Key points are not available for this paper at this time.
Accurate damage localization in plate-like structures is essential for transitioning from scheduled inspections to condition-based structural health monitoring (SHM). Guided wave methods offer whole-area coverage through permanently bonded sensor networks, yet the nonlinear mapping from measured signals to damage coordinates remains challenging due to wave dispersion, boundary reflections, and environmental variability. This study develops a damage localization framework for aluminum plate structures based on guided wave imaging, using systematically optimized Radial Basis Function Networks (RBFN) to predict 2D damage positions. The hybrid approach, termed ShmGwi-RBFN (Structural Health Monitoring Guided Wave Imaging with RBFN), integrates Delay-and-Sum (DAS) beamforming with RBFN regression optimized through a multistage hyperparameter search. A 500 × 500 mm aluminum plate instrumented with 12 piezoelectric transducers in a pitch-catch arrangement generates 66 unique signal paths. On an independent augmented test dataset, the Euclidean RMSE is 13.07 mm with 93.1% of estimates within 20 mm of the true damage location. On the original experimental dataset, the mean RMSE is 5.40 mm with 100% of estimates within 20 mm. Parameter importance analysis shows that kernel type selection has the greatest effect on localization accuracy, with approximately 69% of total variance attributed to kernel type. These results demonstrate effective joint use of physics-based imaging with machine learning for practical structural health monitoring. The proposed framework offers a computationally efficient, interpretable, and easily deployable alternative to deep-learning-based SHM methods, with direct applicability to aerospace panels, civil infrastructure plates, and other thin-walled metallic structures requiring real-time damage assessment.
Drissi et al. (Wed,) studied this question.