CycleGAN for undamaged-to-damaged domain translation for structural health monitoring and damage detection

The advances in data science in the last few decades have benefitted many other fields, including Structural Health Monitoring (SHM). Artificial Intelligence (AI), such as Machine Learning (ML) methods for vibration-based damage diagnostics of civil structures, have been utilized extensively due to the observed high performances in learning from complex data structures. AI-based data-driven techniques used for damage diagnostics and prognostics applications are centered on historical data of the structures and require a substantial amount of data for data-driven prediction models. Although some of these methods are generative-based models, they are used to perform typical ML tasks such as classification, regression, or clustering after learning the data domain. In this study, a variant of Generative Adversarial Networks (GAN), a generative model, Cycle-Consistent Wasserstein Deep Convolutional GAN with Gradient Penalty (CycleWDCGAN-GP) is developed to investigate the domain translation between undamaged and damaged acceleration data (1-D) from one element to the same element as well as to other elements. The outcomes of this study demonstrate that the proposed methodology could be used to generate possible responses of a structure for potentially damaged conditions. In other words, with the proposed methodology, it will be possible to understand and generate the damaged condition while the structure is still healthy.