Super-resolution damage visualisation in laser ultrasonic guided waves array detection based on combined deep learning networks

Artefacts in compact array imaging hinder accurate damage detection. This study presents an integrated deep learning framework combining a global transfer learning network and a local super-resolution network. Scattered waves from simulation models are acquired via the delay-and-sum algorithm and input into the global deep subdomain network for initial damage imaging. A local super-resolution network then refines crack regions, incorporating damage direction information into multidimensional feature maps to enhance resolution. High-resolution real images serve as network outputs, establishing mapping relationships for precise imaging. Tests on aluminum plates with crack damages confirm the method’s effectiveness in accurate damage localization and detailed visualization, improving size and shape estimation. The approach mitigates artefact limitations, offering superior imaging accuracy.