Design Spectrum-Compatible Synthesis of ArtificialAccelerograms Utilizing Generative Adversarial Networks

Recent advancements in Deep Learning (DL) have seen its application extended to addressnumerous challenges within the domains of civil and earthquake engineering. However, the dearth ofdependable data pertinent to earthquake engineering presents an obstacle that may undermine theprecision of DL-derived outcomes. In response to this impediment, Generative Adversarial Networks(GANs) emerge as a promising solution. Originally conceptualized to enhance the training ofgenerative models, GANs have demonstrated their prowess and adaptability, particularly in the realmof image generation, earning significant recognition from the academic community. In the realm ofstructural engineering, the generation of artificial ground accelerograms that adhere to a predefinedtarget response spectrum is a prerequisite for conducting nonlinear dynamic analyses. The paper athand introduces an efficacious algorithm for spectral matching, enabling the synthesis of a multitudeof artificial spectrum-compatible earthquake accelerograms from a limited collection of groundmotion records.

Keywords:

Artificial Accelerograms , GAN , AI , Neural Networks , Deep Learning