Seismic Intensity Feature Construction Based on the Hilbert–Huang Transform

An increasing number of seismic parameters for the representation of the severity of earthquake signals have been proposed in recent years. Considering the complex nature of seismic accelerograms, there is an obvious need for a more effective representation. The destructiveness of a seismic wave cannot be always described using a single feature of the examined ground motion. Moreover, the level of structural damage caused by a severe earthquake depends on the characteristics of the oscillated structure. The aim of this study is to propose new crucial ground-motion parameters, which reflect the damage potential of the seismic excitations. The Hilbert-Huang transform (HHT) is applied to a set of 70 natural accelerograms. Through the HHT, the complex earthquake signals are decomposed into several simple components called intrinsic mode functions. After a thorough analysis, eight new seismic parameters are extracted. The key novelty of this paper is that the frequency-based new seismic parameters are associated with the eigenfrequency of the oscillated structures. By studying the correlation between the new seismic parameters and well-known damage indices (DIs), their interdependence is confirmed. The achieved high levels of correlation indicate that the proposed parameters are directly related to structural damage. Furthermore, the application of the HHT improves the correlation coefficients between the widely used classical earthquake parameters and DIs. The quality of the proposed HHT-based parameters is consolidated by the use of minimal-redundancy-maximal-relevance feature selection. A subset of six seismic parameters is tested by the use of a support vector machine classifier, providing a 91.4% classification performance. Results indicate that the new seismic features along with the proposed optimized classical parameters can be considered highly competitive descriptors of the seismic damage potential.