Guided wave sensors are commonly used in the oil and gas industry to inspect pipework for
damage. Early sensor were detachable and designed for single inspections, but current
best practice is to permanently attach sensors and take repeat readings. Known as a
monitoring configuration, this approach offers potential benefits in terms of cost, defect
detection capability and safety. There are however implementation challenges that are
still to be resolved, particularly the need to compensate collected data for changes in
environmental and operational conditions. The interaction of guided waves with
general corrosion is also poorly understood.
This thesis seeks to address both issues, beginning with a finite element study of
the interaction of guided waves with rough surfaces. These rough surfaces are a basic
model for general corrosion and this study shows that attenuation seen in commercial
inspections can be explained by significant scattering of waves by the rough surface. This
scattering is shown to be a function of the rough surface characteristics, pipe diameter
and inspection wave frequency. This study was computationally expensive and made
possible by recently developed simulation tools, particularly graphics processing unit
based solvers. The development and construction of these simulation tools is described in detail.
The second major theme of this thesis is improved signal processing for monitoring
data based on Independent Component Analysis [ICA]. ICA was applied to numerous
example signals and found to correctly separate structural anomalies from surrounding
noise. This technique was then compared to best practice techniques from industry
(residual processing and singular value decomposition [SVD]) and it was found that SVD
and ICA performed much better than residual processing under a range of conditions. In
general ICA performed better than SVD, but was less robust to outlier behavior. The
framework used for this comparison is itself novel and is described.