The field of structural health monitoring and digital forensics is moving towards leveraging generative models to address data scarcity and enhance the reliability of automated systems. Researchers are exploring the use of customized Generative Adversarial Network models to generate high-quality synthetic data, such as spectrograms and fingerprint images, that mimic real-world events and enhance dataset diversity and robustness. These models have shown great potential in producing high-resolution, temporally consistent data that align closely with real-world data, indicating a scalable and cost-effective solution for monitoring scenarios where rare events suffer from data scarcity. Additionally, the combination of signal processing techniques and deep learning is being used to detect subtle periodic artifacts in GAN-generated images, enhancing digital forensics and strengthening the trustworthiness of industrial AI systems. Noteworthy papers include:

• A study introducing a customized Generative Adversarial Network model for generating short-time Fourier transform spectrograms, which outperforms baseline models in producing high-resolution, temporally consistent spectrograms.
• A paper presenting a novel approach for generating synthetic fingerprint images using conditional StyleGAN2-ADA and StyleGAN3 architectures, achieving robust performance and strong privacy preservation.