UGW-3Mat-2SN: Ultrasonic Guided Wave Dataset from Three Composite Plates with Two Sensor Networks

This database contains ultrasonic guided wave (UGW) data acquired from three composite plates made of different materials: K8, G16, and K2G4S. Each plate was instrumented with two piezoelectric (PZT) sensor arrays, arranged in either a circular or rectangular network configuration. The dataset is organised into two main categories: Before MPCA: Raw UGW data. After MPCA: Data processed using Multilinear Principal Component Analysis (MPCA), which was applied as a domain adaptation (DA) technique to support transfer learning (TL) tasks. The After MPCA category is further divided into three subfolders: Circular – Contains data for TL between different materials while keeping the circular sensor network fixed. Rectangular – Contains data for TL between different materials while keeping the rectangular sensor network fixed. Sensor Network – Contains data for TL between different sensor network geometries (circular ↔ rectangular) while keeping the material fixed. Within each subfolder, data are organised into directories named using the format: SourceDomain_TargetDomain, where SourceDomain and TargetDomain refer to either material names (e.g., G16_K8) or sensor network types (e.g., Circular_Rectangular), depending on the case study. Each of these directories contains subfolders corresponding to the percentage of variance retained by MPCA (e.g., 97%, 99%, 99.9%), allowing for the analysis of dimensionality effects on domain adaptation performance. These subfolders contain the following data: sourceDomainProjected.mat (source domain data after MPCA domain adaptation) targetDomainProjected.mat (half of the target domain data after MPCA domain adaptation, randomly selected) targetValDomainProjected.mat (target domain data after MPCA domain adaptation) X_sr.mat (original source domain data) X_tg.mat (half of the original target domain data, randomly selected) X_tg_val.mat (original target domain data) Y_sr.mat (damage positions in the source domain) Y_tg.mat (damage positions in half of the original target domain data, randomly selected) Y_tg_val.mat (damage positions in the target domain)