Modelling of guided waves in a composite plate through a combination of physical knowledge and regression analysis

The use of high-frequency guided waves, such as Rayleigh and Lamb waves (actively) or acoustic emissions (passively), has become increasingly prominent in engineering applications, particularly for structural health monitoring (SHM) and, more traditionally, non-destructive evaluation (NDE). In comparison to low-frequency analysis, guided waves have the additional benefit of being able to locate damage with finer spatial resolution (controlled by the diffraction limit). This paper looks into developing a health-monitoring strategy for fibre-reinforced polymer structures using ultrasonic guided waves (UGWs); part of the remit is to determine a methodology for modelling of UGWs propagation. As fibres within such a material act as a secondary guide for these waves, time-space modelling of the waves is difficult. Presented here is a novel methodology utilising a physics-incorporated, data-driven model to determine the feature-space of UGW propagation. The method uses Gaussian processes and in this paper is made a comparison between different kernel-based methods. By careful consideration of these machine learning techniques, more robust and generalised models can be generated.