Machine learning for transmission spectra prediction on gradient seismic metastructure

Seismic metastructure based on phononic crystal theory provides a possible solution to accurately manipulating surface acoustic waves. However, the prediction of gradient seismic metastructure for transmission spectra in clay remains a significant challenge due to the damping characteristics of actual soils and practical engineering factors, which become a research hotspot in recent years. Based on finite element analyses and machine learning techniques, this work proposed a data-driven method for building a general prediction model of embedded pillar seismic metastructure with different multi-resonator gradients in the clayed soil. We employed a multilayer perceptron (MLP) model, with the multi-resonator gradients of the metastructure as the input, to predict the transmission spectra. To achieve input standardization, we applied an Autoencoder (AE) to construct a unified representation of the inputs. Due to the inherent non-linearity and variability in soil-structure interactions, the attenuation zones prediction results can only offer approximate engineering applications under specific conditions. By utilizing machine learning, our method achieves better generalization and can be adapted to a wider range of metastructure configurations. This research not only advances the gradient seismic metastructure design framework but also opens new avenues for practical applications in surface acoustic wave management.