Reliability-aware physics-guided enhancement for sparse-path Lamb-wave defect imaging

Sparse-path Lamb-wave imaging remains challenging because the available pitch–catch paths contribute unequally to defect localisation, and simple equal-weight fusion often produces diffuse and unstable hotspots. This study proposes a reliability-aware physics-guided framework for sparse-path Lamb-wave defect imaging. The method combines unified preprocessing and scattering-envelope extraction, delay-constrained single-path elliptical imaging, two-stage path-reliability weighting, and lightweight image refinement under a soft physics prior. Experimental validation is performed using paired intact–damaged measurements from an aluminium plate with a controlled through-hole defect, which serves as a representative compact scattering source for evaluating the sparse-path imaging chain. Physically aligned scattering-envelope model (SEM)-assisted auxiliary datasets are used for refinement learning and statistical assessment under the same geometry and delay-mapping convention. The results show that single-path imaging is strongly underdetermined, whereas multi-path fusion reduces localisation error from over 100 mm to the order of tens of millimetres. On the SEM-100 benchmark, the trained refinement stage further reduces the mean localisation error from 19.0 to 4.8 mm while substantially improving image quality. The proposed framework therefore provides a practical balance between physical interpretability, sparse sensing, and data-assisted enhancement for guided-wave inspection with limited reliable paths and scarce labelled data.