forecast the performance of RDC by means of the artificial intelligence (AI) techniques in the form of artificial neural networks (ANNs) and linear genetic programming (LGP), which have already been proven to be successful in a wide variety of forecasting applications in geotechnical engineering aspects. This study is focussed solely on the 4-sided, 8 tonne impact roller (BH-1300) and the AI-based models incorporate comprehensive databases consisting of in situ soil test data; specifically cone penetration test (CPT) and dynamic cone penetration (DCP) test data obtained from many ground improvement projects involving RDC. Thus, altogether, two distinct sets of optimal models: two involving ANNs – one for the CPT and the other for the DCP; and two LGP models – again, one for the CPT and the other for the DCP – are presented. The accuracy and the reliability of the optimal model predictions are assessed by subjecting them to various performance measures. Furthermore, each of the selected optimal models are examined in a parametric study, by which the generalisation ability and the robustness of the models are confirmed. In addition, the performance of the optimal ANN and LGP-based models, as well as other aspects, are compared with each other in order to assess the suitability and shortcomings of each. Consequently, a recommendation has been made of the most feasible approach for predicting the effectiveness of RDC in different ground conditions with respect to CPT and DCP test data. The models have also been disseminated via a series of mathematical formulae and/or programming code to facilitate their application in practice. It is demonstrated that the developed optimal models are accurate and reliable over a range of soil types, and thus, have been recommended with confidence. As such, the developed models provide preliminary estimates of the density improvement in the ground based on the subsurface conditions and the number of roller passes. Therefore, it is considered that the models are beneficial during the pre-planning stages, and may replace, or at the very least augment, the necessity for RDC field trials prior to fullscale construction. In addition, the analyses demonstrate that the AI techniques provide a feasible approach for non-linear modelling involving many parameters, which in turn, further encourages future applications in the broader geotechnical engineering context. Finally, a comprehensive set of guidelines for each of the AI techniques employed in this research, i.e. ANN and LGP, is provided, with the intention of assisting potential and current users of these techniques.
Advisor: Jaksa, Mark
Dissertation Note: Thesis (Ph.D.) -- University of Adelaide, School of Civil, Environmental and Mining Engineering, 2017
Keywords: Rolling dynamic compaction
artificial neural networks
linear genetic programming
Provenance: This electronic version is made publicly available by the University of Adelaide in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. This thesis may incorporate third party material which has been used by the author pursuant to Fair Dealing exceptions. If you are the owner of any included third party copyright material you wish to be removed from this electronic version, please complete the take down form located at:
Appears in Collections:Research Theses

Files in This Item:
File Description SizeFormat 
Ranasinghe2017_PhD.pdf5.78 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.