NEW VARIATIONAL MODE DESCOMPOSITION ALGORITHM APPLIED TO IMPROVED EARTHQUAKE ACCELEROGRAMS ANALYSIS

Strong motion earthquake records are a great source of information regarding the local soil dynamic response and the seismic input acting to the structures located around the device location. In addition, a set of those signals recorded in several locations around the epicenter of the same earthquake event could give important information regarding seismic wave propagation and attenuation in the region, useful for engineering design purposes. Many time-frequency domain algorithms have been developed in order to perform signal processing and wave characterization analyses. Among them, Fourier transform has been one of the most used algorithms for decomposing signals into its constituent frequency components. Decomposition methods have also appeared later as improved techniques that decompose signals in intrinsic mode functions (IMF), which are amplitude-modulated-frequency-modulated signals easier to analyze, demonstrating some advantages regarding resolution and adaptability to different type of signals, over traditional approaches. In this work, a novelty technique, the variational-mode decomposition (VMD) has been applied to the analysis of the 2016 Mw7.8 Pedernales earthquake records (the most powerful earthquake ever recorded in Ecuador). For the first time in the country, multi-directional earthquake accelerograms from 20 stations located between 30 to 500 km far from the epicenter of a great earthquake are available for the analysis. The VMD method decomposes a signal into an ensemble of band-limited modes. The band-limited intrinsic modes about a center frequency are obtained by using the Hilbert Transform, shifting and normalizing the signal data to create a frequency translated analytic signal. This process permits to have an accurate analysis of the signal peaks in different frequency windows and its distribution in the space covered by the device array. Spatial analysis permitted to create peak heat maps of the region where records were obtained in order to show those results graphically. Analysis of the results after applying this technique to the mentioned earthquake records were compared with traditional Fourier spectra results in terms of amplitude-frequency peaks, while amplification and attenuation of seismic waves in distance were also mapped, depending on the propagation direction. Results showed different patterns of wave attenuation in the region and corroborated soil dynamic amplification of some of the recording locations, demonstrated that the VMD signal analysis algorithm is more robust than other approaches for earthquake records, especially for interpretation of multi-component signals. This algorithm also improved signal noise reduction and gave amplitude peaks in the frequency and time domain with higher resolution.