In order to decrease risk, downtime, and costs; modern structures often employ strategies to monitor their state and determine the existence, and characteristics, of any damage present. Ultrasonic guided waves may offer a convenient and practical approach to this problem. UGWs, offer a number of distinct advantages, such as; long range, accurate sizing potential, greater sensitivity and cost effectiveness. However, the current state-of-the-art for applications of UGWs in fibre composites is juvenile in comparison to their application in isotropic and homogeneous materials. The aim of this work was to develop advanced, novel tools for deeper understanding of ultrasonic guided waves, so that they can help and enhance NDE/SHM strategies for fibre-composite materials.
In particular, three tools have been developed, the details and results of which are presented in this thesis. The first of these tools is a physics-informed approach to machine learning of guided-wave feature spaces. In order to assess damage in a structure, and implement any NDE or SHM strategy, knowledge of the behaviour of a guided wave throughout the material/structure is important. Determining this behaviour is extremely difficult in fibre-composites, where unique phenomena such as continuous mode conversion takes place. This thesis introduces a novel method for modelling the feature-space of guided waves in a fibre-composite material. This technique is based on a data-driven model, where prior physical knowledge can be used to create structured machine-learning tools; where constraints are applied to provide said structure. The method makes use of a Gaussian process, a full Bayesian analysis tool. Experimental data of an energy-based Lamb wave feature over a fibre-composite plate were collected. This data was then fed through multiple learning algorithms, each with increasing levels of prior knowledge embedded. The work has shown how physical knowledge of the guided waves can be utilised in modelling using an ML tool, and that by careful consideration when applying machine-learning techniques, more robust models can be generated, which offer advantages such as extrapolation, physical interpretation, and increased performance.
The second tool developed was a Bayesian approach to decomposition of single-source, multi-mode signals. This tool was tested on a localisation problem, where decomposition of single-source signals is required to provide information on signals reflected from the damage. A simulation method which can model complex multi-mode wave interaction was used to demonstrate the capability of the decomposition tool. The tool shown here has a distinct advantage in that it produces quantified results for the uncertainty in the decomposed signal, which lends well to any NDE/SHM strategy utilising probabilistic approaches for detection. Furthermore, it was shown that the method inherently produces parametric features which are indicative of the physical behaviour of the wave. The proposed decomposition method was shown to allow localisation of damage accurate to within 1mm in many sensor configurations.
The final tool shown in this thesis, is a Bayesian method for material identification using UGWs. This tool was assessed with the objective of determining accurate group velocity, as is required for localisation of damage. The method uses a Markov chain Monte Carlo procedure to simulate samples of the distribution for each of the parameters. Observations of dispersion-curve data were measured, and used to assess the material properties using a computationally-efficient solution to dispersion curves in orthotropic materials. The results showed the importance and capability of the method having freedom to generate the posterior distribution with respect to both shape and multivariate dependencies. Furthermore, the work presented shows that the method performs well for the objective of determining accurate dispersion-curve information - i.e. group velocity curves.
