In the next generation of power stations, high temperatures (up to 700 °C) will be used to improve efficiency. As creep deformation is of great concern at such high temperatures, materials with large grains are desirable due to their high creep strength. However, this brings a challenge in ultrasonic monitoring of the plants, because when the grain size is comparable with the wavelength, scattering caused by the acoustic impedance contrast on the grain boundaries can strongly attenuate the signals from defects and significantly increase the noise, which often results in low signal to noise ratios (SNRs) and makes the defect detection difficult.

In this thesis, the potential of two techniques to overcome the low SNR problem, baseline subtraction with repeat scanning and optimising the transducer and placement for monitoring with permanently installed sensors, is investigated. In order to find the optimal permanently installed monitoring system, it is desirable to understand the noise behaviour in highly scattering regimes, which cannot be fully described with existing theoretical models. Therefore, 3D FE modelling, which can achieve it, is proposed as an alternative approach for noise prediction.

In Chapter 2, the feasibility and reliability of comparing repeat tests using baseline subtraction to reduce grain noise is studied. The feasibility of setting the transducer with the same angle and stand-off relative to the testpiece, registering the C-scans to locate the same position and using compensation methods to compensate for the temperature and transducer frequency response changes in repeat scans is investigated experimentally. Results show that modest improvement in SNR can be obtained when the temperature and transducer frequency response changes are relatively small, and the dominant factors restricting the improvement are the temperature and transducer frequency response changes.

In Chapter 3, 3D FE modelling is proposed to study the backscattering noise behaviour. Firstly, the 3D FE approach is validated by comparing the FE predicted longitudinal-to-longitudinal backscattering amplitude from a single grain with the theoretical solution given by the established theories. Then noise levels are predicted by the FE approach and compared with the theoretical predictions, which are obtained following a similar way to that used to develop the Independent Scatterer Model. From the comparison, it can be seen that the theoretical model can give a rough estimation of the grain noise level very efficiently. Therefore, considering the high computational costs of FE simulations, the theoretical model should be used to give a first estimate of grain noise level, and the FE approach can be used to provide a more accurate prediction where required.

In Chapter 4, a procedure of how to find the optimal transducer parameters and placement in a permanently installed monitoring system is presented. A combined analytical-and-numerical method is adopted to generate a SNR map and the influence of various parameters of the transducer, excitation and material on SNR is discussed. Then representative examples are presented to show how to solve the optimisation problem under different possible criteria.