A regional damage localization method for composite plate based on guided waves and semi-supervised generative adversarial networks

Abstract

Data-driven structural health monitoring is promising for complex structures but usually requires large-scale and high-quality labeled data. In practice, damaged samples are costly to obtain and difficult to annotate accurately. To overcome these limitations, this work introduces a semi-supervised generative adversarial network (SGAN) tailored for identifying damage-prone regions in carbon-fiber-reinforced polymer (CFRP) plates using guided-wave signals. The monitoring area is divided into K subregions, and the localization task is reformulated as a multi-class classification problem. A one-dimensional SGAN is developed to jointly exploit a small labeled subset and a larger unlabeled dataset. During training, the generator enhances distribution learning, while the discriminator is retained as the final regional classifier. Experimental evaluations demonstrate that the proposed approach delivers reliable regional damage identification under limited-label conditions without relying on complex physics-based modeling. The proposed SGAN outperformed the baseline models even when the annotation ratio was only 9.41%.