Parallel distributed processing in rock engineering systems
File(s)
Author(s)
Millar, Dean Lee
Type
Thesis
Abstract
Rock Engineering Systems are a collection of ideas, mathematical tools and computer
technology all of which are designed to solve problems in rock engineering with
interacting components. The interactions between components can be complex and the
rock engineering problems themselves contain a high degree of uncertainty. The
research described in this thesis investigates the incorporation of computational
techniques known as parallel distributed processing methods into the disciplines of
rock mechanics and rock engineering. Two main applications of parallel distributed
processing methods in rock engineering are investigated in this thesis.
1) Multilayered perceptron artificial neural networks are used successfully to
encapsulate the laboratory behaviour of rocks under triaxial compression. Trained
artificial neural networks are then used to replace conventional constitutive models
within finite difference geomechanical numerical modelling codes.
2) Two multilayered perceptron artificial neural networks are developed to assist
in the task of discrimination of rock fracture presence within digital imagery of rock
exposures. The first is trained using samples of the image that contain fracture image
content and samples that do not, and provides a probability-like measure of fracture
presence. It was sufficiently successful to permit estimation of fracture intensity
parameter , . The second was developed specifically to identify fracture termination
condition by matching samples to a set of fracture termination condition templates.
Seven original contributions to the rock mechanics and rock engineering
disciplines have resulted across the three application areas. These contributions are
itemised, with details, at the beginning of the final Chapter of the thesis.
technology all of which are designed to solve problems in rock engineering with
interacting components. The interactions between components can be complex and the
rock engineering problems themselves contain a high degree of uncertainty. The
research described in this thesis investigates the incorporation of computational
techniques known as parallel distributed processing methods into the disciplines of
rock mechanics and rock engineering. Two main applications of parallel distributed
processing methods in rock engineering are investigated in this thesis.
1) Multilayered perceptron artificial neural networks are used successfully to
encapsulate the laboratory behaviour of rocks under triaxial compression. Trained
artificial neural networks are then used to replace conventional constitutive models
within finite difference geomechanical numerical modelling codes.
2) Two multilayered perceptron artificial neural networks are developed to assist
in the task of discrimination of rock fracture presence within digital imagery of rock
exposures. The first is trained using samples of the image that contain fracture image
content and samples that do not, and provides a probability-like measure of fracture
presence. It was sufficiently successful to permit estimation of fracture intensity
parameter , . The second was developed specifically to identify fracture termination
condition by matching samples to a set of fracture termination condition templates.
Seven original contributions to the rock mechanics and rock engineering
disciplines have resulted across the three application areas. These contributions are
itemised, with details, at the beginning of the final Chapter of the thesis.
Version
Open Access
Date Issued
2008-05
Date Awarded
2008
Copyright Statement
Attribution NoDerivatives 4.0 International Licence (CC BY-ND)
Advisor
Hudson, John
Harrison, John
Sponsor
United States. Army. Corps of Engineers
Publisher Department
Earth Science and Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)