Wavelet scattering transformation for sub-surface damage detection and localization in PZT sensor

Structural Health Monitoring (SHM) systems typically employ piezoelectric sensors for the monitoring of variations in structural members. However, subsurface deterioration or damage within the sensors can impair the system's overall reliability and precision. This drawback can be addressed using a suggested reference-free detection method and subsurface defect localization for Lead zirconate titanate (PZT) sensor defects, combining signal processing and machine-learning-based feature extraction techniques. Ultrasonic wave interactions with the PZT material are measured experimentally by a Coulomb coupling-based excitation scheme, resulting in high-resolution but noisy wavefield images. An advanced Block-Matching 3D (BM3D) filtering algorithm is employed to enhance data quality, which can filter out noise while preserving essential structural details. These denoised images are then processed using the Wavelet Scattering Transform (WST), resulting in stable, hierarchical, and translation-invariant representations of the acoustic patterns embedded within. Critically, the technique enables unsupervised detection of defects - no reference signal from an undamaged sensor is needed - by identifying spatial anomalies that break the expected symmetry of the material. This approach not only correctly locates subsurface anomalies but also assists with sensor self-diagnostics, delivering a strong, interpretable, and non-invasive solution for increasing SHM reliability.