Interpolation techniques for ultrasonic data.

Many applications where ultrasound is used for diagnostics exist where limited data is preventing a particular approach from being fully exploited; for example, sufficient data availability would allow the qualification of non-destructive evaluation (NDE) methods in-silico, and would potentially also enable the training of machine learning algorithms related to ultrasound and its applications. Real, experimental ultrasonic data is often scarce, and while it is already known that finite element (FE) modelling produces data which is sufficiently realistic to augment real data, the computational cost associated with its generation at the scales required for the aforementioned purposes is often prohibitive. In this work, we propose the use of interpolation techniques in combination with results from FE modelling to rapidly generate more data without the need to solve additional FE models. We present the relevant methods to achieve this, and validate them through four exemplary cases of increasing complexity. Validation is achieved through the comparison of interpolation-generated results to those generated by full FE modelling, demonstrating that our method is capable of producing results for different physical setups and signals of various degrees of complexity. The results were typically within less than 1% away from the expected, but generated at a fraction of the typical computational cost, and, while the validation cases examined are of interest to the NDE community, the method extends to other fields where ultrasonic data is of interest.