A comparative study of singular spectrum analysis and empirical mode decomposition for ultrasonic NDE

In this investigation, a comparative study of two time-series analysis techniques, singular spectrum analysis (SSA) and empirical mode decomposition (EMD), is carried out for ultrasonic NDE applications. Unlike transform-based approaches, the SSA and EMD are fully driven in the time domain by the data itself. The SSA method captures the trend of signals via eigenvalues from a trajectory matrix. Whereas EMD searches intrinsic mode functions (IMF) via local maxima and minima, then the instantaneous frequency of IMFs is used to disclose the signal trends. Simulation shows that the trend tracking becomes problematic for defect assessment when signal trends become too complicated. As a remedy, a technique combing these two methods is proposed. Experimental and simulation results reveal the effectiveness of the collective effort from EMD and SSA. The study demonstrates that the data-driven nature is unique and attractive to characterize non-stationary ultrasonic signals where scattering, absorption and dispersion effects become dominant.