Abstract
Scanning laser Doppler vibrometer (SLDV) is widely used to obtain the full wavefields of propagating guided waves, particularly Lamb waves. Signal processing is applied on such full wavefields in order to reveal the defects in the inspected structure. However, obtaining the full wavefields of guided Lamb waves is very time-consuming.
To tackle this problem, one possible solution is to acquire the guided Lamb waves in a low-resolution form and then apply a compressive sensing (CS) or a deep learning-based super-resolution approach to that low-resolution form of full wavefields data.
In this research, we applied a deep learning-based super-resolution technique on a large synthetic dataset of full wavefields of propagating Lamb waves in low-resolution format on a carbon fibre-reinforced polymer (CFRP) plate. The developed deep learning approach is elaborated, and the results acquired from the deep learning-based process are compared with the conventional CS approach. The developed deep learning-based super-resolution approach is evaluated with two evaluation metrics namely peak signal-to-noise ratio (PSNR) and Pearson correlation coefficient (Pearson CC) and has achieved promising results. It is concluded that the developed deep learning-based method outperforms the traditional CS approach, and the deep learning-based approach for super-resolution can improve the speed of data acquisition by SLDV.
