Reliable detection and classification of defects is a major concern for pipeline operators; with pipeline failures being both ecological and financial disasters and because pipeline repairs are so expensive, it is crucial to accurately detect and classify defects in order to take the correct course of action, repair for critical defects and monitoring for non-critical defects. Remote location and difficulty of access make excavation of the pipeline for defect verification impractical, so little ground truth is known with regard to the specific geometries of defects growing in pipelines operating under specific conditions such as buried in an alkaline soil or in salty water. A recent study on Australian pipelines showed that more than 80% of the cracks were tilted, which is an otherwise uncommon occurrence.
This thesis specifically investigates, among other parameters, the effect of tilt on crack reflection and transmission coefficients using a shear horizontal guided wave inspection below the first order cut-off frequency. A quasi-2D model is first used to simulate the interaction between SH0 and tilted surface-breaking cracks, varying both the tilt and depth of the defect. The incident wave interacts with the tilted crack to generate a transmitted wave, a reflected wave and a wave trapped below the crack. It is shown that the direction of the tilt of the crack relative to the incident wave direction does not affect the reflection and transmission behaviour. Additionally, the axial extent of the crack plays a major role in the reflectivity of the crack, leading to transmission nulls in some configurations. These transmission nulls appear for all crack depths, the frequency range over which the transmission is significantly reduced increasing with crack depth. This behaviour is shown to be analogous to the acoustic energy flow in a duct when a Helmoltz resonator is introduced. The null is not seen when the frequency is above the SH1 cut-off as the propagating signals are no longer mono-modal. The existence of a transmission null and corresponding reflection maximum is promising for the detection of small defects and measurement of the frequency at which the null occurs will assist with defect characterisation. Experimental validations of the key results are also presented. These results are then extended to a full 3D model, investigating the effect of both sensor and defect size on the transmission null. It is found the locations of the frequencies being removed is also a weak function of size. As the defect gets smaller than the inspection sensors, the ratio of diffracted waves from the edges to trapped waves below the crack increases, thus size also affects the existence of the null. A similar model is then used to observe the influence of skew on wave propagation and mode conversion. It is shown that skew has little influence over the location of the transmission null or the wave propagation in general for small values of skew (below 10°), which correspond to typical axially-aligned defects of interest. For larger values of skew, the direction of propagation of the reflected signal is so different from its original path that it is not captured by the pulse-echo sensor, failing to detect the defect.
Next, performance analysis of a detection and classification algorithm based on a guided wave-based Pipeline Inspection Gauge (PIG) by Baker Hughes using a simple inspection setup with two sensor types, detection nominally using SH0 and discrimination nominally using S0, is performed over a range of wall thicknesses ranging from 5 mm to 15 mm. Results show that keeping the same sensor design across a large wall thickness yields little control over the mode mixture excited by the sensor. Although all critical defects are detected at all wall thicknesses, some non-critical defects are either not detected or detected but classified as critical. In conclusion, the range of operating points makes efficient inspection very challenging as wavelength greatly varies in that region. The discussion concludes by highlighting the potential advantages and disadvantages of adapting sensor design to wall thickness to improve both sensor and algorithm performances.